Datasets:

Hendrik-a

/

arxiv

like 0

these lectures concern two topics that are becoming increasingly important in the analysis of high energy physics ( hep ) data : bayesian statistics and multivariate methods . in the bayesian approach we extend the interpretation of probability to cover not only the frequency of repeatable outcomes but also to include a degree of belief . in this way we are able to associate probability with a hypothesis and thus to answer directly questions that can not be addressed easily with traditional frequentist methods . in multivariate analysis we try to exploit as much information as possible from the characteristics that we measure for each event to distinguish between event types . in particular we will look at a method that has gained popularity in hep in recent years : the boosted decision tree ( bdt ) .

1012.3589

when a high - energy physics experiment enters the phase of data collection and analysis , the daily tasks of its postgraduate students are often centred not around the particle physics theories one is trying to test but rather on statistical methods . these methods are the tools needed to compare data with theory and quantify the extent to which one stands in agreement with the other . of course one must understand the physical basis of the models being tested and so the theoretical emphasis in postgraduate education is no doubt well founded . but with the increasing cost of hep experiments it has become important to exploit as much of the information as possible in the hard - won data , and to quantify as accurately as possible the inferences one draws when confronting the data with model predictions . despite efforts to make the lectures self contained , some familiarity with basic ideas of statistical data analysis is assumed . introductions to the subject can be found , for example , in the reviews of the particle data group @xcite or in the texts @xcite . in these two lectures we will discuss two topics that are becoming increasingly important : bayesian statistics and multivariate methods . in section [ sec : bayes ] we will review briefly the concept of probability and see how this is used differently in the frequentist and bayesian approaches . then in section [ sec : linefit ] we will discuss a simple example , the fitting of a straight line to a set of measurements , in both the frequentist and bayesian approaches and compare different aspects of the two . this will include in section [ sec : mcmc ] a brief description of markov chain monte carlo ( mcmc ) , one of the most important tools in bayesian computation . we generalize the treatment in section [ sec : fitsys ] to include systematic errors . in section [ sec : multivariate ] we take up the general problem of how to distinguish between two classes of events , say , signal and background , on the basis of a set of characteristics measured for each event . we first describe how to quantify the performance of a classification method in the framework of a statistical test . although the neyman pearson lemma indicates that this problem has an optimal solution using the likelihood ratio , this usually can not be used in practice and one is forced to seek other methods . in section [ sec : bdt ] we look at a specific example of such a method , the boosted decision tree . using this example we describe several issues common to many classification methods , such as overtraining . finally , some conclusions are mentioned in section [ sec : summary ] . in this section we look at the basic ideas of bayesian statistics and explore how these can be applied in particle physics . we will contrast these with the corresponding notions in frequentist statistics , and to make the treatment largely self contained , the main ideas of the frequentist approach will be summarized as well . we begin by defining probability with the axioms written down by kolmogorov @xcite using the language of set theory . consider a set @xmath0 containing subsets @xmath1 . we define the probability @xmath2 as a real - valued function with the following properties : 1 . for every subset @xmath3 in @xmath0 , @xmath4 ; 2 . for disjoint subsets ( i.e. , @xmath5 ) , @xmath6 ; 3 . @xmath7 . in addition , we define the conditional probability @xmath8 ( read @xmath2 of @xmath3 given @xmath9 ) as @xmath10 from this definition and using the fact that @xmath11 and @xmath12 are the same , we obtain _ bayes theorem _ , @xmath13 from the three axioms of probability and the definition of conditional probability , we can derive the _ law of total probability _ , @xmath14 for any subset @xmath9 and for disjoint @xmath15 with @xmath16 . this can be combined with bayes theorem ( [ eq : bayesthm ] ) to give @xmath17 where the subset @xmath3 could , for example , be one of the @xmath15 . the most commonly used interpretation of the subsets of the sample space are outcomes of a repeatable experiment . the probability @xmath18 is assigned a value equal to the limiting frequency of occurrence of @xmath3 . this interpretation forms the basis of _ frequentist statistics_. the subsets of the sample space can also be interpreted as _ hypotheses _ , i.e. , statements that are either true or false , such as `` the mass of the @xmath19 boson lies between 80.3 and 80.5 gev . '' in the frequency interpretation , such statements are either always or never true , i.e. , the corresponding probabilities would be 0 or 1 . using _ subjective probability _ , however , @xmath18 is interpreted as the degree of belief that the hypothesis @xmath3 is true . subjective probability is used in _ bayesian _ ( as opposed to frequentist ) statistics . bayes theorem can be written @xmath20 where ` theory ' represents some hypothesis and ` data ' is the outcome of the experiment . here @xmath21 is the _ prior _ probability for the theory , which reflects the experimenter s degree of belief before carrying out the measurement , and @xmath22 is the probability to have gotten the data actually obtained , given the theory , which is also called the _ likelihood_. bayesian statistics provides no fundamental rule for obtaining the prior probability ; this is necessarily subjective and may depend on previous measurements , theoretical prejudices , once this has been specified , however , eq . ( [ eq : bayesthm3 ] ) tells how the probability for the theory must be modified in the light of the new data to give the _ posterior _ probability , @xmath23 . ( [ eq : bayesthm3 ] ) is stated as a proportionality , the probability must be normalized by summing ( or integrating ) over all possible hypotheses . the difficult and subjective nature of encoding personal knowledge into priors has led to what is called _ objective bayesian statistics _ , where prior probabilities are based not on an actual degree of belief but rather derived from formal rules . these give , for example , priors which are invariant under a transformation of parameters or which result in a maximum gain in information for a given set of measurements . for an extensive review see , for example , ref . @xcite . in section [ sec : linefit ] we look at the example of a simple fit in both the frequentist and bayesian frameworks . suppose we have independent data values @xmath24 , @xmath25 , that are each made at a given value @xmath26 of a control variable @xmath27 . suppose we model the @xmath24 as following a gaussian distribution with given standard deviations @xmath28 and mean values @xmath29 given by a function that we evaluate at the corresponding @xmath26 , @xmath30 we would like to determine values of the parameters @xmath31 and @xmath32 such that the model best describes the data . the ingredients of the analysis are illustrated in fig . [ fig : linefit](a ) . ( 10.0,6.5 ) ( 0.5,0 ) ( 8.5,0 ) ( 0.,5.5)(a ) ( 15.5,5.5)(b ) now suppose the real goal of the analysis is only to estimate the parameter @xmath31 . the slope parameter @xmath32 must also be included in the model to obtain a good description of the data , but we are not interested in its value as such . we refer to @xmath31 as the parameter of interest , and @xmath32 as a _ nuisance parameter_. in the following sections we treat this problem using both the frequentist and bayesian approaches . our model states that the measurements are gaussian distributed , i.e. , the probability density function ( pdf ) for the @xmath33th measurement @xmath24 is @xmath34 where @xmath35 . the _ likelihood function _ is the joint pdf for all of the @xmath24 , evaluated with the @xmath24 obtained and regarded as a function of the parameters . since we are assuming that the measurements are independent , the likelihood function is in this case given by the product @xmath36 in the frequentist approach we construct estimators @xmath37 for the parameters @xmath38 , usually by finding the values that maximize the likelihood function . ( we will write estimators for parameters with hats . ) in this case one can see from ( [ eq : yilikelihood ] ) that this is equivalent to minimizing the quantity @xmath39 where @xmath40 represents terms that do not depend on the parameters . thus for the case of independent gaussian measurements , the maximum likelihood ( ml ) estimators for the parameters coincide with those of the method of least squares ( ls ) . suppose first that the slope parameter @xmath32 is known exactly , and so it is not adjusted to maximize the likelihood ( or minimize the @xmath41 ) but rather held fixed . the quantity @xmath41 versus the single adjustable parameter @xmath31 would be as shown in fig . [ fig : linefit](b ) , where the minimum indicates the value of the estimator @xmath42 . methods for obtaining the standard deviations of estimators the statistical errors of our measured values are described in many references such as @xcite . here in the case of a single fitted parameter the rule boils down to moving the parameter away from the estimate until @xmath41 increases by one unit ( i.e. , @xmath43 decreases from its maximum by @xmath44 ) as indicated in the figure . it may be , however , that we do not know the value of the slope parameter @xmath32 , and so even though we do not care about its value in the final result , we are required to treat it as an adjustable parameter in the fit . minimizing @xmath45 results in the estimators @xmath46 , as indicated schematically in fig . [ fig : chi2contour](a ) . now the recipe to obtain the statistical errors , however , is not simply a matter of moving the parameter away from its estimated value until the @xmath41 goes up by one unit . here the standard deviations must be found from the tangent lines ( or in higher - dimensional problems , the tangent hyperplanes ) to the contour defined by @xmath47 , as shown in the figure . ( 10.0,6.5 ) ( 0.5,0 ) ( 8.5,0 ) ( 0.,5.5)(a ) ( 15.5,5.5)(b ) the tilt of the contour in fig . [ fig : chi2contour](a ) reflects the correlation between the estimators @xmath42 and @xmath48 . a useful estimate for the inverse of the matrix of covariances @xmath49 $ ] can be found from the second derivative of the log - likelihood evaluated at its maximum , @xmath50 more information on how to extract the full covariance matrix from the contour can be found , for example , in refs . the point to note here is that the correlation between the estimators for the parameter of interest and the nuisance parameter has the result of inflating the standard deviations of both . that is , if @xmath32 were known exactly , then the distance one would have to move @xmath31 away from its estimated value to make the @xmath41 increase by one unit would be less , as one can see from the figure . so although we can improve the ability of a model to describe the data by including additional nuisance parameters , this comes at the price of increasing the statistical errors . this is an important theme which we will encounter often in data analysis . now consider the case where we have a prior measurement of @xmath32 . for example , we could have a measurement @xmath51 which we model as following a gaussian distribution centred about @xmath32 and having a given standard deviation @xmath52 . if this measurement is independent of the other @xmath24 values , then the full likelihood function is obtained simply by multiplying the original one by a gaussian , and so when we find the new @xmath41 from @xmath53 there is an additional term , namely , @xmath54 as shown in fig . [ fig : chi2contour](b ) , the new ( solid ) contour of @xmath55 is compressed relative to the old ( dashed ) one in the @xmath32 direction , and this compression has the effect of decreasing the error in @xmath31 as well . the lesson is : by better constraining nuisance parameters , one improves the statistical accuracy of the parameters of interest . to treat the example above in the bayesian framework , we write bayes theorem ( [ eq : bayesthm ] ) as @xmath56 here @xmath35 symbolizes the hypothesis whose probability we want to determine . the likelihood @xmath57 is the probability to obtain the data @xmath58 given the hypothesis , and the prior probability @xmath59 represents our degree of belief about the parameters before seeing the outcome of the experiment . the posterior probability @xmath60 encapsulates all of our knowledge about @xmath38 when the data @xmath61 is combined with our prior beliefs . the denominator in ( [ eq : bayesthmlinefit ] ) serves to normalize the posterior pdf to unit area . the likelihood @xmath57 is the same as the @xmath62 that we used in the frequentist approach above . the slightly different notation here simply emphasizes its role as the conditional probability for the data given the parameter . to proceed we need to write down a prior probability density @xmath63 . this phase of a bayesian analysis , sometimes called the _ elicitation of expert opinion _ , is in many ways the most problematic , as there are no universally accepted rules to follow . here we will explore some of the important issues that come up . in general , prior knowledge about one parameter might affect knowledge about the other , and if so this must be built into @xmath64 . often , however , one may regard the prior knowledge about the parameters as independent , in which case the density factorizes as @xmath65 for purposes of the present example we will assume that this holds . for the parameter of interest @xmath31 , it may be that we have essentially no prior information , so the density @xmath66 should be very broad . often one takes the limiting case of a broad distribution simply to be a constant , i.e. , @xmath67 now one apparent problem with eq . ( [ eq : pit0 ] ) is that it is not normalizable to unit area , and so does not appear to be a valid probability density . it is said to be an _ improper prior_. the prior always appears in bayes theorem multiplied by the likelihood , however , and as long as this falls off quickly enough as a function of the parameters , then the resulting posterior probability density can be normalized to unit area . a further problem with uniform priors is that if the prior pdf is flat in @xmath38 , then it is not flat for a nonlinear function of @xmath38 , and so a different parametrization of the problem would lead in general to a non - equivalent posterior pdf . for the special case of a constant prior , one can see from bayes theorem ( [ eq : bayesthmlinefit ] ) that the posterior is proportional to the likelihood , and therefore the mode ( peak position ) of the posterior is equal to the ml estimator . the posterior mode , however , will change in general upon a transformation of parameter . a summary statistic other than the mode may be used as the bayesian estimator , such as the median , which is invariant under a monotonic parameter transformation . but this will not in general coincide with the ml estimator . for the prior @xmath68 , let us assume that our prior knowledge about this parameter includes the earlier measurement @xmath51 , which we modelled as a gaussian distributed variable centred about @xmath32 with standard deviation @xmath52 . if we had taken , even prior to that measurement , a constant prior for @xmath32 , then the `` intermediate - state '' prior that we have before looking at the @xmath24 is simply this flat prior times the gaussian likelihood , i.e. , a gaussian prior in @xmath32 : @xmath69 putting all of these ingredients into bayes theorem gives @xmath70 where @xmath71 represents the constant prior in @xmath31 and the equation has been written as a proportionality with the understanding that the final posterior pdf should be normalized to unit area . what bayes theorem gives us is the full joint pdf @xmath72 for both the parameter of interest @xmath31 as well as the nuisance parameter @xmath32 . to find the pdf for the parameter of interest only , we simply integrate ( marginalize ) the joint pdf , i.e. , @xmath73 in this example , it turns out that we can do the integral in closed form . we find a gaussian posterior , @xmath74 where @xmath42 is in fact the same as the ml ( or ls ) estimator found above with the frequentist approach , and @xmath75 is the same as the standard deviation of that estimator @xmath76 . so we find something that looks just like the frequentist answer , although here the interpretation of the result is different . the posterior pdf @xmath77 gives our degree of belief about the location of the parameter in the light of the data . we will see below how the bayesian approach can , however , lead to results that differ both in interpretation as well as in numerical value from what would be obtained in a frequentist calculation . first , however , we need to pause for a short digression on bayesian computation . in most real bayesian calculations , the marginalization integrals can not be carried out in closed form , and if the number of nuisance parameters is too large then they can also be difficult to compute with standard monte carlo methods . however , _ markov chain monte carlo _ ( mcmc ) has become the most important tool for computing integrals of this type and has revolutionized bayesian computation . in - depth treatments of mcmc can be found , for example , in the texts by robert and casella @xcite , liu @xcite , and the review by neal @xcite . the basic idea behind using mcmc to marginalize the joint pdf @xmath78 is to sample points @xmath79 according to the posterior pdf but then only to look at the distribution of the component of interest , @xmath31 . a simple and widely applicable mcmc method is the metropolis - hastings algorithm , which allows one to generate multidimensional points @xmath38 distributed according to a target pdf that is proportional to a given function @xmath60 , which here will represent our posterior pdf . it is not necessary to have @xmath60 normalized to unit area , which is useful in bayesian statistics , as posterior probability densities are often determined only up to an unknown normalization constant , as is the case in our example . to generate points that follow @xmath60 , one first needs a proposal pdf @xmath80 , which can be ( almost ) any pdf from which independent random values @xmath38 can be generated , and which contains as a parameter another point in the same space @xmath81 . for example , a multivariate gaussian centred about @xmath81 can be used . beginning at an arbitrary starting point @xmath81 , the hastings algorithm iterates the following steps : 1 . generate a value @xmath38 using the proposal density @xmath80 ; 2 . form the hastings test ratio , @xmath82 $ ] ; 3 . generate a value @xmath83 uniformly distributed in @xmath84 $ ] ; 4 . if @xmath85 , take @xmath86 . otherwise , repeat the old point , i.e. , @xmath87 . if one takes the proposal density to be symmetric in @xmath38 and @xmath81 , then this is the _ metropolis_hastings algorithm , and the test ratio becomes @xmath88 $ ] . that is , if the proposed @xmath38 is at a value of probability higher than @xmath81 , the step is taken . if the proposed step is rejected , hop in place . methods for assessing and optimizing the performance of the algorithm are discussed , for example , in refs . one can , for example , examine the autocorrelation as a function of the lag @xmath89 , i.e. , the correlation of a sampled point with one @xmath89 steps removed . this should decrease as quickly as possible for increasing @xmath89 . generally one chooses the proposal density so as to optimize some quality measure such as the autocorrelation . for certain problems it has been shown that one achieves optimal performance when the acceptance fraction , that is , the fraction of points with @xmath90 , is around 40% . this can be adjusted by varying the width of the proposal density . for example , one can use for the proposal pdf a multivariate gaussian with the same covariance matrix as that of the target pdf , but scaled by a constant . for our example above , mcmc was used to generate points according to the posterior pdf @xmath91 by using a gaussian proposal density . the result is shown in fig . [ fig : mcmc ] . ( 10.0,5 ) ( 0.,0 ) ( 5.3,0 ) ( 10.8,0 ) ( 4.3,4.)(a ) ( 9.6,4.)(b ) ( 14.9,4.)(c ) from the @xmath92 points in the scatter plot in fig . [ fig : mcmc](a ) we simply look at the distribution of the parameter of interest , @xmath31 [ fig . [ fig : mcmc](b ) ] . the standard deviation of this distribution is what we would report as the statistical error in our measurement of @xmath31 . the distribution of the nuisance parameter @xmath32 from fig . [ fig : mcmc](c ) is not directly needed , although it may be of interest in some other context where that parameter is deemed interesting . in fact one can go beyond simply summarizing the width of the distributions with the a statistic such as the standard deviation . the full form of the posterior distribution of @xmath31 contains useful information about where the parameter s true value is likely to be . in this example the distributions will in fact turn out to be gaussian , but in a more complex analysis there could be non - gaussian tails and this information can be relevant in drawing conclusions from the result . the posterior distribution of @xmath31 obtained above encapsulates all of the analyst s knowledge about the parameter in the light of the data , given that the prior beliefs were reflected by the density @xmath63 . a different analyst with different prior beliefs would in general obtain a different posterior pdf . we would like the result of a bayesian analysis to be of value to the broader scientific community , not only to those that share the prior beliefs of the analyst . and therefore it is important in a bayesian analysis to show by how much the posterior probabilities would change upon some reasonable variation in the prior . this is sometimes called the _ sensitivity analysis _ and is an important part of any bayesian calculation . in the example above , we can imagine a situation where there was no prior measurement @xmath51 of the parameter @xmath32 , but rather a theorist had told us that , based on considerations of symmetry , consistency , aesthetics , etc . , @xmath32 was `` almost certainly '' positive , and had a magnitude `` probably less than 0.1 or so '' . when pressed to be precise , the theorist sketches a curve roughly resembling an exponential with a mean of 0.1 . so we can express this prior as @xmath93 with @xmath94 . we can substitute this prior into bayes theorem ( [ eq : bthmlf ] ) to obtain the joint pdf for @xmath31 and @xmath32 , and then marginalize to find the pdf for @xmath31 . doing this numerically with mcmc results in the posterior distributions shown in fig . [ fig : piexp](a ) . ( 10.0,6.5 ) ( 0.5,0 ) ( 8.5,0 ) ( 0.,5.5)(a ) ( 15.5,5.5)(b ) now the theorist who proposed this prior for @xmath32 may feel reluctant to be pinned down , and so it is important to recall ( and to reassure the theorist about ) the `` if - then '' nature of a bayesian analysis . one does not have to be absolutely certain about the prior in eq . ( [ eq : pi1exp ] ) . rather , bayes theorem simply says that _ if _ one were to have these prior beliefs , _ then _ we obtain certain posterior beliefs in the light of the data . one simple way to vary the prior here is to try different values of the mean @xmath95 , as shown in fig . [ fig : piexp](a ) . we see here the same basic feature as shown already in the frequentist analysis , namely , that when one increases the precision about the nuisance parameter , @xmath32 , then the knowledge about the parameter of interest , @xmath31 , is improved . alternatively ( or in addition ) we may try different functional forms for the prior , as shown in fig . [ fig : piexp](b ) . in this case using a uniform distribution for @xmath68 with @xmath96 or gaussian with @xmath97 truncated for @xmath98 both give results similar to the exponential with a mean of @xmath99 . so one concludes that the result is relatively insensitive to the detailed nature of the tails of @xmath68 . we can now generalize the example of section [ sec : linefit ] to explore some further aspects of a bayesian analysis . let us suppose that we are given a set of @xmath100 measurements as above , but now in addition to the statistical errors we also are given systematic errors . that is , we are given @xmath101 for @xmath102 where the measurements as before are each carried out for a specified value of a control variable @xmath27 . more generally , instead of having @xmath103 it may be that the set of measurements comes with an @xmath104 covariance matrix @xmath105 corresponding to the statistical errors and another matrix @xmath106 for the systematic ones . here the square roots of the diagonal elements give the errors for each measurement , and the off - diagonal elements provide information on how they are correlated . as before we assume some functional form @xmath107 for the expectation values of the @xmath24 . this could be the linear model of eq . ( [ eq : muofx ] ) or something more general , but in any case it depends on a vector of unknown parameters @xmath38 . in this example , however , we will allow that the model is not perfect , but rather could have a systematic bias . that is , we write that the true expectation value of the @xmath33th measurement can be written @xmath108 = \mu(x_i ; \boldvec{\theta } ) + b_i \;,\ ] ] where @xmath109 represents the bias . the @xmath109 can be viewed as the systematic errors of the model , present even when the parameters @xmath38 are adjusted to give the best description of the data . we do not know the values of the @xmath109 . if we did , we would account for them in the model and they would no longer be biases . we do not in fact know that their values are nonzero , but we are allowing for the possibility that they could be . the reported systematic errors are intended as a quantitative measure of how large we expect the biases to be . as before , the goal is to make inferences about the parameters @xmath38 ; some of these may be of direct interest and others may be nuisance parameters . in section [ sec : genfreq ] we will try to do this using the frequentist approach , and in section [ sec : genbay ] we will use the bayesian method . if we adopt the frequentist approach , we need to write down a likelihood function such as eq . ( [ eq : yilikelihood ] ) , but here we know in advance that the model @xmath107 is not expected to be fully accurate . furthermore it is not clear how to insert the systematic errors . often , perhaps without a clear justification , one simply adds the statistical and systematic errors in quadrature , or in the case where one has the covariance matrices @xmath105 and @xmath106 , they are summed to give a sort of ` full ' covariance matrix : @xmath110 one might then use this in a multivariate gaussian likelihood function , or equivalently it could be used to construct the @xmath41 , @xmath111 which is then minimized to find the ls estimators for @xmath38 . in eq . ( [ eq : chi2multi ] ) the vector @xmath58 should be understood as a column vector , @xmath112 is the corresponding vector of model values , and the superscript @xmath113 represents the transpose ( row ) vector . minimizing this @xmath41 gives the generalized ls estimators @xmath37 , and the usual procedures can be applied to find their covariances , which now in some sense include the systematics . but in what sense is there any formal justification for adding the covariance matrices in eq . ( [ eq : fullcov ] ) ? next we will treat this problem in the bayesian framework and see that there is indeed some reason behind this recipe , but with limitations , and further we will see how to get around these limitations . in the corresponding bayesian analysis , one treats the statistical errors as given by @xmath105 as reflecting the distribution of the data @xmath61 in the likelihood . the systematic errors , through @xmath106 , reflect the width of the prior probabilities for the bias parameters @xmath109 . that is , we take @xmath114 \ ; , \\*[0.2 cm ] \label{eq : pib } \pi_b(\boldvec{b } ) & \propto & \exp \left [ - \begin{matrix } \frac{1}{2 } \end{matrix } \boldvec{b}^t v_{\rm sys}^{-1 } \boldvec{b } \right ] \ ; , \quad \quad \pi_{\theta}(\boldvec{\theta } ) = \mbox{const . } \ ; , \\*[0.2 cm ] \label{eq : bayesthetab } p(\boldvec{\theta } , \boldvec{b } | \boldvec{y } ) & \propto & l(\boldvec{y } | \boldvec{\theta } , \boldvec{b } ) \pi_{\theta}(\boldvec{\theta } ) \pi_{b}(\boldvec{b } ) \;,\end{aligned}\ ] ] where in ( [ eq : bayesthetab ] ) , bayes theorem is used to obtain the joint probability for the parameters of interest , @xmath38 , and also the biases @xmath115 . to obtain the probability for @xmath38 we integrate ( marginalize ) over @xmath115 , @xmath116 one finds that the mode of @xmath117 is at the same position as the least - squares estimates , and its covariance will be the same as obtained from the frequentist analysis where the full covariance matrix was given by the sum @xmath118 . so this can be taken in effect as the formal justification for the addition in quadrature of statistical and systematic errors in a least - squares fit . if one stays with the prior probabilities used above , the bayesian and least - squares approaches deliver essentially the same result . an advantage of the bayesian framework , however , is that it allows one to refine the assessment of the systematic uncertainties as expressed through the prior probabilities . for example , the least - squares fit including systematic errors is equivalent to the assumption of a gaussian prior for the biases . a more realistic prior would take into account the experimenter s own uncertainty in assigning the systematic error , i.e. , the ` error on the error ' . suppose , for example , that the @xmath33th measurement is characterized by a reported systematic uncertainty @xmath119 and an unreported factor @xmath120 , such that the prior for the bias @xmath109 is @xmath121 \pi_s(s_i ) \ , ds_i \;.\ ] ] here the ` error on the error ' is encapsulated in the prior for the factor @xmath122 , @xmath123 . for this we can take whatever function is deemed appropriate . for some types of systematic error it could be close to the ideal case of a delta function centred about unity . many reported systematics are , however , at best rough guesses , and one could easily imagine a function @xmath124 with a mean of unity but a standard deviation of , say , @xmath125 or more . here we show examples using a gamma distribution for @xmath123 , which results in substantially longer tails for the prior @xmath126 than those of the gaussian . this can be seen in fig . [ fig : pib ] , which shows @xmath127 for different values of the standard deviation of @xmath124 , @xmath128 . related studies using an inverse gamma distribution can be found in refs . @xcite , which have the advantage that the posterior pdf can be written down in closed form . ( 10.0,6 ) ( 1.,-0.2 ) ( 9.0,.8 ) ( 6,4)[b ] using a prior for the biases with tails longer than those of a gaussian results in a reduced sensitivity to outliers , which arise when an experimenter overlooks an important source of systematic uncertainty in the estimated error of a measurement . as a simple test of this , consider the sample data shown in fig . [ fig : outlier](a ) . suppose these represent four independent measurements of the same quantity , here a parameter called @xmath129 , and the goal is to combine the measurements to provide a single estimate of @xmath129 . that is , we are effectively fitting a horizontal line to the set of measured @xmath130 values , where the control variable @xmath27 is just a label for the measurements . in this example , suppose that each measurement @xmath24 , @xmath131 , is modelled as gaussian distributed about @xmath129 , having a standard deviation @xmath132 , and furthermore each measurement has a systematic uncertainty @xmath133 , which here is taken to refer to the standard deviation of the gaussian component of the prior @xmath134 . this is then folded together with @xmath135 to get the full prior for @xmath109 using eq . ( [ eq : pieoe ] ) , and the joint prior for the vector of bias parameters is simply the product of the corresponding terms , as the systematic errors here are treated as being independent . these ingredients are then assembled according to the recipe of eqs . ( [ eq : equivbayes])([eq : margprob ] ) to produce the posterior pdf for @xmath129 , @xmath136 . results of the exercise are shown in fig . [ fig : outlier ] . in fig . [ fig : outlier](a ) , the four measurements @xmath24 are reasonably consistent with each other . figure [ fig : outlier](b ) shows the corresponding posterior @xmath136 for two values of @xmath137 , which reflect differing degrees of non - gaussian tails in the prior for the bias parameters , @xmath134 . for @xmath138 , the prior for the bias is exactly gaussian , whereas for @xmath139 , the non - gaussian tails are considerably longer , as can be seen from the corresponding curves in fig . [ fig : pib ] . the posterior pdfs for both cases are almost identical , as can be see in fig . [ fig : outlier](a ) . determining the mean and standard deviation of the posterior for each gives @xmath140 for the case of @xmath141 , and @xmath142 for @xmath139 . so assuming a 50% `` error on the error '' here one only inflates the error of the averaged result by a small amount . ( 10.0,8.5 ) ( 1.5,5.5 ) ( 8.5,5.5 ) ( 0.5,7.5)(a ) ( 15.,7.5)(b ) ( 1.5,0 ) ( 8.5,0 ) ( 0.5,3.5)(c ) ( 15.,3.5)(d ) now consider the case where one of the measured values is substantially different from the other three , as shown in fig . [ fig : outlier](c ) . here using the same priors for the bias parameters results in the posteriors shown in fig . [ fig : outlier](d ) . the posterior means and standard deviations are @xmath143 for the case of @xmath141 , and @xmath144 for @xmath145 . when we assume a purely gaussian prior for the bias ( @xmath146 ) , the presence of the outlier has in fact no effect on the width of the posterior . this is rather counter - intuitive and results from our assumption of a gaussian likelihood for the data and a gaussian prior for the bias parameters . the posterior mean is however pulled substantially higher than the three other measurements , which are clustered around @xmath147 . if the priors @xmath134 have longer tails , as occurs when we take @xmath139 , then the posterior is broader , and furthermore it is pulled less far by the outlier , as can be seen in fig . [ fig : outlier](d ) . the fact is that the width of the posterior distribution , which effectively tells us the uncertainty on the parameter of interest @xmath129 , becomes coupled to the internal consistency of the data . in contrast , in the ( frequentist ) least - squares method , or in the bayesian approach using a gaussian prior for the bias parameters , the final uncertainty on the parameter of interest is unaffected by the presence of outliers . and in many cases of practical interest , it would be in fact appropriate to conclude that the presence of outliers should indeed increase one s uncertainty about the final parameter estimates . the example shown here can be generalized to cover a wide variety of model uncertainties by including prior probabilities for an enlarged set of model parameters . in these lectures we have seen how bayesian methods can be used in parameter estimation , and this has also given us the opportunity to discuss some aspects of bayesian computation , including the important tool of markov chain monte carlo . although bayesian and frequentist methods may often deliver results that agree numerically , there is an important difference in their interpretation . furthermore , bayesian methods allow one to incorporate prior information that may be based not on other measurements but rather on theoretical arguments or purely subjective considerations . and as these considerations may not find universal agreement , it is important to investigate how the results of a bayesian analysis would change for a reasonable variation of the prior probabilities . it is important to keep in mind that in the bayesian approach , all information about the parameters is encapsulated in the posterior probabilities . so if the analyst also wants to set upper limits or determine intervals that cover the parameter with a specified probability , then this is a straightforward matter of finding the parameter limits such that the integrated posterior pdf has the desired probability content . a discussion of bayesian methods to the important problem of setting upper limits on a poisson parameter is covered in ref . @xcite and references therein ; we will not have time in these lectures to go into that question here . we will also unfortunately not have time to explore bayesian model selection . this allows one to quantify the degree to which the the data prefer one model over the other using a quantity called the bayes factor . these have not yet been widely used in particle physics but should be kept in mind as providing important complementary information to the corresponding outputs of frequentist hypothesis testing such as @xmath148-values . a brief description of bayes factors can be found in ref . @xcite and a more in - depth treatment is given in ref . in the second part of these lectures we will take a look at the important topic of multivariate analysis . in - depth information on this topic can be found in the textbooks @xcite . in a particle physics context , multivariate methods are often used when selecting events of a certain type using some potentially large number of measurable characteristics for each event . the basic framework we will use to examine these methods is that of a frequentist hypothesis test . the fundamental unit of data in a particle physics experiment is the ` event ' , which in most cases corresponds to a single particle collision . in some cases it could be instead a decay , and the picture does not change much if we look , say , at individual particles or tracks . but to be concrete let us suppose that we want to search for events from proton proton collisions at the lhc that correspond to some interesting ` signal ' process , such as supersymmetry . when running at full intensity , the lhc should produce close to a billion events per second . after a quick sifting , the data from around 200 per second are recorded for further study , resulting in more than a billion events per year . but only a tiny fraction of these are of potential interest . if one of the speculative theories such as supersymmetry turns out to be realized in nature , then this will result in a subset of events having characteristic features , and the susy events will simply be mixed in randomly with a much larger number of standard model events . the relevant distinguishing features depend on what new physics nature chooses to reveal , but one might see , for example , high @xmath149 jets , leptons , missing energy . unfortunately , background processes ( e.g. , standard model events ) can often mimic these features and one will not be able to say with certainty that a given event shows a clear evidence for something new such as supersymmetry . for example , even standard model events can contain neutrinos which also escape undetected . the typical amount and pattern of missing energy in these events differs on average , however , from what a susy event would give , and so a statistical analysis can be applied to test whether something besides standard model events is present . in a typical analysis there is a class of event we are interested in finding ( signal ) , and these , if they exist at all , are mixed in with the rest of the events ( background ) . the data for each event is some collection of numbers @xmath150 representing particle energies , momenta , etc . we will refer to these as the _ input variables _ of the problem . and the probabilities are joint densities for @xmath151 given the signal ( s ) or background ( b ) hypotheses : @xmath152 and @xmath153 . to illustrate the general problem , consider the scatterplots shown in fig . [ fig : scatter ] . these show the distribution of two variables , @xmath154 and @xmath155 , which represent two out of a potentially large number of quantities measured for each event . the blue circles could represent the sought after signal events , and the red triangles the background . in each of the three figures there is a decision boundary representing a possible way of classifying the events . ( 10.0,5 ) ( 0.,0 ) ( 5.3,0 ) ( 10.8,0 ) ( 4.4,4.1)(a ) ( 9.6,4.1)(b ) ( 15.1,4.1)(c ) figure [ fig : scatter](a ) represents what is commonly called the ` cut - based ' approach . one selects signal events by requiring @xmath156 and @xmath157 for some suitably chosen cut values @xmath158 and @xmath159 . if @xmath154 and @xmath155 represent quantities for which one has some intuitive understanding , then this can help guide one s choice of the cut values . another possible decision boundary is made with a diagonal cut as shown in fig . [ fig : scatter](b ) . one can show that for certain problems a linear boundary has optimal properties , but in the example here , because of the curved nature of the distributions , neither the cut - based nor the linear solution is as good as the nonlinear boundary shown in fig . [ fig : scatter](c ) . the decision boundary is a surface in the @xmath100-dimensional space of input variables , which can be represented by an equation of the form @xmath160 , where @xmath161 is some constant . we accept events as corresponding to the signal hypothesis if they are on one side of the boundary , e.g. , @xmath162 could represent the acceptance region and @xmath163 could be the rejection region . equivalently we can use the function @xmath164 as a scalar _ test statistic_. once its functional form is specified , we can determine the pdfs of @xmath164 under both the signal and background hypotheses , @xmath165 and @xmath166 . the decision boundary is now effectively a single cut on the scalar variable @xmath130 , as illustrated in fig . [ fig : teststat ] . ( 10.0,5 ) ( 1.5,0 . ) ( 9.0,.8 ) ( 6,4)[b ] to quantify how good the event selection is , we can define the _ efficiency _ with which one selects events of a given type as the probability that an event will fall in the acceptance region . that is , the signal and background efficiencies are @xmath167 \varepsilon_{\rm b } & = & p ( \mbox{accept event } | \mbox{b } ) = \int_{\rm a } f(\boldvec{x } | \mbox{b } ) \ , d \boldvec{x } = \int_{-\infty}^{y_{\rm cut } } p(y | \mbox{b } ) \ , dy \ ; , \end{aligned}\ ] ] where the region of integration a represents the acceptance region . dividing the space of input variables into two regions where one accepts or rejects the signal hypothesis is essentially the language of a frequentist statistical test . if we regard background as the ` null hypothesis ' , then the background efficiency is the same as what in a statistical context would be called the significance level of the test , or the rate of ` type - i error ' . viewing the signal process as the alternative , the signal efficiency is then what a statistician would call the power of the test ; it is the probability to reject the background hypothesis if in fact the signal hypothesis is true . equivalently , this is one minus the rate of ` type - ii error ' . the use of a statistical test to distinguish between two classes of events ( signal and background ) , comes up in different ways . sometimes both event classes are known to exist , and the goal is to select one class ( signal ) for further study . for example , proton proton collisions leading to the production of top quarks are a well - established process . by selecting these events one can carry out precise measurements of the top quark s properties such as its mass . in other cases , the signal process could represent an extension to the standard model , say , supersymmetry , whose existence is not yet established , and the goal of the analysis is to see if one can do this . rejecting the standard model with a sufficiently high significance level amounts to discovering something new , and of course one hopes that the newly revealed phenomena will provide important insights into how nature behaves . what the physicist would like to have is a test with maximal power with respect to a broad class of alternative hypotheses . for two specific signal and background hypotheses , it turns out that there is a well defined optimal solution to our problem . the _ neyman pearson _ lemma states that one obtains the maximum power relative for the signal hypothesis for a given significance level ( background efficiency ) by defining the acceptance region such that , for @xmath151 inside the region , the _ likelihood ratio _ , i.e. , the ratio of pdfs for signal and background , @xmath168 is greater than or equal to a given constant , and it is less than this constant everywhere outside the acceptance region . this is equivalent to the statement that the ratio ( [ eq : lratio ] ) represents the test statistic with which one obtains the highest signal efficiency for a given background efficiency , or equivalently , for a given signal purity . in principle the signal and background theories should allow us to work out the required functions @xmath152 and @xmath169 , but in practice the calculations are too difficult and we do not have explicit formulae for these . what we have instead of @xmath152 and latexmath:[$f(\boldvec{x } sample @xmath151 to produce simulated signal and background events . because of the multivariate nature of the data , where @xmath151 may contain at least several or perhaps even hundreds of components , it is a nontrivial problem to construct a test with a power approaching that of the likelihood ratio . in the usual case where the likelihood ratio ( [ eq : lratio ] ) can not be used explicitly , there exists a variety of other multivariate classifiers that effectively separate different types of events . methods often used in hep include _ neural networks _ or _ fisher discriminants_. recently , further classification methods from machine learning have been applied in hep analyses ; these include _ probability density estimation ( pde ) _ techniques , _ kernel - based pde _ ( _ kde _ or _ parzen window _ ) , _ support vector machines _ , and _ decision trees_. techniques such as ` boosting ' and ` bagging ' can be applied to combine a number of classifiers into a stronger one with greater stability with respect to fluctuations in the training data . descriptions of these methods can be found , for example , in the textbooks @xcite and in proceedings of the phystat conference series @xcite . software for hep includes the tmva @xcite and statpatternrecognition @xcite packages , although support for the latter has unfortunately been discontinued . as we will not have the time to examine all of the methods mentioned above , in the following section we look at a specific example of a classifier to illustrate some of the main ideas of a multivariate analysis : the boosted decision tree ( bdt ) . boosted decision trees exploit relatively recent developments in machine learning and have gained significant popularity in hep . first in section [ sec : dectree ] we describe the basic idea of a decision tree , and then in section [ sec : boosting ] we will say how the the technique of ` boosting ' can be used to improve its performance . a decision tree is defined by a collection of successive cuts on the set of input variables . to determine the appropriate cuts , one begins with a sample of @xmath171 training events which are known to be either signal or background , e.g. , from monte carlo . the set of @xmath100 input variables measured for each event constitutes a vector @xmath172 . thus we have @xmath171 instances of @xmath151 , @xmath173 , as well as the corresponding @xmath171 true class labels @xmath174 . it is convenient to assign numerical values to the labels so that , e.g. , @xmath175 corresponds to signal and @xmath176 for background . in addition we will assume that each event can be assigned a weight , @xmath177 , with @xmath178 . for any subset of the events and for a set of weights , the signal fraction ( purity ) is taken to be @xmath179 where s and b refer to the signal and background event types . the weights are not strictly speaking necessary for a decision tree , but will be used in connection with boosting in section [ sec : boosting ] . for a decision tree without boosting we can simply take all the weights to be equal . to quantify the degree of separation achieved by a classifier for a selected subset of the events one can use , for example , the _ gini coefficient _ @xcite , which historically has been used as a measure of dispersion in economics and is defined as @xmath180 the gini coefficient is zero if the selected sample is either pure signal or background . another measure is simply the misclassification rate , @xmath181 the idea behind a decision tree is illustrated in fig . [ fig : miniboonebdt ] , from an analysis by the miniboone neutrino oscillation experiment at fermilab @xcite . ( 10.0,6 ) ( .5,-2 ) ( 9.0,.8 ) ( 6,4)[b ] one starts with the entire sample of training events in the root node , shown in the figure with 52 signal and 48 background events . out of all of the possible input variables in the vector @xmath151 , one finds the component that provides the best separation between signal and background by use of a single cut . this requires a definition of what constitutes ` best separation ' , and there are a number of reasonable choices . for example , for a cut that splits a set of events @xmath182 into two subsets @xmath183 and @xmath184 , one can define the degree of separation through the weighted change in the gini coefficients , @xmath185 where @xmath186 and similarly for @xmath187 and @xmath188 . alternatively one may use a quantity similar to ( [ eq : delta ] ) but with the misclassification rate ( [ eq : misclassrate ] ) , for example , instead of the gini coefficient . more possibilities can be found in ref . @xcite . for whatever chosen measure of degree of separation , @xmath189 , one finds the cut on the variable amongst the components of @xmath151 that maximizes it . in the example of the miniboone experiment shown in fig . [ fig : miniboonebdt ] , this happened to be a cut on the number of pmt hits with a value of 100 . this splits the training sample into the two daughter nodes shown in the figure , one of which is enhanced in signal and the other in background events . the algorithm requires a stopping rule based , for example , on the number of events in a node or the misclassification rate . if , for example , the number of events or the misclassification rate in a given node falls below a certain threshold , then this is defined as a terminal node or ` leaf ' . it is classified as a signal or background leaf based on its predominant event type . in fig . [ fig : miniboonebdt ] , for example , the node after the cut on pmt hits with 4 signal and 37 background events is classified as a terminal background node . for nodes that have not yet reached the stopping criterion , one iterates the procedure and finds , as before , the variable that provides the best separation with a single cut . in fig . [ fig : miniboonebdt ] this is an energy cut of @xmath190 . the steps are continued until all nodes reach the stopping criterion . the resulting set of cuts effectively divides the @xmath151 space into two regions : signal and background . to provide a numerical output for the classifier we can define @xmath191 -1 & \quad \boldvec{x } \mbox { in background region } . \end{array } \right.\ ] ] equation ( [ eq : ftree ] ) defines a decision tree classifier . in this form , these tend to be very sensitive to statistical fluctuations in the training data . one can easily see why this is , for example , if two of the components of @xmath151 have similar discriminating power between signal and background . for a given training sample , one variable may be found to give the best degree of separation and is chosen to make the cut , and this affects the entire further structure of the tree . in a different statistically independent sample of training events , the other variable may be found to be better , and the resulting tree could look very different . boosting is a technique that can decrease the sensitivity of a classifier to such fluctuations , and we describe this in the following section . boosting is a general method of creating a set of classifiers which can be combined to give a new classifier that is more stable and has a smaller misclassification rate than any individual one . it is often applied to decision trees , precisely because they suffer from sensitivity to statistical fluctuations in the training sample , but the technique can be applied to any classifier . let us suppose as above that we have a sample of @xmath171 training events , i.e. , @xmath171 instances of the data vector , @xmath192 , and @xmath171 true class labels @xmath193 , with @xmath175 for signal and @xmath194 for background . also as above assume we have @xmath171 weights @xmath195 , where the superscript @xmath196 refers to the fact that this is the first training set . we initially set the weights equal and normalized such that @xmath197 the idea behind boosting is to create from the initial sample , a series of further training samples which differ from the initial one in that the weights will be changed according to a specific rule . a number of boosting algorithms have been developed , and these differ primarily in the rule used to update the weights . we will describe the adaboost algorithm of freund and schapire @xcite , as it was one of the first such algorithms and its properties have been well studied . one begins with the initial training sample and from it derives a classifier . we have in mind here a decision tree , but it could be any type of classifier for where the training employs the event weights . the resulting function @xmath198 will have a certain misclassification rate @xmath199 . in general for the @xmath89th classifier ( i.e. , based on the @xmath89th training sample ) , we can write the error rate as @xmath200 where @xmath201 if the boolean expression @xmath202 is true , and is zero otherwise . we then assign a score to the classifier based on its error rate . for the adaboost algorithm this is @xmath203 which is positive as long as the error rate is lower than 50% , i.e , . the classifier does better than random guessing . having carried out these steps for the initial training sample , we define the second training sample by updating the weights . more generally , the weights for step @xmath204 are found from those for step @xmath89 by @xmath205 where the factor @xmath206 is chosen so that the sum of the updated weights is equal to unity . note that if an event is incorrectly classified , then the true class label @xmath24 and the value @xmath207 have opposite signs , and thus the new weights are greater than the old ones . correctly classified events have their weights decreased . this means that the updated training set will pay more attention in the next iteration to those events that were not correctly classified , the idea being that it should try harder to get it right the next time around . after @xmath208 iterations of this procedure one has classifiers @xmath209 , each with a certain error rate and score based on eqs . ( [ eq : errorrate ] ) and ( [ eq : alphak ] ) . in the case of decision trees , the set of new trees is called a _ forest_. from these one defines an averaged classifier as @xmath210 equation ( [ eq : boostave ] ) defines a boosted decision tree ( or more generally , a boosted version of whatever classifier was used ) . one of the important questions to be addressed is how many boosting iterations to use . one can show that for a sufficiently large number of iterations , a boosted decision tree will eventually classify all of the events in the training sample correctly . similar behaviour is found with any classification method where one can control to an arbitrary degree the flexibility of the decision boundary . the user can arrange it so that the boundary twists and turns so as to get all of the events on the right side . in the case of a neural network , for example , one can increase the number of hidden layers , or the number of nodes in the hidden layers ; for a support vector machine , one can adjust the width of the kernel function and the regularization parameter to increase the flexibility of the boundary . an example is shown in fig . [ fig : overtraining](a ) , where an extremely flexible classifier has managed to enclose all of the signal events and exclude all of the background . ( 10.0,6.5 ) ( 0.5,0 ) ( 8.5,0 ) ( 0.,5.5)(a ) ( 15.5,5.5)(b ) of course if we were now to take the decision boundary shown in fig . [ fig : overtraining](a ) and apply it to a statistically independent data sample , there is no reason to believe that the contortions that led to such good performance on the training sample will still work . this can be seen in fig . [ fig : overtraining](b ) , which shows the same boundary with a new data sample . in this case the classifier is said to be _ its error rate calculated from the same set of events used to train the classifier underestimates the rate on a statistically independent sample . to deal with overtraining , one estimates the misclassification rate not only with the training data sample but also with a statistically independent test sample . we can then plot these rates as a function of the parameters that regulate the flexibility of the decision boundary , e.g. , the number of boosting iterations used to form the bdt . for a small number of iterations , one will find in general that the error rates for both samples drop . the error rate based on the training sample will continue to drop , eventually reaching zero . but at some point the error rate from the test sample will cease to decrease and in general will increase . one chooses the architecture of the classifier ( number of boosting iterations , number of nodes or layers in a neural network , etc . ) to minimize the error rate on the test sample . as the test sample is used to choose between a number of competing architectures based on the minimum observed error rate , this in fact gives a biased estimate of the true error rate . in principle one should use a third validation sample to obtain an unbiased estimate of the error rate . in many cases the bias is small and this last step is omitted , but one should be aware of its potential existence . in some applications , the training data is relatively inexpensive ; one simply generates more events with monte carlo . but often event generation can take a prohibitively long time and one may be reluctant to use only a fraction of the events for training and the other half for testing . in such cases , procedures such as _ cross validation _ ( see , e.g. , refs . @xcite ) can be used where the available events are partitioned in a number of different ways into training and test samples and the results averaged . boosted decision trees have become increasingly popular in particle physics in recent years . one of their advantages is that they are relatively insensitive to the number of input variables used in the data vector @xmath151 . components that provide little or no separation between signal and background are rarely chosen as for the cut that provides separation , i.e. , to split the tree , and thus they are effectively ignored . decision trees have no difficulty in dealing with different types of data ; these can be real , integer , or they can simply be labels for which there is no natural ordering ( categorical data ) . furthermore , boosted decision trees are surprisingly insensitive to overtraining . that is , although the error rate on the test sample will not decrease to zero as one increases the number of boosting iterations ( as is the case for the training sample ) , it tends not to increase . further discussion of this point can be found in ref . @xcite . the boosted decision tree is an example of a relatively modern development in machine learning that has attracted substantial attention in hep . support vector machines ( svms ) represent another such development and will no doubt also find further application in particle physics ; further discussion on svms can be found in refs . @xcite and references therein . linear classifiers and neural networks will no doubt continue to play an important role , as will probability density estimation methods used to approximate the likelihood ratio . multivariate methods have the advantage of exploiting as much information as possible out of all of the quantities measured for each event . in an environment of competition between experiments , this can be a natural motivation to use them . some caution should be exercised , however , before placing too much faith in the performance of a complicated classifier , to say nothing of a combination of complicated classifiers . these may have decision boundaries that indeed exploit nonlinear features of the training data , often based on monte carlo . but if these features have never been verified experimentally , then they may or may not be present in the real data . there is thus the risk of , say , underestimating the rate of background events present in a region where one looks for signal , which could lead to a spurious discovery . simpler classifiers are not immune to such dangers either , but in such cases the problems may be easier to control and mitigate . one should therefore keep in mind the following quote , often heard in the multivariate analysis community : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ keep it simple . as simple as possible . not any simpler . _ a. einstein _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ to this we can add the more modern variant , _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ if you believe in something you do nt understand , you suffer , _ stevie wonder _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ having made the requisite warnings , however , it seems clear that multivariate methods will play an important role in the discoveries we hope to make at the lhc . one can easily imagine , for example , that 5-sigma evidence for new physics from a highly performant , and complicated , classifier would be regarded by the community with some scepticism . but if this is backed up by , say , 4-sigma significance from a simpler , more transparent analysis , then the conclusion would be more easily accepted , and the team that pursues both approaches may well win the race . in these lectures we have looked at two topics in statistics , bayesian methods and multivariate analysis , which will play an important role in particle physics in the coming years . bayesian methods provide important tools for analysing systematic uncertainties , where prior information may be available that does not necessarily stem solely from other measurements , but rather from theoretical arguments or other indirect means . the bayesian framework allows one to investigate how the posterior probabilities change upon variation of the prior probabilities . through this type of sensitivity analysis , a bayesian result becomes valuable to the broader scientific community . as experiments become more expensive and the competition more intense , one will always be looking for ways to exploit as much information as possible from the data . multivariate methods provide a means to achieve this , and advanced tools such as boosted decision trees have in recent years become widely used . and while their use will no doubt increase as the lhc experiments mature , one should keep in mind that a simple analysis also has its advantages . as one studies the advanced multivariate techniques , however , their properties become more apparent and the community will surely find ways of using them so as to maximize the benefits without excessive risk . i wish to convey my thanks to the students and organizers of the 2009 european school of high - energy physics in bautzen for a highly stimulating environment . the friendly atmosphere and lively discussions created a truly enjoyable and productive school . probabilistic inference using markov chain monte carlo methods _ , technical report crg - tr-93 - 1 , dept . of computer science , university of toronto , available from www.cs.toronto.edu/~radford/res-mcmc.html . b. roe _ et al . _ , boosted decision trees as an alternative to artificial neural networks for particle identification , _ nucl . methods phys . _ * a543 * ( 2005 ) 577584 ; h.j . yang , b. roe and j. zhu , studies of boosted decision trees for miniboone particle identification , _ nucl . methods phys . _ * a555 * ( 2005 ) 370385 .

spatial interferometry , based on the measurement of total absorbed power , can be used to determine the state of coherence of the electromagnetic field to which any energy - absorbing structure is sensitive . the measured coherence tensor can be diagonalized to give the amplitude , phase , polarization patterns , and responsivities of the individual electromagnetic modes through which the structure can absorb energy . because the electromagnetic modes are intimately related to dynamical modes of the system , information about collective excitations can be found . we present simulations , based on the discrete dipole approximation ( dda ) , showing how the dynamical modes of systems of surface dipoles can be recovered . interactions are taken into consideration , leading to long - range coherent phenomena , which are revealed by the method . the use of dda enables the interferometric response of a wide variety of objects to be modeled , from patterned photonic films to biological macromolecules .

1208.1381

understanding the processes by which molecules and particles on surfaces absorb and scatter electromagnetic radiation is an area of considerable scientific importance . information may be needed because of an intrinsic desire to study surface physics , because an active surface forms part of an electronic device , or because of the need to locate defects and identify contaminants . examples include the dynamics of adsorbed biomolecules @xcite , the adhesion and movement of droplets @xcite , the plasmonic behaviour of nanoparticles @xcite , and the photonic response of patterned films @xcite . surface processes have traditionally been studied through infrared and optical microscopy , spectroscopy , and scattering , with afm playing a key role in recent years . often , the information needed has to be acquired indirectly through measurements made at inappropriate wavelengths , or through models that are hard to extrapolate . crucially , it is necessary to measure the state of coherence of the field to which a system is sensitive in order to fully characterize its behaviour : for example when calculating radiative heat transfer between nanoparticles @xcite . measurements made with a single source only gain access to a subset of the information available . of particular importance are those cases where the ability of a surface to absorb energy is central to the operation of a device . for example , complex schemes are being explored to increase the efficiency of photovoltaics @xcite , to optimize the sensitivity of infrared and optical sensors @xcite , and to develop frequency selective surfaces , high - power mirrors , and hyperspectral black absorbers . in all cases , the physics of the absorption process is intimately related to the existence of coherent phenomena such as resonant dipole interactions , surface plasmon coupling @xcite , and surface phonon exchange . even in the case of apparently simple structures , such as planar dielectric and metallic interfaces , the near - field spatial and polarimetric correlations contain a wealth of information @xcite . a typical set of questions is as follows : what are the scale sizes and forms of the dynamical modes responsible for absorbing energy ? how many modes are involved in the absorption process ? what are their responsivities ? what physical mechanisms mediate the collective behaviour ? in this paper we argue that it is possible to measure the state of coherence of the electromagnetic field to which any system is sensitive , and thereby determine the spatial forms , polarizations , and relative strengths of the individual modes responsible for absorbing energy . because the electromagnetic modes are intimately related to dynamical modes of the system , it is possible to accrue direct information about the dynamical modes themselves , and thereby reveal the existence of collective phenomena . the basic technique can be used at any wavelength , and can be implemented in a number of different ways . the basic idea is to illuminate the structure under test ( sut ) with a pair of phase - locked , near - field or far - field point sources . as the differential phase between the sources is rotated , the _ total _ absorbed power displays a fringe . if the complex visibility of this fringe is measured for different pairs of source locations , the dynamical forms of the modes responsible for absorbing the energy can be found . in the case of detectors and photovoltaics , the scheme is particularly easy to implement because the degrees of freedom responsible for generating the measured output are those of practical significance . although the technique is applicable to a wide range of problems , we illustrate the method by showing how it provides information about the collective behaviour of systems of surface dipoles . the sut is modeled by using the discrete dipole approximation ( dda ) @xcite , together with a polarizability that allows for bound states and damping to the substrate . dda was originally devised @xcite to model light extinction by astrophysical particles , but has since been developed in areas as diverse as modeling the aggregation of irregular particles from colloids @xcite , the scattering of light by particles on surfaces @xcite , the scattering of light by periodic targets @xcite , and the optical properties of viruses , proteins , dna , and blood cells @xcite . formulating our measurement procedure in terms of dda opens the way to predicting the interferometric response of a wide variety of complex objects such as biomolecules , droplets , and nanoparticles , and thereby creating representative dynamical models of measured interferometric data . other simulation methods , such as mom , can be used . as an aside , a variant of eai can be used with dda purely as a numerical tool for determining the state of coherence , and therefore the incident electromagnetic modes , to which any irregular dielectric body is sensitive . the scheme differs from the usual method for determining the state of coherence of thermally radiated fields , through @xmath0 where @xmath1 is the electric green s dyadic of the whole system including scattering , because ( [ a ] ) requires the object to be in thermal equilibrium with its surroundings , and therefore includes the blackbody environment and scattered fields . as a consequence , even lossless dielectric objects display near - field coherence effects , as discussed by yannopapas in the context of chains of spheres @xcite . in contrast , eai only includes those degrees of freedom responsible for absorbing energy . in section [ sec : basic_scheme ] we present the theory of energy absorption interferometry ( eai ) , and then in section [ sec : simulations ] illustrate how it can be used to recover the dynamical modes of ring - like and linear chains of coupled dipoles , with and without defects . linear chains of dipoles @xcite , spheres @xcite , and cylinders @xcite are of practical importance , and exhibit complex forms of spatial and spectral coherence . we show how eai can reveal large - scale collective behaviour , and how the response of a discrete system transitions to that of a continuous system when the sut contains many dipoles per wavelength . in section [ sec : practical ] , we discuss practical implementation . a collection of particles on a surface . the individual particles have dipole moments @xmath2 . the system is illuminated by a pair of phase - locked sources , which can have any polarisation and position . the experiment measures the complex visibility of the absorbed power as the differential phase between the sources is varied . ] consider a collection of particles whose electromagnetic behaviour can be modeled as a system of dipoles . although eai can be applied to any three - dimensional distribution , assume that the particles are bound to a surface , fig . the dynamical response of dipole @xmath3 can be described by the lorentz model @xmath4 where @xmath5 is the effective separation vector of charge @xmath6 from its partner pinned at @xmath7 . the dyadic product @xmath8 constrains @xmath6 to move in the @xmath9 direction , but other dyads could be used to place more general constraints on the degrees of freedom of motion . @xmath10 is the resonant frequency due to binding , @xmath11 the mass , and @xmath12 the momentum relaxation rate of dipole @xmath3 . the damping arises from a transfer of energy to the phonon or electron system of the substrate . ( [ a1 ] ) does not include a radiation - reaction term , because we are only interested in the absorbed power , and the radiation - reaction term has little effect on behaviour . @xmath13 is the applied field , which includes the incident field , @xmath14 , and the scattered field from all other dipoles , @xmath15 . in the parlance of dda , @xmath13 is the ` local field ' because it does not include the self field of the dipole at @xmath7 . invoking time - harmonic behaviour , @xmath16 , leads to a polarization of @xmath17 the anisotropic susceptibility tensor , @xmath18 , defined through @xmath19 , becomes @xmath20 it will be convenient to describe behaviour in terms of polarisation current density @xmath21 , giving @xmath22 , where @xmath23 large currents can flow at the resonance frequencies , @xmath10 , and an increase in either binding , @xmath10 , or damping , @xmath12 , decreases the low - frequency conductivity . the permittivity and conductivity tensors are valid in both the high - frequency and quasi - static regimes . to simulate eai , it is necessary to calculate the total time - averaged power , @xmath24 , transferred to the substrate . the power can be calculated by treating the system as a dielectric ( [ a3 ] ) , or conductor ( [ a4 ] ) . adopting the latter , @xmath25 giving @xmath26^{2 } + \gamma_{i}^{2 } } \left| { \bf e}({\bf r}_{i } ) \cdot \hat{\bf x}_{i } \right|^{2 } \mbox{.}\ ] ] it is convenient to present ( [ a6 ] ) in the symmetric form @xmath27 where @xmath28 is diagonal in position , which emphasizes its local form . when applied to the system of dipoles , @xmath29 where @xmath30 is a discretized conductivity tensor . ( [ a6 ] ) and ( [ a8 ] ) can not be used unless the total electric field , @xmath31 is known , but the total electric field is the sum of the illuminating field , @xmath32 , and the scattered field , @xmath33 , giving @xmath34 \mbox{.}\ ] ] the scattered field can be calculated through the green s dyadic of the environment , @xmath35 where all multiplicative factors have been included in @xmath36 . green s dyadics of dipoles on , and under , dielectric surfaces have been studied extensively , including multilayers @xcite and surface roughness @xcite . ordinarily , we might use the green s dyadic corresponding to the dielectric half space shown in fig . [ fig1 ] , but for illustrative purposes , and consistency with dda , we shall use the green s dyadic of free space , ignoring the complex refractive index of the substrate . of course , the surface could also be modelled through dda . in the case of particles on very thin , say sin , membranes , this assumption is valid without further work , but in any case , it does not alter the substance of what follows . the same functional forms shall be used to describe scattering by the dipoles and the excitation of the system by external point sources . the free - space green s dyadic is @xmath37 g({\bf r}| { \bf r } ' ) \mbox{,}\ ] ] where @xmath38 is the scalar green s function @xmath39 \mbox{.}\ ] ] @xmath40 denotes the gradient with respect to the observation point @xmath41 . the first term in ( [ a14 ] ) emerges from the magnetic vector potential , and the second term from the electric scalar potential , and therefore @xmath42 includes coupling due to current flow and the exposure of charge . in order to perform numerical simulations , it is convenient to express @xmath42 in cartesian coordinates . the green s dyadic has near - field ( nf ) @xmath43 , intermediate - field ( if ) @xmath44 , and far - field ( ff ) @xmath45 , contributions : @xmath46 setting @xmath47 , and evaluating ( [ a14 ] ) gives @xmath48 } { 4 \pi r } \frac{1}{k^{2}r^{2 } } \left [ \overline{\overline{\bf i } } - \frac{3 { \bf r}{\bf r}}{r^{2 } } \right ] \\ \nonumber \overline{\overline{\bf g}}_{if } & = & i \omega \mu_{0 } \frac { \exp\left [ i k r \right ] } { 4 \pi r } \frac{i}{k r } \left [ \overline{\overline{\bf i } } - \frac{3 { \bf r}{\bf r}}{r^{2 } } \right ] \\ \nonumber \overline{\overline{\bf g}}_{ff } & = & i \omega \mu_{0 } \frac { \exp\left [ i k r \right ] } { 4 \pi r } \left [ \overline{\overline{\bf i } } - \frac{{\bf r}{\bf r}}{r^{2 } } \right ] \mbox{,}\end{aligned}\ ] ] where @xmath49 . by expressing the green s dyadic in this way , individual terms can be turned on and off during simulations to gain a better understanding of behaviour . it can be shown that as @xmath50 , @xmath51 leads to the quasi - static polarization of the dipoles , and therefore electrostatic coupling is included . we do not include the depolarization term @xmath52 , consistent with the use of the localized rather than total field . in dda , where each dipole represents a small cell of a large dielectric body , care in needed to distinguish between the total and localized fields @xcite , particularly if power is dissipated in the dielectric itself . eai is equally applicable to this case , but for presentational purposes , we have assumed that each particle is small enough that it can be represented by a discrete dipole . there are several possible ways forward : substitute ( [ a13 ] ) in ( [ a12 ] ) and attempt to solve for @xmath53 , or substitute ( [ a12 ] ) in ( [ a13 ] ) and attempt to solve for @xmath31 . the second approach is well conditioned , and so @xmath54 then @xmath55^{-1 } \cdot { \bf e}^{i}({\bf r } ) \ , d^{3}{\bf r } \mbox{,}\ ] ] where , symbolically , @xmath56^{-1}$ ] is the inverse kernel corresponding to @xmath57 \mbox{.}\ ] ] because only a finite number of dipoles is involved , the inverse operator ( [ a19 ] ) exists . substituting @xmath58 in ( [ a18 ] ) , and noting that it is only necessary to calculate the total field at discrete points in order to calculate the dissipated power , @xmath59 where @xmath60 \mbox{.}\ ] ] in dda , parameters in the green s and conductivity tensors are combined to express the scattered field in term of a modified green s tensor and the susceptibility tensor . the physics is , however , the same . it is convenient to write ( [ a22 ] ) in the form @xmath61 \mbox{,}\ ] ] where the tilde denotes normalization with respect to the resonant frequency of the bound dipole : @xmath62 and @xmath63 . @xmath64 is the green s dyadic introduced previously , but with the prefactor @xmath65 changed to @xmath66 . according to ( [ a3 ] ) , @xmath67 is the electrostatic , @xmath50 , polarizability of dipole @xmath3 . the clausius - mossotti relationship can be used for simple shapes : @xmath68 v \mbox{,}\ ] ] where @xmath69 is the volume of the particle , having shape factor @xmath70 . because the scheme is fully discretized , it is convenient to introduce matrix notation . if @xmath71 and @xmath72 are column vectors containing the complex amplitudes of the incident and total cartesian field components at the positions of the dipoles , then ( [ a21 ] ) can be solved : @xmath73 . in matrix notation , ( [ a10 ] ) becomes @xmath74 \\ \nonumber & = & \frac{1}{2 } \ , \mathsf{e}^{\dagger } \mathsf{\sigma}^{r } \mathsf{e } \\ \nonumber & = & \frac{1}{2 } \ , \mathsf{e}^{i \dagger } \left [ \mathsf{m}^{-1 } \right]^{\dagger } \mathsf{\sigma}^{r } \left [ \mathsf{m}^{-1 } \right ] \mathsf{e}^{i } \\ \nonumber & = & \mathsf{e}^{i \dagger } \mathsf{l } \mathsf{e}^{i } \mbox{,}\end{aligned}\ ] ] where @xmath75 is the matrix equivalent of ( [ a11 ] ) , and @xmath76 is the real part . taking the trace of each side of the last line of ( [ a23 ] ) , noting that the trace is invariant to cyclic permutation , and calculating the expectation value gives @xmath77 = \mbox{tr } \left [ \mathsf{e}^{i } \mathsf{l } \right ] \mbox{.}\ ] ] @xmath78 is hermitian , and contains the correlations between the cartesian components of the incident field at the locations of the dipoles . ( [ a24 ] ) gives the expectation value of the absorbed power in terms of the state of coherence of the incident field and a response matrix @xmath79 . not only is ( [ a24 ] ) computationally convenient , it is conceptually important . the trace of the product of a matrix with the adjoint of another is an inner product in the abstract vector space of complex matrices . ( [ a24 ] ) therefore describes the projection of the state of coherence of the incident field onto the state of coherence of the field to which the system is responsive . in the case of continuous absorbers , an integral formulation of the same concept is available @xcite . @xmath78 and @xmath79 are hermitian , and can be diagonalized : @xmath80 , @xmath81 , which can be substituted in ( [ a24 ] ) to reveal @xmath82 ( [ a25 ] ) describes the way in which the modes of the incident field @xmath83 project onto the modes of the system , @xmath84 . the @xmath85 are proportional to the levels of excitation of each of the incident modes , whereas the @xmath86 correspond to the relative responsivities with which the system absorbs energy in each of its modes . according to ( [ a23 ] ) , the response matrix @xmath79 contains two effects : the intrinsic ability of the dipoles to absorb energy , @xmath87 , and the appearance of collective behaviour due to scattering , @xmath88^{-1}$ ] . the electromagnetic green s dyadic induces a wide variety of coherent behaviour , but additionally an elastic green s dyadic could have been included in ( [ a1 ] ) to describe acoustic coupling through the excitation of surface waves . the dynamical modes of interest might then be mediated by long - range phonon exchange . the overall principle , however , remains the same . also , because dda has proven to be an effective tool for modelling electromagnetic scattering from complex objects , including internal retardation , it use here for describing eai means that the interferometric response of macroscopic inhomogenious objects , such as particulate material or droplets of a liquid , can be calculated . in this case , each macroscopic object is represented by a large collection of dipoles , rather than only having one dipole per object . in an experiment , the sut is illuminated by a pair of phase - locked sources , and the absorbed power recorded as the differential phase between the sources is rotated . the absorbed power displays a fringe , and the complex visibility of this fringe is measured . in reality , the frequencies of the sources can be offset slightly , and the amplitude and phase of the fringe measured directly by using lock - in techniques . if the complex visibility is recorded for different pairs of source locations , either in the near field or far field , or a mixture of both , the response matrix @xmath79 can be found . suppose that two sources are used , one producing the incident field @xmath89 , and the other producing @xmath90 . according to ( [ a23 ] ) , when both sources illuminate the sample simultaneously , @xmath91 \mathsf{l } \left [ \mathsf{e}^{i}_{1 } e^{j \phi_{1 } } + \mathsf{e}^{i}_{2 } e^{j \phi_{2 } } \right ] \mbox{.}\ ] ] ensemble averaging is not required because the sources are fully coherent . ( [ b1 ] ) has four terms , @xmath92 where @xmath93 is the phase difference between the sources . the first two terms , @xmath94 and @xmath95 , are real , and give the powers dissipated by the sources individually . the second two terms are complex conjugates of each other @xmath96 , and produce a fringe when the differential phase is rotated . it follows that @xmath97 can be found experimentally . if the field vectors @xmath89 and @xmath90 corresponded to point sources at the positions of the dipoles , the response matrix @xmath79 could be found directly , but in reality only distributed fields are possible . although the response matrix can be computed for different forms of illumination , plane waves , spherical waves , gaussian beams , etc . , here we assume that point sources , in the form of miniature current probes , are used . the current density of a point source at @xmath98 is @xmath99 where @xmath100 is the vector - valued current - moment . this source produces an incident field of the form @xmath101 where @xmath102 is the green s dyadic that couples the source current to the incident field at the dipoles . substituting ( [ b3 ] ) in ( [ b4 ] ) leads to the discretized form @xmath103 for a single , simply - polarized source , @xmath104 has one non - zero entry corresponding to the position and polarization of the probe . using ( [ b5 ] ) and calculating the induced current moment , @xmath105 at each dipole by integrating ( [ a12 ] ) gives @xmath106 the dissipated power can be written @xmath107 \\ \nonumber & = & \mathsf{i}^{p \dagger } \mathsf{g}^{p \dagger } \left [ \mathsf{m}^{-1 } \right]^{\dagger } \mathsf{\sigma}^{r } \left [ \mathsf{m}^{-1 } \right ] \mathsf{g}^{p } \mathsf{i}^{p } \\ \nonumber & = & \mathsf{i}^{p \dagger } \mathsf{g}^{p \dagger } \mathsf{l } \mathsf{g}^{p } \mathsf{i}^{p } \\ \nonumber & = & \mathsf{i}^{p \dagger } \mathsf{h } \mathsf{i}^{p } \mbox{.}\end{aligned}\ ] ] @xmath79 is the intrinsic response matrix with respect to the incident field at the positions of the dipoles , whereas @xmath108 is the response matrix with respect to current moments at the positions of the external sources . to carry out interferometry , two sources are needed . imagine that the current moment corresponding to source 1 is @xmath109 , and the current moment corresponding to source 2 is @xmath110 . @xmath111 and @xmath112 are unit vectors having a single real entry corresponding to a single current source with a single cartesian component . when both sources illuminate the sample simultaneously , @xmath113 \mathsf{h } \left [ \mathsf{i}^{p}_{1 } e^{j \phi_{1 } } + \mathsf{i}^{p}_{2 } e^{j \phi_{2 } } \right ] \mbox{.}\ ] ] ensemble averaging is not required because the sources are coherent . ( [ b8 ] ) has four terms @xmath114 the first two terms , @xmath115 and @xmath116 , are real , and give the powers dissipated by the sources individually . the second two terms satisfy @xmath117 , and produce a fringe when the differential phase is rotated . thus by measuring the power absorbed when each source is applied separately , and the amplitude and phase of the fringe when the two sources are applied simultaneously , @xmath118 and @xmath119 can be found . now move the two sources around within some sampling volume , stepping from one sample position to another . usually the sample points will lie on a line or plane in the near field or far field , but more complicated paths can be used depending on the sampling needed . construct the matrix @xmath120 , where column @xmath121 , @xmath122 , corresponds to the source being at position @xmath121 . guided by ( [ b7 ] ) , form the new matrix @xmath123^{\dagger } \mathsf{\sigma } \left [ \mathsf{m}^{-1 } \right ] \mathsf{g}^{p } \mathsf{i}^{p } \mbox{.}\ ] ] @xmath124 is not hermitian , @xmath125 . consider the hermitian part of @xmath124 , @xmath126 & = & \mathsf{i}^{p \dagger } \mathsf{g}^{p \dagger } \left [ \mathsf{m}^{-1 } \right]^{\dagger } \mathsf{\sigma}^{r } \ , \left [ \mathsf{m}^{-1 } \right ] \mathsf{g}^{p } \mathsf{i}^{p } \\ \nonumber & = & \mathsf{i}^{p \dagger } \mathsf{g}^{p \dagger } \mathsf{l } \mathsf{g}^{p } \mathsf{i}^{p } \\ \nonumber & = & \mathsf{i}^{p \dagger } \mathsf{h } \mathsf{i}^{p } \mbox{.}\end{aligned}\ ] ] according to ( [ b11 ] ) , the @xmath121th element of the leading diagonal of @xmath108 corresponds to the total power absorbed when single source @xmath121 illuminates the sample . the off - diagonal terms describe the fringe when a pair of phase coherent sources @xmath121 and @xmath127 are used . it is therefore possible to populate @xmath108 experimentally . if one source is held fixed at @xmath121 , and the other moved through all other positions @xmath127 , the complex visibility of the spatial fringe is @xmath128 it is possible to plot out @xmath129 , with respect to reference point @xmath121 , by moving along the @xmath121th row of @xmath108 . these plots reveal the transverse and longitudinal coherence lengths , areas , and volumes of the polarized point - source excitation to which the structure is sensitive . @xmath108 is the response matrix of the system with respect to point source excitation in the scanning volume . once it has been populated experimentally , it can be diagonalized : @xmath130 substituting ( [ b13 ] ) in the last line of ( [ b7 ] ) , shows that if the source vector corresponds to each of the normalized eigenvectors in turn , @xmath131 , then the eigenvalues are the relative responsivities of the system . the @xmath132 correspond to source vectors that excite modes of the system . following ( [ b6 ] ) , the dipole moment vector @xmath133 of mode @xmath134 is @xmath135 \mathsf{g}^{p } \mathsf{w}_{m } \mbox{.}\ ] ] no attempt is made to recover the displacements , as generally the charge at each site is not known . response matrix @xmath108 combines three effects : ( i ) the generation of distributed fields by point current sources , described by @xmath136 ; ( ii ) scattering interactions between the dipoles , described by @xmath137 $ ] ; and ( iii ) the intrinsic ability of each dipole to absorb energy , described by @xmath76 . the second and third contributions , which generally can not be separated , are intrinsic to the system , whereas the first term is a feature of the measurement . one could calculate @xmath138 and although these are field modes at the positions of the dipoles , they still relate to the original source volume . to calculate the modes at the positions of the dipoles , it is necessary to ` deconvolve ' the illumination patterns of the sources . using ( [ b13 ] ) , the deconvolved field modes are found by diagonalising @xmath139^{-1 } \mathsf{h } \left [ \mathsf{g}^{p } \right]^{-1 } = \sum_{n } \mathsf{u}_{n } \beta_{n } \mathsf{u}_{n}^{\dagger } \mbox{.}\ ] ] generally , the green s matrix will be singular , or poorly conditioned , depending on how well the sample points and polarizations span the vector space of fields at the dipoles . for poorly conditioned mappings , the generalized inverse should be used to recover those degrees of freedom that are available to the experiment . in reality , it is only necessary to include enough sample points to determine the significant degrees of freedom contained in @xmath79 , which may be fewer than the number of dipoles . one approach is to keep on adding , during an experiment , more and more points until the measurement has converged . incremental singular value decomposition can be used to diagonalise @xmath79 in real time until all of the available degrees of freedom have been found . finally , the incident fields can be mapped onto the dynamical modes of the dipoles through @xmath140 \mathsf{z}_{m } \mbox{.}\ ] ] total power absorbed ( arbitrary units ) as a function of frequency for two parallel dipoles , 0.1 mm apart , each having @xmath141 = 300 ghz , and @xmath142 = 20 ghz . the lines correspond to polarizabilities of ( solid ) @xmath143 = 0.0005 mm@xmath144 , ( dotted ) @xmath143 = 0.005 mm@xmath144 , ( dashed ) @xmath143 = 0.01 mm@xmath144 . ] a typical simulation of an experiment proceeds as follows . establish a system of dipoles , and calculate the conductivity matrix @xmath75 using ( [ a11 ] ) . calculate the scattering operator @xmath145 using ( [ a22b ] ) , and invert it . calculate @xmath79 as defined by ( [ a23 ] ) , and diagonalize it to find the eigenvalues and eigenvectors @xmath84 and @xmath86 . these are the collective dynamical modes of the system responsible for absorbing energy , which we would like to find . choose a selection of source positions and polarizations , calculate the source green s matrix @xmath136 , and the response matrix @xmath108 through ( [ b13 ] ) . determine the amplitudes and phases that would be measured experimentally using ( [ b7 ] ) , reverse the process using ( [ b16 ] ) , and calculate the apparent dynamical modes using ( [ b17 ] ) . compare the actual modes with the recovered modes . information may be lost due to the sampling used , the introduction of experimental errors , and the complete loss of certain fringe visibilities due to noise . during the work we found it instructive to produce ` movies ' showing the time evolution of the actual and recovered modes . it should be stressed that in all of the simulations that follow , only the total power absorbed by the complete system was regarded as measurable . first , consider the resonant coupling of two parallel dipoles . figure [ fig2 ] shows the total power absorbed , in arbitrary units , as a function of frequency for two @xmath146-polarized dipoles , placed 0.1 mm apart on the @xmath147-axis , each having @xmath141 = 300 ghz , and @xmath142 = 20 ghz . the @xmath146-polarized source was located at @xmath147=@xmath148=10 mm to ensure that symmetric and antisymmetric modes could be excited . the lines correspond to different polarizabilities , thereby changing the degree of coupling , and the individual conductivities . the solid line shows the case when the polarizability is low , @xmath143 = 0.0005 mm@xmath144 . the dipoles respond independently , giving rise to a single spectral feature at @xmath149 ghz , having a width of @xmath142 = 20 ghz . because the conductivity is low , the absorbed power is small . the dotted line shows the case when the polarizability is increased to @xmath143 = 0.005 mm@xmath144 . the dipoles interact , and split into an antisymmetric mode at 240 ghz , having a width of @xmath150 , and a symmetric mode at 340 ghz , having a width of @xmath151 . the delocalized excitation is mediated mostly by near - field coupling , and the splitting is caused by a decrease ( increase ) in the effective spring constant of the antisymmetric ( symmetric ) mode as a consequence of the interaction . the absorbed power is larger than before because of the increased conductivity . the dashed line shows the case when the polarizability is high , @xmath143 = 0.01 mm@xmath144 , and the splitting large . the relative strength of the peaks depends on the position of the source , indicating that spectral information alone can be misleading . although not shown , when the polarizability is increased to @xmath143 = 0.05 mm@xmath144 , the ` high - conductivity ' limit is approached , and the absorbed power falls to a very low level as the dipoles screen each other : the scattered field at each dipole is equal in magnitude and opposite in sign to the incident field . total power absorbed ( arbitrary units ) as a single source is swept in the @xmath147-direction , at a distance @xmath152 mm from two parallel dipoles , 0.1 mm apart , each having @xmath141 = 300 ghz , @xmath142 = 20 ghz , and @xmath143 = 0.005 mm@xmath144 . the source frequencies were ( dashed ) 240 ghz , ( solid ) 300 ghz , and ( dotted ) 340 ghz . ] figure [ fig3 ] shows the total power absorbed as a single source ( using the near - field green s dyadic ) is swept in the @xmath147-direction , at a distance @xmath152 mm from the dipoles , with @xmath143 = 0.005 mm@xmath144 . according to fig . [ fig2 ] , the anti - symmetric mode peaks at 240 ghz , whereas the symmetric mode peaks at 340ghz . the plots in fig . [ fig3 ] correspond to source frequencies of ( dashed ) 240 ghz , ( solid ) 300 ghz , and ( dotted ) 340 ghz . although the dipoles are not resolved spatially at this scan distance , the anti - symmetric mode has a null on - axis , whereas the symmetric mode has a peak on axis . from these absorption patterns one could infer the nature of the modes present at a single , intermediate frequency , say 300 ghz , which is the solid line in fig . [ fig3 ] , but what happens if wish to determine the modal content directly using only a single - frequency source . to this end , two sources were held at fixed positions ( source 1 at @xmath153 0 mm , @xmath154 5 mm , and source 2 at @xmath153 5 mm , @xmath154 0 mm ) , and the differential phase rotated . the fringe visibility was recovered from the simulation , and used to assemble the @xmath108 matrix , which was then converted to the @xmath79 matrix by ` deconvolving ' the beam patterns of the sources , and then diagonalized to find the natural modes and their responsivities . the simulated experiment was repeated for the polarizabilities and frequencies used in figs . [ fig2 ] and [ fig3 ] . in all cases , the spatial dynamical forms and responsivities of the anti - symmetric and symmetric modes recovered were precisely the same as those calculated directly . the same results were achieved regardless of whether the source probes were described by the near - field , far - field , or full green s dyadics . at 240 ghz and @xmath1550.005 mm@xmath144 , the technique revealed a dominant anti - symmetric mode , corresponding to the peak in fig . [ fig2 ] . at 300 ghz , the modes were present in roughly equal amounts , whereas at 340 ghz , the symmetric mode was dominant . crucially , the forms and relative responsivities of the two modes present at 300 ghz could be determined using a simulated interferometric measurement , even though the local absorption spectrum does not indicate the presence of two modes . in the case where @xmath1550.00005 mm@xmath144 , and over a wide range of frequencies , eai correctly revealed the existence of two degenerate modes , corresponding to the oscillation of each dipole independently . similar simulations were carried out for different source types and locations , and the dynamical forms and responsivities of the two modes could always be recovered . this simple example illustrates the nature of eai , but care is required because the method is based on the ability to ` deconvolve ' the beam patterns of the sources , and so the accuracy of the final result will depend on the precision with which the experiment can be carried out . total power absorbed ( arbitrary units ) as a function of frequency for a variety of ring - like configurations : ( solid ) 3 dipoles having @xmath143 = 0.005 mm@xmath144 placed 0.1 mm apart at the corners of a triangle ; ( dotted ) 4 dipoles having @xmath143 = 0.003 mm@xmath144 placed 0.1 mm apart at the corners of a square ; ( dashed ) 8 dipoles having @xmath143 = 0.0004 mm@xmath144 placed 0.0383 mm apart at the corners of an octagon . the power for the 8-dipole case was scaled by a factor of 10 before plotting . ] next , a series of simulations were performed on ring - like configurations of parallel dipoles . by using rings , boundary conditions at the ends are eliminated , and dispersion relationships can be found . figure [ fig11 ] shows the total absorbed power as a function of frequency for the following cases : ( solid ) 3 dipoles having @xmath143 = 0.005 mm@xmath144 placed 0.1 mm apart at the corners of a triangle ; ( dotted ) 4 dipoles having @xmath143 = 0.003 mm@xmath144 placed 0.1 mm apart at the corners of a square ; ( dashed ) 8 dipoles having @xmath143 = 0.0004 mm@xmath144 placed 0.0383 mm apart on the corners of an octagon . the polarizabilities were chosen to give similar spectral splittings , but the power for the 8-dipole case was scaled by a factor of 10 before plotting . the source was positioned at @xmath15310 mm , @xmath1540 mm , and operated in near - field mode . the absorbed power revealed a low - level modulation as the source was rotated around the center at fixed distance , but the polar patterns are not shown for brevity . in all cases , the spectra acquired 2 peaks as a consequence of resonant interaction . interactions take place not only around the ring , but also across the ring . in the case of 3 dipoles , the peak at 378 ghz has a single dominant eigenvalue , corresponding to all dipoles oscillating in phase with equal amplitude . the peak at 240 ghz , is two - fold degenerate and accounts for the symmetric and antisymmetric motion of the 2 remaining degrees of freedom . in the case of 4 dipoles , the largest eigenvalue of the peak at 357 ghz corresponds to the collective delocalized in - phase oscillation of all dipoles , whereas the peak at 289 ghz is 2-fold degenerate and accounts for opposite pairs being out of phase . the system of 8 dipoles produces 2 peaks , with the one at 454 ghz being the in - phase collective oscillation of the whole system . the one at 390 ghz is two - fold degenerate , and corresponds to sinusoidal and cosinusoidal standing waves on the ring . even by comparing the full spectra in fig . [ fig11 ] , it is not possible to gain much information about the nature of the underlying modes . moreover , the relative strengths of the peaks depends on the position of the source . if one is only interested in , and has access to , a limited frequency range , then there is little information in the polar power patterns that helps understand the number and forms of the modes . eai can , however , reveal this information . to simulate eai , two probes were rotated around the center of the system at a fixed distance of 5 mm , and the complex fringe visibility calculated for certain source pairs . [ fig12 ] shows the continuous forms of typical visibilities . remember that these are not spatial fringe patterns , but the complex visibility as a function of angle for a particular reference angle . the various lines are for the ( solid ) triangular , ( dotted ) square , and ( dashed ) octagonal arrangements describe previously . the reference angle was 90@xmath156 in all cases , corresponding to a vertex point of the triangular and octagonal arrangements , and a face of the square arrangement . figure [ fig12 ] shows that the peak visibility is reduced when the sources are on opposite sides due to the ` leading ' edge screening the ` trailing ' edge . to recover the modes , a small number of sample points was used in each case ; typically one more than the number of dipoles , avoiding the obvious degeneracies . it was found that the forms of the modes and their relative responsivities could be recovered easily at randomly chosen frequencies , without much care in selecting the source positions . also , the recovery process depended little on whether the sources were described by the near - field , far - field , or full green s dyadics . of course , only the full green s dyadic describes the physical reality , but the tolerance to the precise form of the illumination pattern indicates that , in principle , it should be possible to carry out interferometry using a variety of different near - field and far - field probes . ( a ) amplitude and ( b ) phase of the continuous fringe visibility of ( solid ) the triangular system at 240 ghz , ( dotted ) the square system at 289 ghz , and ( dashed ) the octagonal system at 390 ghz . the sources were swept around in a circle 5 mm from the centre . the reference angle was 90@xmath156 . ] the octagonal system was considered again , but with the resonant frequency of one dipole increased to 400 ghz , corresponding to a reduction in the mass of the oscillating charge . although not shown here , the principle effect on the spectrum is to split the original low - frequency line at 390 ghz into two separate frequencies , 387 ghz and 413 ghz , reflecting the breaking of the degeneracy of the sinusoidal and cosinusoidal modes . there is also a quasi - static mode at 290 ghz . the dominant modal forms are shown in fig . [ fig13 ] , at 413 ghz , as the reduced mass is moved around the octagon . in each case , the partner mode at 387 ghz has a standing wave that leaves the dipole with the reduced mass stationary . the corresponding visibility functions are shown in fig . [ fig14 ] , where the reference angle was 90@xmath156 . the visibility plots , particularly the phase , clearly show the movement of the defect dipole . ( color online ) modal forms of the 413 ghz resonance of octagonal systems of dipoles with the resonant frequency of one dipole increased to 400 ghz . the reduced - mass dipole was moved around the chain ( solid ) none , ( dashed ) position 1 , ( dotted ) position 2 , ( dot - dashed ) position 3 . ] ( a ) amplitude and ( b ) phase of the recorded fringe visibility at 413 ghz of octagonal systems , with the resonant frequency of one dipole changed to 400 ghz . the lines correspond to ( solid ) none , ( dashed ) position 1 ( corresponding to the 90@xmath156 reference angle ) , ( dotted ) position 2 , ( dot - dashed ) position 3 . ] a number of discrete source positions where then used in a attempt to detect the movement of the defect dipole around the octagonal structure . the sample points were placed on a circle , 5 mm from the center , 20@xmath156 apart , with the first one opposite a vertex . the modes and their relative responsivities could be recovered precisely regardless of whether the near - field or far - field source green s dyadic was used . the forms of most high - order modes could be recovered precisely with 8 or greater sample points , with some low - level of aliasing with 5 to 7 sample points , and not at all with 4 sample points . the determination of the position of the defect dipole , depended little on the positions of the sample points . overall , the modes could be recovered in detail despite the partial screening of the ` dark side ' of the ring . total power absorbed ( arbitrary units ) as a function of frequency for a linear chain of 5 parallel dipoles , 0.1 mm apart , each having @xmath141 = 300 ghz , and @xmath142 = 20 ghz . the different lines correspond to polarizabilities , ( solid ) @xmath143 = 0.0005 mm@xmath144 , ( dotted ) @xmath143 = 0.005 mm@xmath144 , ( dashed ) @xmath143 = 0.01 mm@xmath144 . ] ( color online ) modal forms of the resonances at ( solid ) 239 ghz , ( dashed ) 293 ghz , and ( dotted ) 332 ghz in fig . all dipoles are either perfectly in phase or out of phase . the lines are drawn to guide the eye only . ] turn to linear arrays . figure [ fig4 ] shows the total power absorbed , in arbitrary units , as a function of frequency for 5 , @xmath146-polarized dipoles , placed 0.1 mm apart on the @xmath147 axis , each having @xmath141 = 300 ghz , and @xmath142 = 20 ghz . the @xmath146-polarized source was located at @xmath147=10 mm , @xmath148=10 mm to ensure that symmetric and antisymmetric modes could be excited . the solid line shows the case when the polarizability is low , @xmath143 = 0.0005 mm@xmath144 , and the dipoles are largely , but not completely , decoupled . the dotted line shows the case when the polarizability is increased to @xmath143 = 0.005 mm@xmath144 . the dipoles interact , and give rise to 5 spectral features at 187 , 239 , 293 , 332 , and 381 ghz ; these alternate between symmetric and antisymmetric forms . it is not necessarily the case that the number of features is equal to or smaller than the number of dipoles , because for large systems spatially harmonic response due to widely separated dipoles may exist . the dashed line shows the case when the polarizability is increased further , @xmath143 = 0.01 mm@xmath144 . the response becomes very similar to that of fig . [ fig2 ] , but now both modes are antisymmetric whereas before one was symmetric and the other antisymmetric . figure [ fig5 ] shows the forms of the dominating modes , for @xmath143 = 0.005 mm@xmath144 , at the resonance frequencies of ( solid ) 239 ghz , ( dashed ) 293 ghz , and ( dotted ) 332 ghz . each filled circle shows the position of the displaced charge at @xmath157 : @xmath158 . dipoles having the same sign in the figure remain perfectly in phase as time progresses , whereas those having opposite sign remain perfectly out of phase . those at zero remain at zero . the lines are drawn to guide the eye only . table [ tabb ] gives the elements of the @xmath108 matrix recovered from the fringes when the 5-dipole structure , centred on @xmath1530.0 mm , was illuminated by 2 @xmath146-polarized sources at 280ghz , which according to the dotted line of fig . [ fig4 ] means that the sample was not resonant in any of its natural modes . a total of 5 source positions were used , @xmath1531 , 2 , 3 , 4 , 5 mm , with @xmath154 5 mm in all cases . this set of data was then corrected for the beam patterns , and the dynamical modes calculated . the eigenvalues of the recovered modes are given in the next to last line of table [ tabb ] . the forms of the recovered modes were then reconstructed and compared with the modes associated with the peaks in fig . [ fig4 ] , 3 of which are shown in fig . the corresponding identifications are listed on the last line of table [ tabb ] , which are reasonable given the source frequency , and the frequencies of the resonances . the positions of the test sources were moved randomly , and it was always found that 5 source positions were needed to recover all of the dynamical modes accurately . if more than 5 were used , the recovered modes and eigenvalues did not change , as would be expected for a well - conditioned problem . the conclusion is that a simple interferometric measurement , at a single frequency , based on measuring the total power absorbed , has enabled the forms and responsivities of the dynamical modes to be found . .[tabb ] fringe parameters of the 5 dipole structure described in the text . two @xmath146-polarized , 280 ghz , sources were placed in pairs at the positions @xmath1531.0 , 2.0 , 3.0 , 4.0 , and 5.0 mm , with @xmath154 5.0 mm in all cases . the top block of data gives the amplitudes of the elements of the @xmath108 matrix , and the middle block gives the corresponding phases . all values were recovered from the ` measured ' fringes . the penultimate line gives the eigenvalues of the natural modes at the dipoles , and the bottom line gives the frequency ( ghz ) at which the corresponding mode is resonant . [ cols="^,^,^,^,^ " , ] total power absorbed ( arbitrary units ) as a function of frequency for 11 parallel dipoles , 0.1 mm apart , each having @xmath141 = 300 ghz , and @xmath142 = 20 ghz . all of the dipoles except the central one had a polarizability of @xmath143 = 0.0005 mm@xmath144 , whereas the central one had ( solid ) @xmath143 = 0.001 mm@xmath144 , ( dotted ) @xmath143 = 0.003 mm@xmath144 , and ( dashed ) @xmath143 = 0.005 mm@xmath144 . ] figure [ fig6 ] shows the total power absorbed , in arbitrary units , as a function of frequency for 11 , @xmath146-polarized dipoles , placed 0.1 mm apart , each having @xmath141 = 300 ghz , and @xmath142 = 20 ghz . all of the dipoles except the central one had a polarizability of @xmath143 = 0.0005 mm@xmath144 , whereas the central one had ( solid ) @xmath143 = 0.001 mm@xmath144 , ( dotted ) @xmath143 = 0.003 mm@xmath144 , and ( dashed ) @xmath143 = 0.005 mm@xmath144 . the @xmath146-polarized source was located at @xmath147=0 and @xmath148=1 mm , and operated in the near - field mode . even when the polarizability is low , there is evidence of interaction , because the peak is at 310 ghz , not 300 ghz . as the polarizability of the central dipole is increased , the spectrum splits into strong features at 276 ghz and 323 ghz , indicating the excitation of two resonant modes . figure [ fig7 ] shows the form of the strongest mode at each of 276 , 300 , and 323 ghz . at 276 ghz , the highly polarized dipole induces a localized excitation , with the nearest neighbors 180@xmath159 out of phase with the central dipole . at 323 ghz , the mode is still localized , but now the nearest neighbors are in phase with the central , exciting dipole . there is no spectral feature at 300 ghz , and the dominant mode is less localized , with the nearest neighbors being stationary , and the next nearest neighbors being out of phase . the excitation is influenced by the boundary conditions at the ends of the chain . the modal forms change significantly over the 280 - 320 ghz range , and can take on forms where the central dipole launches traveling , albeit rapidly decaying , waves . beating can also be seen between different spatial frequencies . ( color online ) modal forms of the strongest absorptive modes at ( solid ) 276 ghz , ( dashed ) 323 ghz , and ( dotted ) 300 ghz of the dashed spectrum shown in fig . all dipoles are either perfectly in phase or perfectly out of phase , and the lines are drawn to guide the eye only . ] ( a ) amplitude and ( b ) phase of the recorded fringe visibility at frequencies of ( solid ) 276 ghz , ( dashed ) 323 ghz , and ( dotted ) 300 ghz . the sources were swept along a line , 1 mm from the chain of dipoles , and the visibilities are referenced to the central position . ] figure [ fig8 ] shows the amplitudes and phases of the continuous visibilities when two sources are swept along a parallel line 1 mm from the chain . the sources were operated in near - field mode , and the plots correspond to frequencies of ( solid ) 276 ghz , ( dashed ) 323 ghz , and ( dotted ) 300 ghz . the fringes give a clear indication of the changing form of the localized mode . a set of visibilities were recorded with only 11 source positions , at the same distance and spanning the spatial range shown in fig . it was found that at each frequency the forms of all of the dynamical modes and their responsivities could be recovered , regardless of whether the probes were operated in the near - field or far - field mode . as the total number of source positions was increased from 2 , the recovered modes were in error until 11 source positions were used , at which point all of the modal forms were found precisely . as the number of source positions increased above 11 , no change was seen , showing that convergence could be reached , and recognized . ( a ) amplitude and ( b ) phase of the recorded fringe visibility at 300 ghz . a highly polarizable dipole was placed at ( solid ) @xmath1530.0 mm , ( dashed ) @xmath1530.1 mm , ( dashed ) @xmath1530.2 mm . the sources were swept along a line , 1 mm from the chain of dipoles , and the visibilities are referenced to the central position . ] figure [ fig9 ] shows fringe visibilities at 300 ghz , when the position of the highly polarizable dipole was shifted one place at a time : ( solid ) @xmath1530.0 mm , ( dashed ) @xmath1530.1 mm , ( dashed ) @xmath1530.2 mm . the sources were swept along a line 1 mm from the chain of dipoles , and the visibilities referenced to the central position . the fringe is sensitive to the position of the localized mode . moreover , all of the modal forms could again be recovered with 11 source positions , allowing the position of the highly polarisable dipole to be located . ( color online ) modal forms of the strongest absorptive modes at ( solid ) 270 ghz , ( dashed ) 330 ghz , and ( dotted ) 284 ghz of a system comprising 11 dipoles . all of the dipoles had a polarizability of @xmath143 = 0.0005 mm@xmath144 , except the ones at @xmath153 - 0.2 mm and @xmath153 + 0.2 mm ( dipoles 4 and 8) , which had a polarizability of @xmath143 = 0.005 mm@xmath144 . ] more complicated linear systems were then considered . for example , fig . [ fig10 ] shows the modal forms of two interacting localized excitations . all of the dipoles had a polarizability of @xmath143 = 0.0005 mm@xmath144 , except the ones at @xmath153 - 0.2 mm and @xmath153 + 0.2 mm ( dipoles 4 and 8) , which had a polarizability of @xmath143 = 0.005 mm@xmath144 . the two highly polarized dipoles now oscillate either in phase or out of phase , with the interaction mediated by the background chain . again , we were able to recover all of the modes and responsivities at any frequency using eai . total power absorbed ( arbitrary units ) as a single probe is scanned along a line parallel to a linear chain of 21 dipoles placed 0.2 mm apart , having @xmath160300 ghz , @xmath16120 ghz , and @xmath1620.5 mm@xmath163 the scan distance was 1 mm . the lines correspond to frequencies of ( dotted ) 42 ghz , ( dashed ) 205 ghz , and ( solid ) 500 ghz : the powers have been divided by 5@xmath16410@xmath165 , 1@xmath16410@xmath166 , and 1@xmath16410@xmath167 respectively before plotting . ] finally consider a linear array of 21 dipoles , placed 0.2 mm apart , having @xmath160300ghz , @xmath16120 ghz , and @xmath1620.5 mm@xmath163 . the spectrum , not shown , has high - q resonances at 42 ghz and 205 ghz , each of which has a single dominant eigenvalue . figure [ fig16 ] shows the power absorbed as a single probe was swept parallel to the chain , 1 mm from it , thereby producing a near - field scan . the two sweeps at ( dotted ) 42 ghz and ( dashed ) 205 ghz , show strong evidence of end effects , whereas the sweep at ( solid ) 500 ghz , where no spectral features are present , has a form typical of a continuous nanowire or sheet of resistive material . figure [ fig17 ] shows the amplitude of the fringe visibility at ( top ) 42 ghz , ( middle ) 205 ghz , ( bottom ) 500 ghz . the reference positions are ( solid ) 0 mm , ( dashed ) -2.5 , and ( dotted ) -5.0 mm in all cases . the ends of the chain are at @xmath1682 mm . amplitude of the continuous fringe visibility for an array of 21 dipoles placed 0.2 mm apart , @xmath160300ghz , and @xmath16120 ghz . ( a ) 42 ghz , ( b ) 205 ghz , ( c ) 510 ghz . the reference positions are ( solid ) 0 mm , ( dashed ) -2.5 , and ( dotted ) -5.0 mm in all cases . the ends of the chain are at @xmath1682 mm . ] ( color online ) modal forms of the strongest absorptive modes at ( dotted ) 42 ghz , ( dashed ) 205 ghz , and ( solid ) 500 ghz of a linear chain of 21 dipoles having placed 0.2 mm apart , having @xmath160300ghz , @xmath16120 ghz , and @xmath1690.5 mm@xmath163 . the lines are drawn to guide the eye only . ] using pairs of dipoles at the same scan distance , we recovered the dynamical modes using eai , and compared them with those calculated directly . a limited number of test - source positions , typically 20 , were used , but no differences were seen when over 1000 were used . the behaviour at 42 ghz could always be recovered easily , and the behaviour at 205 ghz and 510 ghz could be recovered for linear scans @xmath1701 mm from the chain . at greater distances , recovery became uncertain , because of difficulty exciting dipoles near the ends . by always placing 2 sources in line with the dipoles , but beyond the ends , the problem could be solved . the situation is similar to edge absorption in continuous films . figure [ fig18 ] shows the recovered modes in those cases where sufficient source positions were used to recover the modes accurately . the mode at ( dotted ) 42 ghz is a quasi - static solution , dominated by near - field coupling ; where half of the dipoles are stationary , and the other half oscillate symmetrically , in perfect antiphase , about the centre of the chain . the fringe visibility , plotted in the top pane of fig . [ fig17 ] , shows directly that dipole interaction is strong enough to produce highly coherent extended behaviour along the whole length of the structure . near - field interactions on linear chains are being considered for possible application as sub - wavelength waveguides . the mode at ( dashed ) 205 ghz is mediated by far - field retarded interaction . the dipoles at the ends are lightly bound , because they only have neighbors on one side , and oscillate with high amplitude ; they form waves that travel inwards and interfere to produce a standing wave at the centre . the visibility plots shown in the middle pane of fig . [ fig17 ] reveal the existence of this large - scale collective behaviour . the bottom pane of fig . [ fig17 ] shows the visibility at 500 ghz , where no obvious spectral features are present . the two largest eigenmodes are degenerate , each corresponding to a high amplitude end wave that travels towards the centre : only one of them is shown in fig . [ fig18 ] ( solid ) . crucially , however , the effects at opposite ends are not coupled , as they are at 205 ghz . this behaviour is apparent in the visibility plots . near the centre , the system behaves as an almost incoherent absorber , whereas at the ends , highly coherent absorption is seen . the absorption process at one end is incoherent with respect to the absorption process at the other . compare the bottom pane of fig . [ fig17 ] with the middle and top panes . the various forms seen are typical of thin - film planar absorbers , and therefore the linear chain is becoming large enough to take on the characteristics of a continuous absorber , and macroscopic multipolar expansions can be used to describe behaviour @xcite . many other simulations were performed during the course of the work . for example , we looked at arrangements where a single , lightly coupled dipole was placed next to a ring - like or linear chain , and found that we were able to separate out the behaviour of the lightly bound dipole from the collective behaviour of the chain . also , we could identify the existence of ` grain - boundary - like ' defects on long chains . although this paper focuses on the theory of eai , it beneficial to comment on its practical application . the central requirements are ( i ) the ability to phase - lock two sources , ( ii ) the ability to scan the sources within a near - field or far - field volume , ( iii ) the ability to measure the total power absorbed , or at least the part of it that constitutes the operation of the device . because the key elements are so simple , the scheme can be used at any wavelength . crucially , high - power sources can be used , perhaps at low temperatures , extending the method well beyond anything that can be achieved by measuring the correlations in thermally radiated fields @xcite . it is certainly straightforward to assemble phased - locked sources for radio , infrared , and optical wavelengths . in fact , we have already demonstrated that it is possible to measure the state of coherence of the reception patterns of multimode power detectors over the range 195 - 270 ghz using absorbed power only @xcite . in that work , the frequencies of the two sources were offset slightly to produce a time varying output that contained the fringe information in real time . in some cases , a single - frequency source would be used , say in the case of a laser and power splitter , and then @xmath171 path - length switching would be needed to determine the quadrature components of the fringe . there are many ways in which the sut could be illuminated with different basis sets to span the dynamical modes of interest . for example , far - field probes in the form of optical fibres , or near - field probes in the form of coaxial afm - like tips , could be scanned across or rotated around the sut . in magnetic measurements , coils could be swept around and scanned across the surface of interest . in general , polarisation switching is needed , but this can be done in a straightforward manner . a crucial question is ` how many samples must be taken to gain access to the degrees of freedom of interest ? ' we have already identified efficient real - time techniques for adding in more and more sample points until convergence is reached , and the significant degrees of freedom identified . eai is , in general terms , quite similar to aperture synthesis interferometry , used extensively in astronomy , and many of the data processing techniques developed for astronomical imaging , such as closure phase , could be incorporated here . the final requirement is a way of determining the total power absorbed . in many applications , such as photovoltaics and optical sensors , the output of the device is already a direct measure of the power being absorbed through the degrees of freedom of interest . in these cases , one might not even be interested in the surface physics per se , but a detailed characterisation of the nature , number , and responsivities of the electromagnetic modes to which the device is sensitive . additionally , representative models based on dda can be used to understand how the electromagnetic modes relate to the dynamical behaviour of the absorbing surface . in some cases , such as patterned films or biomolecules on surfaces , a way of monitoring the total power absorbed is required . our own preferred approach is to attach the sut to a micromachined sin membrane , which also supports a thin - film temperature sensor . such devices have been developed extensively for astronomical detectors , and when used at low temperatures can achieve extreme sensitivity @xcite . our detectors @xcite can detect absorbed powers of much less than 1 fw with noise levels of 1x10@xmath172 whz@xmath173 , making extreme thermometry of thin - film structures entirely possible . we have shown that by illuminating an energy - absorbing surface or structure with a pair of phase - locked sources , it is possible to measure the state of coherence of the electromagnetic field to which the structure is sensitive . the tensor that describes the coherence can be diagonalized to find the amplitude , phase , and polarization patterns of the individual electromagnetic modes through which the structure can absorb power . in some cases , such as photovoltaics and detectors , this information is already sufficient to understand and characterize behaviour . it is also possible , however , to map the electromagnetic modes back onto the surface , or object , to gain representations of the dynamical modes responsible for absorption . these modes are intimately related to the existence of coherent phenomena , such as elastic waves and surface plasmons . in a continuous formulation of the method , not yet published , we have shown that it is possible to detect the presence of subsurface voids . the method can be implemented easily , and because high - power sources are used , can far outperform any experiment that relies on measuring the correlations in themally radiated fields . more speculatively , the use of high power sources , together with a space - frequency formulation , may open the way to studying non - linear surface effects . this work was carried out while prof . withington was a visiting fellow at all souls college , oxford . withington would like to thank the warden and fellows for their generosity and hospitality during his stay .

we present photometry at 324 for all known members of the upper scorpius association ( myr ) based on all images of these objects obtained with the _ spitzer space telescope _ and the _ wide - field infrared survey explorer_. we have used these data to identify the members that exhibit excess emission from circumstellar disks and estimate the evolutionary stages of these disks . through this analysis , we have found new candidates for transitional , evolved , and debris disks . the fraction of members harboring inner primordial disks is% for b g stars ( ) and increases with later types to a value of% atm5 ( ) , in agreement with the results of previous disk surveys of smaller samples of upper sco members . these data indicate that the lifetimes of disks are longer at lower stellar masses , and that a significant fraction of disks of low - mass stars survive for at least myr . finally , we demonstrate that the distribution of excess sizes in upper sco and the much younger taurus star - forming region ( myr ) are consistent with the same , brief timescale for clearing of inner disks .

1209.5433

identifying the stars that harbor circumstellar disks within young clusters and associations is an essential step in using disks to study star and planet formation . for instance , thorough surveys for disks provide well - defined samples of targets for detailed measurements of disk properties . in addition , by comparing the prevalence of disks in different evolutionary stages within a given cluster and among clusters across a range of ages , one can estimate the lifetimes and clearing timescales of disks , which in turn constrain models of disk evolution and planet formation . the best available tool for disk surveys has been mid - infrared ( ir ) imaging with the _ spitzer space telescope _ _ spitzer _ has offered the sensitivity to detect the photospheres of young brown dwarfs in nearby clusters ( @xmath10 pc ) , the field of view to efficiently map these clusters ( @xmath11 ) , the angular resolution to resolve most of their members ( @xmath12 ) , and a suite of mid - ir filters that could roughly classify the evolutionary stages of disks ( 3.624 ) . although _ spitzer _ has focused on relatively compact clusters ( @xmath13 deg@xmath14 , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , it has also imaged more widely distributed populations in areas as large as @xmath1 deg@xmath14 @xcite . the upper scorpius subgroup in the scorpius - centaurus ob association is an attractive target for disk surveys . it contains one of the nearest and richest populations of young stars ( @xmath15165 pc , @xmath16 , * ? ? ? * ) , making it possible to measure disk properties down to low stellar masses with good statistical accuracy . perhaps most importantly , upper sco is at an age where both primordial disks and second - generation debris disks are abundant @xcite , and thus can potentially provide vital constraints on disk evolution . most studies have adopted a mean age of 5 myr for upper sco @xcite , but @xcite recently derived a significantly older value of 11 myr , which affects the timescales of disk evolution derived from any census of disks in the region . the ages of upper sco and its neighboring subgroups remain a subject of debate @xcite . _ spitzer _ has been used to identify and classify disks in upper sco based on mid - ir photometry @xcite and to study the mineralogy of dust in those disks with mid - ir spectroscopy @xcite . in one of the most widely - cited results from the _ spitzer _ data , @xcite found that , in a sample of @xmath17 members , nearly all of the stars more massive than the sun have lost their primordial disks while a significant fraction of the low - mass stars ( @xmath181/5 ) have retained them , indicating that disk lifetimes may be longer at lower stellar masses . however , because upper sco is spread across a large area ( @xmath19 deg@xmath14 ) , _ spitzer _ imaged members individually rather than through a map of the association , and only a subset of the stellar population was observed . as a result , the potential of the association for constraining the properties of disks was not fully exploited . the _ wide - field infrared survey explorer _ ( _ wise _ , * ? ? ? * ) has presented an opportunity to significantly improve the census of disks in upper sco . in 2010 , the satellite imaged the entire sky at 3.4 , 4.6 , 12 , and 22 , providing mid - ir photometry for the areas of upper sco that were not observed by _ spitzer_. although they have lower sensitivity and angular resolution than _ spitzer _ images , the data do detect every known member of upper sco ( that is resolved ) in the most sensitive bands , including the brown dwarfs . photometry has already been used to search for disks around members of upper sco at b , a , and f types @xcite and mid- to late - m types @xcite . to obtain the most thorough census of disks in upper sco that is possible with available data , we have compiled all photometry between 324 from both _ spitzer _ and for all known members of the association ( section [ sec : data ] ) . we use these data to identify the members that exhibit significant mid - ir excesses ( section [ sec : exc ] ) and to estimate the evolutionary stages of the detected disks ( section [ sec : class ] ) . we then measure the distribution of excesses as a function of spectral type , wavelength , disk class , and excess size and discuss the resulting implications for disk evolution ( section [ sec : global ] ) . to characterize the population of disks in upper sco , we began by constructing a list of all published sources that are likely to be members of the association based on proper motions or diagnostics of youth . we also have included new members that we have uncovered through an ongoing survey of the association . the basis for inclusion in our membership catalog will be described in a separate study ( k. luhman , in preparation ) . we have treated known binaries as single sources if they are unresolved by both the two micron all - sky survey ( 2mass , * ? ? ? * ) and the united kingdom infrared telescope ( ukirt ) infrared deep sky survey ( ukidss , * ? ? ? thus , secondaries that have been resolved only by high - resolution imaging with the _ hubble space telescope _ or adaptive optics do not receive separate entries in our catalog of members @xcite . our resulting list of known members of upper sco contains 863 objects . we present their names and spectral types in table [ tab : all ] . our compilation of spectral classifications is not fully exhaustive ; additional measurements have been reported in the literature for some of the stars , particularly at earlier types , but they generally agree with the types in table [ tab : all ] . for our census of disks in upper sco , we make use of images at 3.6 , 4.5 , 5.8 , and 8.0 obtained with _ spitzer _ s infrared array camera ( irac ; * ? ? * ) and images at 24 obtained with the multiband imaging photometer for _ spitzer _ ( mips ; * ? ? ? these bands are denoted as [ 3.6 ] , [ 4.5 ] , [ 5.8 ] , [ 8.0 ] , and [ 24 ] . the fields of view were @xmath20 and @xmath21 for irac and the 24 band of mips , respectively . the cameras produced images with fwhm@xmath22@xmath23 from 3.6 to 8.0 and fwhm=@xmath24 at 24 . several previous studies have reported _ spitzer _ photometry for samples of members of upper sco @xcite . in this work , we consider all irac and mips 24 observations of all members in our catalog . these data were obtained through guaranteed time programs 19 ( j. houck ) , 40 , 58 ( g. rieke ) , 72 ( f. low ) , and 84 ( m. jura ) , director s discretionary time program 248 ( s. mohanty ) , legacy programs 139 , 173 , 177 ( n. evans ) , 148 ( m. meyer ) , and 30574 ( l. allen ) , and general observer programs 20069 ( j. carpenter ) , 20103 ( l. hillenbrand ) , 20146 ( k. gordon ) , 20435 ( r. jayawardhana ) , 20803 ( s. mohanty ) , 30765 ( k. stapelfeldt ) , 50025 ( p. harvey ) , 50554 ( i. song ) , and 80187 ( k. luhman ) . we have measured photometry from these images for the members of upper sco with the methods described by @xcite . we present our measurements of _ spitzer _ photometry for upper sco in table [ tab : all ] . some members were observed at multiple epochs , most of which were separated by a few days . since none of these objects exhibit significant variability in these data , we report the mean of multiple measurements weighted by the inverse square of their flux errors . for each of the irac bands , the camera detected every member that appeared within the field of view and was resolved . this was not the case for the 24 images from mips . all of the non - detections in this band are indicated in table [ tab : all ] . the typical detection limit is @xmath25\sim10 $ ] . we have not measured 24 photometry for hip 80425 and hip 80338 since they are dominated by extended emission . 2mass j16105240 - 1937344 also may be significantly contaminated by extended emission at 24 . we report a tentative measurement for this star , but it is excluded from the analysis in this paper . @xcite noted that their 24 data for several additional stars were unreliable because of extended emission . based on visual inspection of the images , we find that these stars are point sources that are sufficiently bright compared to spatial variations in the background emission so that our aperture photometry should be reasonably accurate . this conclusion is supported by the fact that our 24 data for these stars agree with the fluxes expected from the photospheres . irac and mips observed 381 and 442 members of upper sco , respectively . a total of 484 members were observed by at least one of the cameras , corresponding to 56% of the known members . irac and mips data have not been previously published for 117 and 124 of these sources , respectively , which includes 48 non - detections by mips . in addition to the pointed _ spitzer _ observations , we utilize mid - ir photometry from the all - sky imaging survey by _ wise_. the bands of are centered at 3.4 , 4.6 , 12 , and 22 , which are denoted as w1 through w4 @xcite . the angular resolution of the images is @xmath26 in the first three bands and @xmath27 in w4 . we have retrieved the coordinates and profile - fit photometry of all objects in the preliminary release source catalog and the all - sky source catalog that are within a distance of @xmath28 from members of upper sco . for w1 and w2 , most of the data agree within 5% between the two catalogs , and the catalogs exhibit comparable agreement with the _ spitzer _ data . the width of the sequence of diskless stars in @xmath29 and @xmath30 as a function of spectral type is also similar for the two catalogs ( section [ sec : exc ] ) . we adopt the w1 and w2 data from the all - sky source catalog since it was generated from a newer processing of the images . however , we find that the sequences of diskless stars in @xmath31 and @xmath32 are noticeably tighter with the data from the preliminary release source catalog for the fainter stars , which implies better photometric accuracies in this catalog for w3 and w4 . this result is supported by a comparison of w4 to our measurements of [ 24 ] ; the standard deviation of @xmath33 $ ] is 0.18 and 0.28 mag using the preliminary release and all - sky data , respectively . therefore , we adopt the w3 and w4 measurements from the preliminary release source catalog , with the exception of 2mass j16045716 - 2104160 , 2mass j16133647 - 2327353 , and 2mass j16133688 - 2327298 , for which measurements are available from only the all - sky source catalog . on average , the data from the preliminary release source catalog for upper sco are 0.03 mag brighter in w3 and 0.035 mag fainter in w4 than the data from the all - sky source catalog . we have modified our catalog of data in several ways . since the w2 fluxes for saturated stars are significantly overestimated ( 0.21 mag at w2@xmath34 , * ? ? ? * ) , we have omitted these measurements from our tabulation . these empty entries in the catalog are labeled with the flag err " . biases of this kind are also present in the other bands for the brightest stars , but they are much smaller ( @xmath35 mag ) , so we retain these measurements . although the all - sky source catalog contains data for antares , we label it as saturated in our catalog since it is too bright for reliable photometry . through visual inspection of the images , we have checked for members of upper sco whose photometry may be affected by the psf of another star . each band in which this occurs is labeled with the flag bl " ( for blend " ) . we also have inspected the images to check for unreliable detections and measurements that may be contaminated by extended emission . for the former , we have omitted the photometry and have added the flag false " to our catalog . the photometry is labeled with the flag ext " if extended emission is suspected based on the visual inspection . if measurements are known to be dominated by extended emission based on higher - resolution images from _ spitzer _ , they are omitted from the catalog . some studies have excluded data that have high values of the reduced @xmath36 for the profile fitting in the catalog because this may indicate the presence of extended emission @xcite . however , we find that most of the members of upper sco with high @xmath36 do not show evidence of extended emission in images from _ spitzer _ , when available . therefore , we rely on the steps that we have described rather than the @xmath36 flag for identifying extended emission . finally , for the source in the catalog matched to each member , we have visually checked whether its position differs significantly ( @xmath37 ) among the four bands . by doing so , we can identify erroneous matches in the catalog between one object that dominates at shorter wavelengths and a different object that dominates at longer wavelengths , as in the case of the components of a binary system in which only the secondary has a disk ( e.g. , 2mass j16101888 - 2502325 and 2mass j16101918 - 2502301 ) and stars and red galaxies with small projected separations ( e.g. , sco pms 17 , * ? ? ? we have examined the images of all members that lack counterparts in the catalog . several of the unmatched members are not detected because of their proximity to brighter stars . we label their empty entries with the flag unres " . among the pairs separated by less than @xmath38 ( the fwhm for the shorter bands ) , we have attached bin " ( for binary " ) to the photometry of the brighter components to indicate a possible contribution from the fainter stars . as noted in section [ sec : mem ] , many known close companions are not included in our catalog of members . thus , they represent additional unresolved components in the data that are not marked in our catalog . finally , through visual inspection of the images , we find that did detect four members that are absent from the source catalogs . we found data for two of these stars , 2mass j16071750 - 1820348 and 2mass j16001730 - 2236504 , in the all - sky reject table . we measured photometry for the remaining stars , 2mass j16090407 - 2417588 and 2mass j16081758 - 2348508 , by applying aperture photometry to the images . we present our catalog of photometry for the known members of upper sco in table [ tab : all ] . a dusty circumstellar disk within a few au of a star is brightest at ir wavelengths , becoming brighter than the star beyond @xmath39 . as a result , one can detect the presence of a disk based on ir emission that is greater than that expected from a stellar photosphere . we have applied this technique to the _ spitzer _ and photometry in upper sco to identify the members that have disks . to detect and measure excess emission for each member , we have used the colors produced by the _ spitzer _ and bands relative to @xmath40 ( 2.2 ) . we use @xmath40 for these colors because it is long enough in wavelength so that extinction is low while short enough in wavelength that it is usually dominated by the stellar photosphere in systems with disks . for most stars , we adopt @xmath41 from the 2mass point source catalog or @xmath40 from the seventh data release of ukidss , whose photometric systems are similar @xcite . we use @xmath40 photometry from @xcite and @xcite for a few of the brightest members since their 2mass data have large uncertainties . we focus on the _ spitzer _ bands [ 4.5 ] , [ 8.0 ] , and [ 24 ] and the bands w2 , w3 , and w4 for detecting excess emission . we do not consider [ 3.6 ] and w1 since they should show the smallest excesses among the _ spitzer _ and bands . the data at [ 5.8 ] is discussed only briefly since it is available for a relatively small number of members . antares is excluded from our analysis since it is a red supergiant and it lacks data from _ spitzer _ or _ wise_. the color excess of a star is measured by comparing its observed color to the color expected for its stellar photosphere . because of its richness and moderately advanced age , upper sco contains a large number of diskless stars , providing an accurate determination of the colors of young photospheres . since photospheric colors can vary with spectral type , we plot the ir colors in upper sco as a function of spectral type in figure [ fig : pan ] . the colors are not corrected for extinction , which is typically @xmath422 in this association . we have indicated in figure [ fig : pan ] the known and suspected be stars @xcite and the candidates for disks in different evolutionary classes that are identified in section [ sec : class ] and in previous studies @xcite . for each color in figure [ fig : pan ] , the data exhibit a narrow sequence of blue sources and a broader distribution of redder sources , corresponding to stellar photospheres and stars with disks , respectively . we discuss the measurement of excesses from these colors in the remainder of this section . the results of this analysis are summarized in table [ tab : all ] , where we indicate whether each member of upper sco has excess emission in each of the six _ spitzer _ and bands that we have considered . we examine the data in [ 4.5 ] and w2 together since these bands have similar effective wavelengths ( 4.5 and 4.6 ) . photometry in [ 4.5 ] and w2 is available for 359 and 817 members of upper sco , respectively . the stars that lack [ 4.5 ] were not observed by irac while the stars that lack w2 are too bright for good photometry ( section [ sec : wise ] ) or are blended with other sources . thus , all members were detected in both bands as long as they were observed ( by irac ) and resolved . 26 members have [ 4.5 ] but not w2 , which consist of stars that are too bright for a reliable w2 measurement and companions that are resolved by irac but not _ wise_. as a result , 843 members have data in at least one of the two bands . in the diagrams of @xmath43 $ ] and @xmath30 versus spectral type in figure [ fig : pan ] , we have used the sequences of diskless stars to select boundaries for identifying stars that exhibit significant color excesses . we use the same boundary for both colors since the data in [ 4.5 ] and w2 have an average offset of less than a few percent . however , the photospheric sequences are not well - defined at the latest spectral types because of the small number of members . the bluest l - type members have colors near 1.2 , which is redder than the photospheric values of @xmath43 $ ] estimated by @xcite based on young l dwarfs in the solar neighborhood ( @xmath44 - 100 myr ) . to better define the photospheric colors for l - type members of upper sco , we include in figure [ fig : pan ] data for members of taurus that are later than m9 and that are diskless based on photometry at longer wavelengths @xcite . these data coincide with the bluest sources in upper sco and confirm the rapid increase in @xmath43 $ ] and @xmath30 beyond m9 . the boundaries for @xmath43 $ ] and @xmath30 are defined by lines connecting ( b0 , 0.19 ) , ( k0 , 0.22 ) , ( m1.5 , 0.47 ) , and ( m8.5 , 0.87 ) . since the dependence on spectral type of these colors remains somewhat uncertain at m9l2 , we end the boundaries in figure [ fig : pan ] at m8.5 . measurements of excesses at the latest types are also sensitive to uncertainties in spectral types because of the rapid increase in the photospheric colors . nevertheless , we have inspected the data to check for members later than m8.5 that show obvious excesses . [ lhj2006 ] j163919.15 - 253409.9 has an excess in w2 but not [ 4.5 ] . this discrepancy is a reflection of the fact that w2 is brighter than [ 4.5 ] by 0.35 mag . a similar offset is present between w1 and [ 3.6 ] , indicating that the source is variable . [ lhj2007 ] j160918.69 - 222923.7 exhibits the same characteristics . mid - ir variability of this size is more likely to occur in stars with disks @xcite . therefore , we classify these two objects as having excesses in w2 . we tentatively conclude that the remaining members later than m8 do not have excesses in [ 4.5 ] and w2 . to refine our identifications of excesses , we have checked for sources that are redder than the boundaries for [ 4.5 ] or w2 in figure [ fig : pan ] but have data at longer wavelengths that are consistent with a stellar photosphere . all stars of this kind are only slightly above the excess thresholds in [ 4.5 ] and w2 . therefore , we indicate that they lack [ 4.5 ] and w2 excesses in table [ tab : all ] . these stars consist of 2mass j16072682 - 1855239 , 2mass j16104202 - 2101319 , 2mass j16133688 - 2327298 , 2mass j16152024 - 2333588 , 2mass j16203456 - 2430205 , 2mass j16204144 - 2425491 , 2mass j16370753 - 2432395 , sco pms 48 , hip 78977 , sco pms 42b , and hip 81455 . the latter two stars may have small excesses in [ 24 ] ( section [ sec : exc24 ] ) , but they lack excess emission in w3 . several stars with excesses in [ 4.5 ] or w2 do not have photometry at longer wavelengths , and thus can not be examined in this way , because they were below the detection limit ( w3 , w4 , [ 24 ] ) or were not observed ( [ 5.8 ] , [ 8.0 ] , [ 24 ] ) . we also have checked for members that have excesses in [ 4.5 ] but not w2 , or vice versa . two objects of this kind were discussed earlier in this section in the context of the late - type members . three additional stars show this discrepancy , consisting of 2mass j16090075 - 1908526 , 2mass j16095933 - 1800090 , and rx j1604.3 - 2130a . as in the earlier cases , variability is the likely cause of the differences between @xmath43 $ ] and @xmath30 for these stars given that other bands indicate the presence of disks . for rx j1604.3 - 2130a , previous ir data have exhibited significant variability as well @xcite . however , the variability in this star is rather unusual . its @xmath45-[8.0]$ ] color and its spectrum from @xcite shortward 16 are consistent with a stellar photosphere , but the bands over this same wavelength range ( w1 , w2 , w3 ) show excess emission relative to @xmath41 and each other , which may be caused by changes in the structure of the inner disk @xcite . photometry in [ 8.0 ] has been measured for 335 members of upper sco . 2mass j16075850 - 2039485 is excluded from the diagram of @xmath46 $ ] in figure [ fig : pan ] because its 8 flux is uncertain . as done for @xmath43 $ ] and @xmath30 , we have defined a threshold that varies with spectral type for identifying excesses in @xmath46 $ ] . it is defined by lines connecting ( b0 , 0.2 ) , ( k0 , 0.32 ) , ( m0 , 0.45 ) , and ( m9 , 1.01 ) . excesses from disks are larger at longer wavelengths , resulting in a greater separation between diskless and disk - bearing stars in @xmath46 $ ] compared to @xmath43 $ ] and @xmath30 . in all three of these colors , some of the early - type stars are 0.10.2 mag redder than the sequence of the bluest stars . most of these excesses can be attributed to extinction based on their lack of excesses at longer wavelengths and their red near - ir colors . however , a few of these moderately red stars may have small amounts of emission from debris disks @xcite , even though they appear below our adopted boundaries for excesses in figure [ fig : pan ] . finally , we have checked for discrepancies between apparent excesses in @xmath46 $ ] and data in other bands . 2mass j16210222 - 2358395 is slightly above our excess threshold for @xmath46 $ ] , but [ 8.0 ] does not show an excess relative to irac and wise data at shorter wavelengths . therefore , the excess in @xmath46 $ ] may be due to stellar variability between the observations at @xmath41 and the mid - ir bands . in w3 , w4 , and [ 24 ] , some members of upper sco are not detected or have large uncertainties in their photometry . for identifying excesses in these bands , we consider only measurements with uncertainties less than 0.25 mag . we also exclude stars that are flagged as extended . these criteria are satisfied in w3 by 658 members . our adopted threshold for excesses in @xmath31 is shown in figure [ fig : pan ] , which is defined by lines connecting ( b0 , 0.18 ) , ( g8 , 0.33 ) , and ( m9 , 1.52 ) . for stars that are slightly above the boundary for excesses , we have checked whether excess emission is present in w4 or [ 24 ] . hip 78977 lacks an excess in either band and 2mass j16233234 - 2523485 should have little if any excess in w4 based on its non - detection . therefore , we conclude that these stars probably do not have excesses in w3 . given the limits available at [ 24 ] and w4 , 2mass j16181618 - 2619080 , 2mass j16253672 - 2224285 , 2mass j16053077 - 2246200 , and 2mass j16044303 - 2318258 have colors of @xmath47 in @xmath32 and @xmath48 $ ] , indicating that they could have excess emission at w4 and [ 24 ] . since we can not rule out the presence of excesses at longer wavelengths , we tentatively classify these stars as having excesses in w3 . @xcite found that hip 79878 has excess emission in w3 , but it is slightly below our threshold for @xmath31 . @xcite reported an excess in w3 for 2mass j16185037 - 2424319 , but we rejected this measurement as a false detection . we discuss the bands [ 24 ] and w4 together since they have similar effective wavelengths ( 23.7 and 22 ) . as with w3 , we consider only data with uncertainties less than 0.25 mag and that are not contaminated by extended emission . the resulting sample contains 348 and 260 stars with photometry in [ 24 ] and w4 , respectively . data are available in at least one of the two bands for 428 members . among the _ spitzer _ and bands that we are considering , [ 24 ] and w4 are the only ones in which debris disks are likely to produce noticeable excess emission ( section [ sec : class ] ) . since these excesses can be arbitrarily small , we have defined thresholds in @xmath48 $ ] and @xmath32 that identify the smallest excesses that appear to be significant . the threshold adopted for @xmath32 is higher than that for @xmath48 $ ] since the photometric uncertainties are larger in w4 than in [ 24 ] , as illustrated by the larger width in the sequence of photospheric colors for @xmath32 in figure [ fig : pan ] . as a result , some stars that show small excesses in @xmath48 $ ] are below the threshold for @xmath32 . the thresholds are defined by ( b0 , 0.11 ) , ( k6 , 0.57 ) , and ( m9 , 1.4 ) for @xmath48 $ ] and ( b0 , 0.34 ) and ( m8 , 1.15 ) for @xmath32 . most of our identifications of excesses in [ 24 ] and w4 agree with those from previous studies @xcite , with the following exceptions . our [ 24 ] photometry for hip 81455 is 0.4 mag brighter than that measured by @xcite with the same data . as a result , we find an an excess in this band while @xcite did not . this star is partially resolved from another fainter star at a separation of @xmath49 , which was removed through psf subtraction prior to our aperture photometry . we speculate that the psf photometry of @xcite may have been affected by this second star . @xcite found that hip 78977 has an excess in @xmath48 $ ] , but it is slightly below our thresholds for both @xmath48 $ ] and @xmath32 . neither [ 24 ] nor w4 show an excess relative to w1 , w2 , and w3 while excesses do appear in @xmath30 and @xmath31 . these discrepant colors are likely caused by variability of the star between the near- and mid - ir observations , and we conclude that it does not have excesses in [ 24 ] and w4 . the following stars are above our [ 24 ] thresholds but slightly below those adopted by @xcite : hip 78702 , hip 78099 , hip 76633 , hip 79250 , and rx j1602.8 - 2401b . @xcite used w4 to assess the presence of excess emission for hip 78265 , hip 80019 , hip 80311 , and hip 82319 , but we have rejected these data because they are false detections or too uncertain . we have identified 22 new candidates in upper sco for weak excess emission in [ 24 ] or w4 , which may indicate the presence of debris disks or evolved transitional disks ( see section [ sec : class ] ) . six of these stars were found to lack excesses in previous studies , as described above . the remaining new candidates for weak excesses consist of hd 142506 , hip 80196 , 2mass j16020287 - 2236139 , 2mass j16050231 - 1941554 , 2mass j16052459 - 1954419 , 2mass j16061330 - 2212537 , 2mass j16080555 - 2218070 , 2mass j16094098 - 2217594 , 2mass j16103956 - 1916524 , 2mass j16104202 - 2101319 , 2mass j16114612 - 1907429 , 2mass j16124893 - 1800525 , 2mass j16230783 - 2300596 , rxj1614.6 - 1857 , sco pms 8b , and sco pms 42b . the detection of a w4 excess for hd 142506 is tentative since the photometry may be affected by blending with another star . similarly , 2mass j16235695 - 3312033 is above the threshold for an excess in @xmath32 , but it is blended with another star . because it lacks an excess in the higher - resolution images at [ 24 ] , we classify this star as lacking an excess in w4 . unlike the other _ spitzer _ bands that we consider , the [ 24 ] images did not detect all of the members of upper sco that they encompass . among the members observed in [ 24 ] , 94 stars were not detected or have errors large enough to be excluded from our analysis . three of these stars have early spectral types and lack data because they are projected against bright nebulosity or are dominated by extended emission . the remaining 91 objects have m types , most of which are later than m3 . a few of these stars have poor constraints on [ 24 ] because of bright nebulosity or the psf of another star , but most have limits of @xmath25\sim10 $ ] , indicating that these stars have colors of @xmath50 . the detection limit from w4 is bright enough ( @xmath51 ) that the resulting limits on @xmath32 do not place useful constraints on the presence of disks for most members of upper sco that are not detected in this band . multiple stages of evolution are usually present among the circumstellar disks in young stellar populations ( @xmath52 myr ) . the evolution of a disk is largely characterized by a change in its structure , which is reflected in its ir spectral energy distribution ( sed ) . as a result , one can use multi - band ir data like those we have compiled from _ spitzer _ and for upper sco to estimate the evolutionary stages of disks . in this section , we describe our adopted definitions for these stages and apply them to upper sco . a variety of names and definitions have been devised for the evolutionary stages of disks . we use the ones adopted by @xcite , which are as follows : a _ full disk _ is optically thick at ir wavelengths and has not experienced significant clearing as indicated by its sed . a _ pre - transitional disk _ and a _ transitional disk _ have gaps and holes in their dust distributions , respectively , that are large enough to noticeably affect their seds ( @xmath53 au , @xmath54 au , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? _ evolved disk _ is becoming optically thin and does not have a large gap or hole @xcite . evolved disks also have been named anemic , thin , weak , and homologously depleted @xcite . since we are examining a stellar population that includes relatively advanced stages of disk evolution , we also use the terms _ evolved transitional disk _ , which has optically thin primordial dust and a large inner hole @xcite , and _ debris disk _ , which consists of second - generation dust that is generated by collisions among planetesimals @xcite . all of these stages except debris disks are considered _ primordial disks_. this definition of a primordial disk differs from the one that we adopted in @xcite , where was equivalent to a full disk as defined here . to define our observational criteria for the different evolutionary stages of disks , we begin by describing the qualitative features of their seds . a full disk produces moderate - to - strong emission throughout the ir regime . because it has a large inner hole , a transitional disk is faint at shorter ir wavelengths and is much brighter at longer wavelengths . the emission from an evolved disk does not change abruptly with wavelength since it lacks a large hole or gap , and it is very weak since the disk is becoming optically thin . an evolved transitional disk and a debris disk have similar seds , showing only weak emission at long ir wavelengths . to develop quantitative criteria for classifying disks , one approach has been to identify distinct populations in the ir seds of young stars and associate them with different evolutionary stages of disks . populations of this kind were first found in taurus , where most stars have mid - ir colors that are either consistent with stellar photospheres , or are much redder . the few sources within the gap between these two groups were described as in transition " @xcite . similar gaps are present in the mid - ir colors of other star - forming regions as well ( * ? ? ? * references therein ) . in some studies , the red edge of the gap in colors like @xmath46 $ ] has been adopted as the boundary between full disks and pre - transitional disks on the redward side , and evolved and transitional disks on the blueward side @xcite . this threshold seems reasonable given that it roughly coincides with the colors produced by models of full disks with high degrees of dust settling @xcite . distinctive populations of colors and supporting modeling also have been used to guide the identification of debris disks and evolved transitional disks @xcite . a different approach to classifying disks has been employed by @xcite . they defined transitional disks to include disks with inner holes or gaps and disks that are optically thin , which correspond to transitional , pre - transitional , and evolved disks in our terminology . to identify disks of this kind , @xcite compared observed seds to the seds predicted by disk models . when millimeter measurements were available , they also required that a disk have an estimated mass less than 0.001 @xmath55 in order for it to be considered optically thin . after applying these procedures , @xcite concluded that some of the disks previously classified as settled , optically thick disks ( i.e. , redward of the gap in ir colors described above ) are instead optically thin , and hence qualify as transitional disks as defined in their study . because their definition of transitional disks was broader and their classification criteria encompassed redder disks , @xcite found a higher abundance of transitional disks in regions like taurus than some studies ( e.g. , * ? ? ? * ) . of course , a comparison of the abundances of transitional disks , and the resulting evolutionary timescales ( section [ sec : histo ] ) , is not meaningful when the adopted definitions of transitional disks differ . the classification scheme from @xcite has disadvantages . transitional disks are not defined in the traditional manner ( disks with large holes ) , which can lead to confusion when classifications from different studies are compared . the observational criteria from @xcite rely on theoretical seds , which are subject to imperfections in the physical prescriptions of the models and degeneracies among the disk parameters . the predicted seds also can vary from one set of adopted models to another . for instance , as indicated above , some objects can be reproduced with either an optically thin disk or a settled optically thick disk , depending on the choice of models @xcite . the use of disk mass as a criterion for identifying optically thin disks is also problematic . the requisite millimeter data are unavailable for many members of young clusters and are incomplete at low disk masses , even in taurus @xcite . estimates of disk masses and optical depths are sensitive to both the disk model and the opacities that are adopted @xcite . in addition , @xcite selected a threshold of 0.001 @xmath55 as a lower mass limit for optically thick disks because nearly all optically thick disks in taurus with millimeter data have masses above this value . however , 0.001 @xmath55 is only slightly above the detection limits for the millimeter observations in taurus ( @xmath56 @xmath4 , * ? ? ? * ) , and hence does not represent a measurement of the minimum mass of optically thick disks . deeper millimeter data are needed to reliably determine the distribution of disks masses down to low optical depths . in a critique of the disk classifications based on colors from @xcite , @xcite noted that ux tau a has a disk with a large gap @xcite but did not satisfy the color criteria for a transitional disk from @xcite and instead was classified as a primordial disk ( or a full disk with our current terminology ) . however , this classification was consistent with the definitions adopted by @xcite in which a disk with a gap was considered pre - transitional and a subset of primordial disks , whereas only disks with large holes were named transitional . any set of classification criteria will exclude disks with holes or gaps below a certain size , so the absence of disks with gaps like ux tau a from our definition of transitional disk is not particularly worrisome . we simply use the term transitional disk " to refer to a different range of structural evolution than @xcite . in summary , we have selected a classification scheme for disks in which the evolutionary stages are defined by distinct structural characteristics @xcite . the criteria for identifying members of these classes are based on patterns in the data for disks ( e.g. , gap in colors ) and can be applied to data that are available for large numbers of young stars and brown dwarfs , namely multi - band mid - ir photometry from _ spitzer _ and _ wise_. in @xcite , we used extinction - corrected mid - ir colors to estimate the evolutionary stages of disks in taurus and other regions . for upper sco , we modify this approach by considering instead the color excesses relative to photospheric colors ( e.g. , @xmath57 $ ] ) . we have computed the excess in a given color for each member of upper sco by subtracting the color of a stellar photosphere in upper sco for the spectral type in question , as determined by a fit to the sequence of diskless stars in figure [ fig : pan ] . through this process , the colors for all members are corrected for roughly the same level of extinction , which is the average extinction of the diskless members . the resulting excesses for the colors from figure [ fig : pan ] are presented in figure [ fig : ex ] . the be stars are omitted since they do not require classification . we plot the excesses based on [ 4.5 ] and w2 together because of the similarity of these bands . we do the same for [ 24 ] and w4 . for each object , we show the data in [ 4.5 ] and [ 24 ] when available , and otherwise use w2 and w4 . @xcite used @xmath58 $ ] , @xmath46 $ ] , and @xmath48 $ ] for their disk classifications . since relatively few members of upper sco have been observed in [ 5.8 ] , we begin our classifications with the other two colors ( and @xmath32 ) . in figure [ fig : ex ] , we show our adopted thresholds in @xmath59)$ ] and @xmath57/w4)$ ] for distinguishing between full disks and later evolutionary stages . these thresholds approximate those applied to @xmath46 $ ] and @xmath48 $ ] by @xcite . we have used the objects classified in this way to reveal the boundary between full disks and later stages in @xmath60 vs. @xmath57/w4)$ ] , and we have used this boundary to classify stars that lack data in [ 8.0 ] . to classify the disks bluer than full disks , we apply the following criteria , which are similar to those from @xcite and @xcite : transitional disks have @xmath57/w4)>3.2 $ ] ; evolved disks have @xmath57/w4)<3.2 $ ] , @xmath59)>0.3 $ ] , and @xmath61 ; debris disks and evolved transitional disks have @xmath57/w4)<3.2 $ ] , @xmath59)<0.3 $ ] , and @xmath62 . measurements of gas content are necessary for distinguishing between the latter two classes of disks since they exhibit similar mid - ir seds @xcite , but such data are unavailable for most disks in upper sco . for the early - type stars , the high abundance of these debris / evolved transitional candidates relative to full disks indicates that most are probably debris disks since evolved transitional disks are expected to be short - lived ( section [ sec : histo ] ) . two exceptions to the above criteria are hip 79288 and hip 79977 . both stars are redder than our thresholds for debris / evolved transitional disks , but we classify them as such based on previous studies @xcite . we have attempted to classify the disks that lack photometry in [ 24 ] or w4 , and hence do not appear in figure [ fig : ex ] . for the disks of this kind that are detected in w3 , the excesses in this band are small enough to indicate that they are evolved or transitional . some of those disks are probably evolved rather than transitional based on the available constraints on @xmath57/w4)$ ] . two of the disks without w3 , w4 , or [ 24 ] have data in [ 8.0 ] , both of which are full disks according to their values of @xmath59)$ ] . the remaining disks that lack data in [ 24 ] or w4 are detected only at @xmath63 , most of which are later than m6 . we classify them as full disks for the purposes of this work , although some could be evolved disks . in table [ tab : all ] , we present our disk classifications for all members of upper sco that show mid - ir excess emission . through our classifications , we arrive at 135 full disks , 28 evolved disks , 6 transitional disks , three disks that are evolved or transitional , and 63 debris / evolved transitional disks . previous mid - ir studies have already identified many of these full disks @xcite , evolved and transitional disks @xcite , and debris / evolved transitional disks @xcite . our new candidates for transitional disks consist of 2mass j16082733 - 2217292 , 2mass j16101888 - 2502325 , and 2mass j16151239 - 2420091 . it is likely that a number of full disks have inner holes that are too small to be classified as transitional using our criteria , such as 2mass j16081566 - 2222199 , which has strong excesses in w3 and w4 but no excess in w2 ( see upper panel of figure [ fig : ex ] ) . most of the candidates for evolved disks are newly identified as such in this work . in section [ sec : exc24 ] , we listed our new candidates for debris / evolved transitional disks and a few stars in this category that we classify differently than other studies . we note that hip 81474 has been frequently treated as a herbig ae / be star , but it lacks a clear detection of line emission @xcite . therefore , it is classified as debris / evolved transitional in this work . the distribution of disk excesses as a function of parameters like wavelength , spectral type , cluster age , excess size , and disk class can provide insight into the evolution of disks . to characterize the distribution of excesses in upper sco , we begin by considering the fraction of members that exhibit excesses . in table [ tab : fex ] and figure [ fig : fex ] , we present the excess fraction as a function of spectral type for [ 4.5]/w2 , [ 8.0 ] , w3 , and [ 24]/w4 . to examine the evolution of primordial and debris disks separately , one excess fraction is computed for full , transitional , and evolved disks , and another is computed for debris and evolved transitional disks . the latter disks do contain primordial dust , but they can not be distinguished from debris disks with available data , as discussed in the previous section . the images in w3 , w4 , and [ 24 ] were not sufficiently sensitive to detect all known members of upper sco . the members that were too faint to be detected tend to have later spectral types or smaller excesses . as a result , we can not compute meaningful excess fractions in these bands at the latest types . therefore , the excess fractions for w3 and [ 24]/w4 in figure [ fig : fex ] are shown only for spectral types in which nearly all of the known members are detected . for the joint [ 24]/w4 excess fraction , we adopt the [ 24 ] data when available , which have a high level of completeness down to a spectral type of m4 . among [ 24 ] non - detections earlier than m4 , six stars were not subject to extended emission or blending with other stars , and hence have useful limits on their photometry . the resulting constraints on @xmath48 $ ] indicate that these six stars have little , if any , excess emission , so we include them as non - excess stars in the [ 24]/w4 fraction . among stars that were not observed at [ 24 ] , the w4 images detected all but three stars earlier than m0 . therefore , we use the w4 data of members not observed at [ 24 ] and earlier than m0 in the [ 24]/w4 fraction . the three stars earlier than m0 with non - detections in w4 have sufficient constraints on this band to indicate that excesses are probably not present , so we count them as non - excess stars . it is unnecessary to impose such restrictions on the spectral types included in the excess fractions at [ 4.5 ] , w2 , and [ 8.0 ] since all known members were detected in these bands as long as they were observed ( by irac ) and not blended with another star . the be stars are excluded from these excess fractions . we first discuss the excess fractions for the full , transitional , and evolved disks ( i.e. , all primordial disks except for evolved transitional disks ) . for these disks , the excess fractions are very low at a roughly uniform level as a function of spectral type for b through g types , and then increase with later types from k through m. this trend was discovered by @xcite , supported by additional data from @xcite for late - m members , and now is reinforced further with better number statistics . all of the bands in figure [ fig : fex ] exhibit a similar dependence on spectral type , except that the excess fractions are slightly higher at longer wavelengths due to the presence of a small number of transitional disks . as discussed by @xcite , these data indicate that disk lifetimes are longer for stars at lower masses , providing more time for planet formation . we note that the excess fraction in [ 4.5]/w2 for the last bin in figure [ fig : fex ] at m8l2 ( @xmath64 @xmath4 ) is uncertain because of the small number of objects and the rapidly changing photospheric colors , which makes it difficult to reliably identify excesses . in addition , since excesses in [ 4.5 ] are smaller at later types @xcite , the excess fraction in this band may significantly underestimate the disk fraction . longer wavelengths are needed for more robust detections of disk excesses for brown dwarfs , but the members at m8l2 were too faint for w3 and w4 photometry and were not observed at [ 5.8 ] and [ 8.0 ] . the disk fractions based on the data in figure [ fig : fex ] are compared to similar measurements in younger clusters by @xcite . we now consider the excess fractions for the debris and evolved transitional disks . as mentioned earlier , we can not distinguish between these two types of disks with the mid - ir data , but most of the candidates at earlier spectral types are probably debris disks . it is unclear which class of disk dominates at later types . in figure [ fig : fex ] , the excess fraction for this sample increases rapidly with longer wavelengths . excesses in [ 4.5]/w2 are present for only a few stars , which may be some of the youngest debris disks in the association @xcite . the [ 24]/w4 fraction is highest for a stars ( @xmath65 ) , drops sharply for b stars ( @xmath66 ) , and decreases more gradually with later types , reaching 0.10.2 for g , k , and m stars . these trends are consistent with those previously found in smaller samples of [ 24 ] data in upper sco by @xcite and @xcite . those studies investigated various aspects of their [ 24 ] data , and their results apply to our new measurements as well . for instance , @xcite found that the rapid decrease in [ 24 ] fraction from a to b stars is expected because of increased ejection of dust grains by radiation pressure ( see also * ? ? ? * ) . the [ 24]/w4 excess fractions at later spectral types are more difficult to interpret since they contain an unknown contribution from evolved transitional disks . in addition , many debris disks around late - type stars may be too cool to produce noticeable excesses in these bands . @xcite and @xcite also examined the evolution of debris disks by comparing the [ 24 ] excess fractions among upper sco and other associations and clusters across a range of ages ( see also * ? ? ? * ; * ? ? ? * ; * ? ? ? our data have improved the statistical constraints on excess fractions and sizes in upper sco , particularly at m types , but an analysis like that from @xcite and @xcite for m stars is not possible because sufficiently complete samples of [ 24 ] data are available for only a few regions . the mid - ir colors of the members of a given star - forming region typically exhibit a gap between diskless stars and the bulk of the disk - bearing stars ( see section [ sec : criteria ] ) . this trough in the distribution of excess sizes has been previously interpreted as evidence that once a disk has begun to clear to a certain degree , the remainder of the dust in the inner disk clears rapidly @xcite . in this section , we examine the constraints on the clearing timescale that are provided by the distribution of excesses in upper sco . because the contrast between a stellar photosphere and circumstellar disk increases at longer ir wavelengths , [ 24 ] and w4 offer the largest dynamic range in the sizes of excesses among the photometric bands that we have compiled for upper sco . given that the [ 24 ] observations are more sensitive than those at w4 , we choose the former for constructing the distribution of excesses . to characterize how this distribution changes with cluster age , we compare the data for upper sco to measurements of [ 24 ] in taurus @xcite . upper sco and taurus are the best available regions for such a comparison since they roughly bracket the ages of primordial disks ( @xmath13 and 10 myr ) , contain two of the richest and nearest samples of young stars , and have been thoroughly observed by _ spitzer_. in particular , few other regions have [ 24 ] measurements that reach the stellar photospheres of low - mass stars , which is essential for fully sampling excesses across all sizes . since very few primordial disks are present among the early - type stars in upper sco , we exclude stars earlier than k0 from the distributions of @xmath48 $ ] excesses for upper sco and taurus . we also omit stars later than m5 since most of the diskless stars at these types are below the detection limits of the [ 24 ] images . roughly a dozen stars in each region at k5m5 are not detected in [ 24 ] and could have small excesses based on the limits on @xmath48 $ ] . as a result , the distributions of excesses that we construct are biased slightly against small excesses . the distributions of @xmath57)$ ] for k0m5 members of taurus and upper sco are presented in figure [ fig : histo ] , excluding protostellar sources . the two distributions exhibit similar troughs at @xmath57)\sim0.5$]3 . most observations of other young clusters at [ 24 ] are less complete at low excesses or consider smaller numbers of stars , but they are generally consistent with gaps of this kind @xcite . meanwhile , among stars redward of the trough ( @xmath57)>3 $ ] ) , the average color is bluer in upper sco than in taurus . these characteristics can be explained in the following manner . the youngest systems have highly flared disks , and hence the largest ir excesses . as they evolve ( e.g. , growth of grains , accretion onto the star ) , disks become more settled and produce smaller excesses . after disks have become highly settled and have reached a certain level of initial clearing ( corresponding to @xmath57)\sim3 $ ] ) , they quickly clear their remaining inner disks , resulting in a scarcity of disks with small excesses . because star formation is ongoing in taurus but has ceased in upper sco , the distribution of excesses in the former is more heavily weighted to younger and redder disks . in other words , the youngest disks in upper sco have already evolved substantially , explaining its paucity of very red disks relative to taurus . because the evolutionary stages of disks are not sampled in the same way by the differing star formation histories of taurus and upper sco , a comparison of the n(evolved+transitional)/n(full ) does not indicate the relative timescales of clearing between the two populations @xcite . instead , one must compare the ratio of the number of evolved and transitional disks to the number of sources slightly redward of the trough , i.e. , the shape of the distribution of excesses at @xmath57)\lesssim3.5 $ ] . this property is not tightly constrained in either region because of the small numbers of stars involved , but taurus and upper sco do not show any obvious differences . thus , there is no evidence to indicate that the clearing timescale changes from the age of taurus to the age of upper sco ( @xmath1310 myr ) . the data in figure [ fig : histo ] do reveal a higher abundance of candidates for evolved transitional disks or debris disks in upper sco . if the lifetime of the evolved transitional stage is similar to that of the transitional and evolved stages , then most of these candidates in upper sco are probably debris disks . it is expected that upper sco would contain more debris disks than taurus given their relative ages . as discussed in section [ sec : criteria ] , the definition of transitional disks adopted by @xcite encompasses some of the objects redward of the gap in the mid - ir colors in regions like taurus and upper sco . as a result , they have derived higher abundances of transitional disks in young clusters , and hence a longer timescale for inner disk clearing , than found here and in other studies that use a more restrictive definition for transitional disks @xcite . of course , it is not meaningful to compare abundances of transitional disks and the resulting timescales among studies that do not define this evolutionary stage in the same manner . we contend that the clearing timescales from @xcite are less useful and practical since they are based on a definition of transitional disks that is subject to uncertainties and degeneracies in model seds and that does not rely on distinct observational signatures ( i.e. , gap in colors ) . @xcite suggested that the gap in mid - ir colors in taurus is not evidence of a short lifetime for the evolutionary stage of disks within this gap , contrary to the original interpretation @xcite . instead , they argued that the gap is present because few members of taurus are old enough to have reached that stage , and that the diskless stars in taurus have lost their disks due to binary companions rather than advanced age . in @xcite , we noted that a rather large fraction of the taurus population is diskless ( 40% among non - protostars ) and that these stars are older than disk - bearing members , on average , based on their wider spatial distribution , demonstrating that a significant number of stars in taurus have aged sufficiently to fully clear their disks , and hence pass through the evolved and transitional stages . @xcite reiterated the same points from their previous study , but also conceded that many of the diskless stars in taurus may indeed be older than their counterparts with disks . here , we once again contend that if a significant number of stars have aged sufficiently to have fully cleared their disks , then the evolved and transitional stages should be fully sampled as well , and therefore their scarcity demonstrates that they are short - lived relative to full disks . this conclusion is supported by the fact that older regions like upper sco exhibit the same gap in their colors ( figure [ fig : histo ] ; * ? ? ? * references therein ) . we have performed a census of the circumstellar disk population of the upper sco association ( @xmath0 myr ) using mid - ir images obtained with the _ spitzer space telescope _ and _ wise_. the results of this study are summarized as follows : 1 . we have analyzed all images of known members of upper sco at 3.6 , 4.5 , 5.8 , 8.0 , and 24 that have been collected by _ spitzer _ to date . among the 863 known members , 484 have been observed in at least one of these bands . we have presented a catalog of photometry and non - detections for these objects . we have compiled all photometry at 3.4 , 4.6 , 12 , and 22 ( w1w4 ) for members of upper sco from the mission , thoroughly vetting these data for false detections , blends with neighboring stars , and extended emission . all resolved , unblended members were detected by in at least 3.4 and 4.6 . we have used colors computed between @xmath41 and six of the _ spitzer _ and bands to identify the members that exhibit excess emission from circumstellar disks . we have estimated the evolutionary stages of the detected disks using the sizes of the excesses among these bands . these stages are defined based on distinct structural characteristics that are apparent in the seds ( * ? ? ? * references therein ) , and they consist of the following : optically thick disks without large holes or gaps ( _ full _ ) , optically thick disks with large holes ( _ transitional _ ) , optically thin disks without large holes or gaps ( _ evolved _ ) , optically thin disks with large holes ( _ evolved transitional _ ) , and disks of second - generation dust ( _ debris _ ) . through these classifications , we have found @xmath1 new candidates for transitional , evolved , and debris disks . 4 . based on our census of disks , the fraction of upper sco members with inner primordial disks is @xmath2% for b g stars ( @xmath3 @xmath4 ) and increases with later types , reaching @xmath5% at m5l0 ( @xmath670.2 @xmath4 ) . this trend was first observed by @xcite , and is now reinforced with a sample that is larger and reaches lower stellar masses . the combination of these data and the latest age estimate for upper sco ( @xmath0 myr , * ? ? ? * ) indicate that a significant fraction of primordial disks around low - mass stars survive for at least @xmath8 myr . 5 . k0m5 members of upper sco show a dearth of disks with small excesses ( transitional and evolved disks ) . when the differences in star formation histories of upper sco and taurus ( @xmath9 myr ) are taken into account , the distributions of excesses for both regions indicate a timescale for inner disk clearing that is much shorter than the typical lifetime of primordial disks . we thank catherine espaillat , nuria calvet , and eric feigelson for comments on the manuscript . k. l. was supported by grants ast-0544588 and nnx12ai58 g from the national science foundation and the nasa astrophysics data analysis program , respectively , and e. m. was supported by grant ast-1008908 from the national science foundation . this work makes use of data from the _ spitzer space telescope _ , 2mass , _ wise _ , and the infrared processing and analysis center ( ipac ) infrared science archive ( irsa ) . _ spitzer _ and irsa are operated by the jet propulsion laboratory ( jpl ) and the california institute of technology ( caltech ) under contract with nasa . is a joint project of the university of california , los angeles , and jpl / caltech , funded by nasa . 2mass is a joint project of the university of massachusetts and the ipac / caltech , funded by nasa and the nsf . the center for exoplanets and habitable worlds is supported by the pennsylvania state university , the eberly college of science , and the pennsylvania space grant consortium . ll name & source name + othernames & other source names + sptype & spectral type + r_sptype & spectral type reference + adopt & adopted spectral type + 3.6mag & _ spitzer _ [ 3.6 ] band magnitude + e_3.6mag & error in [ 3.6 ] band magnitude + f_3.6mag & flag on [ 3.6 ] band magnitude + 4.5mag & _ spitzer _ [ 4.5 ] band magnitude + e_4.5mag & error in [ 4.5 ] band magnitude + f_4.5mag & flag on [ 4.5 ] band magnitude + 5.8mag & _ spitzer _ [ 5.8 ] band magnitude + e_5.8mag & error in [ 5.8 ] band magnitude + f_5.8mag & flag on [ 5.8 ] band magnitude + 8.0mag & _ spitzer _ [ 8.0 ] band magnitude + e_8.0mag & error in [ 8.0 ] band magnitude + f_8.0mag & flag on [ 8.0 ] band magnitude + 24mag & _ spitzer _ [ 24 ] band magnitude + e_24mag & error in [ 24 ] band magnitude + f_24mag & flag gon [ 24 ] band magnitude + w1mag & _ wise _ w1 band magnitude + e_w1mag & error in w1 band magnitude + f_w1mag & flag on w1 band magnitude + w2mag & _ wise _ w2 band magnitude + e_w2mag & error in w2 band magnitude + f_w2mag & flag on w2 band magnitude + w3mag & _ wise _ w3 band magnitude + e_w3mag & error in w3 band magnitude + f_w3mag & flag on w3 band magnitude + w4mag & _ wise _ w4 band magnitude + e_w4mag & error in w4 band magnitude + f_w4mag & flag on w4 band magnitude + exc4.5 & excess present in [ 4.5 ] ? + exc8.0 & excess present in [ 8.0 ] ? + exc24 & excess present in [ 24 ] ? + excw2 & excess present in w2 ? + excw3 & excess present in w3 ? + excw4 & excess present in w4 ? + disktype & disk type + llllll b0b8 & 2.818 & 0/25 = @xmath66 & 0/22 = @xmath68 & 0/25 = @xmath66 & 0/26 = @xmath66 + b8a6 & 1.82.8 & 0/45 = @xmath69 & 0/40 = @xmath69 & 1/46 = @xmath70 & 1/45 = @xmath70 + a6f4 & 1.51.8 & 0/20 = @xmath71 & 0/12 = @xmath72 & 1/21 = @xmath73 & 1/21 = @xmath73 + f4g2 & 1.41.5 & 0/18 = @xmath71 & 0/9 = @xmath74 & 0/18 = @xmath71 & 0/17 = @xmath75 + g2k0 & 1.31.4 & 1/25 = @xmath76 & 1/21 = @xmath73 & 1/25 = @xmath76 & 1/25 = @xmath76 + k0m0 & 0.71.3 & 6/67 = @xmath77 & 4/46 = @xmath78 & 8/68 = @xmath79 & 7/60 = @xmath79 + m0m4 & 0.20.7 & 35/231 = @xmath80 & 18/96 = @xmath81 & 41/210 = @xmath82 & 23/108 = @xmath83 + m4m8 & 0.0350.2 & 97/387 = @xmath84 & 22/84 = @xmath85 & & + m8l2 & 0.010.035 & 4/23 = @xmath86 & & & + b0b8 & 2.818 & 0/25 = @xmath66 & 0/22 = @xmath68 & 0/25 = @xmath66 & 0/26 = @xmath66 + b8a6 & 1.82.8 & 0/45 = @xmath69 & 0/40 = @xmath69 & 5/46 = @xmath87 & 20/45 = @xmath88 + a6f4 & 1.51.8 & 1/20 = @xmath89 & 0/12 = @xmath72 & 3/21 = @xmath90 & 7/21 = @xmath91 + f4g2 & 1.41.5 & 0/18 = @xmath71 & 0/9 = @xmath74 & 0/18 = @xmath71 & 3/17 = @xmath92 + g2k0 & 1.31.4 & 0/25 = @xmath66 & 0/21 = @xmath68 & 1/25 = @xmath76 & 4/25 = @xmath93 + k0m0 & 0.71.3 & 0/67 = @xmath94 & 0/46 = @xmath69 & 0/68 = @xmath94 & 8/60 = @xmath95 + m0m4 & 0.20.7 & 0/231 = @xmath96 & 0/96 = @xmath97 & 0/210 = @xmath96 & 13/108 = @xmath98 + m4m8 & 0.0350.2 & 0/387 = @xmath96 & 0/84 = @xmath97 & & + m8l2 & 0.010.035 & 0/23 = @xmath66 & & & +

we study the effect of quenched spatial disorder on the steady states of driven systems of interacting particles . two sorts of models are studied : disordered drop - push processes and their generalizations , and the disordered asymmetric simple exclusion process . we write down the exact steady - state measure , and consequently a number of physical quantities explicitly , for the drop - push dynamics in any dimensions for arbitrary disorder . we find that three qualitatively different regimes of behaviour are possible in 1- disordered driven systems . in the vanishing - current regime , the steady - state current approaches zero in the thermodynamic limit . a system with a non - zero current can either be in the homogeneous regime , chracterized by a single macroscopic density , or the segregated - density regime , with macroscopic regions of different densities . we comment on certain important constraints to be taken care of in any field theory of disordered systems . 0.5 cm pacs numbers : 05.60.+w , 47.55.mh , 64.60.-i , 05.50.+q 2

cond-mat9711302

it is known that quenched disorder can strongly affect the large - scale , long - time behaviour of nonequilibrium driven systems with interacting constituents . the interplay of disorder , interactions and drive opens up the possibility of new regimes of complex and interesting behaviour arising in these systems @xcite . in the theoretical effort to delineate and explore regimes of new behaviour , an important role is played by simple models which capture some features of more complex physical systems . in this paper , we study disordered driven diffusive systems by analysing stochastically evolving lattice gas models , with quenched disordered hopping rates @xcite . driven diffusive systems in the absence of disorder have been studied extensively and are reviewed in @xcite . also , systems with disorder and drive but no interactions between particles are well studied and understood @xcite . but there have been only sporadic studies of disordered driven diffusive systems of interacting particles . it has been argued that strong enough random site dilution can substantially affect the transport properties of particles with hard - core interactions , and can make the system respond nonmonotonically to the driving field @xcite . on the other hand , a low concentration of blocked sites was found numerically not to affect the critical behaviour of a driven lattice gas with additional attractive inter - particle interactions @xcite . finally , a driven lattice gas with a quenched noise distribution was studied using field - theoretic techniques in @xcite , but the connection of this study with particle - conserving disordered lattice gas models is not clear . in this paper , we study disordered lattice gas models with a view towards identifying different sorts of generic behaviour that can arise on large scales as a consequence of disorder . the only interaction included is the hard - core constraint which limits the allowed occupancy of each site . our results pertain mostly , but not exclusively , to one dimension . in the remainder of this introduction , we discuss the different types of behaviour displayed by the lattice gas models under study . we find three distinct regimes in disordered driven diffusive systems in one dimension : in the _ homogeneous _ regime , the state of the system is characterized by a single density and a nonzero current . quenched disorder induces variations of the density on the microscopic scale , of the order of a few lattice spacings . however , the system has a macroscopically homogeneous density . in the thermodynamic limit , the current approaches a finite value . in the _ segregated - density _ regime , the state of the system is characterized by two distinct values of density , and a nonzero current . besides microscopic - scale variations of the density , there are macroscopic regions with differing high and low densities . the state is thus characterized by phase separation of the density , and a spatially constant time - averaged current which remains finite in the thermodynamic limit . in the _ vanishing - current _ regime , the state of the system is characterized by two distinct values of the density , and an essentially zero current . the hallmark of this regime is that the current decreases as the system size increases , and vanishes in the thermodynamic limit . this is a consequence of rare but rate - limiting backbends , or stretches of bonds which disfavour the forward flow of current . the density is inhomogeneous on a macroscopic scale . the density profiles in typical states in each of the three regimes are depicted in figure 1 , while figure 2 shows the variation of the current with system size in the three cases . examples of these behaviours are discussed in this paper for two types of lattice - gas models , namely the disordered drop - push process ( ddpp ) and the disordered asymmetric simple exclusion process ( dasep ) . the models are defined in detail in sections [ sec : ddpp ] and [ sec : dasep ] respectively , but for the purpose of discussion here , it suffices to note that the models are similar in that there is a maximum occupancy of each site in both , and are different in the dynamical moves attempted nearest neighbour jumps in the dasep , and slightly longer - ranged leapfrogging moves in the ddpp . the absence of detailed balance , together with the breaking of translational invariance , in disordered off - equilibrium systems makes the characterization of even the stationary state difficult in general . it is shown that the steady state of the disordered drop - push process can be found explicitly the first such instance we are aware of , in a system with disorder , interactions and drive @xcite . this determination which is based on the condition of pairwise balance @xcite shows that a product measure form is valid in all dimensions . the form reflects the microscopic inhomogeneities coming from the underlying disorder , and results in a macroscopically homogeneous state . for the disordered asymmetric exclusion process , the steady state measure is not analytically characterizable , and we study the problem within a site - wise inhomogeneous mean - field theory and by numerical simulation . the result depends crucially on whether or not the system has backbends , which are stretches of the lattice where the local bias is against the particle flow . in the no - backbend case , when the average particle density is sufficiently away from 1/2 , the spatial profile of the density has microscopic shocks , but is uniform on macroscopic scales ( fig . [ fig : regimes1]a ) . however , in a finite region around half - filling , disorder induces phase separation into macroscopic regions of high and low density ( fig . [ fig : regimes1]b ) . we give approximate arguments to understand the origin and nature of this phase separation , and to obtain the form of the phase diagram in the current - density plane . this sort of behaviour has also been seen earlier in a model with a single weak bond @xcite . we argue that disorder - induced phase separation is a generic feature of systems in which the current @xmath1 versus density @xmath2 shows a maximum at some intermediate density , in the absence of disorder . in the version of the dasep in which the easy direction of hopping is itself a quenched random variable , the model represents a system of hard - core particles in a random potential with an overall downward tilt , but with backbends of arbitrary length . long backbends severely limit the maximum current that can flow through the system , and in fact the current decreases to zero as the system size increases ( fig . [ fig : regimes2 ] ) ; the system is in the vanishing - current regime . although our emphasis in this paper is on the analysis of lattice models , we comment briefly on certain constraints that are important in a continuum description . such a description is expected to be valid for the large - scale , long - time behaviour , and is based on stochastic differential equations involving appropriate coarse - grained variables . it is argued that quenched randomness is manifest in random multiplicative coefficients in a gradient expansion . conservation of particle number which implies spatial constancy of the current in the steady state imposes strong constraints on these terms . in one dimension , using a well known mapping @xcite , the particle models are equivalent to stochastic growth models of a 1-@xmath0 interface moving in a 2-@xmath0 medium . the interface moves with a speed proportional to the current in the particle model . the disordered jump rates now become local growth rates which are disordered in a columnar fashion for the moving interface @xcite . the three principal regimes of behaviour discussed above for the particle models translate into distinct regimes for interface motion , namely ( i ) a moving interface with normal roughness , ( ii ) a moving interface with large segments with different mean slopes , and ( iii ) an interface with different - slope segments , which is stationary in the thermodynamic limit . the paper is organized as follows . in section [ sec : ddpp ] we define and discuss the steady state properties of the disordered drop - push process in arbitrary dimensionality . the disordered asymmetric exclusion process with only forward - easy - direction of hopping , but quenched random rates , is discussed in section [ sec : dasep ] ; the case in which there are some backward - easy - direction bonds is discussed in section [ sec : sinai ] . in section [ sec : continuum ] we discuss the constraints on a continuum description , while section [ sec : height ] discusses the implications of our results for models interface growth in the presence of columnar disorder . section [ sec : conclusion ] is the conclusion . the drop - push process was initially introduced in @xcite as a model of activated flow involving transport through a series of traps of equal depths . the dynamics consists of activated hops together with a cascade of overflows following each move . the disordered version of the model may be considered as a discrete model of activated fluid flow down an inclined rugged slope with lakes of varying depths ; see figs . [ fig : lakes],[fig : ddpp-1d ] . this is similar to above - threshold behaviour of the model considered in @xcite . in this section we show that the steady state and current can be found exactly in all dimensions for the ddpp and its generalizations . the model in @xmath0-dimensions is defined on a hypercubic lattice with periodic boundary conditions along all the @xmath0 axes ( with unit vectors \{@xmath3 } ) . at each site @xmath4 is a well which can hold at most @xmath5 particles ( figs . [ fig : ddpp-1d],[fig : ddpp-2d ] ) with @xmath5 s chosen independently from some probability distribution @xmath6 . the configuration @xmath7 of the system is specified by specifying the set of occupation numbers \{@xmath8 } with ( @xmath9 ) . further , with each site @xmath4 is assigned a set \{@xmath10 } of positive random numbers chosen from some given distribution @xcite . the dynamics is stochastic . in a time interval @xmath11 , with a probability @xmath12 , the topmost particle in the well @xmath4 hops out , and drops into well @xmath13 , i.e. into the adjacent well in the @xmath14th direction . here @xmath15 are a set of site - independent positive numbers satisfying @xmath16 . now , if well @xmath13 is already full , then the particle gets pushed further _ preserving the direction of the initial jump _ to the next site and so on . the cascade of transfers terminates once a partially full well is encountered . note that here the set of jump - rates \{@xmath17 } are site - dependent as well as functions of the occupation numbers . these rates , together with the well - depths \{@xmath5 } , constitute the quenched random variables in the model . the set of probabilities \{@xmath18 } determines the direction of the global bias @xmath19 and , as will be shown in section [ sub : imddpp ] , also the direction and magnitude of the steady state current in the model . however , they do not enter the expression for the normalized invariant measure . though all the results we will discuss holds for any arbitrary choice of the @xmath20 s , in a physical system they should be determined from the details of the trapping mechanisms etc . , e.g. they may be taken to be of the kramers form @xmath21 $ ] for situations where the jumps are activated @xcite . the time evolution of the probability @xmath22 for the system to be in configuration @xmath23 is given by the master equation @xcite @xmath24 here the @xmath25 s are the transition matrix elements identified with the rates @xmath20 s defined in the model . e.g. if the transition @xmath26 involves moving the topmost of the @xmath27 particles at @xmath4 to @xmath28 along the @xmath29th direction , then @xmath30 . the steady - state or the invariant measure of the dynamics is the set of time independent weights @xmath31 satisfying ( [ eq : master ] ) above . hence the problem of finding the invariant measure reduces to that of finding a set of positive weights @xmath31 such that the total incoming flux into any configuration @xmath32 ( the first sum in ( [ eq : master ] ) ) equals the total flux out of @xmath32 ( the second sum in ( [ eq : master ] ) ) . the uniqueness of the invariant measure is ensured by the connectedness property of the @xmath25-matrix , i.e. every configuration can be reached from any other by a sequence of transitions @xcite . we claim that the ( unnormalized ) measure of configuration @xmath33 in the steady state has the product form @xmath34 here @xmath35 are the single - site weights defined as @xmath36 where @xmath37 with @xmath38 being a microscopic rate . to show that ( [ eq : imddpp ] ) is indeed the invariant measure for the ddpp we show that it is possible to associate configuration @xmath39 in one - to - one correspondence with every @xmath40 obtained from @xmath41 by an elementary transition such that @xmath42 the above is the condition of _ pairwise balance _ @xcite which ensures that the terms in the two sums on the right hand side of ( [ eq : master ] ) cancel in pairs . pairwise balance has been used earlier to find steady states of translationally invariant systems @xcite . we now see that it can be used effectively to deduce the steady state of a disordered system as well . suppose the transition @xmath26 involves hopping a particle at site @xmath4 to a site @xmath43 with all wells in between along the @xmath29th axis full ( fig . [ fig : pwb-2d ] ) . also suppose the well @xmath44 is not full but all wells between @xmath45 and @xmath4 are full . the configuration @xmath39 is constructed such that it is identical to @xmath23 at all sites except at the sites @xmath45 and @xmath4 , at which @xmath46 . with @xmath39 thus defined , the pairwise balance condition ( [ eq : pwb ] ) @xmath47 reduces to @xmath48 since @xmath23 and @xmath39 differ only at the sites @xmath45 and @xmath4 . as can be explicitly checked , condition ( [ eq : pwb - check ] ) is satisfied in view of the form of the weights @xmath49 . the prescription above ensures one - to - one correspondence between configurations @xmath39 and @xmath26 transitions . in the limit of large @xmath50 we use the grand canonical description and write the normalized probability of configuration @xmath7 in the steady state as @xmath51 where the site generating functions are given by @xmath52 and @xmath53 is the number of particles in the configuration . here @xmath54 is the fugacity and will be shown to be directly related to the steady - state current . the mean particle density at site @xmath4 is given as @xmath55 @xcite . thus the fugacity @xmath54 is related to the mean particle density @xmath2 of the system through @xmath56 the steady state is characterized by a spatially uniform @xmath54 but inhomogeneous site densities . for simplicity of presentation , we first derive a closed form expression for the steady - state current @xmath57 for the 1-@xmath0 fully asymmetric model . then we use these results to write down the current for the general @xmath0-dimensional case . \(i ) _ @xmath58 , with forward jumps _ : in this case , for notational convenience , we replace the lattice site index @xmath4 by a single integer index @xmath59 . further , we allow jumps only in one direction ( @xmath60 ) . for an infinite system we write the current @xmath61 in the bond @xmath62 as @xmath63 where @xmath64 is the probability that site @xmath59 has occupancy @xmath65 . note that the product form of the measure ( [ eq : imddpp ] ) has been exploited to write the above expression with decoupled joint probabilities . it is cumbersome , and not very instructive , to perform the sum in ( [ eq : ji - i1 ] ) to obtain the current in a closed form . instead we use the spatial constancy of the current in the steady state to arrive at the result . we note that in the above expression the first term represents contribution of jumps originating from site @xmath66 to the left of site @xmath59 with all sites in between full . the second sum represents jumps from the site @xmath59 itself contributing to @xmath61 . the first sum may be rewritten as @xmath67 $ ] . the quantity within square brackets [ @xmath68 is immediately recognised to be @xmath69 by comparing it with ( [ eq : ji - i1 ] ) . physically , this means that all jumps contributing to @xmath69 , for which site @xmath59 is completely full , also contribute to @xmath61 . hence we have the recursion @xmath70 relating @xmath61 to @xmath69 . now , since @xmath71 are the steady - state site probabilities , and since in the steady state all the bond currents must be equal ( i.e. @xmath72 ) , from ( [ eq : jrecur ] ) we obtain @xmath73 where ( [ eq : imddpp ] ) and ( [ eq : site - prob ] ) have been used in the second step above . note that the steady - state current does not depend upon the detailed spatial arrangement of wells . it is only a function , through the fugacity @xmath54 , of the density and the total number of wells of different types in a particular realization of disorder . \(ii ) _ @xmath58 , with jumps in both directions : _ we can write the current @xmath61 in bond @xmath62 as the difference between the current @xmath74 due to the rightward jumps and the current @xmath75 due to the leftward jumps . this can be done since in each cascade the direction of the initial jump is preserved . now , we use the result for the fully asymmetric case above to each of these currents separately to obtain @xmath76 \(iii ) _ @xmath77 _ : to generalize the above results to @xmath77 we note that for ddpp in @xmath77 , the invariant measure ( [ eq : imddpp ] ) is the same if we single out a particular direction , say @xmath29 , and allow jumps only along that direction . together with the direction preservation of individual jumps , this allows us to write the expression for the current in any dimension : @xmath78 \epsilon_0 z = \epsilon_0 z { { \vec e}}\label{eq : j0-dd}\end{aligned}\ ] ] where @xmath79 is the external drive . as in the @xmath58 case , the magnitude and the direction of the steady - state current does not depend upon the detailed arrangements of the wells . because of the product form of the measure , the connected part of the equal - time density - density correlation function @xmath80 vanishes identically for @xmath81 . consequently , the fluctuation of the number of particles in @xmath82 consecutive sites along a straight line can be computed exactly : @xmath83 ^ 2 \right\rangle \nonumber\\ & = & \sum_{j = i+1}^{i+r } \left ( z{\partial\over \partial z}\right)^2 \ln z_j . \label{eq : hhcor - ddpp}\end{aligned}\ ] ] the second step follows from the product form of the measure . for @xmath84 , a standard mapping discussed in section vi introduces height variables defined by @xmath85 evidently , @xmath86 is the equal - time height - height correlation @xmath87 . averaging over the disorder distribution gives @xmath88 implying that the ` roughness ' exponent @xmath89 ( defined by @xmath90 ) is @xmath91 . let us consider a drop - push model where the maximum occupancy of each site is restricted to one , i.e. @xmath92 , but the hopping rates @xmath93 are disordered , and chosen independently from the binary distribution @xmath94 this model has the essential ingredients of disorder present in the original ddpp , yet it is simple enough that explicit , closed form relations between the mean density @xmath2 and fugacity @xmath54 , and hence the steady - state current @xmath95 , can be written down . let us denote by @xmath96 and @xmath97 the site generating functions for the @xmath98 and the @xmath99 sites respectively . using ( [ eq : sitegen - ddpp ] ) and ( [ eq : imddpp ] ) , these are given by @xmath100 and @xmath101 . now , since the fractions of @xmath98 and @xmath99 sites are @xmath102 and @xmath103 respectively , ( [ eq : implicit ] ) reduces to @xmath104 this can be easily inverted to obtain @xmath54 as a function of @xmath2 , e.g. for @xmath105 and @xmath106 we obtain @xmath107 since @xmath108 is known , the steady - state current is trivially obtained from ( [ eq : j0-dd ] ) . finally , the correlation function @xmath86 of ( [ eq : hhcor - ddpp ] ) , upon disorder averaging , may be written as @xmath109 \nonumber \\ & = & { 1\over 2}\left[{z\over(1+z)^2}+{qz\over(q+z)^2}\right ] r\end{aligned}\ ] ] where @xmath54 is given by ( [ eq : fugacity ] ) above . we may consider a generalized version of the drop - push process in which , in addition to the particle moves , independent hole moves are also allowed . for simplicity we restrict ourselves to the generalized version of the single occupancy ddpp introduced above . this generalized model may be regarded as the disordered lattice gas analogue of the toom interface dynamics in the low - noise limit @xcite ; see section [ sec : height ] . the techniques developed for the ddpp may be used to obtain the exact steady - state measure and other quantities such as current and static correlations provided a certain condition [ ( [ eq : constr - gddp ] ) below ] is met . the model in @xmath0 dimensions is defined on a hypercubic lattice with periodic boundary conditions along all the @xmath0 axes ( with unit vectors \{@xmath3 } ) , similar to the ddpp . each site @xmath4 of the lattice can hold either a particle ( @xmath110 ) or a hole ( @xmath111 ) . the configuration @xmath7 of the system is specified by specifying the occupation number of each well \{@xmath8 } with ( @xmath112 ) . further , to each site @xmath4 is assigned a pair of positive random numbers @xmath113 chosen from some distribution . the dynamics is stochastic and is very similar to that for the ddpp dynamics : in a time interval @xmath11 , a particle at site @xmath4 is exchanged with the closest hole in the @xmath14th direction with a probability @xmath114 ( fig . [ fig : gddp - model ] ) . for identical particles this move is equivalent to a cascade of particle moves terminating at the first vacant site as in the drop - push dynamics . likewise , in interval @xmath11 , a hole at site @xmath4 is exchanged with the closest particle along the @xmath29th direction with probability @xmath115 . this can be looked upon of as a cascade of hole - moves analogous to the cascade of particle moves . here , as in the ddpp , the @xmath116 s and @xmath117 s are all non - negative and satisfy @xmath118 . further we chose @xmath89 s and @xmath119 s such that @xmath120 where @xmath121 is a constant independent of @xmath4 . as we will see below , this particular choice of the jump - rates allows the exact determination of the invariant measure of the model . physically this choice is quite reasonable since it implies that the sites which act as traps for particles ( low @xmath122 ) are more transparent to holes ( high @xmath123 ) and vice versa . as pointed out earlier the non - disordered version of this model , i.e. \{@xmath124 } in 1-@xmath0 , is the lattice gas equivalent of the low - noise driven toom interface dynamics @xcite . the master equation ( [ eq : master ] ) governing the time evolution of the system now includes terms corresponding to hole moves as well . since each micro - step involves either only particle moves or hole moves , we use the principle of pairwise balance for the particle moves and hole moves separately . we work in the grand canonical picture in the thermodynamic limit . if only particle moves were allowed the invariant measure would be given by @xmath125 the single site weights @xmath35 are given by @xmath126 introducing the fugacity @xmath54 and site generating functions @xmath127 we can write the normalized single site probabilities as @xmath128 and @xmath129 . similarly , with only the hole moves , the invariant measure has the product form @xmath130 where @xmath131 ( @xmath132 ) refers to the presence ( absence ) of a hole . the single site weights @xmath133 are given by @xmath134 introducing the fugacity @xmath135 for holes and site generating functions @xmath136 we can write the normalized single - site probabilities as : @xmath137 and @xmath138 . now , since each site is occupied either by a particle or a hole , we must have @xmath139 and @xmath140 . using the detailed forms of @xmath141 s and @xmath142 s , we arrive at the condition ( [ eq : constr - gddp ] ) with @xmath143 . if this condition is satisfied then the invariant measure for gddp is given by either ( [ eq : muc - p ] ) or ( [ eq : muc - h ] ) , since both are equivalent . in a similar manner as for the ddpp the current due to the particle moves and hole moves may be computed . the total particle current due to both types of moves , in @xmath0 dimensions , is given by @xmath144 as for the ddpp , static density - density correlations in the steady state vanish identically on account of the product form of the steady - state measure . in @xmath58 the height - height correlation is given by ( [ eq : hhcor - ddpp ] ) . the asymmetric simple exclusion process ( asep ) is a prototype model for studying nonequilibrium phenomena in the context of lattice gases @xcite . when discussing the effect of quenched disorder , it is important to distinguish between cases in which ( a ) the easy direction of hopping in each bond is the same but hopping rates are random , and ( b ) the easy direction is itself a random variable . the latter case is studied in section [ sec : sinai ] . in this section , we consider a 1-@xmath0 system with disorder of type ( a ) and show that quenched disorder can induce macroscopic phase separation . using a variety of arguments we sketch the phase coexistence curve in the current ( @xmath57 ) - mean density ( @xmath2 ) plane . this agrees qualitatively with the results obtained from the monte carlo ( mc ) simulations . in one dimension , we define the disordered asymmetric simple exclusion process on a ring of @xmath50 sites . each site can hold either @xmath145 or @xmath132 particle . to each bond @xmath62 of the lattice is assigned a quenched random rate @xmath146 chosen independently from some probability distribution @xmath147 . the dynamics is stochastic : in a time interval @xmath11 a particle at site @xmath59 attempts to hop , with probability @xmath148 , to site @xmath149 . the move is completed if and only if site @xmath149 is unoccupied ( see fig . [ fig : dasep-1d ] ) . for the model defined above no analytically exact characterization of the steady - state measure could be obtained . a simpler model in which there is only one defect bond has been studied in detail by janowsky and lebowitz @xcite , but in this case too the exact steady - state measure is not known . we use monte carlo simulations and a mean - field approximation to demonstrate some striking effects of quenched disorder . figure [ fig : dasep - jrho ] shows the steady - state current @xmath57 vs mean density @xmath2 plot , obtained from mc simulations , for a system of size @xmath150 and the rates @xmath89 chosen from the binary distribution @xmath151 here @xmath103 is the fraction of weak bonds , and @xmath82 is a measure of the strength of the weak bonds . we used the values @xmath152 in our numerical work . for a specified mean density @xmath2 , a random initial distribution of @xmath153 particles on @xmath50 sites is chosen and the system is allowed to settle into a steady state by evolving it for a sufficiently large number of mc steps . then the current across each bond is obtained by counting the total number of jumps across that bond , over a large number of mc steps . an average of all the bond currents thus computed is taken as @xmath57 , as currents across all bonds are equal in the steady state . @xmath57 is a symmetric function of density around @xmath154 as a result of a certain symmetry with respect to particle - hole interchange ( see appendix [ apx : phsym ] ) . as may be expected , the current for the disordered system lies between the corresponding values of the two pure reference systems with @xmath155 and @xmath156 on all the bonds . the more striking qualitative difference between the disordered and pure systems is the appearance of a plateau ( regime b in fig . [ fig : dasep - jrho ] ) for a range of densities @xmath157 . in this regime , the current is independent of the mean density and equals the maximum allowed current in the system . the approximate size @xmath158 of regime b , which is a function of @xmath82 and @xmath103 , is obtained in subsection [ sub : fsm ] below . we studied the steady state density profiles characterized by the set of site densities @xmath159 in both regimes a and b , using mc simulations . we found that in regime a the system is homogeneous on a macroscopic scale , while in regime b it shows macroscopic density segregation . figure [ fig : dasep - profiles ] shows the steady state density profiles for three representative mean densities one from regime a and two from regime b. evidently there is a large qualitative difference between the profiles in the two regimes . in regime a ( fig . [ fig : dasep - profiles]a ) , there are density variations ( shocks ) only over microscopic scales ; coarse - graining over a few lattice spacings leads to a spatially uniform density . in contrast to this , in the profiles corresponding to regime b ( fig . [ fig : dasep - profiles]b ) , there are density inhomogeneities over length scales comparable to the system size @xmath50 , in addition to the shocks on a microscopic scale . this segregation into high and low density phases , with large shocks separating them , is reminiscent of phase separation , and occurs over the full range of mean particle densities corresponding to regime b. a qualitatively similar phenomenon has been found in a system with one defect bond , studied in @xcite . the _ number _ of different large - scale regions of high and low density shows fluctuations from one realization of disorder to another . as the size of the system is increased , we monitored the mean number of these regions , and found that it is nearly constant , or perhaps increases very mildly certainly , much less fast than linearly in the size of the system . this implies that the characteristic length scale of density segregation increases indefinitely in the thermodynamic limit . we now turn to a mean - field approximation , which assumes no correlations between site densities in the steady state . we will see that it captures most of the steady - state features found in the mc simulations above , including the steady - state density profile . the time - averaged steady - state current @xmath61 in the bond @xmath62 is given by @xmath160 . in view of the mean - field approximation @xmath161 this reduces to @xmath162 to compute the steady - state current @xmath57 as a function of the mean particle density @xmath2 for a given realization of disorder , we use the two iteration schemes described below , which yield equivalent results . \(i ) _ constant current iteration scheme . _ for a given system of size @xmath50 and for a fixed value @xmath57 for the current , we iterate the following set of equations @xmath163 around the chain starting with , say , some value @xmath164 ( periodic boundary conditions imply @xmath165 ) . if @xmath57 is less than a certain value @xmath166 , which is the maximum current supported by the system within the mean - field approximation , then the iteration scheme converges , i.e. we get all the site densities in the physically acceptable range @xmath167 $ ] . the average of these site densities gives the mean density of the system corresponding to the stationary current @xmath57 . there are in general two values of the mean particle density corresponding to an allowed value of @xmath57 and the iteration scheme converges to one or the other depending on the initial value of the density @xmath164 . however , in this scheme we find that the number of iterations required for convergence increases without bound , as the trial value @xmath57 gets closer to @xmath166 from below . this divergence of the iteration scheme is presumably due to the existence of the plateau in the @xmath168 curve , i.e. there exist many values of @xmath2 for @xmath57 very close to @xmath166 . hence to obtain the density profile for @xmath2 in the density segregated regime we resort to the _ constant density _ iteration scheme described below . \(ii ) _ constant density iteration scheme . _ in this scheme we begin by assigning site densities @xmath169 to the lattice sites subjected to the constraint @xmath170 . the site densities are updated in parallel according to : @xmath171 where @xmath172 $ ] in view of ( [ eq : mfa ] ) . we refer to this as the constant density scheme , since in each iteration the total density remains unchanged , i.e @xmath173 . after a sufficient number of iterations , which depends upon the starting mean density @xmath2 , the set of site densities converge to a set of numbers \{@xmath174 } and the current on each bond converges to the steady state current @xmath57 . the steady - state density profiles and the @xmath175 plot ( @xmath176 ) for a given configuration of disorder obtained using these schemes is shown in figs . [ fig : dasep - profiles ] and [ fig : dasep - jrho ] respectively . it is evident that the mean - field approximation ( [ eq : mfa ] ) reproduces quite well not only the @xmath177 relationship , but also the density profile , including the locations of shocks , though not the shapes of individual shocks . although the mean - field approximation of the previous subsection successfully reproduces many features in the steady state , it does not yield a simple understanding of the phase separation ( fig . [ fig : dasep - profiles ] ) or the plateau in the @xmath1@xmath2 curve ( fig . [ fig : dasep - jrho ] ) in terms of the macroscopic parameters of the model . we conjecture that underlying the behaviour of the dasep in different regimes is a maximum current principle : for a given mean density , the system settles into a steady - state which maximizes the stationary current . such a maximum current principle has been used to describe phase separation in the asymmetric simple exclusion process with open boundary conditions in @xcite . to use the maximum - current principle to have a qualitative understanding of the phase separation in dasep , let us assume that the density in each stretch of like bonds is uniform . this approximation is in fact exact in the fully segregated model ( fsm ) discussed in the following subsection . let us denote stretches of @xmath178 bonds by x and stretches of @xmath179 bonds by y. the two parabolas in fig . [ fig : phases - expl ] are the steady - state @xmath168 curves for the two pure reference systems all x and all y. in the disordered system , since the steady - state current has to be spatially constant , the possible densities are given by the four intersections of the line @xmath180 with the two parabolas . if the mean density is in the range @xmath181 ( or @xmath182 ) , then the allowed densities for the x and y stretches are @xmath183 ( or @xmath184 ) respectively . the current is in fact determined by the density constraint @xmath185 . the variation of density between @xmath164 and @xmath186 ( or @xmath187 and @xmath188 ) between x and y stretches corresponds to the ` sub - bands ' seen in fig . ( [ fig : dasep - profiles]a ) . on a macroscopic scale , however , the system has uniform density . now consider what happens when the mean density is brought closer to @xmath91 . from fig . [ fig : phases - expl ] , it is evident that the current would tend to increase , and would eventually reach the maximum allowed value @xmath189 ( which equals @xmath190 in the thermodynamic limit , as argued below ) . as the density is increased further , the current remains constant at @xmath189 , in accordance with the maximum current principle , and the excess density is taken care of by converting some of the x stretches of density @xmath164 into ones with @xmath188 ( or vice versa if @xmath191 ) . this conversion takes place adjacent to the largest stretch of y bonds , leading to two macroscopic regions of different mean densities one with densities @xmath192 for the x and y stretches , and the other with @xmath193 . the position of the principal shock separating these regions is at the location of the largest y stretch . in the dasep , the assumption of uniform density in each stretch is not really true , on account of the finite length of most of the stretches . however , the above argument provides a qualitative picture towards understanding the reason for phase separation in the dasep . . in a certain respect , the reason for the phase separation in the dasep is similar to that in the single defect bond model studied in @xcite both have local segments in the system where the maximum allowed current is less than that allowed everywhere else . in the single defect bond model , phase separation takes place when the current carried by the rest of the system , with presumed uniform density , is larger than the maximum current allowed through the weak bond . in the dasep , with an extensive number of weak bonds , the _ largest stretch _ of weak bonds acts as the current - limiting segment . the length of this stretch increases as @xmath194 with system size , and in the thermodynamic limit the maximum allowed current in this stretch is @xmath195 equal to the maximal current in a pure system with only weak bonds . the essential point leading to phase separation is thus the maximum current principle , coupled with localized current - limiting regions in the system . in the dasep , this limit is determined by long stretches of weak bonds , and similar considerations should apply in related models . consequently , we would expect a density - segregated phase in disordered versions of models which , in the absence of disorder , display a maximum in the steady - state current as a function of density . it is useful to define a model for which many of the approximations made in the previous subsection are actually exact . to this end , we study a fully segregated model ( fsm ) , which is obtained from the binary random model above by arranging all like bonds together . thus , in this model , one has _ two _ large stretches of x and y , of lengths @xmath196 and @xmath197 respectively . for the fsm in the thermodynamic limit , the assumption of uniform density within each stretch is justified , as correlations due to the junctions decay with increasing separation , and may be neglected in the bulk @xcite . steady - state mc density profiles for the fsm at three different fillings @xmath198 are shown in fig . [ fig : dp - seg ] symmetry under particle - hole exchange implies analogous behaviour for @xmath199 ( appendix [ apx : phsym ] ) . for low densities ( @xmath200 ) , the two stretches have uniform bulk densities @xmath201 and @xmath202 related to each other by the requirement of equality of the two bulk currents , @xmath203 and the density constraint @xmath204 these three equations determine @xmath205 and @xmath57 uniquely for a given @xmath2 . for @xmath105 we obtain @xmath206 this is analogous to the macroscopically homogeneous state of the fully random system . as the mean density is increased , the density of each stretch increases , until , at a critical density @xmath207 ( the corresponding critical density on the higher side is @xmath208 ) , @xmath209 equals @xmath91 ( fig . [ fig : dp - seg ] ) . at this density , the current equals the maximum possible current in y , namely , @xmath210 . as @xmath2 is increased further , @xmath209 and @xmath57 remain constant at @xmath91 and @xmath211 respectively . the density change is adjusted by creating a density inhomogeneity in stretch x. the two densities @xmath212 and @xmath213 are related by @xmath214 so that the currents in the two phases are equal , i.e. @xmath215 . this implies @xmath216 ( see inset in fig . [ fig : dp - seg ] ) . the fraction of these phases can be determined from a lever rule and are given by @xmath217 . this locking of the density in the y stretch at @xmath218 is a direct consequence of the maximum current principle introduced above : any change of @xmath209 from @xmath91 would reduce the current in y , and hence in the full system . all the arguments above can be applied for @xmath219 because of particle - hole symmetry . thus for @xmath220 the state of the segregated model is analogous to the phase separated regime b of the random model . the size of regime b in the fsm is given by @xmath221 . it closely approximates the size of the b regime in the dasep ( fig . [ fig : dasep - jrho ] ) . for the fsm , as @xmath82 is varied we obtain different @xmath222 curves . the phase - coexistence curve in the current - density plane in the parametric form @xmath223 which is the parabola @xmath224 in fig . [ fig : phase - diag ] . the difference between the phase boundaries of the dasep and fsm fig . [ fig : phase - diag ] comes from the fact that the interspersed weak - bond stretches in the dasep have finite lengths , and the density in these small stretches is not quite equal to @xmath91 . close to the phase separation , we anticipate the mean density in a y - stretch of length @xmath225 in the dasep to be of the form @xmath226 , with @xmath227 . using the distribution of the @xmath228 stretches , namely @xmath229 , we obtain the @xmath82 dependence of the critical density @xmath230 where @xmath231 is the density in the x stretches near the phase transition . comparing with the phase diagram for the fsm in fig . [ fig : phase - diag ] , the correction term @xmath232 seems to be positive for all @xmath82 . further , let us suppose that the current in the fsm is a lower bound to that in the dasep , as suggested by fig . [ fig : phase - diag ] . then the coexistence curve for the random system must be quadratic near the critical point ( @xmath233 ) being bounded by two quadratics , namely , the @xmath177 curve for the pure system @xmath234 , and the coexistence curve of the fsm @xmath235 . figure [ fig : dasep - corln ] shows the monte carlo results for the site - averaged density - density and height - height correlation functions @xmath236 in both the homogeneous and density - segregated regimes of the dasep . @xmath237 is seen to decay rapidly over a few lattice spacings , accounting for the success of the mean field approximation . it is found that @xmath238 grows as @xmath82 implying a roughness exponent @xmath239 . as discussed at the beginning of section [ sec : dasep ] , the introduction of randomness in the easy direction of individual bonds alters the properties of one - dimensional disordered exclusion process in a crucial way . we study this in this section . the model is defined as follows : assign quenched arrows ( pointing either right or left ) independently to each bond of a periodic chain , with probability @xmath240 for left arrows , and @xmath102 for right arrows . an arrow defines the easy direction of hopping on each bond : a particle - hole exchange across a bond occurs with rate @xmath241 if the particle moves along the direction of the arrow , and @xmath242 if it moves opposite to the arrow . since @xmath240 , there is an overall tendency for particles to circulate rightward , and the question is whether there is a nonzero current even in the thermodynamic limit . the model represents a system of hard - core particles in a random potential with a downward tilt . a conglomeration of left pointing arrows constitutes a backbend , where the potential climbs up before going down again . within mean - field theory it is possible to obtain an upper bound @xmath243 on the current that can be carried by mutually excluding particles through a backbend of length @xmath244 @xcite . to this end , consider biased diffusion of hard core particles in the segment [ @xmath245 of a 1-d lattice , with the ` optimal ' boundary conditions @xmath246 and @xmath247 ; these boundary conditions force the largest possible current through the segment , opposite to the bias . the master equation that describes transport is invariant under interchange of particles and holes and simultaneous relabelling of sites in reverse order , i.e. @xmath248 the boundary conditions respect this symmetry , implying that the steady - state density @xmath249 at site @xmath66 satisfies @xmath250 . thus in the steady state the number of particles in the backbend is @xmath251 irrespective of the strength of the bias @xmath252 . the principal effect of increasing @xmath252 is to sharpen the region which marks the transition from the particle - rich half of the backbend to the hole - rich half . the steady - state profile approaches a step function centered at @xmath253 as @xmath254 . the current in the steady state is the number of particles crossing site @xmath225 in unit time . results of a monte carlo study @xcite are consistent with the large-@xmath225 asymptotic behaviour @xmath255 where @xmath256 is a bias - induced length given by @xmath257 . this can be seen by writing the current within a mean field approximation as @xmath258 , and finding the value of @xmath1 for which the boundary conditions @xmath259 , @xmath260 hold . for @xmath261,this leads to @xmath262 @xcite , in agreement with ( [ eq : s1 ] ) when @xmath252 is small . the origin of the factor @xmath263 in the exponent in ( [ eq : s1 ] ) has been discussed in @xcite , and we recount the argument in brief . the transport of a single particle through the backbend involves two ( approximately ) causally independent steps which occur in parallel : ( i ) the topmost particle ( located at site @xmath264 in large fields ) has to be activated a distance @xmath251 , which requires an activation time @xmath265 and , ( ii ) the consequent hole that remains in the steady - state distribution moves to the bottom and is filled up , by moving each of @xmath251 particles up through a lattice spacing . the time required is @xmath266 again . the current @xmath1 is thus proportional to @xmath267 , and consequently follows ( [ eq : s1 ] ) . since , for fixed @xmath252 , the largest current that can flow through a long backbend ( @xmath268 ) is exponentially small in its length , the largest current through the 1-d lattice of length @xmath50 is determined by the length @xmath269 of the largest backbend encountered . since the probability of occurrences of @xmath244 consecutive left - pointing arrows on bonds is proportional to @xmath270 , we may estimate @xmath271 from @xmath272 where @xmath273 is a constant of order unity . substituting in ( [ eq : s1 ] ) , we find that the current falls with increasing lattice size as @xmath274 with @xmath275 . thus the current is expected to decay as a power law in @xmath50 , with a bias - dependent power , and to vanish in the thermodynamic limit . figure [ fig : jl_sinai ] shows the result of monte carlo simulations . as with the milder sort of disorder discussed in section iii , the state is strongly inhomogeneous and shows macroscopic regions of high and low density . figure [ fig : regimes1]c shows the time - averaged density profile for a typical configuration of bonds . there is a large shock around the rate - limiting backbend , which separates the two regions . for fixed lattice size @xmath50 with an associated longest backbend @xmath276 , the current is a nonmonotonic function of @xmath252 . this can be seen as follows . if @xmath252 is low enough that @xmath277 , linear response theory would imply that current @xmath1 grows linearly with @xmath252 . on the other hand , if @xmath252 is large enough that @xmath278 , the current falls with increasing @xmath252 according to ( [ eq : s1 ] ) . in between , @xmath1 achieves a local maximum when @xmath279 , which implies @xmath280 . the argument given above implies that the current carried by a system of hard - core particles through the randomly backbending lattice vanishes in the thermodynamic limit , no matter how small the bias @xmath252 . this is in contrast to the behaviour of _ noninteracting _ particles in the same random environment , where the drift velocity vanishes only if the bias is strong enough @xcite . the difference can be traced to the possibility , in the noninteracting case , of compensation by a large build - up of density at the bottom of a backbend , which then succeeds in driving a finite current over the backbend . this option does not exist once repulsive hard - core interactions come into play , and the current vanishes in the thermodynamic limit . it is interesting to ask whether the behaviours found above in the disordered lattice gases can be reproduced using a continuum description of the problem . though we have not pursued this question to its logical end , we discuss in this section some general constraints that a continuum description should satisfy . the steady - state of the disordered system of interacting particles is described by a spatially varying time - averaged density profile @xmath281 . the time evolutions of fluctuations around the mean density profile are governed by the continuity equation @xmath282 the coarse - grained current @xmath283 may be phenomenologically divided into three parts @xmath284 a local hydrodynamic assumption has been made in writing the _ systematic _ part of the current , @xmath285 as a function of the local density , @xmath286 . the explicit @xmath4 dependence of @xmath285 comes from the breaking of translational invariance by quenched disorder , and its exact form of @xmath285 depends on the microscopic dynamics of the model . the _ diffusive _ current , @xmath287 $ ] , involves a quenched diffusion tensor , @xmath288 . the _ noise _ , @xmath289 , is to mimic , on a mesoscopic scale , the stochastic nature of the evolution . it is usually assumed to be gaussian distributed and @xmath290-correlated in space and time with vanishing spatial and temporal averages . to obtain the time evolution of the density fluctuations @xmath291 , we expand @xmath285 in powers of @xmath292 as @xmath293 and put it in ( [ eq : continuity ] ) . this results in @xmath294 \label{eq : dynamics - ddim}\ ] ] in one spatial dimension , the above reduces to ( with @xmath4 replaced with @xmath295 ) the form @xmath296 \label{eq : dynamics-1dim}\ ] ] which was considered in @xcite . in steady state , the time - averaged current must be independent of @xmath295 . as both @xmath297 and @xmath298 vanish under time averaging , the constraint to be satisfied is @xmath299 which is important to account for , as the coefficients in @xmath300 are explicitly space - dependent , _ i.e. _ , @xmath301 $ ] must vanish . in their attempt to study a continuum model which describes the dasep , becker and janssen ( bj ) @xcite write the current @xmath302 in powers of @xmath303 , the density fluctuation away from the _ overall _ particle density @xmath2 in the system . in one dimension , the form quoted @xcite for the dasep is @xmath304 where @xmath305 is an additive quenched noise term . since the time - averaged density @xmath306 varies in space , it is evident that @xmath307 has a nonzero expectation value @xmath308 . as we have seen in section iii , @xmath306 , and thus @xmath309 , can show strong variations , especially in the density - segregated regime . spatial constancy of the average current demands that the time average of the right hand side of ( [ eq : bj ] ) must satisfy @xmath310 = 0 . \label{eq : bjconstr}\ ] ] even when this condition is satisfied , it is not entirely clear that this continuum theory actually represents the dasep in one dimension . bj argue that if @xmath311 , then disorder is irrelevant on the large scale , a conclusion that is supported by @xcite . however , their conclusion , that the condition @xmath312 holds only if @xmath313 , does not seem to be correct , at least for the dasep in one dimension ; we have seen in section iv , there is an entire range of densities @xmath314 where @xmath315 vanishes , associated with segregation of density on a macroscopic scale . in 1-@xmath0 , both the dasep ( with or without backbends ) and the gddp for which the maximum occupancy per site is 1 are equivalent to stochastic growth models of a 1-@xmath0 interface moving in a 2-@xmath0 medium . corresponding to each particle - hole configuration @xmath316 is assigned an interface profile @xmath317 through @xmath318 @xcite . pictorially this means that each particle corresponds to a @xmath319 downward line segment , while a hole corresponds to an upward one ( fig . [ fig : map ] ) . thus , clusters of particles and holes translate to @xmath320 slope segments , and the interface has a mean slope which vanishes when the particle density is @xmath91 . away from half - filling , periodic boundary condition for the lattice becomes a helical boundary condition for the interface . junctions between adjacent particle and hole clusters correspond to corners in the interface profile . evolution of the interface is dictated by the dynamics of the corresponding particle system . the gddp corresponds to the slice - wise motion of segments of a toom interface in the low noise limit @xcite , while the dasep corresponds to the corner - flip ` single step ' growth model @xcite ( fig . [ fig : map ] ) . in both cases particle movement to the right ( or hole move to left in gddp ) corresponds to local forward growth ( deposition ) of the interface while a leftward particle move ( e.g. in dasep with backbends ) corresponds to local backward growth ( evaporation ) . the quenched jump rates for the particle moves implies _ columnar _ disordered growth - rates for the interface : the normal growth rate of the interface at a fixed @xmath59 is the same irrespective of the location of the interface at successive times . in the long time limit , the mean local forward speed of the interface is the same at all points along the interface being proportional to the spatially constant steady - state current of the corresponding particle model . we now discuss the various qualitatively different regimes which arise in the interface growth models . figure [ fig : interface ] shows time averaged steady - state interface profiles @xmath321 corresponding to the three regimes of driven particle systems illustrated in fig . [ fig : regimes1 ] in section [ sec : introduction ] . in all the three cases we started from an initially uniform profile corresponding to a random distribution of particles on the lattice . figure [ fig : interface]a is an interface with a non - zero mean tilt , which has net forward growth rate at all points . the interface has a uniform slope on a macroscopic scale and moves with a finite non - zero speed preserving its mean tilt . this case corresponds to the _ homogeneous _ regime depicted in fig . [ fig : regimes1]a . on a microscopic scale the interface has frozen - in roughness ( fig . [ fig : interface]a , inset ) corresponding to the microscopic shocks in the steady - state density profile of fig . [ fig : regimes1]a . if the mean tilt vanishes , but the interface still has a net forward growth rate at all points , then the initial uniform profile at @xmath322 coarsens into large segments of different mean slopes at long times ( fig . [ fig : interface]b ) . these segments have frozen roughness on microscopic scales , similar to the non - zero tilt case ( fig . [ fig : interface]b , inset ) . the interface moves with a finite speed preserving its mean shape and mean vanishing tilt . this corresponds to the _ segregated - density _ regime of fig . [ fig : regimes1]b . in addition to the frozen roughness on the microscopic scales , we can define the kinetic roughness as the equal - time mean - square height fluctuations around the steady - state profile . we consider the zero - mean height variables @xmath323 defined in ( [ eq : heightdef ] ) , and define the roughness exponent @xmath89 through @xmath324 . as discussed in subsections [ sec : ddpp]d , e and [ sec : dasep]f , the roughness exponent @xmath239 , in both the cases above . in fig . [ fig : interface]c , which corresponds to the _ vanishing - current _ regime of fig . [ fig : regimes1]c , the profile resembles that corresponding to the segregated - density regime in that it has large segments of different mean slopes . however , in this case the interface is stationary in the thermodynamic limit reflecting the vanishing of the steady - state current in the particle system . in interface language , this situation can be visualized as a case of interface growth where there are local stretches of the interface having a net backward growth ( evaporation ) rate . though , on the average , there are more forward - growth regions than backward - growth ones , in the limit of large system size , arbitrarily long evaporation stretches effectively pin the interface . turning now to continuum description of the dynamics of these interfaces , at least in the cases where the mean speed of growth is non - zero , the sum in the definition of the height variable is replaced by an integral of the coarse - grained particle density , @xmath325 dx'$ ] . the growth equation for @xmath326 is governed by ( spatially ) integrated one dimensional version of the continuity equation ( [ eq : continuity ] ) : @xmath327 . the fluctuations @xmath328 in @xmath329 around the steady - state profile @xmath330 , are governed by @xmath331 obtained by integrating ( [ eq : dynamics-1dim ] ) . the absence of any additive quenched spatial noise term in ( [ eq : height ] ) is due to the spatial constancy of the growth speed of the interface dictated by the same constraint on the steady - state current . in this respect ( [ eq : height ] ) differs from the model discussed in @xcite where such a term arises naturally due to absence of any such constraints . as can be readily verified by power counting , an additive quenched columnar term is highly relevant in the rg sense and leads to much rougher interfaces than the @xmath239 interfaces described by ( [ eq : height ] ) . in this paper , we have studied the stationary current - carrying states of driven lattice gas models with quenched disordered hopping rates . the principal results are of two types first the exact determination of the steady states for a class of disordered models , and second the demarcation of distinct regimes of behaviour on macroscopic length scales , as a result of disorder . in this section , we briefly review these results , and discuss some related open problems . the steady states of a family of disordered models the disordered drop - push process ( ddpp ) , and related models have been found in all dimensions by an application of the condition of pairwise balance . the result is a product measure state , with site - dependent weights , reflecting the microscopic disorder in the model . the current has been computed as well . the system is characterized by a strictly uniform current density , and a coarse - gained particle density that is approximately uniform . on a macroscopic scale , the state is homogeneous . disorder can lead to macroscopically non - homogeneous states , as in the 1-@xmath0 disordered asymmetric simple exclusion process ( dasep ) . our numerical and mean - field results show that a macroscopically density - segregated state occurs in the dasep model with no backbends , for densities in a finite range around half - filling . the origin of density separation is traceable to the existence of a largest current that can be carried by a stretch of weak bonds . this low current can be sustained in the rest of the lattice only by separating the density into distinct large and small values in macroscopic regions of the lattice . backbends introduce a third type of possible behaviour in one dimension . like the stretch of weak forward bonds in the density - segregated regime , a backbend rate - limits the current , leading to density segregation . however , there is an important difference : the longer the stretch of weak forward bonds , the closer is the current to a finite asymptotic value ; by contrast , the longer the stretch of reverse - biased bonds , the smaller the current it decreases exponentially fast with backbend length . since the probability of occurrence of a backbend decreases exponentially with its length , the result is a current that decreases as a bias - dependent power of the overall size of the lattice . the crucial physical point which underlies the behaviour in each of these regimes is the requirement that the steady - state current be constant at all points in the system . continuum field - theoretical approaches must ensure that this local constraint is respected ; while this is automatically assured for translationally invariant systems , it may need special care to guarantee in disordered systems . it would be of interest to generalize our results to higher dimensions and also to include interactions between particles at different sites . a few scattered results along these lines are available . ( i ) for the drop - push class of problems , we have seen in section [ sec : ddpp ] that the exact steady state even in higher dimensions is characterized by inhomogeneous product measure . on large scales , this leads to a homogeneous state . ( ii ) the transport of particles with hard - core interactions through the infinite cluster of a randomly diluted lattice above the percolation concentration has been studied @xcite . in a certain regime of dilution , backbends act as local traps , but unlike the one - dimensional case considered in section [ sec : sinai ] , there exist infinitely long paths on which the length of every backbend is less than a fixed value @xcite . the sub - network of such paths is expected to carry a current which then remains finite in the thermodynamic limit . there is thus no vanishing - current regime in this system . ( iii ) with attractive interactions between particles , the driven lattice gas system with nearest - neighbour hopping is known to undergo phase separation below a certain temperature @xcite . a numerical simulation showed that the addition of a low concentration of blocked sites did not alter the critical behaviour of this system @xcite . more systematic studies of higher - dimensional systems are called for . in particular , it would be interesting to know whether disorder - induced large - scale inhomogeneities , akin to the phase separation found in one dimension , persist in higher dimensions as well . acknowledgements : we thank r. e. amritkar , d. dhar , s. krishnamurthy , r. lahiri , g. i. menon and n. trivedi for useful discussions . in this appendix we discuss two types of symmetry transformations in the dasep which leave the steady - state current invariant . ( i ) the first , which involves flipping the directions of all jumps and interchanging all particles and holes , can be proved exactly . under these operations , the steady - state weights of particle configurations and the magnitude of the current are shown to remain invariant . ( ii ) the second type of symmetry is restricted to 1-@xmath0 systems in which only one way jumps are allowed on each bond . the symmetry transformation consists of flipping the direction of jump on each bond . in this case , we have convincing numerical evidence that the current is invariant , though there appears to be no simple relationship between steady - state weights of various particle configurations . we have no general proof of this result . \(i ) _ easy direction flip and particle - hole interchange _ : consider the dasep , with quenched disordered , unequal forward and backward hopping rates on each bond . below we explicitly deal with the one - dimensional case , but the results are easily generalizable to higher dimensions . let @xmath332 denote a particular realization of disorder and @xmath333 the realization obtained from @xmath332 by flipping the easy direction of jumps on all the bonds : if @xmath332 denotes the set of transition rates \{@xmath334 } , then @xmath333 corresponds to the set \{@xmath335}. also , let us denote by @xmath23 and @xmath336 two particle configurations related to each other by particle - hole interchange . clearly , if @xmath23 is an allowed configuration of the system at filling @xmath2 , then @xmath336 corresponds to a filling @xmath337 , and thus there is a one - to - one correspondence between the configurations at the two fillings . now , let @xmath338 be the configuration obtained from @xmath23 by exchanging the occupation numbers at the sites @xmath59 and @xmath66 . it is easy to see that the two transition rates , @xmath339 in realization @xmath332 , and @xmath340 in realization @xmath333 are equal ( fig . [ fig : phsym ] ) , i.e. the transition matrices @xmath341 and @xmath342 in the two realizations have identical entries . identification of the two @xmath343-matrices implies that the invariant measures @xmath344 and @xmath345 satisfy @xmath346 using this , together with the identity @xmath347 , we can relate @xmath348-point correlations at fillings @xmath2 and @xmath337 . in particular , the site densities at the two fillings are related as @xmath349 further , the steady - state currents in @xmath332 and @xmath333 at the two fillings are equal in magnitude : @xmath350 the negative sign in the last step reflects the fact that the direction of the current is opposite in the two systems . in the case of the fsm ( subsection [ sub : fsm ] ) the realizations @xmath332 and @xmath351 are identical ( apart from a reflection ) . hence in this case the above arguments imply that the magnitudes of the currents at the two fillings @xmath2 and @xmath337 are equal . a similar result holds for the single defect bond case studied by janowsky and lebowitz @xcite . \(ii ) _ easy direction flip only ( for 1-@xmath0 , forward hopping ) _ : above , we found that the current is invariant when the filling is changed from @xmath2 to @xmath337 , provided that the disorder realization is changed from @xmath332 to @xmath333 . here we observe ( based on numerical evidence ) that in 1-@xmath0 , with only forward hopping , the result is also true for realization @xmath332 on its own , i.e. @xmath352 in view of ( [ eq : strong_phsym ] ) this is equivalent to @xmath353 the claim is easily verified for the single particle ( or single hole ) case using an explicit form for the current : @xmath354 @xcite . also , ( [ eq : phsym3 ] ) is true at @xmath154 since ( [ eq : phsym2 ] ) is an identity at this filling . we do not have a proof for ( [ eq : phsym2 ] ) or ( [ eq : phsym3 ] ) in the general case , but they seem to be borne out numerically . for instance , we studied the validity of ( [ eq : phsym3 ] ) for a system of size @xmath355 with @xmath356 particles . we took @xmath332 to be the set @xmath357 . the invariance of the current is verified up to @xmath145 part in @xmath358 . we also studied the steady - state probabilities of all the @xmath359 particle configurations for each of @xmath332 and @xmath333 , both by monte carlo and by lanczos iteration of the stochastic evolution operator . there seems to be no straightforward correspondence between the two sets of steady - state weights . this suggests that there should be a proof of the invariance of the current which does not rely on identifying the weights of individual configurations .

we present a comparison of halo properties in cosmological simulations of collisionless cold dark matter ( cdm ) and self - interacting dark matter ( sidm ) for a range of dark matter cross sections . we find , in agreement with various authors , that cdm yields cuspy halos that are too centrally concentrated as compared to observations . conversely , sidm simulations using a monte carlo-body technique produce halos with significantly reduced central densities and flatter cores with increasing cross section . we introduce a concentration parameter based on enclosed mass that we expect will be straightforward to determine observationally , unlike that of navarro , frenk & white , and provide predictions for sidm and cdm . sidm also produces more spherical halos than cdm , providing possibly the strongest observational test of sidm . we discuss our findings in relation to various relevant observations as well as sidm simulations of other groups . taking proper account of simulation limitations , we find that a dark matter cross section per unit mass of is consistent with all current observational constraints . 3m_^-3 200r_200 200v_200

astro-ph0006218

the cold dark matter ( cdm ) family of cosmological models provides an excellent description of a wide variety of observational results , from the earlier observable epochs detected via microwave background fluctuations to present - day observations of galaxies and large - scale structure . a concordance model " with roughly one - third matter and two - thirds vacuum energy , either a cosmological constant or quintessence ( @xcite ) , is consistent with almost all current observations on scales @xmath2 mpc ( @xcite ) . recently , improving observations and numerical techniques have enabled a comparison of cdm scenarios to observations on galactic scales of @xmath3 few kpc . the results have not been encouraging . there are a number of distinct observations that may be in conflict with predictions of cdm : * the density profile of galaxies in the inner few kiloparsecs appears to be much shallower than predicted by numerical simulations ( navarro , frenk & white 1996 , hereafter @xcite ) . for density profiles characterized by @xmath4 as @xmath5 , cdm predicts @xmath6 with little scatter ( @xcite 1999 ) , while current h@xmath7 observations suggest @xmath8 with significant scatter ( @xcite ; @xcite , though see @xcite ) . * the central density of dark matter halos is observed to be @xmath9 roughly independent of halo mass ( @xcite ) , while cdm predicts halos with @xmath10 at dwarf galaxy masses , increasing to larger masses ( @xcite 1999 ) . * the number of dwarf galaxies in the local group is an order of magnitude fewer than predicted by cdm simulations , with the discrepancy growing towards smaller masses ( @xcite ; @xcite ) . * hydrodynamic simulations produce galaxy disks that are too small and have too little angular momentum , yielding a tully - fisher relation whose zero - point is off by several magnitudes from observations ( @xcite ) . * the robustness of rapidly rotating bars in high surface brightness spiral galaxies implies lower density cores than predicted by cdm ( @xcite ) . * cluster cl 0024 + 1654 is nearly spherical with a large , soft core , while cdm typically predicts triaxial clusters with cuspy cores ( @xcite , though see miralda - escud 2000 for a counterexample ) . each piece of evidence taken individually is perhaps not convincing enough to claim that cdm has failed on galactic scales . for instance , until recently there was controversy amongst simulators regarding inner profiles ( @xcite ) , but more careful simulations have converged on a consistent prediction ( @xcite ) . observationally , inner galactic profiles are uncertain due to beam smearing effects in observations ( @xcite ; @xcite ) , though samples of high - resolution h@xmath7 observations continue to show shallower profiles than predicted by cdm ( @xcite ) . the number of observed local group dwarf galaxies may be reconciled with cdm via plausible scenarios for suppressed galaxy formation ( e.g. @xcite ) , or else compact high - velocity clouds could represent the missing satellites " that are seen in @xmath0-body simulations ( @xcite ) . hydrodynamic simulations of disk galaxy formation are fraught with the usual concerns about the effects of feedback , artificial viscosity and resolution , though it appears the discrepancies above are due to the underlying dark matter distribution ( @xcite ) . it becomes more interesting to consider alternatives to conventional cdm when one recognizes that all these discrepancies may be symptomatic of a single cause : _ dark matter halos in cdm simulations appear to be more centrally concentrated than observed . _ recognizing this , various authors have recently forwarded a plethora of alternative dark matter theories that suppress the central concentration of dark matter in galaxy halos . among such theories are that the dark matter is warm ( @xcite ; @xcite ; @xcite ) , repulsive ( @xcite ) , fluid ( @xcite ) , fuzzy ( @xcite ) , decaying ( @xcite ) , annihilating ( @xcite ) , and the alternative we investigate here , self - interacting ( sidm ; spergel & steinhardt 2000 ) . interestingly , all theories may be tuned to solve the problems mentioned above ( at least in analytic approximations ) , all theories may be motivated from particle physics considerations , and all theories retain the desirable properties of cdm on extragalactic scales ( though warm dark matter is non - trivially constrained by this requirement ; see @xcite ) . sidm is governed by a single free parameter , the cross section per unit mass @xmath11 of the interacting dark matter particle . @xcite suggested @xmath12 in order to reduce the central concentration of galaxy halos by a sufficient amount to alleviate the above problems . intriguingly , this value is close to the cross section of ordinary hadrons , motivating some particular particle physics candidates for sidm ( steinhardt et al . , in preparation ) . if @xmath11 is significantly smaller than this range , then the optical depth at galactic densities is much less than unity , implying that sidm would have a negligible effect on the dark matter distribution in halos . a qualitative picture of the evolution of an sidm halo is as follows : at early times there is no difference between sidm and cdm since the densities and peculiar velocities are sufficiently low that collisions are rare ; hence sidm makes identical predictions to cdm regarding cosmic microwave background fluctuations and the lyman alpha forest . as the halo forms and grows via gravitational instability , the central density increases . eventually , collisions are so frequent that dark matter particles scatter out of the center as fast as they are accreted , and the density growth is halted , forming a core . such a limit is not present in the cdm model , where the central density grows unchecked . the sidm core then begins to extend while retaining constant central density . heat transfer from the outer parts of the halo raises the temperature in the halo core . if the halo is truly isolated , then eventually the core thermalizes with the exterior resulting in an isothermal halo with a steep density profile . this initiates gravothermal collapse , where the direction of heat transfer is reversed and the exterior begins to cool the halo center . however , in a realistic cosmological setting , galaxies constantly accrete material , keeping the outer halo hot and heat flowing inwards , thus delaying core collapse . the interplay between collisional heat transfer and accretion determines whether a halo will undergo core collapse in a hubble time . it is important to appreciate that the transport behavior does not change monotonically with @xmath11 . for small cross - sections , heat transfer increases with @xmath11 since the frequency of collisions increases ; however , for large cross - sections , the conductivity @xmath13 and no heat transfer occurs . thus , as we discuss in [ sec : comparison ] , the fluid approximation is a poor decription of sidm in the moderate cross - section regime proposed by spergel and steinhardt . furthermore , the behavior in the moderate cross - section regime can not be surmised by interpolating between the fluid and the non - interacting cdm regimes . a proper treatment of the sidm proposal , which includes the interplay of accretion and heat transfer , its non - monotonic dependence @xmath11 , and the effects of merging demands numerical simulations designed to explore the moderate cross - section regime . in this paper we investigate the statistical properties of halos in sidm and cdm in cosmological @xmath0-body simulations . our spatial and mass resolutions are sufficient to probe the inner regions ( @xmath14 ) of small halos ( @xmath15 few@xmath16 ) , while maintaining sufficient volume so as to have a significant sample of such halos . we use a monte carlo technique similar to @xcite ( 2000 ) to model collisions . the primary difference between our simulations and prior investigations ( discussed in more detail in [ sec : comparison ] ) is that we model a cosmologically - significant random volume of the universe with self - interaction cross sections in the range favored by @xcite , enabling us to characterize the statistical properties of halos as we vary @xmath11 . [ sec : sim ] describes our initial conditions and simulation techniques using a monte carlo @xmath0-body approach . in [ sec : structure ] we compare the structural properties of halos in cdm versus sidm models with several cross sections . in particular we examine their central densities , inner profile slopes @xmath7 , the mass dependence of @xmath7 , concentrations , phase space densities , and ellipticities , and where possible compare to observations . in [ sec : res ] we use lower resolution simulations to test the effects of finite particle numbers in our monte carlo method . in [ sec : subhalo ] we examine the subhalo population around the largest halo in our simulations . in [ sec : comparison ] we compare our findings to the simulations of other groups who have conducted numerical studies of self - interacting dark matter , and examine results from a wider range of @xmath11 . we summarize our results and discuss observational constraints in [ sec : disc ] . we find that sidm with @xmath1 produces halos that are in better agreement than collisionless cdm for a wide variety of observations . we use a modified version of gadget ( @xcite ) , a publically - available treesph code for distributed - memory parallel machines . here we only employ the gravitational @xmath0-body portion . we evolve a @xmath17 randomly - chosen volume of a @xmath18cdm universe with @xmath19 , @xmath20 , @xmath21 , and @xmath22 , similar to the concordance model " in agreement with a wide variety of observations ( @xcite ) . we generate initial conditions using cosmics ( @xcite ) at @xmath23 , where our particle distribution first becomes nonlinear , and evolve to @xmath24 . we employ @xmath25 dark matter particles in each run , resulting in a dark matter particle mass of @xmath26 , and a spline kernel softening of @xmath27 ( i.e. force is newtonian at @xmath28 ) . to test resolution effects , we also run a suite of simulations with @xmath29 particles and @xmath30 . their initial conditions have an initial density field identical to the @xmath25 runs , constructed by sampling at alternate grid points . while our @xmath17 box is small , well below the nonlinear scale at @xmath24 , we are interested here in the behavior on scales of a few kpcs , and it is unlikely that the missing large - scale power would have a significant effect on the inner portions of halos . in addition , our primarily conclusions are based on a _ comparative _ study between collisionless and collisional dark matter for individual halos , so we expect these results to be robust to volume effects . lc|cccc|cccc & & @xmath31 & @xmath32 & @xmath33 & @xmath34 & @xmath31 & @xmath32 & @xmath33 & @xmath34 cdm & 0 & 670 & @xmath35 & 1.95 & 8.0 & 111 & @xmath36 & 1.91 & 6.7 sidm & @xmath37 & 647 & @xmath38 & 0.22 & 5.6 & 106 & @xmath39 & 0.36 & 4.8 sidm & @xmath40 & 566 & @xmath41 & 0.030 & 2.6 & 89 & @xmath42 & 0.027 & 1.8 observed & & & @xmath43 & @xmath44 & & & @xmath43 & @xmath44 & we have modified gadget to include self - interactions using a monte carlo @xmath0-body technique to probabilistically incorporate collisions , along the same lines as @xcite ( 2000 ) and @xcite , closer to the latter as we use @xmath45 from individual particles colliding rather than setting @xmath45 to be the particle s velocity ; see the discussion in @xcite . each pair of particles with positions and velocities @xmath46 and @xmath47 , separated by @xmath48 and @xmath49 , interact with a probability given by @xmath50 where @xmath51 is the timestep , @xmath52 and @xmath53 where @xmath54 is the cubic spline kernel used in gadget . this geometrical factor weights the probability of interaction by the product of spline kernel - weighted density distributions of the two particles at their given separation . the normalization @xmath0 is set by requiring that @xmath55 which ensures that when a particle has interacted with all its neighbors within @xmath28 , the resulting probability is equivalent to @xmath56 where @xmath57 is the local dark matter density . in our code , the scatterings are performed between individual particles at the time that the acceleration between those particles is being computed ( i.e. during the treewalk " ) . in order to ensure that all possible scatterings are considered , a tree cell is opened whenever it is within @xmath28 of a particle , regardless of the opening criterion . if two particles scatter , their velocities are randomly re - oriented , keeping the magnitudes of their velocities fixed . in practice , a running sum is kept of the change in velocity due to the interactions that a given particle undergoes on every processor , and at the end of the step the velocity change for each particle is summed over all processors and added to that particle s velocity . in this way , energy and momentum are explicitly conserved , even if the scattered particles are on different processors , or a particle undergoes more than one scatter in a single timestep ( which is very rare for the cross sections considered here ) . we consider @xmath58 ( collisionless ) , @xmath37 , and @xmath59 . we also examined @xmath60 and @xmath61 in a @xmath29 simulation , which we will examine in [ sec : comparison ] . the total number of collisions per particle in our simulations are 1.01 for @xmath62 , and 6.05 for @xmath63 , with slightly lower numbers ( 0.9 and 5.3 ) for the @xmath29 runs . note that a factor of ten increase in @xmath11 translates only to a factor of six increase in the number of collisions , since the lowered central densities ( [ sec : rhoc ] ) partially compensates for the increase in @xmath11 . all runs were performed on fluffy , a 32-processor beowulf - class machine at princeton , with each @xmath25 run taking approximately one week . we identify dark matter halos using skid ( spline kernel interpolative denmax ; see @xcite ) , with a linking length of @xmath28 . we only consider halos containing 64 or more particles , to ensure a roughly complete sample of such halos in our simulations ( @xcite ) . table 1 lists the number of halos for identified in these simulations . a specific resolution issue arises from the finite number of particles used to probabalistically model collisions in the sidm simulations : the number of particles in a given halo must be high enough to properly monte carlo sample the distribution . as we will show in [ sec : res ] , halos with @xmath64 particles at @xmath24 seem to be accurately represented with this technique for the simulations considered here . this is quite restrictive , but still permits a significant sample of halos ( roughly 30 in each @xmath25 run ) with which to compute statistics . we also use the full sample of halos to examine certain aspects , but we will be cautious about interpretations made from halos below this monte carlo resolution limit " . we determine halo profiles @xmath65 by spherical averages over radii @xmath66 , in 20 equal intervals of @xmath67 . a sample of 16 halo profiles from our @xmath25 simulations is shown in figure 1 . each panel shows a halo profile for @xmath58 ( solid ) , @xmath37 ( short dashed ) , and @xmath59 ( long dashed ) . dotted line segments emanating from the innermost radius point of the @xmath58 curve indicate slopes @xmath68 and @xmath69 for comparison . the same corresponding halos are chosen from each simulation , allowing a case - by - case comparison of the effect of sidm . the outer halo profiles ( @xmath70 ) are virtually identical for each halo , showing that the effects of self - interactions are limited to the inner few kpcs of halos , and confirming that the same halos are being compared in the different simulations . the halos in the leftmost column are the four most massive ones in our simulations , while the halos in other columns are chosen randomly from a descending range of masses . note that rightmost column shows halos with roughly 300 particles each , below our monte carlo resolution requirement of @xmath64 particles ( i.e. @xmath71 ) , thus the effects of self - interactions are not necessarily accurately represented in these cases . from figure 1 it is immediately evident that sidm produces halos that have enlarged central cores and shallower inner profiles . cdm halos are almost all cuspy ( @xmath72 typically ) , while most @xmath63 cores are close to flat . @xmath62 leads to profiles that are intermediate between these two . in some cases , non - cuspy cdm halos are seen , especially at lower masses . in these cases the halo may have undergone recent merging activity that temporarily lowers the central density , which is particularly effective in smaller mass halos . additionally , recent mergers that have not relaxed make it difficult to unambiguously identify the halo center about which to compute profiles , typically making profiles appear shallower . we make no cut in regards to the merging history or isolatedness " of halos , but we do note that the missing large - scale power in our simulations will tend to generate fewer mergers , and make the largest objects in our simulations appear more isolated . + figure 1 : 16 selected halo profiles for collisionless ( solid ) , @xmath73 ( dashed ) , and @xmath59 ( dot - dashed ) . corresponding halos from each simulation are presented , allowing a direct comparison of the effect of sidm on a halo - by - halo basis . the total halo mass in @xmath74 for the @xmath58 halo is shown in the lower right ; the sidm halo masses are typically within 20% . the columns are ordered by mass , with the four highest mass halos shown in the leftmost column . dotted lines from innermost point show reference slopes of @xmath75 . 0.1 in at high masses , the effect of sidm is very prominent . the upper leftmost halo is milky way - sized ( @xmath76 ) , and shows a large core of @xmath77 for @xmath63 ( @xmath78 for @xmath62 ) . the core size and difference in inner slope become less prominent to lower masses , though this could be due to the increasing effects of unrelaxed halos , as well as the monte carlo resolution issues discussed earlier . we examine these issues quantitatively in [ sec : alphamass ] . no evidence is seen for sidm halo profiles that are isothermal , as would be expected if the cross section was so large that core collapse would occur on timescales significantly shorter than a hubble time . this supports the analytic estimates of @xcite that core collapse on a hubble time would not occur until @xmath79 . overall , there is a clear trend on a case - by - case basis that sidm results in a reduced central density and shallower inner slope of the dark matter halo , with increasing @xmath11 having a greater such effect . figure 2 shows the central density of dark matter halos @xmath80 , taken to be the density at our innermost resolved radius @xmath81 , as a function of halo mass . here we only consider halos with more than 1000 particles , where our monte carlo technique has sufficient numbers to represent the collisional behavior ( as we will discuss in [ sec : res ] ) . the central halo density of galaxies is observed to be @xmath82 ( @xcite ) , and is consistent with being independent of halo mass . the observed range of halo densities is shown as the hatched region , with a majority of their data falling towards the lower end of that region . the arrow in the upper left indicates the increase in @xmath80 projecting the profile from @xmath83 in to 500 pc , typical of observations of dwarf and low surface brightness galaxy central densities , using the slope shown . sidm @xmath63 halos are in good agreement with these observations , while @xmath62 produces inner densities that are a few times higher , but still marginally consistent with observations . in addition , @xmath80 in sidm models show little trend with halo mass , in agreement with observations , because the core density is set by collisional physics . + figure 2 : @xmath80 vs. @xmath84 for cdm ( left panel ) and sidm ( right panel ) . for sidm , crosses show @xmath63 , while open squares show @xmath62 . only halos with 1000 or more particles are shown . the hatched region indicates the range of observed @xmath80 compiled by @xcite . arrow in upper left indicates how much each value of @xmath80 would increase if measured at 500 pc ( instead of @xmath83 ) , typical of observations , assuming a profile with the slope shown . 0.1 in conversely , the more massive halos in cdm have central densities that are too high by at least an order of magnitude already at @xmath83 , and because of their cuspy profile the disagreement would be much worse at smaller radii , as indicated by the arrow in the upper left . moreover , cdm halos have central densities that increase with mass , in conflict with observations . as such , sidm halos appear to agree better with observations . table 1 lists the median central density of halos with more than 1000 particles in our various models . this shows that our simulations reproduce the observed central halo densities for @xmath85 . we estimate the inner halo profile slope @xmath7 as the slope between the innermost resolved radii , @xmath86 . figure 3 shows a histogram of this slope for the collisionless ( solid line ) , @xmath73 ( dashed ) , and @xmath59 ( dot - dashed ) cases , for all halos that have more than 1000 particles . the number of such halos in each simulation is indicated in the legend . the qualitative impression from figure 1 that sidm produces shallower inner profiles is quantified in figure 3 . the median values of @xmath7 are indicated by the arrows from the upper @xmath87-axis , and are listed in table 1 . cdm produces halos that have cuspy cores , with @xmath88 . this is consistent with the work of @xcite , among others . 25 of the 28 cdm halos have @xmath89 , indicating that cuspy cores are a common feature of cdm models . conversely , the inner slopes in sidm models are significantly shallower . for @xmath62 , @xmath90 , while for @xmath63 , @xmath91 , with no halos having @xmath89 . this latter case has a median @xmath7 close to the value preliminarily suggested by h@xmath7 observations of low surface brightness galaxies ( @xcite ) , though a definitive value awaits a more thorough analysis of observational biases . + figure 3 : histogram of inner slopes @xmath7 for collisionless ( solid ) , @xmath73 ( dashed ) , and @xmath59 ( dot - dashed ) . only halos with @xmath92 ( @xmath93 particles ) are included ; the number of such halos is indicated in parantheses in the legend . the median values of @xmath7 are indicated by the arrows from the top edge of the plot , and are listed in table 1 . 0.1 in the scatter in @xmath7 is mostly real . there is some scatter due to discreteness effects in measuring the inner slope as we do . while we could obtain an inner slope from fitting a general halo shape ( @xcite ; @xcite ) , with 5 free parameters the inner slope would be poorly constrained by 20 correlated data points , thus we choose our simpler definition . further scatter arises from recently merged halos that temporarily have shallower profiles until relaxed . however , neither of these effects is very significant for the large mass halos plotted in figure 3 . still , we choose to quote the median @xmath7 rather than the mean , in order to quantify typical " halos in these models and reduce sensitivity to outliers , although the mean is similar . we note that a significant scatter in inner slopes is also seen in the observations ( e.g. @xcite ; @xcite ) . our cdm profiles are , at face value , in better agreement with the analytic profile of @xcite ( 1999 ) , with an asymptotic slope of @xmath94 , rather than an @xcite profile having @xmath95 . however , profile fitting is a tricky business ( as discussed in @xcite ) . by reducing the scale radius in the @xcite profile ( i.e. increasing the concentration ) , one can push the transition to a slope of @xmath96 to a radius smaller than @xmath83 where we can not resolve the profiles ( the cusp - core degeneracy " ; see @xcite ) . thus we suspect that our cdm profiles can also be adequately fit by an @xcite profile having a large concentration parameter . as such , we do not argue for or against either profile form . our simulations can only predict the slope at @xmath97 , and that is what should be compared to observations . sidm appears to be in better agreement with observations of the inner slopes of dark halo profiles than cdm . at face value , @xmath98 is preferred , but given uncertainties in observations and simulation techniques , @xmath99 is probably also consistent . a similar value of @xmath11 also reproduces the observed central density of galaxies . such a coincidence is not expected _ a priori _ , and may represent a significant success of the sidm scenario . figure 4 shows a plot of inner halo slope @xmath7 vs. halo mass @xmath100 for all halos in our cdm ( left panel ) and sidm ( right panel , @xmath63 ) simulations . the curve shows the running median value of @xmath7 in bins of @xmath101 . + figure 4 : @xmath7 vs. @xmath84 for cdm ( left panel ) and sidm ( @xmath63 ; right panel ) . all halos with 64 or more particles are shown . the line shows a running median of the @xmath7 distribution , with a variance computed from all halos within each mass bin . 0.1 in the cdm case shows almost no trend with mass , with the median slope always around @xmath102 . the largest halo does have a slightly shallower slope , consistent with the trend seen in very high resolution cdm simulations of individual halos ( @xcite ) . the scatter increases to low mass due to discreteness and merging effects described in [ sec : alpha ] . the sidm case shows dramatically different behavior , suggesting at face value that smaller halos have steeper inner profiles . however , caution must be used in interpreting this result . first , smaller halos have smaller cores , meaning that a slope measured at a fixed radius ( not scaled to the halo core size ) will result in a steeper slope . for cdm this effect is less significant , since the slope remains similar from the outer to the inner halo . second , the small numbers of particles in the low - mass halos makes the monte carlo technique less effective in modeling collisions , thereby making sidm appear more like cdm ; we investigate this issue further in [ sec : res ] . thus we make no claim regarding a trend of @xmath7 with @xmath84 . as seen in figure 1 , sidm appears to have the desired effect of reducing the concentration of dark matter halos . in this section we quantify this effect using a concentration parameter , which we define differently than previous authors in order to facilitate a more direct comparison with observations . the canonical definition of a concentration parameter is given by @xcite as the ratio between the virial radius @xmath103 ( taken to be the radius at which the halo density is @xmath104 the cosmic mean ) and the scale radius of the halo @xmath105 in the @xcite profile . this concentration parameter , however , is difficult to compute unambiguously in the case of non - isolated halos , and difficult to compare directly to observations that seldom extend out to @xmath103 . furthermore , @xmath105 is only defined within the context of the specific @xcite model , and profile fits are typically degenerate between @xmath105 and concentration ( @xcite ) . coln et al . ( 2000 ) circumvent some of these issues by defining the concentration as the ratio of the minimum of @xmath103 and the halo radius to the radius that encloses 20% of the halo mass . however , this inner radius is dependent on knowing the total halo mass , something which is difficult to determine observationally . instead , we choose to define a _ mass concentration parameter _ @xmath106 , based on enclosed mass rather than radii , and restrict the scales in our definition to those where observations are available , typically @xmath107 . we define @xmath108 with @xmath109 where @xmath110 is the circular velocity of the halo . the choice of @xmath111 is arbitrary ; here we base it on the milky way , as it is convenient and results in observationally accessible scales . the scaling with @xmath110 is that expected for self - similar halos following the tully - fisher relation . the normalization factor of 27 results in a uniform density distribution having a mass concentration of unity . a flat rotation curve between @xmath111 and @xmath112 implies @xmath113 , resulting in @xmath114 . in our simulations , we take @xmath110 to be the maximum circular velocity of the halo as output by skid . + figure 5 : histogram of mass concentrations @xmath106 for collisionless ( solid ) , @xmath73 ( dashed ) , and @xmath59 ( dot - dashed ) . only halos with @xmath92 ( @xmath93 particles ) are included . the median values of @xmath106 are indicated by the arrows from the top edge of the plot , and are listed in table 1 . 0.1 in figure 5 shows a histogram of @xmath106 for all halos with more than 1000 particles . as expected , there is a clear trend for cdm to have more concentrated halos than sidm , with the amount of concentration decreasing with increasing @xmath11 . note that the difference between sidm with @xmath73 and @xmath59 is exaggerated relative to the difference between the inner slopes of those models ( cf . figure 3 ) . this is because the concentration is increased in @xmath73 relative to @xmath59 both due to the increased inner slope , as well as the reduced core radius . we also examined the mass dependence of @xmath106 and found no obvious trend , but our range of masses is small . for comparative purposes , our mass concentration parameter @xmath106 may be analytically related to the @xcite concentration parameter @xmath115 . from @xcite , @xmath116 where @xmath117 , @xmath118 is the circular velocity at @xmath87 , and @xmath119 is the circular velocity at @xmath103 . our @xmath110 is taken to be the maximum halo circular velocity , which may obtained by maximizing equation [ eqn : vc ] ; this occurs at @xmath120 ( though we compute it exactly for the results shown below ) . let @xmath121 . in appropriate units , @xmath122 ( see @xcite , equation a2 ) . thus @xmath123 implying @xmath124 and @xmath125 are solely functions of @xmath115 ( note that this arises because we defined @xmath126 ) . using @xmath127 , @xmath128 which is purely a function of @xmath115 . the resulting relationship is shown in figure 6 . + figure 6 : mass concentration parameter @xmath106 , defined in equation [ eqn : cm ] , vs. @xcite concentration parameter @xmath115 . 0.1 in figure 6 shows that @xmath129 , typical of halos in our cdm model , corresponds to @xmath130 . this value is in agreement with expectations for dwarf galaxies in a @xmath18cdm model . conversely , @xmath131 , which is the median value for sidm with @xmath73 , corresponds to @xmath132 . note that the minimum value of @xmath106 for an @xcite halo is 3 . thus sidm with @xmath59 , with @xmath133 , produces halos that typically can not be described properly by @xmath115 . this is because these halos have @xmath134 typically , so @xcite profiles with @xmath95 are a poor fit . this further illustrates why @xmath115 is a poor way to describe halos in general . the largest halo in our simulations has a mass comparable to the milky way s , @xmath135 . the concentrations of this halo are 6.7 , 4.4 and 2.3 in cdm , @xmath62 and @xmath63 , respectively . @xmath106 of the milky way halo is somewhat uncertain , because of the uncertainty in the rotation curve outside the solar circle ( @xmath136 ) and the effect of baryonic mass within @xmath136 , but we make a rough estimate here . if the rotation curve is flat , @xmath114 as stated before . there are suggestions that the rotation curve rises somewhat beyond the solar circle ( though this is uncertain ; see @xcite ) , in which case @xmath106 is reduced ; let us take @xmath137 as a working estimate . the rotation curve measures the total mass , so we must correct for the baryons to compare with our simulated @xmath106 . if we take the fraction of baryonic mass to be 50% inside @xmath136 and 20% inside @xmath138 , then @xmath106 reduces to 5 . in addition , baryons adiabatically compress the dark matter as they dissipate , so we must correct the milky way @xmath106 further downwards to compare to our dissipationless halos . from the analysis of @xcite , this reduction factor is @xmath139 , resulting in the milky way halo having @xmath140 . thus after reasonable corrections , the milky way mass concentration appears to be in better agreement with sidm than cdm . the rapid rotation of bars also suggests a lower concentration for the milky way - sized galaxies than that predicted by cdm ( @xcite ) . a more direct comparison with simulations may be obtained from rotation curves of dark matter - dominated , low surface brightness galaxies , where baryonic corrections are smaller . we expect that this mass concentration measure @xmath106 will be relatively straightforward to compute from such rotation curves ( e.g. @xcite ) , so we look forward to comparisons . @xmath106 has the advantage that it is independent of halo fitting parameters , as the enclosed mass can be obtained directly from the observed circular velocity with modest assumptions . in this sense , it is a more robust comparison than the inner slope and the @xcite concentration parameter , which are degenerate and sensitive to scales outside those typically observed ( @xcite ) , and the central density , which depends on an uncertain contribution from baryons . a recently popularized measure of the concentration of dark matter halos is the central phase space density . @xcite find that observed phase space densities @xmath141 scale as @xmath142 , where @xmath143 is the velocity dispersion , from dwarf spheroidals up to clusters of galaxies . observations compiled by @xcite suggest a similar relation , albeit with a large scatter , and he uses them to argue against _ any _ form of collisionless dark matter ( though see @xcite ) . figure 7 shows the phase space density @xmath144 of dark matter within @xmath111 as a function of @xmath143 , for halos with @xmath145 . we calculate @xmath143 as the velocity dispersion around the group center of mass velocity , within @xmath111 ( cf . equation [ eqn : rin ] ) . open circles show cdm halos , crosses indicate sidm ( @xmath63 ) halos . the dashed and dotted lines show @xmath146 and @xmath147 , respectively , that bracket the observations , reproduced from @xcite . + figure 7 : phase space density @xmath144 vs. velocity dispersion @xmath143 for cdm ( open circles ) , and sidm with @xmath63 ( crosses ) . dashed and dotted lines bracket observations , showing scalings of @xmath146 and @xmath147 , respectively . 0.1 in sidm generally shows lower phase space densities than cdm . sidm is in somewhat better agreement with observations , falling in the middle of the observed range . measuring @xmath144 in galaxies is a difficult task , because the stellar velocity dispersion is not necessarily that of the dark matter . furthermore , in rotationally supported galaxies the dubious assumption of an isothermal spherical halo is used to relate circular velocity to dispersion . thus @xmath144 is perhaps not among the most useful observational discriminants between cdm and sidm . an interesting remark from figure 7 is that the scaling of @xmath148 is roughly the same in both models , roughly @xmath146 . @xcite argue that such a scaling results from the dynamical assembly of halos , and is not expected based on simple phase packing arguments . this further motivates simulations of sidm that include the cosmological growth of halos via dynamical processes of merging and accretion . sidm produces halos that are more spherical than cdm , because of the isotropic nature of the collisions ( @xcite ) . this is a generic feature of sidm with _ any _ significant cross section , since in the inner portions of halos where collisions are frequent , the velocity ellipsoid is quickly isotropized . thus the shapes of dark matter halos provide an important observational discriminant between cdm and sidm . we compute axis ratios of our halos using the prescription outlined in @xcite . they define a tensor @xmath149 where @xmath150 is the elliptical radius , and @xmath151 are the axis ratios , and the sum is over all particles with distances @xmath152 from the halo center along the axes of the ellipsoid . then , @xmath153 where @xmath154 are the eigenvalues of @xmath155 . @xmath156 represents the axisymmetry of the halo , while @xmath157 measures the halo flattening . since @xmath150 depends on @xmath156 and @xmath157 , the calculation of @xmath158 must be iterated until convergence , which we take to be better than 0.01 in @xmath156 and @xmath157 . this scheme weights particles roughly equally regardless of distance from center , unlike a moment of inertia tensor which weights the outskirts heavily , and thus better represents the ellipticity of the density distribution , as shown in @xcite . + figure 8 : axisymmetry @xmath156 and flattening @xmath157 vs. @xmath159 for a @xmath160 halo ( left panel ) and a @xmath161 halo ( right panel ) . solid lines are cdm , dashed lines are sidm , @xmath62 , and dot - dashed lines are sidm , @xmath63 . lower of two curves is @xmath157 . 0.1 in figure 8 shows axis ratios as a function of radius in our two most massive halos , having masses @xmath135 ( left panel ) and @xmath162 ( right panel ) . solid line is the cdm halo , dashed line is sidm with @xmath62 , and dot - dashed line is sidm with @xmath63 . @xmath156 is the upper of two curves for a given model . cdm halos are fairly triaxial , while sidm produces halos that are much closer to spherical . the effect is dependent on radius , as in the outer regions sidm and cdm become more similar , since the effect of self - interactions is confined to the inner parts of halos . still , even at @xmath163 there are significant differences between sidm and cdm . while not stated , this trend with radius is also evident from figure 1 of @xcite . figure 9 shows histograms of axis ratios @xmath156 and @xmath157 at @xmath164 ( top panels ) and @xmath165 ( bottom panels ) for all halos with more than 1000 particles . the median value for each model is indicated by the corresponding tick mark on the top axis . the difference between cdm and sidm is more pronounced at small radii , where cdm produces significantly triaxial halos while sidm halos remain spherical . at large radii there is a much milder trend to more spherical halos with increasing @xmath11 . @xmath157 also shows more differences than @xmath156 . figure 9 shows that while cdm produces halos are typically more spherical , there is still significant non - sphericity in many sidm halos . in particular , there is a tail in the distributions of both cdm and sidm to smaller axis ratios . this may be due to asymmetric infall that temporarily distorts the shape of the density in some halos , particularly smaller ones . this also may just be an artifact of finite number of collisions in smaller halos . note that the two largest halos shown in figure 8 show greater differences at @xmath165 than suggested by the statistics in figure 9 . + figure 9 : histogram of axis ratios @xmath156 ( left panels ) and @xmath157 ( right panels ) , at @xmath164 ( upper panels ) and @xmath165 ( lower panels ) . only halos with more than 1000 particles are included . tick marks at the upper axis show median values . 0.1 in a comparison with observations of halo shapes is as yet inconclusive . in the inner portions of dark halos the shape of the potential is likely to be dominated by baryons , so a comparison to these simulations is not straightforward . further out , perhaps the most direct observations of axisymmetry are those for galaxies with hi rings such as ic 2006 , which suggests a very axisymmetric halo , @xmath166 at @xmath167 kpc ( @xcite ) . other observations ( see @xcite ) are more dependent on observational and theoretical uncertainties such as viewing angle and potential modeling , but persistently suggest @xmath168 at @xmath169 kpc . both cdm and sidm halos are consistent with these observations . lensing maps of galaxies and clusters offer the best hope for mapping the mass potential in the inner halos , which should place strong constraints on sidm . conversely , observations of @xmath157 from polar ring galaxies ( e.g. @xcite ) and x - ray isophotes ( e.g. @xcite ) suggest a substantial amount of flattening , @xmath170 at @xmath171 kpc in the density distribution . such a flattening , if confirmed , may prove troublesome for sidm . the baryonic component would provide a flattened contribution , but is not expected to be significant at those radii . it is not immediately evident how these discrepancies may be resolved , but we note that the problem is almost as severe for cdm as sidm in our simulations . it is worth mentioning that our small simulation volume results in significantly reduced tidal distortion of large halos , so our simulations may not accurately represent the ellipticities of the outer portions of halos . our spatial resolution and mass resolution are well - understood . however , another resolution issue arises due to the monte carlo modeling of self - interactions . a monte carlo method must be sufficiently well sampled , resulting in a separate criterion for the number of particles in a halo to be well - represented by our simulation technique . in this section , we determine this criterion using our suite of lower resolution simulations with @xmath29 particles described in [ sec : sim ] . since we are most concerned with the inner parts of halos , we focus on the inner slope as a function of mass as the best measure for examining this monte carlo resolution limit . figure 10 shows a plot similar to right panel of figure 4 , except for the @xmath29 simulation of the @xmath63 sidm model . the top axis shows the number of particles in these halos . here we compute @xmath7 at @xmath164 since that is the spatial resolution of our @xmath29 runs . the dashed line is the running median @xmath7 from the @xmath25 simulation , computed at @xmath164 . figure 10 shows that for halos with @xmath64 particles in the @xmath29 run , the median value of @xmath7 is within @xmath172 of that of the @xmath25 run , though consistently lower . by 300 particles , the value of @xmath7 is significantly lower in the @xmath29 run . the reason it is lower is because with few particles , the monte carlo procedure results in too few interactions to make the profile depart significantly from the collisionless cdm case . thus to lower masses , sidm looks increasingly like cdm when modeled using this technique . + figure 10 : @xmath7 vs. @xmath84 for sidm with @xmath63 , for our @xmath29 run . @xmath7 here is measured at @xmath164 , the resolution of our @xmath29 runs . solid line shows a running median of the @xmath7 distribution . dashed line with error bars shows a similar curve from @xmath25 run . deviations between two at a level @xmath173 occur for halos having somewhere between 300 and 1000 particles . 0.1 in note that this limit is specific to our simulation parameters , redshift , and @xmath11 , and is not a general statement about the monte carlo @xmath0-body technique . the limit becomes higher as @xmath11 is lowered , since collisions become less frequent , but even for @xmath62 , @xmath174 particle halos also appear convergent . we therefore take our monte carlo resolution limit to be @xmath64 particles . we list the values of @xmath32 and @xmath34 for the @xmath29 runs in table 1 for comparison with the @xmath25 results , where the median values here are computed for all halos with more that 500 particles in these smaller runs ( roughly 10 in each simulation ) . note that @xmath32 is computed at @xmath164 instead of @xmath83 , partially explaining the steeper slopes even at the highest masses . in general , the trends indicated by the @xmath25 runs are reproduced at this lower resolution , suggesting that discrete particle effects do not significantly affect our conclusions . we have also examined the @xmath25 statistics presented previously using a limit of 300 particles instead of 1000 , and our overall conclusions remain the same . self - interacting dark matter is predicted to significantly lower the population of subhalos orbiting around large halos , thereby bringing simulation predictions into better agreement with observations of the local group dwarf population . there are two reasons why sidm has this effect : ( 1 ) the lowered central concentration and larger core radius makes small halos more susceptible to tidal disruption , and ( 2 ) dark matter is ram - pressure stripped out of small galaxies as they move through the large central halo . in our @xmath25 simulations , we have one halo that is roughly milky way sized , having @xmath175 . in this section we examine the subhalo population around this large halo . + figure 11 : subhalo positions within @xmath176 the largest object in our volume , in cdm ( left panel ) and sidm with @xmath63 ( right panel ) . central cross is the position of the large halo . 0.1 in figure 11 shows a projected plot of halos within @xmath176 of the largest halo in our volume , indicated by the central cross . circle sizes are scaled as @xmath177 , with the smallest circles representing halos with @xmath178 . left panel shows cdm , right panel shows sidm with @xmath63 . the positions of subhalos are different due to the accumulated differences of chaotic orbits within a highly nonlinear potential well . thus a halo - by - halo comparison for these small halos is not possible . a careful examination reveals that sidm has fewer subhalos than the cdm distribution , especially the smallest ones . + figure 12 : number of halos within @xmath176 of the largest halo , histogrammed by mass , for collisionless ( solid ) , @xmath73 ( dashed ) , and @xmath59 ( dot - dashed ) . vertical dotted line indicates our 1000-particle monte carlo resolution limit . 0.1 in figure 12 quantifies this effect , showing the mass function of halos within @xmath176 of our largest halo , for our three @xmath25 simulations . there is a clear trend that sidm suppresses the subhalo population at the smallest masses . for @xmath179 , cdm has 56 neighboring halos , @xmath62 has 40 , and @xmath63 has 29 . while sidm does reduce the population of smallest halos , the reduction is not nearly by the order of magnitude required to obtain agreement with local group dwarf galaxy counts ( @xcite ) . however , the effect of subhalo suppression in these sidm simulations should be regarded as a lower limit to the true strength of the effect . the reason is that almost all these subhalos are well below our monte carlo resolution limit ( dotted line in figure 12 ) , therefore their concentrations are approaching those in cdm models . thus tidal disruption of these halos is not much stronger in sidm as in cdm , and is increasingly similar to lower masses . furthermore , ram - pressure stripping is reduced in effectiveness for the same reason that the monte carlo technique is less effective in these small halos . hence the numbers of small sidm halos are not as significantly suppressed relative to cdm as they should be . we conclude that the simulations considered here suggest a weak trend in reducing the number of subhalos with increasing @xmath11 , but due to resolution effects we can make no robust quantitative estimates . what is required is to simulate a large halo with incredibly high resolution , having subhalos containing thousands of particles to properly model the effects of self - interactions . such a simulation is unfortunately beyond the scope of our current computational resources . alternatively , more sophisticated algorithms are necessary to model self - interactions in small halos moving through large ones , which is an avenue we are currently pursuing . a number of authors have investigated sidm using @xmath0-body simulations . the literature divides into two subsets : those that model interactions in the fluid approximation , effectively employing a large cross section , and those that model self - interactions in the optically thin regime as suggested by @xcite . in both cases , there is disagreement over whether sidm makes halos less or more concentrated than cdm . @xcite and @xcite simulate a galaxy cluster within a cosmological context , using a treecode with sph to model interactions in the fluid approximation . both studies resulted in halos that had isothermal profiles and were more centrally concentrated than cdm halos . this may be because the large effective cross section increases heat transfer efficiency , though as mentioned in the introduction , for a sufficiently large cross section one expects heat transfer to be diminished . @xcite suggested that intermediate cross sections would likely yield results that were intermediate between steep cdm profiles and steeper isothermal profiles , and thereby argued against sidm . however , this is contradicted by @xcite , as well as the results presented here , confirming that the intermediate case results in halos that have long - lived shallow profiles . in contrast , bryan ( private communication ) uses a adaptive mesh hydrodynamics code to model self - interactions in a cosmological volume , and finds that even in the fluid limit sidm produces sizeable , long - lived cores . it is not clear why different hydrodynamic codes give different results when they should be operating in the same regime . perhaps the effective cross section is larger in the adaptive mesh code due to algorithmic differences , reducing heat transfer . another possibility is that a numerical effect in sph in which cold clumps moving through hot halos have their drag significantly overestimated ( @xcite ) makes objects rapidly sink into a dense , isothermal core . it is beyond the scope of this paper to resolve these issues . we simply note that the highly optically thick limit is not the relevant scenario to test the cross section range proposed by @xcite . @xcite ( 2000 ) and @xcite ( 2000 ) simulated isolated halos with sidm having a cross section closer to the range of @xcite . they begin with a fully formed cuspy galaxy halo and study the evolution after interactions are turned on . they both find that halos develop a shallow core for some length of time , and then undergo core collapse . @xcite and @xcite disagree on the timescales of core collapse ; @xcite finds @xmath180 in agreement with estimates from two - body relaxation , while @xcite finds a much shorter collapse timescale of @xmath181 for the same dark matter cross section . @xcite explain this difference by arguing that @xcite s method underestimates collisions of slow moving particles . we note that our method does not suffer from this concern , as it is more like @xcite . we suggest some possible reasons why the results of @xcite and ( to a lesser extent ) @xcite are at odds with ours . the first is that they begin with a cuspy hernquist profile . this halo evolves rapidly initially ( as seen in figure 3 of @xcite ) , resulting in an artificially large amount of heat transfer . second , they simulate an isolated halo , ignoring the accretion of dynamically hot material during the formation process that would keep the outer halo hot and delay core collape . we note that their dimensionless cross section @xmath182 converts to ours by a factor of @xmath183 for @xmath184 and @xmath185 ( cf . figure 1 ) . @xcite s simulation with @xmath186 already produces halos that maintain cores over many dynamical times ( cf . top panel of their figure 2 ) . we suggest that a somewhat smaller @xmath187 might be consistent with observations as well as their limits on core collapse timescales , even without considering the effects of accretion and merging . @xcite also point out , as we have , that the monte carlo @xmath0-body technique requires a large number of particles for accurate modeling , and show that @xmath188 particles is sufficient . we note that while most of our halos do not have that many particles , our largest ( milky way sized ) halo has roughly @xmath189 particles , and its properties are consistent with those of smaller halos . @xcite have now performed a cosmological cluster simulation with a cross sections @xmath190 using a monte carlo @xmath0-body method . they find shallower central slopes and less concentrated cores with sidm . our combined results span the range from dwarf galaxies to clusters , and are in broad agreement with each other if we focus purely on the numerical results . for instance , they find that @xmath191 produces a cluster having a core size of @xmath192 . if we extrapolate our largest halo s core size ( in our @xmath63 run ) to cluster scales using the expected @xmath193 , we predict a core size of @xmath194 for their cluster . for @xmath195 , they get @xmath196 core , while we would predict @xmath197 for @xmath62 , again in good agreement . they then proceed to scale down from clusters to dwarf galaxies using a simplistic argument based on number of collisions , but their scaled result is contradicted by our direct simulations . we estimate that this may be because they use a cdm value for the scale radius and @xmath115 of dwarfs , and compare them to sidm values for their cluster . we find that @xmath62 produces @xcite scale radii that are double that of cdm ( cf . figure 6 and discussion ) ; such a factor would go a long way towards alleviating the discrepancy . taking this into account , we find the simulations of @xcite to be broadly consistent with ours . + figure 13 : halo profile of the largest halo in our @xmath29 simulations , for a range of @xmath11 values . halos are progressively less concentrated and have larger cores with increasing @xmath11 . 0.1 in in order to explore the high-@xmath11 limit , we ran @xmath29 simulations of sidm with @xmath198 . the most illustrative result is to compare the density profile of the largest halo in all our @xmath29 simulations , as shown in figure 13 . as seen in figure 1 , there is a smooth trend of increasing core radius with @xmath11 . sidm with @xmath60 is quite similar to cdm , though it may also have a core below our @xmath164 resolution limit . increasing @xmath11 to @xmath199 , we continue to see no evidence for the development of an isothermal core due to accelerated heat transfer . the reason is because the collisions are so frequent in the outer portion of the halo that a dense core can not develop . instead , collisions randomize the dark matter velocities and prevent a smooth radial inflow required to generate a dense core . as dynamically hot material accretes onto the halo , heat keeps flowing inward and a large core is maintained . our results are in better agreement with bryan as opposed to @xcite and @xcite . this also illustrates why simulating sidm beginning with an isolated cuspy hernquist profile may not be appropriate for large @xmath11 ; one should at least begin with a halo profile that is self - consistently stable for a few dynamic times . we present a set of cosmological self - interacting dark matter simulations having cross - sections in the range favored by @xcite ( 2000 ) . our simulations include the growth of halos from linear fluctuations in a random volume of the universe , with sufficient volume and resolution to obtain a statistical sample of galactic halos resolved to @xmath83 . we compare the resulting halos on a case - by - case basis to those in a collisionless cdm simulation having the same initial conditions . 1 . the inner slopes of sidm with @xmath200 typical halos have @xmath201 at @xmath202 , with some scatter in @xmath7 . our cdm halos have @xmath6 , in agreement with previous studies ( e.g. @xcite 1999 ) . sidm with @xmath203 is intermediate between these cases , with median @xmath204 . sidm is in better agreement with a preliminary analysis of h@xmath7 rotation curves of low surface brightness galaxies ( @xcite ) . 2 . sidm with @xmath200 produces central densities @xmath205 at @xmath83 , and shows no trend with halo mass . sidm with @xmath206 has somewhat higher @xmath80 values , but remains fairly independent of mass . conversely , @xmath80 in cdm halos is much larger than observed , typically @xmath207 at @xmath83 , and shows a strong trend with halo mass . with their steep profiles , cdm halos are in significantly worse agreement at smaller radii . sidm is thus is in better agreement with observations , as has also been argued by @xcite . 3 . simulations with sidm having @xmath206 are intermediate between cdm and sidm with @xmath200 , indicating a smooth increase in the effect of sidm with cross section , a result that extends ( using lower - resolution simulations ) from @xmath208 . in particular , the generation of singular isothermal halos is not seen in any of the massive halos simulated , even for @xmath209 . this suggests that the dynamical process of halo growth in a cosmological setting helps keep outer regions of halos hot and prevents core collapse in a hubble time . we introduce a new mass concentration parameter @xmath106 based on a more directly observable quantity , the enclosed mass within tens of kpc . this halo concentration is significantly lower in sidm models as compared to cdm , providing an observationally accessible discriminant that is not dependent on fitting a particular profile form . a rough estimate of @xmath106 for the milky way , with large corrections for baryonic effects , favors sidm over cdm . the central phase space density is lower in sidm vs. cdm mostly due to the reduction in @xmath80 . the velocity dispersions in the inner regions are quite similar . both sidm and cdm are consistent with observations shown in @xcite , though sidm is mildly favored . sidm produces halos that are more spherical , especially in their inner regions , as compared to cdm . in principle , this is one of the strongest tests of the sidm paradigm , as near the center _ any _ value of @xmath11 that has a non - negligible effect on the dark matter distribution will increase the core sphericity , while cdm cores are almost always significantly triaxial . however , baryons are likely to dominate the shapes of the inner parts of halos , complicating a direct comparison , and in the outer parts the differences between sidm and cdm are less pronounced . the number of subhalos around our largest ( milky way - sized ) halo is somewhat reduced with increasing @xmath11 , but due to discreteness effects in our monte carlo @xmath0-body technique , we can not put robust quantitative estimates on the strength of this effect . based on these simulations , our currently favored value for @xmath11 is somewhere between @xmath59 and @xmath73 . such a cross section simultaneously reproduces both the observed central density and inner slope , as well as being consistent with various observations considered here , which is non - trivial . in contrast , for instance , warm dark matter has difficulty simultaneously reproducing the observed central densities , inner slopes and subhalo population ( @xcite ) . as stated before , inner halo shapes may provide a strong discriminant between cdm and sidm . on galactic scales , they are difficult to observe and confused by baryonic contributions . conversely , clusters provide a cleaner test because they have large cores that are not baryon - dominated , and their mass distributions are directly observable via lensing . @xcite uses the asphericity of cluster ms 2137 - 23 to ( analytically ) argue that @xmath210 , effectively ruling out sidm as a solution to halo concentration problems . on the other hand , cl 0024 + 1654 is very spherical , much more so than cdm models generally predict ( @xcite ) . our simulations can not directly address the shapes of clusters , as we have no cluster - sized objects in our volume . however , sidm shows some range of halo shapes due to asymmetric infall and unrelaxed mass distributions , so it is unclear whether a single object can definitively rule out sidm . support for this statement is provided by @xcite , whose cluster has enough triaxiality to be consistent with ms 2137 - 23 even for @xmath195 ( their model s1wb ) , contradicting miralda - escud s scaling argument . we note that halo shapes are unaffected by annihilating ( @xcite ) or decaying ( @xcite ) dark matter , thus they also provide a discriminant between these variants and sidm . the cluster core sizes in the simulations of @xcite are larger than observed ( @xmath211 , @xcite ; @xcite ) , certainly for @xmath191 ( @xmath192 ) , and probably even for @xmath195 ( @xmath196 ) . so it may be that sidm has difficulty matching observations at both dwarf galaxy and cluster scales . however , adiabatic contraction of baryons during the formation of the cd galaxy will reduce the cluster core radius from @xmath0-body predictions , so the discrepancy may not be that large . in any case , sidm with @xmath212 comes remarkably close to matching dwarf galaxies , @xmath213 galaxies and clusters given the @xmath214 range in mass scales , so we reserve judgement pending a more careful comparison with observations . @xcite mention that sidm core sizes would be in better agreement with observations if @xmath215 , which would result in the effects of self - interactions being diminished in hot cluster environments as compared to galaxies . such a scenario occurs naturally if the dark matter dark matter scattering has low - lying resonance or bound state contributions , as is the case for ordinary nucleons . another theoretical avenue explored in relation to sidm has been modeling the tully - fisher relation . hydrodynamic simulations show that the simulated tully - fisher zero point may be brought into agreement with observations only if halos are less centrally concentrated than predicted by cdm ( @xcite ) . @xcite determine that a cross section of @xmath216 would produce a correct tully - fisher relation for @xmath217 km / s halos . thus it is conceivable that the cross section preferred from halo structure constraints may also alleviate the tully - fisher discrepancies . @xcite suggested that dark matter interactions in the centers of halos naturally produce central black holes with a mass scaling @xmath218 , in agreement with observations ( @xcite ; @xcite ) , and he estimates @xmath219 in order to avoid central black holes that are too large . however , this estimate is based on an @xmath220 profile all the way in to the black hole . if such a dense center never arises because collisions inhibit its formation , then the limits on @xmath11 are weakened considerably . observations of dark matter halos promise to improve significantly in the coming years , particularly constraints on halo core shapes from lensing and on inner profiles and concentrations of halos from h@xmath7 rotation curves . if the inner parts of dark matter halos are found to be generically triaxial , this would be the high place of sacrifice for self - interacting dark matter ; conversely , spherical halos would provide strong support for this scenario . the main modeling work yet to be is done is an improved examination of the subhalo populations in sidm , as well as simulations of a larger range of mass scales . the @xmath0-body monte carlo approach has difficulty achieving a large dynamic range due to the stringent monte carlo resolution limit ( i.e. discreteness effects in the probabalistic description of collisions ) , therefore a different approach may be necessary . the sidm simulations presented here are a first attempt at examining the effect of self - interacting dark matter within the context of a realistic halo formation scenario . the results are quite encouraging that this simple variant of the cold dark matter paradigm will alleviate a wide range of difficulties faced by cdm on galactic scales . we look forward to further investigations and comparisons with observations . we thank julianne dalcanton , lars hernquist , jerry ostriker , penny sackett , scott tremaine , martin white , and naoki yoshida for helpful discussions . we thank greg bryan for sharing his unpublished results . weinberg , d. h. , dav , r. , gardner , j. p. , hernquist , l. , & katz , n. 1999 in `` photometric redshifts and high redshift galaxies '' , eds . r. weymann , l. storrie - lombardi , m. sawicki & r. brunner ( sf : asp conf series )

we analyze hst / ghrs spectra of ab doradus , the prototypical ultra - rapidly rotating k dwarf . we observed chromospheric ( ) and transition region ( , , , and ) lines periodically throughout the stellar rotation period , and provide a low dispersion stellar atlas of 78 emission lines . the quiescent line profiles of the chromospheric and transition region lines show narrow cores superposed on very broad wings . the broad wings of the & lines and of the transition region lines can be explained by emission from gas co - rotating with the star and extending out to near the keplerian co - rotation radius ( 2.8 stellar radii ) . while this is not a unique solution , it is consistent with previous studies of h emission that are naturally explained by large co - rotating prominences . we find no evidence for rotational modulation of the emission line fluxes . the density diagnostics suggest that the transition region is formed at constant pressure , with an electron density 23 10 at a temperature of 3k . the electron pressure is about 100 times larger than that for the quiet sun . the emission measure distribution shows a minimum between log(t ) = 5 and 5.5 . the line exhibits three interstellar absorption components along the 15 pc line of sight . we identify the lowest velocity component with the g cloud , but the other components are not identified with any interstellar clouds previously detected from other lines of sight .

astro-ph0012487

we study magnetically - active solar - like stars because the characteristics of such stars may provide insights into the nature of solar - like magnetic activity . while we accept the solar paradigm , that stellar activity is a consequence of magnetic fields , it is not at all clear that any simple scaling of solar coronal structures can account for the level of activity seen in the most active stars . indeed , the very assumption of solar - like structures may bias our interpretation of the observations . walter & byrne ( 1997 ) , and walter ( 1999 ) proposed a non - solar paradigm for active stars , based in part on observations of ab doradus . in this picture , a quasi - dipolar global magnetic field may dominate the large - scale activity , while the hotter coronal gas is confined by solar - like magnetic structures to a small scale height . an active stellar atmosphere would then consist of a compact , solar - like chromosphere / transition region / corona and an extended , co - rotating envelope of large volume . ayres et al . ( 1998 ) posit a similar picture for the magnetospheres of active hertzsprung gap giants . ultraviolet spectra sample the chromosphere and transition region , at temperatures from about 10@xmath6k through 2@xmath310@xmath7k . while emission in the cooler lines may result in part from heating by acoustic fluxes , the transition region line emission is almost certainly from hot plasma trapped in magnetic loops . time - resolved spectra obtained over a stellar rotation can yield important constraints on magnetic filling factors , global asymmetries , and atmospheric scale heights . velocity - resolved line profiles can be inverted to yield spatial resolution on these stars , and images of their surfaces . our goal , through detailed observations of highly active stars , is to test our understanding of stellar coronae and chromospheres , of the morphology of the magnetic field , and of its relation to the observed activity . ab doradus ( hd 36705 ; k0 - 2 iv v ) , the brightest ( v=6.7 ) of the ultra - rapid rotators ( p@xmath8=0.51479 d ; @xmath9 sin @xmath10=91 km s@xmath11 ) , is the quintessential active young single star . in addition to large flares ( e.g. , robinson & cameron 1986 ) , large starspots ( anders et al 1992 ) , and large coronal and chromospheric fluxes ( pakull 1981 ; vilhu , gustafsson , & walter 1991 ) , ab dor also possesses co - rotating material at 2@xmath125 stellar radii ( cameron & robinson 1989 ) , in the form of cool prominences or h@xmath2 clouds ( cameron et al . its brightness and activity levels make it an ideal target to study the extremum of stellar magnetic activity . among the recent studies are those by rucinski et al . ( 1995b ) ; mewe et al . ( 1996 ) ; krster et al . ( 1997 ) ; schmitt , cutispoto , & krautter ( 1998 ) vilhu et al . ( 1998 ) , and ake et al . ( 2000 ) . the parallax ( guirado et al . 1997 ) places the star at 15 pc , slightly above the zero - age main sequence ( cameron & foing 1997 ) , with a probable age close to that of the pleiades . ab dor is a member of a multiple star system . the dme4 star rossiter 137b ( ab dor b ) is a common proper motion companion at an angular distance of 10 arcsec ( innis et al . 1985 ; innis , thompson , & coates 1986 ; vilhu et al . guirado et al . ( 1997 ) detected a low mass astrometric companion , ab dor c , with an inferred separation of a few tenths of an arcsec . r 137b is outside the ghrs aperture ; ab dor c falls within the aperture , but is unlikely to make any significant contribution to the observed flux . we observed ab dor with the goddard high resolution spectrograph ( ghrs ; brandt et al . 1994 ; heap et al . 1995 ; robinson et al . 1998 ) aboard the hubble space telescope , as part of a multiwavelength campaign in november 1994 ( walter et al . 1995 ) . our goal was to obtain simultaneous spectroscopic and photometric observations at x - rays , uv , optical , and radio wavelengths over at least one stellar rotation period , in order to correlate the coronal , chromospheric , and optical behaviors , and to come up with a 3-dimensional picture of the atmosphere of a very active star . the optical spectroscopy and photometry , and the doppler images , have been reported by cameron et al . ab dor behaved normally ( for ab dor ) . there was a prominent photometric wave , indicating a highly asymmetric starspot distribution . we observed no large flares ( i.e. , flares with durations in excess of about an hour ) . cameron et al . ( 1999 ) report a number of strong absorption events due to cool material ( extended prominences ) co - rotating at high altitude . the doppler image ( cameron et al . 1999 ) shows a dark feature at high latitudes , with some low latitude spottedness . vilhu et al . ( 1998 ) report a continuous ghrs observation of the line which immediately preceeded our observations . we will report on the variability of the chromosphere and corona in a subsequent paper ( walter et al . , in preparation ) . here we present an analysis of the quiescent chromosphere of ab dor as viewed over one stellar rotation period by the ghrs instrument on the hst . we observed ab dor ( program 5181 ) with the hst / ghrs on 1994 november 14 - 15 while the target was in the continuous viewing zone . the observations began near the end of stellar rotation cycle 10440 . our strategy was to maximize spectral and temporal coverage by rapidly changing gratings to obtain a sequence of exposures during the 12.3 hour rotation period of ab dor . for 12 hours we alternated between 256 second integrations of the lines , using the echelle - b ( r@xmath13 100,000 ) , and 1228 second integrations of transition region lines using the g160 m first order grating ( r@xmath1420,000 ) . this strategy was driven by the one dimensional format of the ghrs detectors ; at r=20,000 we observe only about 35 at a time . the main observation was split into 6 sequences . each sequence began with a peakup in the large aperture , to ensure that the target remained well - centered . each sequence then consisted of 3 observations of the line alternating with observations of the @xmath151393,1402 doublet , the @xmath151548,1551 doublet , and either the @xmath151238,1242 doublet , the @xmath151334,1335 doublet , or the density - sensitive ] , ] @xmath151892,1908 intersystem lines . we concluded these six sequences with a final exposure sandwiched between two observations . all these observations utilized the d2 detector . at the end of these sequences , we changed to the d1 detector and observed the full short - wavelength spectrum ( 1150 - 1750 ) at low dispersion in two g140l exposures . the details of the exposures are given in table [ tbl - obslog ] . four observations were lost when the carrousel failed to lock . each science observation was preceeded by a spectral calibration lamp exposure which established the accurate wavelength scale . occasional spybal observations of the wavelength calibration lamps were also interspersed by the scheduling software . the target was observed through the large science aperture ( lsa ) using substep pattern 5 . we calibrated the raw data using the ghrs team software procedure calhrs ( blackwell et al . each observation is a single spectrum , but each transition region spectrum consists of 4 independent spectra , providing about 5 minute temporal resolution . the analysis of the line profiles ( excluding the ism analysis in section [ sec - mgism ] ) was carried out using the icur software package . all the measurements ( fluxes , centroids , gaussian fits ) use standard techniques . mewe et al . ( 1996 ) fit the euve and asca spectra , and found a best fit absorption column of [email protected] @xmath310@xmath17 @xmath18 . since the reddening correction is less than 1% , we do not apply any reddening corrections . the summed low dispersion spectrum is shown in figure [ fig - g140l ] . two spectra centered at 1305 and 1571 cover almost all the useful wavelength range of the g140l grating ( 1162 - 1714 ) with 20 of overlap in the middle . the resolution of this spectrum , r@xmath142000 , is adequate to resolve most astrophysically - important blends , including ( marginally ) the @xmath191334,1335 doublet . we obtained this spectrum primarily for the purpose of constructing the line atlas ( table [ tbl - lines ] ) , and to determine the emission measure distribution ( section [ sec - emd ] ) of the chromosphere of this very active star . the spectrum shows a wealth of detail , including weak lines of neutral and low - ionization species , the strong transition region lines , and the coronal line . we have identified the emission lines using the solar line lists of burton & ridgeley ( 1970 ) and feldman et al . other possible identifications are found in the chianti ( dere et al . 1997 ) database . the brightest stellar emission line is @xmath191640 ( the h i lyman @xmath2 emission is geocoronal ) . identifications for some of the weaker lines and blends are uncertain , and will require higher dispersion spectra for definitive identification . there are 63 lines in table [ tbl - lines ] with secure identifications . the mean difference between the observed and tabulated ( rest ) wavelengths is [email protected] , corresponding to a radial velocity of 33@xmath165 km s@xmath11 , in good agreement with the previously - measured photospheric value of 30 km s@xmath11 . much of the scatter lies , as expected , in the weaker lines . the mean radial velocity of the 34 strongest lines ( fluxes @xmath201.0@xmath310@xmath21 erg @xmath18 s@xmath11 ) is also 33@xmath165 km s@xmath11 . this agreement with the photospheric radial velocity verifies both that the wavelength scale is accurate and that the line identifications are most likely correct . figure [ fig-1900 ] shows the mean spectrum in the @xmath191900region . the two g160 m observations were obtained at rotational phases 0.336 and 0.800 , and sample opposite hemispheres of the star . the @xmath191900 continuum is 46% brighter at phase 0.8 . the contemporaneous @xmath22 and @xmath23 band photometry ( rucinski , garrison , & duffee 1995a ; cameron et al . 1999 ) shows that the star was faintest at about rotation phase 0.5 , and that in the @xmath22 band ab dor is about 15% brighter at phase 0.8 than at phase 0.34 . there is no significant change in the emission line fluxes between the two observations . in addition to the expected ] @xmath191892 and ] @xmath191908 intersystem lines , we see four narrow lines of and a line of ( table [ tbl-1900 ] , figure [ fig-1900 ] ) . the possible feature at 1907.3 ( 1907.1 rest wavelength ) is unidentified . the ] @xmath191908 line is much broader than the other lines ( see section [ sec - dens4 ] ) . the line widths listed in table [ tbl-1900 ] are measured by fitting gaussians to the lines . a line broadened by the stellar rotation would have a fwhm of approximately 0.8 in this spectral region ( assuming a 40% overestimate of @xmath9 sin @xmath10 from the gaussian fit : see discussion in section [ sec - trp ] ) . we monitored the transition region doublets of , , , and . the and lines were each observed six times , and exhibited variability . each observation consisted of four @xmath14 5 minute integrations , which permits us to examine the data for line profile variations or flaring on timescales of 5 to 20 minutes . we examined the individual integrations for evidence of flaring ( rapid flux variations over the 4 integrations , varying line profiles , or fluxes significantly in excess of the median ) . we generated quiescent line profiles by summing those spectra which had fluxes near the median , symmetric line profiles , and no evidence for flaring . in figure [ fig - tr ] we compare the 4 quiescent line profiles . here we discuss only the quiescent line profiles ; we will discuss the flaring spectra elsewhere . we initially fit the emission lines as gaussians or sums of gaussians because this approach is commonly used and the resulting fits are generally quite acceptable . vilhu et al . ( 1998 ) found that the emission lines could be fit as a sum of narrow and broad gaussian components ( as found in general by wood , linsky , & ayres 1997 ) . the narrow components are significantly broader than the expected rotational broadening , and vilhu et al . interpreted this nonthermal broadening in terms of solar - like nonthermal broadening mechanisms ( e.g. , dere & mason 1993 ) , perhaps related to microflares . unlike the case of the sun , we do not know much about the spatial distribution of the emitting gas . it probably is not confined to a thin region , given the evidence for extended prominences , in which case the atmosphere is extended and interpretation of line profiles is not straightforward . we simulated line profiles from photospheres ( with limb darkening parameters @xmath121.0 @xmath241.0 ) and from optically - thin extended co - rotating atmospheres , convolved them with the expected thermal broadening and with a gaussian instrumental response , and fit the lines as gaussians . in the case of @xmath25=0.6 we recover the well - known rotational profile ( e.g. , gray 1992 ) ; @xmath25=0.0 is a uniform disk , and @xmath260 yields limb - brightened profiles . transition region lines are more likely optically thin emission from an extended atmosphere . we also examined cases where the emission is confined to low latitudes . limb - brightened atmospheres tend to produce flat - peaked line profiles ( for modest extents of @xmath2710% of the stellar radius ) , whereas limb - brightened atmospheres with emission confined to low latitudes produce emission cusps for sufficiently large @xmath9 sin @xmath10 . although the intrinsic line profiles are not gaussian , significant noise and low resolving power allow gaussian fits to be acceptable in many cases . the widths of the gaussian fits always exceed @xmath9 sin @xmath10 . we find that for photospheric line profiles , the gaussian fits overestimate @xmath28sin @xmath10 by up to 40% , depending on the value of @xmath25 and the intrinsic @xmath28sin @xmath10 . gaussian fits to limb - brightened profiles ( containing a few thousand counts ) overestimate the @xmath28sin @xmath10 by 40 - 80% . dere & mason ( 1993 ) showed that optical thickness will also tend to broaden the line profiles ; in slightly thick but effectively thin lines , photons from the line core will be scattered into the wings . we caution the reader that the interpretation of emission line profiles is highly model dependent . consequently , we do not interpret the width of the narrow emission component of the transition region lines . in figure [ fig - mg2s ] we show the subordinate ( 3p @xmath29p@xmath30 3d @xmath29d@xmath31 ) line at @xmath19 2791.6 , which is almost certainly optically thin . the profile is flat - topped , and at resolution r@xmath1410@xmath32 the shape of the line profile is clearly not gaussian . the rotation profile for a limb - darkened photosphere is much more sharply peaked . figure [ fig - mg2s ] demonstrates that this line can be produced by a limb - brightened , optically thin atmosphere with a height of a few percent of the stellar radius . this fit is not unique : within the uncertainties one could also fairly well reproduce the profile with , for example , a rotating photospheric profile ( @xmath25=0.6 ) with no emission above about 30@xmath33 latitude , or a uniform disk profile ( @xmath25=0.0 ) with emission confined below about 60@xmath33 latitude . the strong resonance lines of the transition region clearly can not be fit with a single component . instead we fit the lines as sums of narrow and broad gaussians . we find that we get better fits with three components : a broad component and two similar narrow gaussians displaced from line center by about @xmath16 0.5 @xmath9 sin @xmath10 . this model gives better fits because the peaks of the and are flatter than a gaussian profile . physically , this is because in an extended limb brightened atmosphere the centroid of the emission is displaced to higher velocities where the path length through the atmosphere is longer . we also have had success in fitting the line wings with emission from an optically thin , extended , rigidly - rotating atmosphere ( see section [ sec - trd ] ) . [ [ section ] ] we identified 17 of the 24 integrations as likely to represent the non - flaring , quiescent transition region . the flux ( the sum of the fluxes in the @xmath151548,1551 lines ) during quiescence is less than 1.5@xmath310@xmath34 erg @xmath18 s@xmath11 . vilhu et al . ( 1998 ) noted that the non - flaring flux level just prior to our observation ranged between 1.0 and 1.5@xmath310@xmath34 erg @xmath18 s@xmath11 , and the average flux during our entire observation is 1.2@xmath310@xmath34 erg @xmath18 s@xmath11 . the @xmath151549/1551 line ratio of [email protected] is close to but less than the optically thin limit . the summed quiescent line profile ( figure [ fig - civ ] ) should represent the phase - averaged transition region . although we can fit the line with a sum of two gaussians with fit parameters similar to those of vilhu et al . ( 1998 ) , we find that the data are better fit with three components . we plot neither the 2- or 3-component gaussian fits to this or any line in figures [ fig - civ ] to [ fig - mg2k ] . rather , the fits to the high velocity wings shown in figure [ fig - civ ] are for optically thin , constant density atmospheres extending out 2.5 stellar radii . forty percent of the total flux is in the extended component . [ [ section-1 ] ] the mean @xmath191393 line flux is 2.2@xmath310@xmath35 erg @xmath18 s@xmath11 . from a visual examination of the line profiles and the line fluxes , we identified 20 quiescent spectra , and summed these ( figure [ fig - siiv ] ) . the @xmath151393/1402 line ratio of [email protected] is identical to that in the lines . although close to the optically thin limit , both line ratios are significantly less than 2 . this probably indicates that at least some of the emission is from optically thick regions . because the lines are well separated and because the blue wing of the @xmath191402 line is blended with ] @xmath191401.34 , we fit only the @xmath191393 line . as with the lines , the line profile is better fit with a broad component centered on the rest velocity of the star plus two narrow emission components than it is with a two - gaussian model . if we fit the high velocity wings as emission from extended gas ( see figure [ fig - siiv ] ) , we find that the emission extends out to 3 stellar radii , which is comparable to what we found for . about 25% of the emission is in the extended component . [ [ section-2 ] ] we obtained a single g160 m observation of the doublet at phase 0.5 . the line fluxes are 95% of those seen in the low dispersion spectrum . the line ratio of [email protected] is consistent with the line s being optically thin . there is no evidence for the broad component seen in the and lines , perhaps because of the low s / n in this integration . however , there is a small flux excess on the red side of the @xmath191242 line . this is not seen on the blue side of the line , nor is it obvious in the @xmath191238 line . the cause could be a downflow , or perhaps a small flare on the receding limb of the star . [ [ section-3 ] ] we obtained a single g160 m observation of the doublet at phase 0.65 . the emission lines are severely blended , and we were unable to fit the various components uniquely . high velocity wings are evident . there is a strong interstellar absorption feature in the @xmath191334.5line , with a radial velocity of 9@xmath163 km s@xmath11 . given the resolution of the ghrs in this mode , it is not possible to identify the absorption with any particular component of the interstellar medium ( see section [ sec - mgism ] ) . corresponding ( though much weaker ) absorption is seen near the top of the @xmath191335.7 line . this spectrum also shows the @xmath191323.52 , @xmath191329.1 , and @xmath191351.66 lines with fluxes similar to those in table [ tbl - lines ] . the transition region lines of active stars are generally modeled as the sum of broad and narrow gaussians . the width of the narrow component is generally significantly broader than the stellar @xmath9 sin @xmath10 , and this is interpreted as the convolution of the stellar rotational profile with some other broadening mechanism . wood , linsky , & ayres ( 1997 ) showed that among dwarf stars , the excess broadening appears to decrease with increasing surface gravity ( @xmath36 ) . the broad component is often interpreted as evidence for high velocity gas associated with microflaring . we have difficulty interpreting the line profiles of ab dor in this manner . after deconvolving the 90 km s@xmath11 rotational velocity from the narrow central emission component we find that the velocity of the excess broadening mechanism is about 120 km s@xmath11 , which places ab dor well above the trend found by wood , linsky , & ayres ( 1997 ) for less active stars . and while microflaring may indeed exist and contribute to the width of the broad component , the evidence for spatially extended gas leads us to propose a fundamentally different mechanism for the broad line component . we propose that the high velocity wings are due to spatially - extended gas co - rotating with the star , and not to some unknown broadening mechanism . the existence of extended prominences seen in the light of h@xmath2 is well established ; it is likely that this same gas accounts for the extended wings of the @xmath37 lines . the wings of the transition region lines have very similar shapes , suggesting that these lines also have a similar origin . corroborating evidence for extended hot gas hot gas exists : walter et al . ( 1995 ; 2001 in preparation ) show that c iv undergoes absorption events associated with h@xmath2 prominences . we model the extended gas as a simple optically thin volume co - rotating with the star , and assume that the density is uniform between some inner and outer radii . the emission from each point is broadened by a thermal velocity appropriate to the ion in question . the computed profiles do not depend sensitively on the value of the assumed inner radius , since the emission from gas close to the star is hidden by the central emission component , but the outer radius does determine the width of the broad line component . we vary only the outer radius to match the observations . we can also vary the extent of the gas in latitude , but we are not very sensitive to this because gas at high latitudes has smaller velocities , and is buried under the central emission . the inferred radial extent of the extended gas is close to the keplerian co - rotation radius of 2.8 stellar radii . we favor a model in which the gas is trapped by large - scale quasi - dipolar magnetic fields ( walter 1999 ) . such gas will be most stable near the co - rotation radius ( e.g. , the `` slingshot prominences '' of cameron & robinson 1989 ) . assuming that the gas is optically thin and in rigid rotation , we can use the high velocity wings to estimate the mean electron density in the extended gas . the mean electron density n@xmath38 is given by 0.8 @xmath39 , where l is the velocity - resolved luminosity of the line , p is the power in the line , v is the velocity - resolved volume element , and the 0.8 factor accounts for the fact that hydrogen is largely ionized . the mean density is about 5@xmath310@xmath40 @xmath5 at the temperature of the transition region lines for a filling factor of 1 ; the presence of discrete prominences suggests that the filling factor of this extended region is very low , and the density in the prominences must be commensurately larger . we see no evidence for any rotational modulation of the fluxes in the , , or lines . the photosphere is faintest at rotation phase 0.5 ( cameron et al . following solar analogy , the dark regions ( starspots ) are likely regions of larger photospheric magnetic flux and the bright chromosphere and transition region should be located above the starspots . if the chromosphere and transition region are spatially extended , as is likely the case here , then any modulation would be diluted . our data let us place a limit of 5% on the amplitude of the rotational modulation in the chromospheric and transition region lines . we observed the chromospheric @xmath0 & @xmath1 lines 19 times , approximately every 35 - 50 minutes , with a resolving power of about 10@xmath32 . the mean profile of the two lines is presented in figure [ fig - mg2k ] . the @xmath0 & @xmath1 resonance lines are well - exposed , and the subordinate lines are detected as well . both the @xmath0 & @xmath1 lines exhibit extended blue wings , to velocities of about -320 km s@xmath11 with respect to the star . the red wings can not be measured because the @xmath0 line is blended with the subordinate lines , and the red wing of the @xmath1 line falls off the edge of the detector . the brightness ratio near line center is 1.23 , confirming that the line cores are optically thick , and the @xmath0 line has brighter line wings as expected for the higher opacity line . over 10% of the flux in the line core is lost to the narrow interstellar absorption lines ( cf . section [ sec - mgism ] ) . since is formed in the chromosphere , at about the same temperature as h@xmath2 , one might expect to see absorption events in similar to those seen in h@xmath2 ( cameron et al . 1990 , 1999 ) , and indeed we do see such absorption events . these will be discussed elsewhere ( walter et al . 2001 , in preparation ) . the same prominences seen in absorption against the stellar disk should be seen in emission off the limb , and may account for the broad wings of the @xmath0 & @xmath1 lines . the maximum velocity is consistent with emission from gas in co - rotation at heights of up to 3.5 stellar radii . in figure [ fig - mg2k ] we overplot the expected emission line profile , scaled to the data , for a co - rotating atmosphere extending to a height of 3 stellar radii . this is neither a fit to the data , nor it is a unique description of the data , but it demonstrates that the extended wings could arise from an extended atmosphere without assuming non - thermal broadening mechanisms . this extended emission accounts for 20% of the total @xmath0 & @xmath1 emission , after correcting for the interstellar absorption . the mean density in the extended atmosphere at chromospheric temperatures is about 3@xmath310@xmath41 @xmath5 ( for a filling factor of 1 ) . this is about a factor of 6 higher than the density inferred from the transition region lines , and within the uncertainties in consistent with a constant pressure atmosphere . the mean density at a given temperature provides clues to the gross morphology of the stellar atmosphere , since the emission measure ( @xmath42n@xmath38n@xmath43 dv ) depends on both the density and the volume . even at very low spatial resolution , the observed density can help us discriminate between compact high density emitting regions and large extended structures . there are a number of density diagnostics in this region of the spectrum ( e.g. , doschek et al . we use the line strengths in tables [ tbl - lines ] , [ tbl-1900 ] , and [ tbl-1400 ] to compute the density sensitive line ratios , and use the chianti software package to convert the line ratios to densities . we used 3 lines and the sum of the @xmath151294 - 1301 lines to form 4 line ratios . these consistently yield densities between 2 and 3@xmath310@xmath4 @xmath5 at a temperature of about 3@xmath310@xmath6k ( table [ tbl - dens ] ) . cook & nicolas ( 1979 ) discuss using the lines to determine densities . we use the @xmath191175 , 1247 , and 1908 lines in this analysis . the densities are not mutually consistent ( table [ tbl - dens ] ) . cook & nicolas ( 1979 ) note that in the sun the density ratios involving the @xmath191175 line are consistently off , perhaps because the @xmath191175 line intensity is reduced by optical depth effects . an increase of about a factor of two in the @xmath191175 line flux would bring the @xmath44 density into agreement with the @xmath45 density , at about 10@xmath46 @xmath5 . the @xmath47 density would then become about 5@xmath310@xmath46 @xmath5 . using the @xmath48 line ratio , schmitt , cutispoto , & krautter ( 1998 ) and ake et al . ( 2000 ) both infer densities at the high density limit , @xmath2010@xmath46 @xmath5 , which is consistent with our estimates . the densities inferred from the lines ( formed at 6@xmath310@xmath6k ) are smaller than those inferred from ( formed at about 3@xmath310@xmath6k ) : in a constant pressure atmosphere they should have about half the density indicated by the lines . this would be consistent with a model in which the bulk of the transition region emission arises in constant pressure loops . cook & nicolas ( 1979 ) also estimate densities from ratios of ] @xmath191892 and @xmath191402 to ] @xmath191908 . these densities are valid so long as the relative abundances are solar . the @xmath49 ratio has traditionally been used as a density diagnostic in iue spectra ( doschek et al . we find this ratio to be about 1.3 , which suggests a density near 3@xmath310@xmath50 @xmath5 , but the @xmath51 ratio implies a density nearer 3@xmath310@xmath46 @xmath5 , using cook & nicolas model atmosphere ratios . note that the ] @xmath52 and ] @xmath191908 line profiles are very different ( figure [ fig-1900 ] ) , with the ] line nearly twice as broad , suggesting that they might not be formed co - spatially . the @xmath191400 region ( figure [ fig-1400 ] ) includes a number of density - sensitive ] ( cook et al . 1995 ) and ] ( dufton et al . 1982 ) intercombination lines . these lines are much weaker than the nearby resonance lines . the @xmath191399.8 and @xmath191401.2 ] lines are clearly visible in the low dispersion spectrum , blended with the blue wing of @xmath191402.8 . we estimated their fluxes first by fitting all 3 lines , forcing the wavelengths and line widths of the ] lines to their expected values . since the two lines should have identical profiles , we then shifted , scaled , and subtracted the @xmath191393.8 from the @xmath191402.8 line profile , and fit the residuals . the @xmath191407.4 ] line is also visible in the low dispersion spectrum . these fluxes can be found in table [ tbl - lines ] . we also searched for these lines in the summed g160 m spectrum ( total exposure time = 7368s ) . this spectrum is a global average over all rotational phases . although flaring affects the strong resonance line profiles and fluxes , we find no evidence of enhanced emission in the weaker lines during flares . as in the low dispersion image , there is clear evidence for extra emission on the blue wing of the @xmath191402.8 line . using the same multi - line fitting procedure , we measure the line fluxes given in table [ tbl-1400 ] . we note that high dispersion does not improve the flux measurements signficantly , because the lines are over - resolved due to rotational line broadening , and the rotational line broadening causes line blending . we measure the @xmath53 line ratio to be [email protected] , which is marginally consistent with the expected value of 0.993 . due to the poor accuracy of the flux determinations , the density diagnositics are inconclusive . similarly , the line ratios are inconclusive . the @xmath191423.9 line is blended at low resolution with a stronger line . although there is a weak line near @xmath191416.9 in the low dispersion spectrum , no line of similar strength is seen in the higher resolution spectrum . the densities listed in table [ tbl - dens ] range from about 10@xmath54 @xmath5 to 3@xmath310@xmath4 @xmath5 . the densities derived from the four line ratios using the lines are self - consistent but larger than the other diagnostics , while those involving ] @xmath191908 are systematically low . as discussed above , an arbitrary doubling of the @xmath191175 line flux ( justified because it appears to works for the sun ) brings the @xmath47 density into better agreement with the densities ( table [ tbl - dens ] ) , while leaving all the densities based on ] @xmath191908 consistently lower by about an order of magnitude . the systematically lower densities indicated by the ratios involving the ] @xmath191908 line can be reconciled if the ] @xmath191908 line is preferentially formed in the extended , low density region near the co - rotation radius . this would explain both the significant broadening of the ] @xmath191908 line , and the enhanced line strength . the fwhm of the ] @xmath191908 line ( table [ tbl-1900 ] ) corresponds to about 3 @xmath9 sin @xmath10 , and so is consistent with formation in a region close to the co - rotation radius . we believe that the most reliable density estimate is n@xmath38=23@xmath310@xmath4 @xmath5 from the line ratios because of the consistency among the four ratios . the electron pressure at 30,000k would then be p@xmath38=n@xmath38t=69@xmath310@xmath55 @xmath5k , about 100 times the value for the quiet sun . we present an emission measure distribution based on the observed line fluxes in figure [ fig - emd ] . we use the line emissivities tabulated by landini & monsignori fossi ( 1990 ) , interpolated with a cubic spline . landini & monsignori fossi computed the line emissivities in the low density limit . since much of the flux seems to arise in a high density atmosphere , we used the chianti database to determine the relative emissivities at low and high densities , and used this to determine proper emissivities for the few density - sensitive lines . the emission measure curves in figure [ fig - emd ] assume n@xmath38=10@xmath4 @xmath5 . in this diagram , we have overplotted the t@xmath56 extrapolation of the coronal emission measure , for @xmath42n@xmath38n@xmath43dv=3@xmath310@xmath57 @xmath5 at log t = 6.6 ( rucinski et al . 1995b ; mewe et al . we have also added the far uv line fluxes from orfeus ii ( schmitt , cutispoto , & krautter 1998 ) , after multiplying the orfeus ii fluxes by 1.1 to bring their @xmath191175 and @xmath191206 fluxes into agreement with our measurements . the extrapolation of the emission measure of the cooler coronal component is in good agreement with the fluxes of the , , , and lines , especially considering the non - simultaneity of the uv , fuv , and x - ray observations . a number of lines seem to have emission measures above the minimum . the cool c , si , and s lines may either form at temperatures below log(t)=4.2 , or they may derive much of their flux from a large extended volume . three other lines are clearly out of place . * ] @xmath191908 implies an emission measure 1.5 orders of magnitude larger than that of the other lines , which are consistent with the trends from other species . no more than a few percent of the ] @xmath191908 flux can be generated at these high densities due to strong collisional de - excitation ; the remainder of the flux may arise in an extended lower density region . * @xmath191640 is far too strong to be formed in collisional equilibrium at a temperature near 10@xmath32 k ( jordan 1975 ) , given the emission measure at this temperature . this line is more likely formed during recombination following ionization of by coronal x - rays shining down on the lower chromosphere at t@xmath1410@xmath6 k ( zirin 1975 ; laming & feldman 1993 ; wahlstr@xmath58 m & carlsson 1994 ) . * the @xmath191371 line suggests a very small emission measure . the line emissivity in landini & monsignori fossi ( 1990 ) suggests that the line should be an order of magnitude _ brighter _ than is observed . the emission measure is inconsistent with the and line emission measures . note that the strong transition region resonance lines of , , and suggest emission measures somewhat above the minimum values indicated by other ions . we suggest that the excess flux , that above the minimum in the emission measure diagram , may arise in the spatially - extended gas near the co - rotation radius . there is considerable evidence that the sun lies inside a small cloud , called the local interstellar cloud ( lic ) , of warm partially ionized gas moving with a single valued bulk flow velocity of 26 km s@xmath11 from galactic coordinates @xmath59 and @xmath60 , roughly the direction of the sco cen association ( crutcher 1982 ; lallement et al . 1995 ; redfield & linsky 2000 ) . in spectra of nearby stars , narrow interstellar absorption features with velocities equal to the projected velocity of the lic vector are identified as absorption from this cloud along the line of sight to the star . other nearby clouds that have been identified include the g cloud , so named because it lies in the direction of the galactic center ( lallement & bertin 1992 ) , and clouds in the direction of the north galactic pole and south galactic pole ( redfield & linsky 2000 ) . observations of @xmath2 cen and 36 oph indicate that the interface between the lic and the g cloud is located @xmath61 0.19 pc from the sun roughly in the direction of the galactic center ( linsky & wood 1996 ; wood , linsky , & zank 2000 ) . a three dimensional model of the lic was developed by redfield & linsky ( 2000 ) based on 32 lines of sight using hst , euve , and ground based spectra . in their model , the lic extends about 5 pc in the anti galactic center direction , and a fraction of a parsec in the galactic center direction , so that the sun is located just inside the very edge of the lic . for the line of sight toward ab dor , the model predicts the presence of very little lic material . the lines of ab dor contain two broad interstellar absorption features . we fit the ism features in the mean @xmath1 and @xmath0 lines simultaneously to maximally constrain the heliocentric velocity @xmath9 ( km s@xmath11 ) , the doppler parameter @xmath62 ( km s@xmath11 ) , and the column density @xmath63 ( @xmath64 ) . the best fit is illustrated in figure [ mgismfig ] . to estimate the `` continuum '' against which the absorption features are measured , we use observed fluxes on either side of the absorption features and fluxes derived from mirroring the red side of the line together to fit a fifth degree polynomial . this `` continuum '' is shown in figure [ mgismfig ] as the thin solid line . for a given set of interstellar parameters ( @xmath9 , @xmath62 , @xmath65 ) , the profiles for the interstellar absorption features were computed and then convolved with the echelle b lsa instrumental profile ( gilliland & hulbert 1993 ) . in figure [ mgismfig ] the absorption profiles of the individual ism components are shown as dotted lines , and the convolution of these components with the instrumental profile is shown as the thick solid line . a scattered light correction was made by simply subtracting a small flux level across the whole spectrum . numerous fits were made with different estimates of the scattered light correction factors in order to estimate the systematic errors . for the most part , the uncertainties are dominated by the random error . the derived parameters for these interstellar absorption components are listed in table [ mtbl ] . the two broad interstellar absorption features in the lines can be resolved into three different components one with a heliocentric velocity of 5.17 km s@xmath11 and the other two with velocities of 14.5 and 19.6 km s@xmath11 . a two component model inaccurately fits the high velocity feature in the @xmath0 line , so a third component was added to adequately fit the flattened appearance of the high velocity absorption feature between 14 and 19 km s@xmath11 . since the quality of fit does not improve significantly with the addition of a fourth component , the data do not require more than three components . with a galactic longitude of @xmath66 and a latitude of @xmath67 , the projected lic velocity toward ab dor should be 4.25 km s@xmath11 and the corresponding g cloud velocity should be 5.29 km s@xmath11 . because the projected velocities of the lic and g cloud are so similar in value , it is difficult to interpret the observed 5.17 km s @xmath11 component as being associated with only one or the other cloud . the fit itself seems to favor the g cloud projected velocity vector , and based on the lic model and observations of other nearby stars ( @xmath2 cen and 36 oph ) we expect very little lic material along this line of sight . the lic model predicts a hydrogen column density of @xmath68 = 16.88 ( redfield & linsky 2000 ) . using a typical magnesium depletion for the lic of d(mg ) @xmath14 1.1 @xmath16 0.2 from piskunov et al . ( 1997 ) , the predicted column density contribution of the lic in the direction of ab dor is only @xmath63 @xmath69 , where @xmath70 . therefore , the lic should be a small contributor to the observed 5.17 km s@xmath11 interstellar absorption component . however , if this absorption component is associated with the g cloud , the large doppler parameter , @xmath71 km s@xmath11 , implies a large turbulent velocity . if we assume a temperature typical for the g cloud , @xmath72 , based on observations of @xmath2 cen and 36 oph ( linsky & wood 1996 ; wood , linsky , & zank 2000 ) , the turbulent velocity is then @xmath73 , where we have used the equation @xmath74 with @xmath75 for magnesium . other turbulent velocities measured for the g cloud by linsky & wood ( 1996 ) and wood , linsky , & zank ( 2000 ) have significantly lower magnitudes . this high turbulent velocity could be caused by a shearing of the gas , possibly near the interface of the cloud . however , it is more likely that because the lic and g cloud projected velocities are so close , we are unable to differentiate between them in the spectrum . both clouds may be contributing to the broad absorption feature , leading to the appearance of a large doppler parameter and turbulent velocity if the absorption is ascribed to only one cloud . the observed 14.5 and 19.6 km s@xmath11 components have not yet been identified with any previously detected cloud . we have presented the overall quiescent characteristics of ab doradus as it appeared during our 1994 november hst / ghrs observations . the transition region and line profiles provide evidence for significant amounts of material co - rotating with the star out to at least 3 stellar radii with temperatures extending from the lower chromosphere to at least 10@xmath32k . this picture supports a model with extended gas confined near the keplerian co - rotation radius ( 2.8 stellar radii ) by large - scale magnetic fields . we surmise that most of the detected emission , the @xmath1460@xmath1280% emitted in the narrow line component , arises in high density , small scale height regions close to the photosphere . the extended material may well be at a lower density , as indicated by the low density derived from the @xmath76}~\lambda 1892}{\rm{c iii]}~\lambda 1908}$ ] line ratio , since both lines are depleted at high density ( with the ] line depleted faster ) . formation of the ] @xmath19 1908 line in the low density extended gas would explain both the broad line profile and the large flux . these quiescent spectra alone contain a wealth of information , and show that high signal - to - noise , high dispersion spectra of magnetically active stellar chromospheres and coronae provide important insights into the spatial morphology of the stellar magnetospheres . in particular , we can determine the emission measure distribution , the atmospheric densities , the spatial extent of the atmosphere , and the properties of the interstellar medium along the line of sight . high signal - to - noise , high dispersion uv spectra of active stars provide our best indication of the conditions from which our present day solar atmosphere evolved . this paper is based on observations made with the nasa / esa _ hubble space telescope _ , obtained at the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas 5 - 26555 . this research was supported primarily by nasa grant nag51862 to suny stony brook . we also acknowledge support to the university of colorado through grant s-56460-d . we thank the many people who contributed to the success of the hst and ghrs programs . ake , t.b . , dupree , a.k . , young , p.r . , linsky , j.l . , malina , r.f . , griffiths , n.w . , siegmund , o.h.w . , & woodgate , b.e . 2000 , , 538 , l87 anders , g.j . , coates , d.w . , & thompson , k. 1992 proc . astron . soc . australia , 10 , 33 ayres , t.r . , simon , t. , stern , r.a . , drake , s.a . , & brown , a. 1998 , , 496 , 428 blackwell , j. , shore , s.n . , robinson , r.d . , feggans , k. , lindler , d.j . , malumuth , e. , sandoval , j. , & ake , t.b . 1993 , a user s guide to the ghrs software ( version 2.1 ; greenbelt md : gsfc ) brandt , j.c . , 1994 , , 106 , 890 burton , w.m . & ridgeley , a. 1970 , , 14 , 3 cameron , a. , duncan , d.k . , ehrenfreund , p. , foing , b.h . , kuntz , k.d . , penston , m.v . , robinson , r.d . , & soderblom , d.r . 1990 , , 247 , 415 cameron , a.c . & foing , b. 1997 , observatory , 117 , 218 cameron , a. & robinson r.d . 1989 , , 238 , 657 cameron , a. , et al . , 1999 , , 308 , 493 cook , j.w . , keenan , f.p . , dufton , p.l . , kingston , a.e . , pradhan , a.k . , zhang , h.l . , doyle , j.g . , & hayes , m.a . 1995 , , 444 , 936 cook , j.w . & nicolas , k.r . 1979 , , 229 , 1163 crutcher , r.m . 1982 , , 254 , 82 dere , k.p . , landi , e. , mason , h.e . , monsignori fossi , b.c . , & young , p.r . 1997 , , 125 , 149 dere , k.p . & mason , h.e . 1993 , , 123 , 41 doschek , g.a . , feldman , u. , mariska , j.t . , & linsky , j.l . 1978 , , 226 , l35 dufton , p.l . , hibbert , a. , kingston , a.e . , & doschek , g.a . 1982 , , 257 , 338 feldman , u. , behring , w.e . , curdt , w. , schle , u. , wilhelm , k. , lemaire , p. & moran , t. 1997 , , 113 , 195 gilliland , r.l . , & hulbert , s.j . 1993 , ghrs instrument sci . rep . 55 gray , d.f . , _ the observation and analysis of stellar photospheres _ , ( cambridge : cambridge university press ) guirado , j.c . 1997 , , 490 , 835 heap , s.r . et al . 1995 , , 107 , 871 innis , j.l . , coates , d.w . , thompson , k. , & robinson , r.d . 1985 , proc . astron . soc . australia , 6 , 156 innis , j.l . , thompson , k. , & coates , d.w . 1986 , , 223 , 183 jordan , c. 1975 , 170 , 429 krster , m. , schmitt , j.h.m.m . , cutispoto , g. , & dennerl , k. 1997 , , 320 , 831 lallement , r. , & bertin , p. 1992 , , 266 , 479 lallement , r. , ferlet , r. , lagrange , a.m. , lemoine , m. , & vidal madjar , a. 1995 , , 304 , 461 laming , j.m . & feldman , u. 1993 , , 403 , 434 landini , m. & monsignori fossi , b.c . 1990 , , 82 , 229 linsky , j.l . , & wood , b.e . 1996 , , 463 , 254 mewe , r. , kaastra , j.s . , white , s.m . , & pallavicini , r. 1996 , , 315 , 170 pakull , m.w . 1981 , , 104 , 33 piskunov , n. , wood , b.e . , linsky , j.l . , dempsey , r.c . , & ayres , t.r . 1997 , , 474 , 315 redfield , s. , & linsky , j.l . 2000 , , 534 , 825 robinson , r.d . & cameron , a. 1986 , proc . astron . soc . australia , 6 , 308 robinson , r.d . 1998 , , 110 , 68 rucinski , s. , garrison , r.f . , & duffee , b. 1995a ibvs 4156 rucinski , s.m . , mewe , r. , kaastra , j.s . , vilhu , o. , & white , s.m . 1995b , 449 , 900 schmitt , j.h.m.m . , cutispoto , g. , & krautter , j. 1988 , , 500 , l25 vilhu , o. , ambruster , c. , neff , j.e . , linsky , j.l . , brandenberg , a. , ilyin , l.v . , & shakovskaya , n.i . 1989 , , 222 , 179 vilhu , o. , gustafsson , b. , & walter , f.m . 1991 , , 241 , 167 vilhu , o. , muhli , p. , huovelin , j. , hakala , p. , rucinski , s.m . , & cameron , a. 1998 , , 115 , 610 wahlstr@xmath58 m , c. & carlsson , m 1994 , , 433 , 417 walter , f.m . 1995 , , 187 , # 103.05 walter , f.m . 1999 , in asp conf . 158 , solar and stellar activity : similarities and differences . c.j . butler & j.g . doyle ( san francisco : asp ) , 87 walter , f.m . & byrne , p.b . 1997 , in asp conf . 154 , cool stars , stellar systems , and the sun 10 , eds . r. donahue & j. bookbinder ( san francisco : asp ) , cd-1458 wood , b.e . , linsky , j.l . , & ayres , t.r . 1997 , , 478 , 745 wood , b.e . , linsky , j.l . , & zank , g.p . 2000 , , 537 , 304 zirin , h. 1975 , , 199 , l63 lrrlrrr z2i60107 & 14 & 23 : 4:20 & ech - b & 256 & 0.930 & 1 + z2i60109 & 14 & 23:12:23 & g160 m & 1228 & 0.941 & 2 + z2i6010b & 14 & 23:36 : 4 & ech - b & 256 & 0.973 & 1 + z2i6010d & 14 & 23:43:51 & g160 m & 1228 & 0.983 & 3 + z2i6010f & 15 & 0 : 8:38 & ech - b & 256 & 1.017 & 1 + z2i6010h & & & g160 m & 1228 & & 7 + z2i6010 m & 15 & 0:59 : 2 & ech - b & 256 & 1.085 & 1 + z2i6010o & 15 & 1 : 7 : 5 & g160 m & 1228 & 1.096 & 2 + z2i6010q & 15 & 1:30:46 & ech - b & 256 & 1.128 & 1 + z2i6010s & 15 & 1:38:33 & g160 m & 1228 & 1.138 & 3 + z2i6010u & 15 & 2 : 5 : 2 & ech - b & 256 & 1.174 & 1 + z2i6010w & & & g160 m & 1228 & & 7 + z2i60111 & 15 & 2:53:50 & ech - b & 256 & 1.240 & 1 + z2i60113 & 15 & 3 : 1:47 & g160 m & 1228 & 1.250 & 2 + z2i60115 & 15 & 3:25:34 & ech - b & 256 & 1.282 & 1 + z2i60117 & 15 & 3:33:21 & g160 m & 1228 & 1.293 & 3 + z2i60119 & 15 & 3:57:16 & ech - b & 256 & 1.325 & 1 + z2i6011b & 15 & 4 : 5:17 & g160 m & 1228 & 1.336 & 4 + z2i6011 g & 15 & 4:48:38 & ech - b & 256 & 1.394 & 1 + z2i6011i & 15 & 4:56:35 & g160 m & 1228 & 1.405 & 2 + z2i6011k & 15 & 5:20 : 8 & ech - b & 256 & 1.437 & 1 + z2i6011 m & 15 & 5:28:17 & g160 m & 1228 & 1.448 & 3 + z2i6011o & 15 & 5:52 : 4 & ech - b & 256 & 1.480 & 1 + z2i6011q & 15 & 5:59:58 & g160 m & 1228 & 1.491 & 5 + z2i6011v & 15 & 6:43:20 & ech - b & 256 & 1.549 & 1 + z2i6011x & 15 & 6:51 : 9 & g160 m & 1228 & 1.560 & 2 + z2i6011z & 15 & 7:14:50 & ech - b & 256 & 1.592 & 1 + z2i60121 & 15 & 7:23:11 & g160 m & 1228 & 1.603 & 3 + z2i60123 & 15 & 7:46:52 & ech - b & 256 & 1.635 & 1 + z2i60125 & 15 & 7:54:47 & g160 m & 1228 & 1.646 & 6 + z2i6012a & 15 & 8:38 : 8 & ech - b & 256 & 1.704 & 1 + z2i6012c & 15 & 8:45:51 & g160 m & 1228 & 1.714 & 2 + z2i6012e & & & ech - b & 256 & & 7 + z2i6012 g & 15 & 9:17:35 & g160 m & 1228 & 1.757 & 3 + z2i6012i & 15 & 9:41:31 & ech - b & 256 & 1.790 & 1 + z2i6012k & 15 & 9:49:21 & g160 m & 1228 & 1.800 & 4 + z2i6012p & 15 & 10:52:38 & ech - b & 256 & 1.885 & 1 + z2i6012r & & & g160 m & 1228 & & 7 + z2i6012 t & 15 & 11:22:52 & ech - b & 256 & 1.926 & 1 + z2i6012v & 15 & 12:45:21 & g140l & 844 & 2.038 & 8 + z2i6012w & 15 & 13 : 3:29 & g140l & 844 & 2.062 & 9 + rrl 1175.86 & 44.0 @xmath16 0.4 & @xmath191175.711 + 1189.86 & 1.43 @xmath16 0.28 & @xmath191190.17 ; @xmath191190.42 + 1194.41 & 1.68 @xmath16 0.29 & @xmath191193.29 ; @xmath151194.06,1194.46 + 1197.56 & 0.40 @xmath16 0.18 & @xmath191197.39 + 1201.13 & 2.77 @xmath16 1.14 & @xmath151200.22,1200.71 ; @xmath151200.97,1201.73 + 1206.68 & 24.9 @xmath16 0.7 & @xmath191206.51 + 1215.91 & & ( @xmath77 ) + 1238.99 & 13.7 @xmath16 0.28 & @xmath191238.82 + 1243.05 & 6.62 @xmath16 0.25 & @xmath191242.80 + 1247.81 & 0.28 @xmath16 0.16 & @xmath191247.38 + 1253.96 & 0.31 @xmath16 0.10 & @xmath191253.80 + 1259.57 & 0.96 @xmath16 0.09 & @xmath191259.53 + 1261.08 & 1.01 @xmath16 0.08 & @xmath191261.3 + 1264.81 & 1.46 @xmath16 0.09 & @xmath191264.74 + 1266.50 & 0.74 @xmath16 0.09 & @xmath191266.42 + 1275.04 & 0.52 @xmath16 0.07 & + 1277.33 & 1.19 @xmath16 0.08 & @xmath191277.5 + 1280.14 & 0.98 @xmath16 0.09 & @xmath191280.5 + 1288.77 & 0.64 @xmath16 0.15 & @xmath191288.42 + 1294.87 & 0.70 @xmath16 0.20 & @xmath191294.54 + 1296.77 & 0.46 @xmath16 0.28 & @xmath191296.72 + 1299.09 & 1.31 @xmath16 0.30 & @xmath191298.89 + 1301.16 & 0.64 @xmath16 0.07 & @xmath191301.15 + 1302.33 & 5.27 @xmath16 0.10 & @xmath191302.16 + 1304.97 & 5.27 @xmath16 0.15 & @xmath191304.86 + 1306.12 & 7.06 @xmath16 0.15 & @xmath191306.03 + 1309.59 & 0.90 @xmath16 0.08 & @xmath191309.28 + 1311.53 & 0.70 @xmath16 0.09 & @xmath191311.36 + 1313.35 & 0.30 @xmath16 0.08 & + 1316.67 & 0.19 @xmath16 0.06 & @xmath191316.54 + 1319.46 & 0.13 @xmath16 0.06 & @xmath191319.00 + 1323.99 & 0.57 @xmath16 0.16 & @xmath191323.52 + 1329.28 & 1.17 @xmath16 0.07 & @xmath191329.1 + 1334.74 & 22.8 @xmath16 0.2 & @xmath191334.53 + 1335.83 & 35.2 @xmath16 0.2 & @xmath191335.71 + 1351.76 & 1.34 @xmath16 0.08 & @xmath191351.66 + 1354.16 & 1.90 @xmath16 0.08 & @xmath191354.08 + 1355.71 & 1.83 @xmath16 0.10 & @xmath191355.60 + 1358.91 & 0.92 @xmath16 0.10 & @xmath191357.13 ; @xmath191358.51 + 1361.63 & 0.27 @xmath16 0.05 & + 1363.27 & 0.29 @xmath16 0.06 & + 1364.44 & 0.37 @xmath16 0.06 & @xmath191364.16 + 1370.27 & 0.14 @xmath16 0.05 & @xmath191369.99 + 1371.53 & 1.06 @xmath16 0.12 & @xmath191371.292 + 1393.89 & 21.3 @xmath16 0.05 & @xmath191393.76 + 1399.72 & 0.65 @xmath16 0.10 & ] @xmath191399.77 + 1401.34 & 1.24 @xmath16 0.15 & ] @xmath191401.16 + 1402.90 & 12.0 @xmath16 0.13 & @xmath191402.77 + 1407.68 & 0.24 @xmath16 0.10 & ] @xmath191407.39 + 1412.59 & 0.34 @xmath16 0.10 & @xmath191412.83 + 1416.71 & 0.25 @xmath16 0.08 & ] @xmath191416.93 + 1424.04 & 0.14 @xmath16 0.07 & ] @xmath191423.9 + 1425.41 & 0.73 @xmath16 0.11 & @xmath191425.03 + 1432.43 & 1.07 @xmath16 0.10 & @xmath191432.11 + 1459.14 & 0.11 @xmath16 0.10 & @xmath191459.03 + 1463.72 & 1.02 @xmath16 0.13 & @xmath191463.33 ; @xmath191463.50 + 1468.00 & 0.79 @xmath16 0.16 & @xmath191467.06 ; @xmath191467.4 + 1473.12 & 0.88 @xmath16 0.12 & @xmath191472.97 + 1474.36 & 1.22 @xmath16 0.13 & @xmath191474.00 + 1481.93 & 1.12 @xmath16 0.13 & @xmath191481.67 + 1483.27 & 0.33 @xmath16 0.08 & @xmath191483.23 + 1486.89 & 0.90 @xmath16 0.13 & @xmath191486.50 ; @xmath191487.15 + 1492.98 & 0.85 @xmath16 0.13 & + 1526.89 & 2.14 @xmath16 0.13 & @xmath191526.71 + 1533.66 & 2.71 @xmath16 0.61 & @xmath191533.43 + 1540.84 & 0.22 @xmath16 0.10 & @xmath191541.03 + 1542.26 & 0.27 @xmath16 0.09 & @xmath191542.18 + 1548.40 & 80.9 @xmath16 0.7 & @xmath191548.20 + 1550.94 & 42.7 @xmath16 0.6 & @xmath191550.77 + 1561.09 & 9.31 @xmath16 0.24 & @xmath191561.00 + 1640.52 & 84.7 @xmath16 0.89 & @xmath191640.48 + 1649.98 & 0.54 @xmath16 0.19 & @xmath151649.42,1649.57 + 1657.33 & 38.0 @xmath16 0.71 & @xmath151556.28,1556.97,1557.38,1558.01 + 1666.70 & 1.52 @xmath16 0.19 & ] @xmath191666.15 + 1670.86 & 5.87 @xmath16 0.25 & @xmath191670.74 + 1686.79 & 0.42 @xmath16 0.19 & @xmath191686.46 + 1697.03 & 2.40 @xmath16 0.34 & @xmath191696.79 + 1713.18 & 1.64 @xmath16 0.13 & @xmath191713.0 + llll 1892.30 @xmath16 0.01 & 1.51 @xmath16 0.01 & 1.28 @xmath16 0.05 & ] @xmath191892.030 + 1893.58 @xmath16 0.02 & 0.09 @xmath16 0.02 & 0.26 @xmath16 0.07 & @xmath191893.252 ? + 1895.39 @xmath16 0.02 & 0.52 @xmath16 0.03 & 0.63 @xmath16 0.09 & @xmath191895.459 + 1900.48 @xmath16 0.02 & 0.34 @xmath16 0.01 & 0.81 @xmath16 0.13 & @xmath191900.270 + 1901.74 @xmath16 0.01 & 0.54 @xmath16 0.01 & 0.94 @xmath16 0.10 & @xmath191901.338 + 1907.34 @xmath16 0.04 & 0.25 @xmath16 0.02 & 0.88 @xmath16 0.16 + 1909.38 @xmath16 0.04 & 1.1 @xmath16 0.1 & 2.09 @xmath16 0.15 & ] @xmath19 1908.734 + 1915.05 @xmath16 0.04 & 0.75 @xmath16 0.03 & 0.87 @xmath16 0.24 & @xmath191914.680 + llll @xmath78}~\lambda 1892}$ ] & 16.5 @xmath16 0.5 & 12.2 @xmath16 0.1 + @xmath79}~\lambda 1892}$ ] & 0.42 @xmath16 0.05 & 12.5 @xmath16 0.1 + @xmath80}~\lambda 1892}$ ] & 2.0 @xmath16 0.2 & 12.2 @xmath16 0.1 + @xmath81 & 0.12 @xmath16 0.01 & 12.2 + @xmath82~\lambda 1908}$ ] & 0.25 @xmath16 0.14 & 11.1 + @xmath83 & 157 @xmath16 89 & 9.1 & 11.8 + @xmath84~\lambda 1908}$ ] & 39 @xmath16 1 & 10.8 @xmath16 0.1 & 11.1 @xmath16 0.1 + @xmath85~\lambda 1908}$ ] & 10.6 @xmath16 0.5 & 11.5 @xmath16 0.1 + @xmath76}~\lambda 1892}{\rm{c iii}]~\lambda 1908}$ ] & 1.3 @xmath16 0.1 & 10.5 @xmath16 0.1 +

we compare the properties of galaxies that form in a cosmological simulation without strong feedback to observations of the galaxy population . we confirm previous findings that models without strong feedback overproduce the observed galaxy baryonic mass function , especially at the low and high mass extremes . through post - processing we investigate what kinds of feedback would be required to reproduce the statistics of observed galaxy masses and star formation rates . to mimic an extreme form of `` preventive '' feedback , such as a highly efficient agn `` radio mode , '' we remove all baryonic mass that was originally accreted from shock - heated gas ( `` hot mode '' accretion ) . this removal does not bring the high mass end of the galaxy mass function into agreement with observations because much of the stellar mass in these systems formed at high redshift from baryons that originally accreted via `` cold mode '' onto lower mass progenitors . an efficient `` ejective '' feedback mechanism , such as supernova - driven galactic winds , must reduce the masses of these progenitors before they merge to form today s massive galaxies . feedback must also reduce the masses of lower mass galaxies , which assemble at lower redshifts and have much lower star formation rates . if we monotonically remap galaxy masses to reproduce the observed mass function , but retain the simulation - predicted star formation rates , we obtain fairly good agreement with the observed sequence of star - forming galaxies . however , we fail to recover the observed population of passive , low star formation rate galaxies , especially at the high mass end . suppressing all hot mode accretion improves the agreement for high mass galaxies , but it worsens the agreement at intermediate masses . reproducing these observations requires a feedback mechanism that dramatically suppresses star formation in a _ fraction _ of galaxies , increasing with mass , while leaving star formation rates of other galaxies essentially unchanged . cooling flows feedback cluster galaxies : evolution galaxies : formation models : semi - analytic models : numerical

0901.1880

one of the oldest challenges in galaxy formation theory is to explain its overall inefficiency , specifically how to prevent too much gas from cooling onto the central galaxies of dark matter halos and forming stars , resulting in higher - mass galaxies at a given number density than observed ( e.g. * ? ? ? a second , related challenge is to explain the bimodality in galactic properties , where massive galaxies are typically red , elliptical and have very little star formation , while lower mass objects are blue , disky , star - forming galaxies ( e.g. @xcite ) . early work on these problems emphasised the need for efficient feedback in low mass galaxies , to stop very early star formation and to be able to match the low mass end of the galaxy mass function @xcite . the mechanism most often invoked at the low - mass end is supernova - driven winds @xcite . these winds not only suppress galaxy masses in low - mass halos , but are also necessary to enrich the intergalactic medium with metals as observed . there has been substantial progress in understanding the basic scaling relations required for supernova - driven outflows : for example , the models most successful in explaining the igm metal distribution rely on scaling relations appropriate for momentum - driven winds @xcite . however , it is not yet clear under which conditions these winds operate in general . if the feedback at low masses is a supernova - driven wind , it is more likely to be efficient at high redshifts , when galaxies had star formation rates that are an order of magnitude or more higher than they are today . at lower redshifts , efficient feedback is also needed in low mass objects , but winds are less likely to be sufficient to remove gas from galaxies @xcite owing to their lower star formation rates . much of the recent focus has been on the properties of massive galaxies . some recent semi - analytic models ( sams ) of galaxy formation ( e.g. @xcite ) are able to reproduce the global properties of massive ellipticals , which appeared too blue and too massive in earlier theoretical models . all these models suppress star formation in massive halos by means of feedback from active galactic nuclei ( agn ) . in several models , supermassive black holes accreting gas that has cooled from the surrounding hot halo atmosphere provide strong feedback to the hot halo gas , which prevents or slows the gas cooling . this type of feedback from agn is often called `` radio mode '' feedback , since it is believed to operate in massive radio galaxies . however , such feedback can not directly change the galaxy morphology , which is believed to be changed by major mergers @xcite , i.e. mergers of galaxies with similar masses . during major mergers the rapid inflow of gas into the central parts can feed the black hole in the galactic centre , which in turn can ionise , heat , and expel the surrounding cold gas . in idealised simulations of such major mergers , this `` quasar mode '' feedback was also successful in preventing the accretion of gas after the merger @xcite . however , these simulations are typically evolved outside of their cosmological environment and do not model the subsequent evolution of halos , so it is still unclear how long this `` quasar mode '' feedback effect lasts before gas can fall in from the intergalactic medium and once again start cooling onto the galaxy . in our previous work ( * k05 , hereafter ) and in the first paper in this series ( * ? ? ? * k09 , hereafter ) , we followed the buildup of galaxies and halo gas in a @xmath1cdm universe ( inflationary cold dark matter with a cosmological constant ) , using hydrodynamic simulations _ without _ any of the strong feedback processes just mentioned . we showed that it is the smooth accretion of intergalactic gas that dominates the global galactic gas supply , not accretion by merging , and that it proceeds via two stages . first , gas is accreted through filamentary streams , where it remains relatively cold before it reaches the galaxy ( * ? ? ? this accretion mode , which we call cold mode accretion , is very efficient because the gas does not need to cool and hence it falls in on approximately a free - fall time . this means that the baryonic growth closely mimics the growth of the dark matter halo , albeit with a slight time delay . cold mode accretion dominates the global growth of galaxies at high redshifts and the growth of lower mass objects at late times . as the dark matter halo grows larger , a larger fraction of the infalling material shock heats to temperatures close to the virial temperature . in the denser , central regions a fraction of this hot gas is able to cool . this latter process is the `` classical '' scenario by which galaxies gain gas ( e.g. * ? ? ? * ; * ? ? ? * ) , and to differentiate it from the previous mode we call it hot mode accretion . the dominant accretion mode depends on halo mass , and the transition between these two regimes occurs at a mass of @xmath2@xmath3 . a similar ( albeit slightly lower ) transition mass between shock heated and non - shock heated halo gas is found in spherically symmetric calculations @xcite , and the physics behind this transition is apparently related to the ratio of post - shock compression and cooling times . in halos with masses near the transition mass , cold filamentary flows still supply the central galaxy with gas even though some of the infalling gas shock heats to near the virial temperature . at higher redshifts cold filaments are even able to survive within halos above this transition mass , i.e. halos dominated by hot halo gas ( k05 , @xcite , k09 ) . overall , the bulk of the baryonic mass in galaxies is accreted through the cold accretion mode and on average no galaxy of any mass acquires more than about 30% of its mass through hot mode accretion ( k09 ) . we can sort the galaxy feedback processes discussed above into two classes : those that prevent gas from entering a galaxy in the first place , `` preventive '' feedback , and those that expel a fraction of the gas that does manage to enter the galaxy , `` ejective '' feedback . agn radio mode heating and photoionisation are examples of preventive feedback , while winds driven by supernovae or agn `` quasar mode '' are examples of ejective feedback . ( the same feedback process could in principle be both ejective and preventive for example if the energy released in a quasar - driven wind heats the surrounding halo gas and thereby prevents it from cooling . ) the effectiveness of these two feedback types should vary depending on the dominant accretion mode in the halo hosting the galaxy . because cold mode halos have very little halo gas outside of the cold dense filaments , it is doubtful whether ejective feedback can drastically affect the accretion of gas . however , ejective feedback could lower the masses of galaxies in these halos by expelling already accreted material . in this case , whether or not the winds escaped into the igm would only be determined by energetics ( assuming winds can not destroy dense filamentary streams of gas ) , since the halos are mostly devoid of gas . such winds , however , might be stopped by the quasi - spherical , hot halos that surround hot mode galaxies , making them ineffective . conversely , preventive feedback like the `` radio mode '' of agn is likely most effective in hot mode halos , where it can prevent the hot , dilute gas that is in quasi - static equilibrium from cooling ( as discussed in k05 ) . the exceptions may be at very low masses , where photoionisation and preheating could prevent gas from getting into the halos in the first place by strongly reducing the gas cooling rates . to understand the role of feedback during galaxy formation and evolution , one also has to understand how galaxies are supplied with baryons and in particular with the gas that provides the fuel for star formation . this is difficult with current sams , as they do not yet accurately track cold mode accretion . on the other hand , simulations currently lack the resolution necessary to accurately simulate either preventive or ejective feedback processes directly within a cosmological environment . therefore , it is important to gain understanding of when and where feedback is necessary to explain the observed properties of galaxies . in this paper we do this from the vantage point of our own cosmological simulations , which do not include _ any _ strong feedback mechanisms . by comparing the observed properties of galaxies with the galaxies that form in our simulations , we confirm the conventional wisdom that one or more strong feedback mechanisms are needed to prevent the excessive accumulation of baryons into galaxies . we draw inferences about where in redshift and galaxy / halo mass gas accretion must be suppressed and what feedback mechanisms are likely to be successful . to mimic an extreme version of preventive feedback , e.g. , a perfectly efficient agn `` radio mode '' , we remove all the gas accreted through hot mode from all galaxies . we show that this extreme feedback scenario only slightly improves the match between simulated and observed galaxy masses , which still disagree at the faint and bright ends . therefore ejective feedback from starburst winds is required . in addition , a selective feedback mechanism , like one that occurs primarily during major mergers , is probably required to explain the bimodality of the galaxy population . a large fraction of the results presented in this paper are presented in slightly different form as a part of d. kere s phd thesis at university of massachusetts , amherst @xcite . in [ sec : simulations ] we describe our new simulations , and the specifics of our procedure for removing hot mode accretion , as well as the `` cold drizzle '' in massive galaxies which we suspect to be numerically enhanced . in [ sec : massfn ] we compare the observed stellar mass function of galaxies to that of simulated galaxies both with and without hot mode accretion . we discuss in [ sec : buildup ] the accumulation of galaxy mass and the stellar component , and in [ sec : ssfr ] we compare the observed star formation rates with the simulations , again with and without hot mode accretion . we discuss the feedback mechanisms needed to bring the masses and specific star formation rates of the simulated galaxies into better agreement with the observations in [ sec : discussion ] and conclude in [ sec : conclusions ] . we described the simulation analysed in this paper in k09 , but for completeness we summarise its main properties here . we adopt a cold dark matter model dominated by a cosmological constant , @xmath1cdm , with the following cosmological parameters : @xmath4 , @xmath5 , @xmath6 , and a primordial power spectrum index of @xmath7 . for the amplitude of the mass fluctuations we use @xmath8 , and for the baryonic density we adopt @xmath9 . all of these cosmological parameters are consistent with the the newest measurements from the wmap team @xcite and with various large scale structure measurementscdm / all values ) ] , except for the primordial power spectrum index which is slightly higher in our simulations . we model a 50.0@xmath10 comoving periodic cube using @xmath11 dark matter and @xmath11 gas particles , i.e. around 50 million particles in total . gravitational forces are softened using a cubic spline kernel of comoving radius @xmath12 , approximately equivalent to a plummer force softening of @xmath13 . using the naming scheme from k09 , we will refer to this simulation as l50/288 later throughout the text . we include the relevant cooling processes using primordial abundances as in @xcite , omitting cooling processes associated with heavy elements or molecular hydrogen . in all of these simulations we include a spatially uniform , extragalactic uv background that heats and ionises the gas . the background redshift distribution and spectrum are slightly different than in k05 . the background flux starts at @xmath14 and is based on @xcite . for more details about the calculation of this uv background see @xcite . we note here that smaller volume simulations with our new uv background and with the version used in k05 showed no noticeable differences in the evolution of the galaxy population in the redshift range @xmath15 of interest in this paper . the initial conditions are evolved using the sph code gadget-2 @xcite . the calculation of the gravitational force is a combination of the particle mesh algorithm @xcite for large distances and the hierarchical tree algorithm @xcite for small distances . the smoothed particle hydrodynamics algorithm @xcite used here is entropy and energy conserving , and it is based on the version used in @xcite . the public version of this code was modified ( both by v. springel and by us ) to include the cooling , the uniform uv background , and the two - phase star formation algorithm . once the particle reaches a density above the star forming threshold , star formation proceeds in a two - phase medium where the supernova energy released by type ii sne during star formation balances cold cloud formation and evaporation by the hot medium as in @xcite . this enables more stable gas rich disks , but it does not produce galactic outflows , i.e. sn feedback only provides disk pressurisation . in this model , the dependence of the star formation rate on density is still governed by a schmidt law @xcite . the model parameters are the same as in @xcite , which were selected to match the @xmath0 kennicutt law @xcite . the threshold density for star formation is the density where the mass - weighted temperature of the two phase medium equals 10,000 k. in practice , this threshold density remains constant in physical units during the simulation and corresponds to a hydrogen number density of @xmath16 . each gas particle within the two - phase medium has an assigned star formation rate but the actual conversion from gas to star particles occurs stochastically @xcite , similar to the algorithm in @xcite . each star particle inherits half of the initial gas mass of the gas particle . to identify bound groups of cold , dense baryonic particles and stars , which represent galaxies , we use the program skid ( see k05 for more details ) . briefly , a galaxy identified by skid contains gravitationally bound groups of stars and gas with an overdensity higher than 1000 relative to the mean baryonic density and @xmath17 . here , we slightly alter these criteria by using an increased temperature criterion at densities where the two - phase medium develops to allow star - forming , two - phase medium particles to be part of a skid group , since at high densities the mass - weighted temperatures in the two phase medium can be higher than @xmath18 . after the identification of a galaxy ( skid group ) , we determine its total stellar mass and star formation rate . the stellar mass is simply the sum of the masses of the star particles and the `` instantaneous '' star formation rate is the sum of the `` instantaneous '' star formation rates of all the gas particles , calculated from the properties of the two - phase medium . the simulation here accounts for mass loss from stars that explode as type ii supernovae , but not for the larger fraction of mass lost from intermediate - mass stars , which can be 0.30.4 at late times for a @xcite or diet salpeter imf @xcite . after beginning the simulation discussed in this paper , we added this effect to the code and tested it in smaller volume simulations . we find very similar rates of stellar mass lockup in the two simulations . in other words , the global density of stars as a function of redshift remains nearly unchanged , but the global star formation rates increase , by up to factor 2 at late times . thus , we expect delayed recycling to have little effect on the masses or stellar ages of a typical galaxy but to boost the individual star formation rates by about a factor 2 . we caution , however , that this test was performed using simulations without strong feedback . ejective feedback processes might succeed in removing the recycled gas before it forms stars , thus resulting in lowered stellar masses and smaller increases in the star formation rate . we will comment in more detail on the possible effects of mass feedback in [ sec : ssfr ] . in k05 , we showed that much of the gas in galaxies entered through cold mode , and we argued that removing a large fraction of the remaining hot mode accretion could bring the simulated galaxies into better agreement with the observed galaxy masses and colours . the simulation method used in k05 in fact suffered from numerically enhanced hot - mode cooling rates @xcite , owing to an enhanced density of the hot phase at hot - cold boundaries . our new simulation has a much larger volume and contains many more galaxies in massive halos , and it also uses an sph algorithm that prevents numerical overcooling . thus we can more reliably test the significance of hot mode accretion . models of preventive feedback suppress the hot mode accretion to varying degrees . for example , in models of `` radio mode '' feedback @xcite , a black hole at the centre of its halo suppresses the accretion only if it contains enough mass to provide a significant input of energy into the halo gas and only if it resides in a halo containing a significant hot atmosphere , and it need not be 100 percent effective . however , the sam of @xcite assumes the complete shut - down of accretion in massive halos , above a critical mass and below a critical redshift . in our case , to mimic the most extreme case of preventive feedback , we choose to completely ignore _ all _ the gas particles that cool from any hot halo , regardless of halo mass or redshift . the method we use to identify hot mode accretion is similar to that used in k05 . starting with particles in galaxies at @xmath0 , we follow the particles back through time and flag all the particles that have ever reached a temperature higher than @xmath19 . ( particles within the star - forming two - phase medium can exceed this threshold owing to the contribution from the hot phase to the mass - weighted temperature . here we ignore these particles , since they clearly have already been accreted into a galaxy . ) then we produce a new , revised list of galaxy stellar , gas , and total masses along with galactic star formation rates , excluding the contribution from particles flagged as accreting through hot mode . in k09 , we showed that in gadget-2 simulations massive halos contain many cold clumps of gas that are not identified by skid as galaxies . such clumps survive within the hot halo gas , and because the ram pressure drag on such objects is likely overestimated , they fall onto the central galaxy in just over a free - fall time , a process we term `` cold drizzle '' . as a result , the contribution of cold mode to the total gas accretion falls to a minimum around a galaxy mass of @xmath20 and then rises again at higher masses ( see figure 8 of k09 ) . whether or not such cold clumps should form and survive in very hot halos is not clear ( see the discussion in k09 ) . however , the accretion rate of these clumps is almost certainly overestimated , since in most sph implementations the effective cross - section for ram pressure is overestimated at low resolution , increasing the drag on a cold clump as it moves through a hot medium @xcite . furthermore , it is possible that many of these clumps would be destroyed by surface instabilities that are hard to model correctly in sph @xcite or by conduction . to bracket the correct solution , therefore , we remove cold mode accretion onto galaxies more massive than @xmath20 to see how the properties of our simulated galaxies change . we use the galaxy s mass at the time the gas particle accretes ( without adjusting its mass for hot mode removal ) . even though this cutoff mass is well into the hot mode regime , a large mass fraction of these massive galaxies is built up by gas initially accreted through cold mode , if it joined the massive galaxy through mergers . we do not remove this indirect cold mode accretion but since some of the sub - resolution skid identified groups are also cold clouds , we also remove all the gas that was accreted from progenitors below our resolution limit of 64 particles to make the effect of cloud removal more extreme . we then recalculate the galaxy properties . = 5.5 in we begin the comparison of simulated and observed galaxy properties with the @xmath21 stellar mass functions ( smf ) . in figure [ fig : mf ] we compare the observed ( _ g_-band derived ) smf from @xcite to the smf from our l50/288 simulation . we plot the smf starting from our adopted resolution limit . the stellar masses in @xcite are based on data from the sdds survey @xcite and the diet salpeter imf , and are shown to be consistent with the masses derived from the near infrared 2mass survey @xcite . the simulated galaxies are overabundant at all galaxy masses , with the largest disagreements occurring at the low and high mass ends . by matching the integrated number density of galaxies above a certain mass in the simulation with the observations , we estimate the difference in mass between the observed and simulated galaxies ( assuming the same rank ordering of galaxies by mass in the simulation and observations ) . we show this difference as a correction factor @xmath22 in the lower panel of figure [ fig : mf ] . the masses of the most massive simulated galaxies are about a factor of 35 higher than those observed . one can immediately conclude that strong feedback , which is not included in our simulations , is needed to decrease the masses of these simulated massive galaxies . the differences decrease at intermediate masses , around the `` knee '' of the observed mass function , where the simulated masses are only high by a factor of 1.52 . at masses lower than several times @xmath23 the differences between the observed and simulated stellar mass function are enormous . to match the number density of observed galaxies requires suppressing the masses of the simulated galaxies by more than an order of magnitude . these differences suggest that a very efficient mechanism must prevent the formation of the majority of low mass galaxies with masses up to several times @xmath24 or that it must drastically lower their masses to bring the simulated galaxies into agreement with the observations . another possibility is that the observed galaxies contain significantly more gas and less stars at similar galaxy masses . of course these conclusion are not new or unique to our work , and we will discuss the required properties of this mechanism in more detail in the discussion section . the total amount of baryons locked in the stellar component in our simulation is 18 percent , which is about a factor of 3 higher than the observed value ( e.g. * ? ? ? * ) . while the differences between the simulated and observed galaxies are large at the high and low mass ends , the stellar masses of galaxies are in relatively good agreement around the `` knee '' of the mass function where a large fraction of the global stellar mass is concentrated , making this disagreement more moderate globally . the galaxy masses in figure [ fig : mf ] do not account for stars dispersed into the intracluster medium from the hierarchically built remnants . observationally , in clusters of @xmath25 , about 2030% of all stars in a halo are part of the icm+central galaxy system , which is dominated by intracluster stars ( ics ) ( e.g. * ? ? ? * ) . in fractional terms this is not far from what we find in our simulated massive halos , which have 2035% of their stars in the ics at halo masses of about @xmath26 . the fraction of ics stars is also subject to the exact definition of where the light of the central ( brightest ) cluster galaxy ends and becomes ics and , therefore , any comparison of this aspect of our simulation to the observations is quite uncertain . however , the _ combined _ mass of the central galaxy and the ics is already higher in our simulations than what is observed . furthermore , most of the massive galaxies in our simulation are in halos of even lower mass , where ics is likely an even less important stellar component . given these facts , it appears unlikely that a large fraction of the factor of @xmath27 larger galaxy mass in the simulations could be explained by having more ics in the simulation . in [ sec : hotremoval ] we discussed removing the hot mode to mimic the extreme case of preventive feedback . we plot the smf with hot mode accretion removed in figure [ fig : mf ] ( dashed line ) . we see that the decrease in galaxy masses caused by this extreme feedback does bring the simulations into better agreement with the observations at high masses , but the change in mass is only modest . the change is largest at intermediate masses , around the `` knee '' in the observed mass function . the typical change is several tens of percent at masses of around @xmath20 and is even smaller at higher masses . the removal of hot mode accretion even changes galaxy masses in objects with masses as small as several times @xmath24 , because there are still some galaxies at these small masses with significant hot mode accretion at low redshifts . at even lower masses , however , there is almost no change in mass after hot mode removal , which is not unexpected since all low mass galaxies are built exclusively through cold mode accretion . the most surprising feature of this curve is the almost negligible change at the high - mass end . we conclude from this plot that the removal of hot mode accretion , to mimic an extreme case of preventive feedback such as an agn radio mode , _ can not _ explain the observed steep drop in the smf at the high mass end . the removal of hot mode accretion only partially succeeds at intermediate masses , where the simulated mass function is indeed quite close to the observed one , but only over a small range of masses . previously , in k09 , we saw that the accretion of many galaxies at low redshift is completely dominated by hot mode accretion and yet their masses are not greatly affected by hot mode removal . we provide an explanation for this seeming contradiction in the next section and later , where we discuss the processes needed to bring the simulated galaxy masses into better agreement with the observations . in [ sec : coldremoval ] we discussed the removal of the cold gas infall in massive galaxies in addition to removing all hot mode , since we suspect this infall to be a numerical artifact . we show the results of this in figure [ fig : mf ] as the dot - dashed line . the effect on massive galaxies is again rather small and , by construction , it only affects the most massive objects . typically , most galaxies of around @xmath28 have their masses lowered by only @xmath29 percent when the cold drizzle is removed . therefore , even without cold mode accretion in massive galaxies , preventive feedback alone ( such as an extreme agn radio mode ) is not sufficient to bring the high mass end of the observed and simulated smfs into agreement . = 3.4 in -15pt = 3.4 in -15pt in figure [ fig : formhist ] ( left ) , we plot lines showing the median redshift at which galaxies formed 25 , 50 , and 75 percent of their @xmath0 stellar mass as a function of the stellar mass of the galaxies . to indicate the scatter from one galaxy to another , we also show the mass weighted stellar formation redshift ( the redshift when 50 percent of the @xmath0 stellar mass was formed ) for each individual galaxy . the rising trend with mass has been described as the `` downsizing '' of galaxy formation . sometimes this trend has been explained as a consequence of particular feedback models , but here we see that it is present in our sph simulations _ without _ strong feedback , and therefore is a natural consequence of hierarchical models of galaxy formation ( see also @xcite ) . even the build - up of dark matter halos proceeds in a similar way , where the bulk of the present halo mass of massive halos was already assembled in lower mass progenitors ( above a given minimum mass ) at earlier times than the bulk of the mass in low mass halos @xcite . for comparison , we also present in figure [ fig : formhist ] ( right ) observational estimates of the stellar formation redshifts from the spectral analysis galaxies from the sloan digital sky survey by @xcite ( triangles ) . to generate these lines , we have taken the sfr as a function of redshift in each of five mass bins from their figure 4 , assumed the sfr to be constant within each redshift bin , extended the minimum redshift to @xmath0 , and interpolated the cumulative stars formed to get the 25 percent , 50 percent , and 75 percent redshifts . the analysis of @xcite , made purely from the integrated spectra of the local galaxy population , is clearly difficult and subject to systematic errors ( from the dust model , surface brightness biases , spectroscopic fibre coverage , or imperfections in the spectral modelling among others ) . the errors increase rapidly with the redshift , so the 25 percent curve is more reliable than the 75 percent one . however , the global behaviour of star formation and stellar mass as a function of redshift is in reasonable agreement with direct observations , and the overall `` downsizing '' picture inferred from this dataset has been confirmed by many previous studies . we also show the observationally inferred mass weighted stellar age from @xcite . their analysis differs in detail from @xcite , and it results in significantly younger stellar ages at all masses , but especially at the low mass end . we show the formation redshift of 50 percent of the galaxy stellar component ( diamonds ) from their figure 9 ( they do not provide 25th and 75th percentiles ) . to compare the simulated galaxies to these observations we correct the simulated galaxy stellar masses by the correction factor from figure [ fig : mf ] . by comparing the stellar formation redshift in the simulations to the observed values , one can infer whether the bulk of such a mass correction , which in nature would happen through some feedback mechanism , needs to occur below the observationally inferred formation redshift , in the case where the observed formation redshifts are higher than those simulated or above the observational redshift in the opposite case . both the observational results and our simulation show that stars in the progenitors of massive galaxies form very early . however , at the high mass end the simulated stellar ages are in rough agreement with the results of @xcite but are much higher than those of @xcite . it is beyond the scope of this paper to track the differences in the observationally inferred stellar ages , but it is clear that the systematic uncertainties in these estimates are still very significant . while the differences between the observed datasets are large at the low mass end , both observational estimates indicate that the simulated galaxies form too early , implying that their formation should be more suppressed at early times ( we discuss this in more detail in [ sec : discussion.masses ] ) . in figure [ fig : cold_fractions ] , we show the contribution of gas initially accreted through cold mode to the final masses of galaxies at @xmath30 and @xmath0 ( the second panel is similar to figure 7 in k09 ) . the dispersion in the cold mode fractions is large at all masses , but the overall trends with galaxy mass ( indicated with the median line ) are clear . we demonstrated in k09 that at @xmath31 , cold mode accretion completely dominates the smooth gas accretion rates at all masses . around @xmath30 , hot mode starts to be an important source of gas in a limited mass range , around few times @xmath23 . however , from figure [ fig : cold_fractions ] one can see that typical high - redshift galaxies at any mass build up 90100 percent of their mass through cold mode accretion , making the hot mode contribution insignificant at these early times . = 5.5 in the situation is a bit more complicated at @xmath0 . as expected , a typical low mass galaxy forms almost completely through cold mode accretion . the contribution of hot mode accretion increases with mass up to a maximum of 2530 percent for galaxies with masses of around @xmath32@xmath33 , but then the trend reverses and the median hot mode contribution decreases in more massive galaxies . ( we emphasise that this hot mode contribution is measured over the entire history of the baryons that make up the galaxy , whereas a plot of the _ current _ accretion would show much larger hot mode fractions . ) the increase in the cold mode contribution at higher masses owes to the increased contribution of mergers , which mostly adds material that was originally accreted through cold mode . this occurs because the merging progenitor galaxies accrete the bulk of their mass at much earlier times when cold mode accretion dominates the buildup of galaxies at all masses . this trend is reinforced by the lack of substantial hot mode accretion in massive halos at @xmath34 . the mass dependence of the cold mode fractions also explains why the simulated mass function is most affected by hot mode removal at masses of @xmath35@xmath36 , since these are the galaxies that had the largest fractions of their total mass accreted through hot mode . in figure [ fig : cold_fractions ] we also plot the median fraction of mass initially accreted in cold mode after the cold mode accretion is removed from massive galaxies as described in [ sec : coldremoval ] . the residual cold drizzle contributes around 20 percent to the galaxy mass , so removing it increases the ratio of hot to cold mode by around 20 percent ; since this ratio is itself small , this change raises the cold mode contribution by several percentage points at most . from figure [ fig : formhist ] , it is clear that bulk of the stellar mass in massive galaxies forms prior to @xmath30 , in much smaller objects whose formation is completely dominated by cold mode accretion . the most massive progenitors of these galaxies at @xmath31 contain typically only @xmath372025 percent of their present mass . these objects later merge to form even more massive galaxies . since in high - redshift galaxies only a small fraction of the baryonic mass is gained through the cooling of hot halo gas , the removal of hot mode accretion can not significantly affect the masses of these galaxies . therefore , any feedback mechanism that aims to lower the masses of the most massive galaxies must actually affect the masses of their lower - mass , high - redshift progenitors as we discuss in [ sec : discussion.masses ] . -40pt = 3.5 in -15pt = 3.5 in the final observable we shall consider is the current star formation rate of individual galaxies , which is most usefully presented relative to its mass as the specific star formation rate ( ssfr ) . quite frequently galaxy colour is used as a proxy for ssfr , although of course a galaxy s colour depends on its current star formation rate , the previous star formation history , and the amount of dust attenuation . colour - magnitude plots incorporating observations of many thousands of galaxies have become commonplace , and the main features of these plots are by now familiar ( e.g. @xcite ) : there is a red sequence of galaxies , mostly populated by massive , early types with low ssfrs , distinctly offset from a blue cloud of late type , lower mass galaxies with high ssfrs . while part of this division can be caused by dust attenuation , most of the differences are clearly caused by varying amounts of recent star formation . however , for comparison to the simulations , a plot of log(ssfr ) versus galaxy stellar mass is more useful . obtaining this requires modelling each observed galaxy s spectral energy distribution to obtain transformations from magnitude to mass and from colours to log(ssfr ) ; the latter transformation in particular is highly non - linear . just such a plot of the low redshift galaxy population was recently obtained by @xcite , using uv data from the galex satellite @xcite combined with _ ugriz _ photometric data from the sloan digital sky survey ( sdss ) @xcite . their analysis builds upon and supersedes earlier optical - only work by @xcite and @xcite . the authors consider a wide range of star formation histories for every observed galaxy and assign each a likelihood , leading to a two - dimensional probability distribution of ssfr and stellar mass for each individual galaxy ( where the ssfr is averaged over the last 100 myr ) . if one simply plots the mean value of this distribution for individual galaxies in the observed sample , it shows an obvious bimodality , as one can see in the left panel of figure [ fig : ssfr_data ] . the `` blue cloud '' galaxies identified in a colour - magnitude diagram now correspond to a tight star - forming sequence in the upper part of the plot , partly because scatter caused by variations in metallicity and dust have been removed by the sed fitting . the tight `` red sequence '' of the colour - magnitude diagram now corresponds to a low - ssfr sequence , broadened by the nonlinear transformation between colour and log(ssfr ) , and there is a significant bridge population between the two sequences . ( the relative prominence of the low - ssfr sequence in the left panel of figure [ fig : ssfr_data ] is partly an artifact : there is a lower limit of @xmath38 allowed in the models of ssfr history used in @xcite causing the mean log(ssfr ) of individual low - ssfr galaxies to cluster around a value somewhat higher than this . ) a fairer representation of observed galaxies in this plot smears each individual galaxy over its full two - dimensional probability distribution in ssfrs and mass , and normalises it by the effective survey volume at each mass , i.e. @xmath39 . we show this result , which is the only one that should be compared with simulations , as the solid contours in the right panels of figure [ fig : ssfr_data ] . this figure is equivalent to the grayscale image shown as the lower panel of figure 15 in @xcite . we generate the contours from a grid with a spacing of 0.05 in log(m ) and 0.1 in log(ssfr ) , and plot contours encompassing 25 , 50 , 75 , and 90 percent of the maximum probability density ( from darkest to lightest shading ) . with these changes , the lower - mass galaxies become more prominent , and the low - ssfr sequence smears out into a flat ledge covering a very wide range of ssfrs that extends downward from the star - forming sequence in contrast to the colour - magnitude diagram , one might now describe the features as a `` blue sequence '' and a `` red cloud '' . the star - forming sequence persists as a tight ssfr vs. galaxy mass relation at relatively high ssfrs , around @xmath40 at stellar masses of @xmath41 . ssfrs in this sequence decrease toward higher masses . the sequence dominates the number density over several orders of magnitude in galaxy mass , up to @xmath42 , although some galaxies have low ssfrs even at intermediate and low galaxy masses . the tightness and weak dependence on mass of this relation was discussed extensively in the literature ( e.g. * ? ? ? * ) . in the same plot , we also show the properties of the simulated galaxies . as we discussed previously , the masses of the simulated galaxies are much higher than those of observed galaxies . hence , even if the present star formation rate in a simulated galaxy were to correspond to the observed values , the ssfrs would be much lower than those observed . to avoid this problem we correct the masses of the simulated galaxies using the correction factors @xmath43 in figure [ fig : mf ] , assuming that the mass rank order of the simulated galaxies is preserved this is similar in spirit to halo and subhalo matching models that try to connect the observed properties of galaxies to the dark matter halos and sub - halos from n - body simulations ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , except in our case we are connecting observed and simulated galaxies . furthermore , to account for the different initial mass function ( imf ) used in @xcite , a diet @xcite imf , and that used in @xcite , a @xcite imf , we lower these corrected masses by @xmath44 . we leave the sfrs of individual galaxies unchanged in this plot , so that the ssfrs increase by a factor @xmath45 . our reasoning is that here we are essentially examining the star formation rates as a function of halo mass , and the adjustment to the mass should just enable a fair comparison of galaxies residing in similar halos . in k09 , we showed that the sfrs of individual galaxies closely follow the smooth gas accretion rates . as we discuss in [ sec : discussion.masses ] , if the mechanism that is responsible for producing the correct mass function is ejective feedback from sn - driven winds , it should not significantly affect the accretion of intergalactic gas at low redshift , when such feedback is inefficient , and hence it would also not likely affect the tight relation between gas accretion and star formation . therefore , the long term mass accumulation of galaxies ( that we are trying to correct with the mass re - normalisation ) should not affect the low - redshift sfrs . since the simulated galaxies have exact values for both their ssfrs and their masses , to compare them with the observational data we add two - dimensional gaussian uncertainties . we choose the mass uncertainties to have @xmath46 and the ssfr uncertainties to have @xmath47 linearly decreasing from 0.6 for the lowest ssfrs to 0.2 for the highest ssfrs , both in rough agreement with the errors present in @xcite . most of the simulated galaxies also reside in a well defined star forming sequence with properties that are similar to the observed star forming sequence . there is , however , a sizeable population of simulated galaxies with zero ssfr , which possibly corresponds to part of the observed red sequence . owing to our limited mass and time resolution , the transition between zero sfr galaxies and star forming galaxies suffers from discretisation effects , with the lowest possible nonzero simulated star formation rate being around @xmath48 . when we plot these zero ssfr simulated galaxies we assign them a very low ssfr of @xmath49 , before adding gaussian uncertainties . while most of the observed red sequence galaxies reside at the high mass end , most of the simulated zero ssfr galaxies reside at the low mass end . interestingly , the mass dependence and scatter of the simulated star forming sequence at @xmath0 is consistent with the observed trends . the median ssfrs of simulated galaxies at a given mass are very close to the observed sequence except around @xmath24 , where it is several tenths of a dex lower . however , as noted in [ sec : simulations ] , the current simulations do not include mass feedback from intermediate - mass stars . on average at least , this effect might boost the typical sfr by up to 0.3 dex , bringing the low - mass star - forming sequence back into agreement with the observed one , but if the same effect operates at high masses it would make their simulated ssfrs too high . to quantify any differences in the way galaxies populate the high and low ssfr regions , approximately corresponding to `` red '' and `` blue '' galaxies , we determine the fraction of galaxies with ssfrs lower and higher than @xmath50 . we plot the fraction of low ssfr galaxies in the lower - right panel of figure [ fig : ssfr_data ] for both observed ( solid line ) and simulated ( dashed line ) galaxies . the observed galaxy sample has about 20 percent of its galaxies in the red sequence for @xmath51 . in the simulation the fraction of low ssfr galaxies is about 2530 percent , which is slightly higher than the observed fraction . however , the differences begin to increase towards higher masses and by around @xmath52 about 60 percent of the observed galaxies have very low ssfrs while all but 15 percent of the simulated galaxies reside on the star forming sequence . the discrepancies get even larger at higher masses . this illustrates a generic problem that occurs for massive galaxies in simulations and sams without feedback : they form too many stars at late times and hence are too blue on average . the differences between the simulated and observed galaxies in figure [ fig : ssfr_data ] thus show some different symptoms than those in the mass function plot ( fig . [ fig : mf ] ) . at low masses the current star formation rates agree well with the observations , whereas the simulation masses are much higher . alternatively , whereas the masses should be reduced at high redshift by ejective feedback from sn - driven winds , perhaps it should not significantly affect the gas supply and star formation rates at low redshift when such feedback is inefficient ( as we discuss in [ sec : discussion.masses ] ) . starting at @xmath53 where the simulated masses are in the best agreement , however , the fraction of star - forming galaxies starts to diverge strongly from the observed fractions , even though the actual star formation rates of those galaxies still on the star - forming sequence are in reasonable agreement . -30pt = 3.5 in -15pt = 3.5 in as we did with the mass function , to mimic an extreme version of preventive feedback in hot halos , we completely remove all the hot mode accretion and plot the resulting galactic properties in the left panels of figure [ fig : ssfr_cleaned ] . we again re - normalise the galaxy masses to match the observed mass function after the removal of the hot mode in the same way as in figure [ fig : ssfr_data ] . this hot mode removal does not change any of the properties at the low mass end because these galaxies accrete their gas almost exclusively through cold mode accretion . as we showed before , even the highest mass galaxies do not significantly change their masses when hot mode accretion is removed , but they do significantly change their ssfrs . up to masses of @xmath54 , the average ssfrs of simulated galaxies change enough to be roughly consistent with the observed galaxies . however , if one looks more closely this leaves too few galaxies to correspond to the observed star forming sequence . even at intermediate masses of around @xmath24 , the removal of the hot mode ruins the relatively good agreement between the observed and simulated star forming sequence . although the fraction of galaxies below our adopted `` red sequence threshold '' is similar to the observed fraction , the typical ssfr of a star forming galaxy is now factor of 35 lower than those observed . even if one were to include mass feedback from intermediate - mass stars as in [ sec : simulations ] , the ssfr would still be low by a factor of about two . this suggests the need for a selective feedback mechanism , which only significantly affects star formation in a fraction of the galaxies at a given mass to preserve the star forming sequence . finally , at masses greater than @xmath55 the discrepancies persist and are as large as when the hot mode is not removed . in the right panels of figure [ fig : ssfr_cleaned ] , we remove both the hot mode accretion and the cold mode accretion in galaxies more massive than @xmath20 ( before the mass re - normalisation ) , i.e. the possibly spurious cold drizzle . we again renormalise the stellar masses of galaxies using the corrections from figure [ fig : mf ] to match to observed mass function . the ssfrs of the galaxies at the massive end are reduced even further and are more similar on average to the observed massive galaxies than when we only remove the hot mode accretion . the rough agreement in the red fraction now extends up to galaxy masses of about @xmath56 with the more massive galaxies remaining discrepant . however , once again there are not enough galaxies on the star forming sequence owing to the nature of our mock feedback that affects approximately all galaxies above a given mass ( in the case of cold drizzle removal ) and above a given halo mass ( in the case of hot mode removal ) . again , the need for a more selective feedback mechanism is obvious from the observed ssfr distribution alone , as a significant fraction of galaxies even at the largest masses appear to be part of a normal star forming sequence . to summarise , removing all the hot mode accretion and the potentially spurious cold mode accretion from massive galaxies , yields a fraction of simulated `` red '' galaxies that is similar to but actually slightly higher than the observed sample around @xmath20 , suggesting that this recipe removes too much late time accretion . furthermore , this procedure moves all galaxies away from the star - forming sequence and therefore is not supported by observations . however , the problem at the very massive end ( above @xmath56 ) remains the same , where only a small fraction of the simulated galaxies have low enough ssfrs . the removal of hot mode accretion and `` cold drizzle '' significantly lowers their ssfrs , but they are still higher than those observed . therefore , preventive feedback mechanisms like radio mode agn alone are probably not sufficient to make the most massive galaxies , those with @xmath57 , red enough . these massive galaxies are typically the central galaxy of a group or cluster and their ssfrs remain too high because fresh gas for star formation arrives through minor mergers . therefore , to keep the central galaxies red enough , an efficient feedback mechanism needs to lower the gas content of the satellite galaxies before they merge , or to prevent star formation from the gas that arrives with the satellites . another possibility is that a fraction of the gas identified as coming from mergers actually comes from stripped galactic disks ( similar to the `` cold drizzle '' ) but which we can not identify as such . because of our limited time spacing between the simulation outputs ( and possibly the numerically enhanced infall of such cold clumps ) , some of the gas stripped between two outputs can end up in the central galaxy of a massive halo by the next simulation output and , therefore , be identified as coming from merger . many authors emphasise the overabundance of massive galaxies and their lack of red colours in simulations and sams ( e.g. k05 , @xcite ) and usually blame excessive gas cooling onto central galaxies in massive halos , i.e. hot mode accretion , for this failure . however , in k09 we demonstrated that many galaxies in massive halos have almost stopped accreting gas from their hot virialised atmospheres and their masses are still a factor of @xmath27 too large compared to the observations , indicating that the problem of massive galaxy formation is more complex . most of the material currently present in the most massive galaxies was originally acquired by smaller galaxies at very early times , as shown in figure [ fig : formhist ] , through cold mode accretion , which dominates gas accretion in all galaxies at early times . the time by which a significant fraction of a current galaxy s mass is already accreted into its progenitors is a strong function of galaxy mass . the most massive galaxies today have their mass accreted into their progenitors , and hence most of their stars form , before @xmath58 . such a scenario is a natural consequence of hierarchical halo and galaxy formation and this `` downsizing '' effect is ubiquitous in both cosmological simulations @xcite and sams of galaxy formation @xcite . this is the reason why preventive feedback alone , e.g. agn activity , can not significantly lower the masses of the most massive galaxies , contrary to the assumptions in some popular current models . this finding also helps us to elucidate the nature of the feedback mechanism needed to fix the problems at the high mass end in the simulated galaxy mass function . feedback in low and intermediate mass objects that form at early times , which subsequently hierarchically merge to make the massive galaxies today , must be very efficient at high redshifts to reduce their masses before they transform that mass into stars . since these galaxies gain most of their mass through cold mode accretion , it is natural that it should be an ejective feedback mechanism like some form of galactic winds . standard ejective feedback mechanisms have little or no effect on cold mode accretion since the gas is already cold and has a high momentum flux . however , since cold mode dominated halos do not possess a hot halo of gas , the winds are free to leave the galaxy unimpeded as long as they have enough energy . in fact , many high redshift galaxies show evidence of very strong , high velocity outflows of gas @xcite . strong feedback in high redshift galaxies could be the consequence of momentum driven winds , which could operate in starburst galaxies @xcite . in this model , driving galactic outflows requires a very efficient starburst with a high star formation rate per unit surface area . at redshifts @xmath59 almost all galaxies have sufficiently high star formation rates to enable this efficient feedback mechanism ( see the discussion in @xcite ) . to match the observed mass function it is probably still necessary to have some preventive feedback , especially to prevent the re - accretion of the necessarily large amounts of ejected gas . once a galaxy s halo grows massive enough , it will gain a hot halo of gas that will impede the winds , quenching the feedback . this could lead to a natural upper mass cutoff to the ejective feedback mechanism at approximately the dividing mass between hot and cold mode halos , @xmath60@xmath3 ( in reality this mass could be up to factor of 3 higher in simulations with strong outflows and metal line cooling ) . another alternative is to somehow keep the high redshift galaxies gas rich , with only a small fraction of their mass converted into stars , then remove the gas during some violent process occurring during galaxy mergers . however , this requires a significantly different star formation algorithm than we use in our simulations , which does not produce such gas - rich massive galaxies . even then , it is not clear if quasar - driven winds or some other mechanism would be sufficient to remove enough of the gas before the progenitors merge into a massive galaxy and convert the gas into stars . the largest differences between the observed and simulated mass functions occur at stellar masses around and below @xmath24 ( simulated mass ) . obviously , the comparison of theoretical and observational stellar mass functions highlights the need for very efficient feedback at the low mass end , but it is far from clear what mechanism actually reduces the masses of these galaxies . it is possible that a mechanism similar to the starburst driven winds that operate at high redshift , which are necessary to lower the mass function at the high mass end , are also able to lower the @xmath0 galaxy masses at the low mass end . however , we demonstrated that most of these galaxies acquire a large fraction of their stellar mass at late times ( after [email protected] ) and , therefore , such feedback needs to be active at late times . in general , feedback from supernova - driven winds @xcite is the most popular candidate and is often used in sams @xcite . simple calculations indicate that such feedback should be effective in halos with circular velocities up to @xmath61 @xcite , which is enough to significantly affect this problematic mass range . however , realistic hydrodynamic simulations show that such feedback has difficulty significantly reducing galaxy masses in galaxies with @xmath62 @xcite , unless the star formation rates are extremely high as at high redshifts @xcite . in some models the winds exhibit a thresholding behaviour : for example , in the momentum driven wind models discussed above the star formation rate surface density in most spiral galaxies at late times is not high enough to drive outflows at all after @xmath63 , except in a small number of starburst galaxies ( see @xcite and references therein ) . therefore , an additional efficient feedback mechanism appears necessary at the low mass end that can affect galaxies forming at much later times . these findings place strong requirements on cosmological simulations that model galaxy formation . the smallest objects that are often poorly resolved need to be resolved in detail at all times to understand the way feedback mechanisms interact with the infalling and galactic gas . uv background heating can affect the formation of galaxies that would form in halos with @xmath64@xmath36 at @xmath0 but it is efficient for only much lower mass galaxies at higher redshifts ( e.g. @xcite ) . other alternatives include pre - heating by gravitational pancaking . in hierarchical galaxy formation models like @xmath1cdm , at low redshifts many intermediate mass halos form in much more massive , flattened structures . in this pre - heating scenario , one assumes that the gas in these structures is shock heated to @xmath65 , a temperature equivalent to the expected infall velocities onto these structures , thus preventing the gas from collapsing into halos of lower virial temperature @xcite . however , some n - body simulations suggest that this scenario does not work in practice because halos in the problematic mass range actually form in pancakes of much lower mass and , therefore , in regions with temperatures lower than previously thought @xcite . the least likely possibility we consider is that a large fraction of the baryons in late forming galaxies are locked in a gaseous component . if the observed gas fraction were substantially higher than in the simulations this would lead to a very different stellar mass function for galaxies with similar total mass . however , we checked this directly and found that our simulated baryonic mass function is no more discrepant with the observed baryonic mass function from @xcite than with the stellar mass function . in addition , the observed mass functions of hi gas @xcite and of the molecular gas @xcite show a smaller number density of objects with gas masses comparable to the stellar masses in the most problematic mass range we analyse , @xmath66 , suggesting that stellar mass dominates a galaxy s baryonic mass in this mass range . only at lower masses does the gas begin to dominate the baryonic mass of galaxies ( e.g. * ? ? ? * ) and hence begin to alleviate the differences between the observed and simulated stellar mass functions . however , even if the baryons in these lower mass objects remained mostly gaseous , they would greatly overproduce the observed hi mass function @xcite . while the need for an ejective feedback mechanism is clear , the effect of such ejecta on the subsequent gas accretion are unknown at the present . depending on the halo mass , wind energy , structure , metallicity and ability to diffuse metals , ejected material can have both `` positive '' and `` negative '' feedback : i.e. while it removes the gas from galaxies and adds the entropy to the surrounding gas @xcite it also increases the amount of available gas in halos and pollutes halos with metals , which could increase the amount of gas that gets accreted at later times . significant feedback at the low mass end at early times could raise the gas fractions in massive halos at later times because a smaller amount of the halo gas will be locked up in galaxies , making this effect an _ intergalactic fountain_. some fraction of the mass that is ejected from low mass galaxies could be reaccreted by higher mass galaxies at later times when such feedback might be less efficient ( e.g. * ? ? ? this could lead to higher halo gas fractions , which in turn could supply more gas , thus leading to higher star formation rates in higher mass galaxies . this would increase the need for efficient preventive feedback in massive halos to remove this added late time accretion . in this sense , both ejective feedback ( e.g. supernova driven winds ) and preventive feedback ( e.g. agn radio mode ) might be needed to yield correct galaxy masses in massive halos . however , in k09 we showed that a large fraction of massive halos naturally develop a gas density core , which might persist even in the case of higher halo gas fractions , possibly alleviating the need for additional feedback . in the previous section we showed that the observed and simulated star forming sequence of galaxies share similar properties . however , the simulations are missing a large fraction of the passively evolving galaxies with very low star formation rates , which dominate the massive end in the observations . removing all the hot mode accretion and the ( potentially spurious ) cold drizzle produces a larger population of passive galaxies , but it makes the fraction of star - forming galaxies too low , especially at intermediate masses . reproducing the observations requires a mechanism that suppresses accretion in a _ fraction _ of galaxies but not in all galaxies , with the fraction itself increasing from intermediate to high masses . in addition , observed red sequence galaxies are preferentially early type ( morphologically ) relative to galaxies on the star forming sequence ( e.g. * ? ? ? * ) and suppressing accretion will not in itself change a galaxy s morphology , though it can help preserve early types by preventing the regrowth of disks . a promising candidate mechanism for moving galaxies to the red sequence is quasar and starburst driven winds , occurring during the mergers of gas rich galaxies @xcite . using the halo occupation distribution and the evolution of a type - separated galaxy mass function , ( * ? ? * a , b ) show that such a model , which assumes that star formation stops after the major mergers of gas rich galaxies , can explain the buildup of the red sequence over time as well as the fraction of red galaxies as a function of galaxy mass . the transformation from the blue to the red sequence during these events is aided by several processes : the development of a shocked hot virialised medium that slows the cooling of gas , quasar mode agn feedback from the growing black holes , and starburst driven winds , which all occur nearly simultaneously . the fraction of galaxies at a given mass that has undergone such a transition is an increasing fraction with increasing galaxy mass , a trend that is required to be consistent with the observations . this model requires that most galaxies do not return to a star forming phase after the star formation initially stops , so a long term preventive feedback mechanism is still necessary to prevent further gas accretion , perhaps `` radio mode '' agn feedback @xcite . even the need for this `` maintenance '' feedback could be avoided in a large fraction of massive halos owing to the natural development of constant density cores that prevent cooling from the hot atmosphere , as we found in k09 . however , if the accretion of cold gaseous clumps also occurs in massive halos ( i.e. if `` cold drizzle '' is not a numerical artifact ) , then this maintenance feedback must also be able to prevent the bulk of this form of accretion . in addition , accounting for metal cooling can enhance the cooling in massive halos where it can dominate at temperatures typical of halo gas . this could make the problem of preventing the accretion and re - accretion of gas in massive halos even more severe . of course the major mergers of late type galaxies have the additional advantage that they result in remnants with early type morphologies @xcite , as required for the majority of galaxies on the red sequence . therefore , most red sequence galaxies in such models would undergo colour and morphological transformations as a consequence of the same astrophysical process . in addition , this violent feedback combination could contribute to a faster termination of the cold accretion mode in massive halos , especially at low redshift . at high redshift this might happen only in very massive halos hosting massive galaxies with modest gas fractions , because the high gas fraction of lower mass galaxies make merger driven central gas flows and therefore starburst and quasar winds less efficient @xcite . in conclusion , agn radio mode feedback in very massive objects , mimicked by hot mode removal , can dramatically lower the ssfrs of massive galaxies but it should not affect every object with hot mode accretion and it should have a galaxy mass dependence . the total removal of hot mode accretion removes too much recent star formation . as a consequence , semi - analytic models that completely remove this accretion in massive halos must adopt a hot mode suppression threshold mass , i.e. the mass above which hot mode is suppressed , that is higher than the mass where hot virialised halos dominate @xcite . however , this approach , without additional assumptions , might have difficulty matching the evolution of the red sequence over time , as shown by ( * ? ? we describe some observational consequences of a cosmological sph simulation of galaxy formation in a @xmath1cdm universe using gadget-2 . the simulation covers a wide dynamic range : it is able to resolve galaxies with masses larger than @xmath67 and contains several cluster size halos with masses @xmath68 . the sph algorithm used in gadget-2 is not prone to numerical overcooling . however , the simulation does not incorporate strong feedback , and , like other simulations and analytic models with weak feedback , it predicts an excessive fraction of baryons in galaxies . detailed analysis of the simulation ( see k09 ) allows us to understand the physics of galaxy assembly in the absence of strong feedback , and the comparison to observations in this paper indicates what form of feedback ( ejective or preventive , dependence on redshift and galaxy mass ) are needed to reconcile @xmath1cdm predictions with observed galaxy populations . we compare the observed galaxy mass function with the simulation results and conclude that the simulations overproduce galaxies at all masses . the problem is most severe at the low and high mass ends . the removal of baryons that were accreted through hot mode , to mimic an extreme form of preventive feedback , only mildly lowers their mass , since the hot mode contribution to the total baryonic mass in galaxies is modest . this form of feedback is not enough to bring the stellar mass function of the simulated galaxies into agreement with the observations , even at the high mass end . this failure owes to cosmic `` downsizing '' , where most of the massive galaxies have already formed most of their stellar mass in smaller objects at high redshift , and these smaller systems accreted their mass through cold mode accretion . these findings suggest that an extremely efficient feedback mechanism is necessary in high redshift galaxies at low and intermediate masses to reduce their masses substantially . then , owing to the hierarchical nature of massive galaxy buildup , this will significantly lower the masses of these massive galaxies at late times . a natural candidate for this feedback mechanism is starburst driven winds , although our analysis does not directly constrain the exact nature of the feedback . however , since cold mode dominated galaxies must predominantly be affected , it must be some form of ejective feedback . once the galaxy halo gains enough in mass to become hot mode accretion dominated and develops a hot atmosphere , this ejective feedback will be naturally quenched . at low redshifts an additional feedback mechanism is necessary to reduce the masses of low mass galaxies . this feedback mechanism must have the property that it substantially reduces the galaxy masses but retains similar sfrs in low and intermediate mass star forming galaxies today . we demonstrate that most of the simulated galaxies are part of the star forming sequence whose properties are similar to the observed star forming sequence . however , while this sequence contains only a small fraction of observed galaxies at high masses , we show that a large fraction of simulated massive galaxies are still part of this star forming sequence at @xmath0 . therefore , these massive galaxies have specific star formation rates that are too high on average , caused by a small amount of residual hot mode accretion , by potentially numerically induced and hence potentially spurious cold mode accretion , and by merging with gas rich smaller objects in massive halos . the removal of the hot mode accretion , which mimics extreme preventive feedback in hot halos , like an extreme form of radio mode agn , is enough to make the massive galaxies red enough on average . such extreme feedback , however , should be selective since the removal of all the hot mode accretion from all galaxies ruins the agreement with the observed star forming sequence . this suggests that the feedback in halos with @xmath69 should affect around half of galaxies with intermediate masses , e.g milky way mass halos , but should affect most of the galaxies in group and cluster size halos . this could be accomplished though feedback that occurs during the gas rich major mergers of galaxies , especially if such mergers occur in hot virialised halos . we are grateful to samir salim for sharing with us stellar masses and ssfrs from his paper and for useful comments . d.k acknowledges the support of the itc fellowship at the harvard college observatory . we also acknowledge support from nsf grant ast-0205969 and from nasa grants nags-13308 and nng04gk68 g .

we study the reionisation of the universe by stellar sources using a numerical approach that combines fast 3d radiative transfer calculations with high resolution hydrodynamical simulations . by supplementing a _ one - step _ radiative transfer code specifically designed for following ionisation processes with an adaptive ray - tracing algorithm , we are able to significantly speed up the calculations to the point where handling a vast number of sources becomes technically feasible . this allows us to study how dim low - mass sources , excluded in previous investigations owing to computational limitations , affect the morphological evolution of the reionisation process . ionising fluxes for the sources are derived from intrinsic star formation rates computed in the underlying hydrodynamical simulations . analysis of numerically converged results for star formation rates and halo mass functions allows us to assess the consequences of not including low - mass objects and enables us to correct for resolution effects . with these corrections , we are able to reduce the effective mass resolution limit for sources to m , which is roughly an order of magnitude smaller than in previous studies of this kind . our calculations reveal that the process by which ionised regions in the igm percolate is complex and is especially sensitive to the inclusion of dim sources . moreover , we find that given the same level of cosmic star formation , the number of ionising photons required to reionise the universe is significantly overestimated if sources with masses below m are excluded . this result stems from the fact that low - mass sources preferentially reside in less clumpy environments than their massive counterparts . consequently , their exclusion has the net effect of concentrating more of the cosmic ionising radiation in regions which have higher recombination rates . we present the results of our reionisation simulation assuming a range of escape fractions for ionising photons and make statistical comparisons with observational constraints on the neutral fraction of hydrogen at derived from the sdss quasar of becker and coworkers . we find that given the amplitude and form of the underlying star formation predictions , an escape fraction near is most consistent with the observational results . in these models , reionisation is expected to have occurred between , although the igm remains fairly opaque until . our method is also capable of handling the simultaneous reionisation of the helium component in the igm , allowing us to explore the plausibility of the scenario where sources with harder spectra are primarily responsible for reionisation . in this case we find that if the sources responsible for reionising hydrogen by had spectra similar to agns , then the helium component of the igm should have been reionised by . we find that such an early reionisation epoch for helium does not necessarily conflict with observational constraints obtained at , but may be challenged by future observations at higher redshifts . the recent wmap measurements of the electron scattering optical depth ( according to the `` model independent '' analysis of kogut et al . ) appear to be inconsistent with the relatively late onset of reionisation by the normal population ii type stars that we consider . in order to simultaneously match the observations from the sdss quasar and the optical depth measurement from wmap with the sources modeled here , we require an evolving escape fraction that rises from near to at . such a steep enhancement in the stellar production rate of ionising photons would be consistent with an imf that becomes more and more top heavy with increasing redshift .

astro-ph0303098

the nature of the sources responsible for ionising the neutral hydrogen in the intergalactic medium ( igm ) has been the subject of a long - standing debate . there are both observational and theoretical indications that some or perhaps all of this ionising background comes from stars and not hard sources , such as quasars . while it seems clear that quasars provide a significant contribution to the ionising background at low redshifts , @xmath14 , the steep drop in the abundance of bright quasars at earlier times brings into question whether they are capable of maintaining the ionised state of the igm at @xmath15 . in an earlier paper ( sokasian et al . 2003 ) , we showed that the observed opacities in both h i and he ii indicate that quasars alone can not produce the required emissivity to match observations at @xmath16 . the situation is exacerbated at higher redshifts where observations by fan et al . ( 2000 ) , becker et al . ( 2001 ) , and djorgovski et al . ( 2001 ) imply that the igm was highly ionised at @xmath7 even though the emissivity from bright quasars was likely unimportant . given this discrepancy , it appears that a substantial portion of the ultraviolet photons at @xmath17 were produced by an additional class of sources . it has been argued that star forming galaxies may yield the required ionising emissivities ( e.g. haehnelt et al . 2001 , bianchi et al . 2001 , steidel et al . 2001 ) , although it is possible that harder low - luminosity active galactic nuclei ( agn ) may also make up the difference ( haiman & loeb 1998 ) . early star formation also provides a possible explanation for the widespread abundance of metals in the igm ( cowie et al . 1995 ) while reionisation by faint agns may provide a more plausible solution if the escape fraction for stellar radiation from galaxies is restrictively small ( schirber & bullock 2002 , wood & loeb 2000 ) . there have been numerous studies of cosmological reionisation using semi - analytical methods ( e.g. haardt & madau 1996 , madau et al . 1999 , chiu & ostriker 2000 , valageas & silk 1999 ) . in these models , the impact of the uv background is treated using a spatially and directionally averaged radiative transfer equation which incorporates the mean luminosity function per unit volume for the source term and the statistical properties of absorbing clouds for the recombination term . inevitably though , to accurately address questions related to the time - dependent propagation of radiation fronts , especially during the early stages of the reionisation process where the radiation field is highly inhomogeneous , a detailed radiative transfer approach is required . this has proved to be an extremely challenging endeavour owing to the high dimensionality of the problem . a brief review of the problem and the various techniques which have been employed in recent years is presented in razoumov et al . ( 2002 ) . one of the major challenges involved with such exercises is determining how to efficiently process the large number of sources that may be responsible for reionisation . in the case of bright quasars , the associated number density is small enough that straightforward approaches can be employed ( e.g. sokasian et al . 2001 , 2002 ) . the situation becomes much more difficult when galaxies are the primary sources . for typical cosmologies , a @xmath18 mpc@xmath19 comoving volume can easily harbour @xmath20 star forming galaxies by @xmath21 . tracking the ionising flux from such a large pool of sources over many time steps inevitably requires the implementation of sophisticated algorithms designed to efficiently process the cumulative effects from each source . by combining a photon conserving algorithm with two independent hierarchies of trees , one for rays ( abel & wandelt 2002 ) and one for the sources themselves , razoumov et al . ( 2002 ) were able to improve the efficiency of the calculations to study the inhomogeneous process of reionisation from star forming sources down to a redshift of @xmath22 . in particular , they were able to carefully follow time - dependent radiative transfer calculations on a uniform @xmath23 cartesian grid representing a volume with a comoving length of @xmath24 mpc on each side which stored the results of pre - computed hydrodynamical simulations of galaxy formation . as a result of their calculations , the authors were able to address a number of questions related to the morphological evolution of the reionisation process . in particular , they showed a picture of stellar reionisation in which photons initially do not travel far from ionising sources , in contrast to images from simulations by gnedin ( 2000 ) . more recently , ciardi et al . ( 2003b ) combined high - resolution n - body simulations and a semi - analytic model of galaxy formation with a monte carlo radiative transfer code to study reionisation in a volume with a comoving length of @xmath25 mpc . in their analysis they assessed the effect of the environment by performing a subsequent simulation on a smaller @xmath26 mpc comoving box centered on a cluster . they found that the environment where the radiation is produced modifies reionisation through the influence of the density dependence of the recombination rates . this conclusion raises the question of whether the exclusion of the copious number of dim low - mass sources which fall below the resolution limit of these simulations may play a more important role in determining the morphological evolution of the reionisation process . the fact that most models of galaxy formation predict that the bulk of the resultant ionising radiation is produced in relatively massive systems has provided some motivation for ignoring the low - mass sources altogether , especially in lieu of the high computational cost of radiative transfer calculations . however , since dim low - mass sources reside in relatively less clumpy environments compared to their brighter counterparts , their overall contribution to the ionising emissivity in the igm may be significant in spite of their intrinsic faintness . we examine the impact of low - mass sources by conducting very fast radiative transfer calculations capable of processing a large number of sources from high resolution hydrodynamic simulations . in particular , by supplementing our previous _ one - step _ radiative transfer code ( see sokasian et al . 2001 ) with an adaptive ray tracing algorithm we are able to significantly speed up the calculations to the point where handling more than @xmath27 sources becomes technically feasible . our cosmological simulation is taken from a high resolution run in a series of calculations performed by springel & hernquist ( 2003a ) , and utilises @xmath28 particles in a box of comoving length @xmath29 mpc . radiative transfer calculations are then performed on a 200@xmath19 grid where we store the corresponding density fields at each redshift . this allows us to simulate reionisation with a source mass resolution that is about an order of magnitude below the limit of the ciardi et al . ( 2003b ) simulations . furthermore , the use of a 200@xmath19 grid offers a factor of @xmath30 improvement in spatial resolution for the same box length . it is also important to point out that by using hydrodynamics to describe the gas physics , rather than an n - body simulation , we are able to estimate the clumping on unresolved scales in our grid by computing the spatial distribution of particles within a given cell ( see sokasian et al . 2001 ) . this represents an important advantage over the density interpolation schemes associated with n - body codes which smooth out density variations on sub - grid scales . as a result we expect to more properly account for the recombination rates in high density regions where clumping is important . in the present paper , which is part of a series examining reionisation , we focus on the redshift interval @xmath31 to @xmath7 with specific interest in how the level of cosmic star formation predicted in the comprehensive set of hydrodynamic simulations conducted by springel & hernquist ( 2003a ) performs in the context of observational constraints on the opacity of the igm at @xmath7 . our method also treats the simultaneous reionisation of helium in the igm , allowing us to explore the plausibility of the scenario where agns are primarily responsible for reionisation . here , we restrict our attention to sources of radiation similar to normal , star - forming galaxies at the present day . in particular , we do not account for massive , population iii stars that may have been prevalent at @xmath32 . tentative measurements of polarisation in the cmb by the wmap satellite ( kogut et al . 2003 ) suggest that much of the igm was ionised earlier than indicated by the sdss quasars ( e.g. , spergel et al . we explore the implications of this result in the context of our present model by allowing the escape fraction to evolve with redshift . in the future , we will explore the combined effect of reionisation by both population ii and population iii stars . in the next section we review some of the technical aspects related to our method , including the approximations inherent to our radiative transfer calculations . we then follow with a description of the cosmological simulation that will be used in our analysis . next , we discuss our methodology for source selection and present an in - depth analysis regarding resolution effects and how to correct for them . finally , we discuss the results of our reionisation simulation and review the properties of the inhomogeneously ionised igm and make comparisons with observational constraints . in general , a description of the evolution of ionised regions around cosmological sources requires a full solution to the radiative transfer equation . such a solution would yield everywhere the monochromatic specific intensity of the radiation field in an expanding universe : @xmath33 , where @xmath34 is a unit vector along the direction of the propagation of a ray with frequency @xmath35 . presently , it is computationally impractical to obtain a complete , multi - dimensional solution for @xmath36 at the high resolution required for cosmological simulations . our approach relies on an algorithm which approximates the full solution from ionising sources by semi - independently processing the net effect of each source separately ( see sokasian et al . 2001 ) . owing to the large number of sources and the fact that the order in which sources are processed is randomised at each step , the departure from the fully self - consistent solution is negligible . ( we have verified this explicitly by running different cases in which the ordering of the sources varies and shown that our results are unaffected . ) by adopting this approximation , we are able to employ a simple jump condition to compute all radiative ionisations from a given source in a single step . this eliminates the need to repeatedly re - cast rays and calculate rates at every time step , thereby greatly speeding up the analysis . ionisation fractions within ionised regions are then computed by assuming photoionisation equilibrium . to further speed up the calculations , the ray casting scheme used in sokasian et al . ( 2001 ) has been modified to incorporate an adaptive scheme based on the method described by abel & wandelt ( 2002 ) . in particular , the transfer of direct source photons is computed on a uniform cartesian grid by casting a set of base rays which subsequently split hierarchically according to a local refinement criterion which involves keeping the ratio of the area of a cell to the area associated with the ray on a given hierarchical level above a fixed threshold value . this criterion ensures that a minimum number of rays pass through any given cell . the underlying pixelisation scheme that governs how rays are split has the property of exactly tiling a sphere with equal areas , ensuring equal photon flux per ray for isotropic sources . rays are propagated until they travel at most @xmath37 times the box length ( assuming periodic boundary conditions ) or the assigned number of photons is exhausted . once the ionisation level in the simulation box has reached significant levels , periodic boundary conditions are turned off and photons reaching the edges are added to a pool associated with the diffuse background . at the end of each interval , the effect of a diffuse ionising background is accounted for by casting rays of ionising radiation ( according to a detailed prescription ) inward from the sides of the simulation box . density fields , clumping factors and information regarding sources is specified from outputs at desired redshifts from the underlying cosmological simulation ( see sokasian et al . 2001 for technical details ) . our approach entails a number of straightforward approximations to simplify the calculations ; we summarise them briefly here . first , our radiative transfer calculations are done on a uniform grid whose scale @xmath38 will always be much smaller than the horizon , @xmath39 , where @xmath40 is the speed of light and @xmath41 is the time dependent hubble constant . this eliminates the need to include doppler shifting of frequencies in line transfer calculations . additionally , if the light crossing time , @xmath42 , is much shorter than the ionisation timescale , the time dependence of the intensities drops out as well . in the volumes we simulate , this will certainly be true and so we make this approximation . next , we assume that the density field will experience negligible cosmological evolution during a single timestep . this assumption allows us to perform all our radiative calculations during a given timestep in a static density field , greatly reducing the complexity of the algorithm . in our analysis we find that a constant time step of @xmath43 yr satisfies the latter assumption and manages to capture the characteristic timescale over which source intensities change . moreover , we find that the chosen timestep allows us to reliably track the morphological evolution of the reionisation process during early epochs when the rate of sources switching on is relatively fast . finally , we ignore the dynamical consequences of thermal feedback into the gas from radiative ionisation , enabling us to decouple our calculation of the radiation field from the hydrodynamical evolution of the gas . this allows us to use existing outputs from cosmological simulations to describe the evolving density field during the reionisation process . in reality , photoheating introduces extra thermal energy into the medium , and as ionisation fronts ( i - fronts ) move from small scales to large scales , there is a corresponding transfer of power from small to large scales through nonlinear evolution . this effect is somewhat accounted for in the underlying cosmological simulation used in this paper which includes a uniform ionising background capable of heating the gas . however , one also expects additional heating due to radiative transfer effects during the reionisation process , and such uniform backgrounds can not reproduce the observed increase in gas temperatures from the extra heating ( abel & haehnelt 1999 and references therein ) . as a result , we expect minor systematic errors to be present in our solutions . however , the main purpose of this paper is to describe the general morphological evolution of the reionisation process on large scales , which is insensitive to these systematic errors . in particular , recombination rates depend weakly on temperature and hence a proper accounting of photons is possible even if the temperatures are not computed precisely . one of the models we consider in this paper includes sources which are capable of ionising both hydrogen and helium simultaneously . in this case , it becomes necessary to couple the corresponding ionisation fractions during the radiative transfer calculations . exact solutions require taking into account the detailed frequency dependences of all the relevant processes and solving the multi - species chemical reaction flows in nonequilibrium ( e.g. anninos et al . owing to the large number of sources which need to be processed in our study , the implementation of a fully consistent nonequilibrium chemistry solver would require computations that are not presently feasible . fortunately , there are a number of simplifying approximations which we are able to employ that greatly speed up the calculations without compromising the main objectives of our study . our analysis of hydrogen and helium reionisation requires us to track three ionisation zones : h ii , he ii , and he iii . in order to simplify our calculations , we decouple the ionisation calculations by applying separate ray tracing calculations for each species . to properly account for the recombination rates implicit in the radiative transfer calculations , we are required to follow a particular ordering for the ray tracing . specifically , since he iii ionisation zones will always reside within h ii zones , we first apply ray tracing calculations for he ii reionisation which incorporate recombination rates based on electron densities that assume complete ionisation in both hydrogen and helium . although the equations for h i and he ii ionisation are , in principle , coupled , they can be separated to good accuracy using the fact that recombination processes from he iii to he ii are able to ionise h i. in particular , recombination emission from he ii l@xmath44 ( @xmath45 ev ) photons , which diffuse slowly owing to resonance scattering , and he ii balmer - continuum photons , which are concentrated close to the h i ionisation threshold , will tend to ionise h i within the he iii zone . for near cosmic abundances , these two processes are just about sufficient to balance h ii recombination within this zone ( e.g. osterbrock 1989 ) . an additional source of h i ionising radiation produced in this zone is he ii two - photon continuum emission for which @xmath46 ev ( the spectrum peaks at 20.4 ev ) . most of these photons are able to escape out of the he iii zone and should , in principle , be added to to the pool of photons which will be responsible for ionising the h i zone . in order to access the maximal contribution from this process relative to the contribution already present from the stellar radiation field , we examine the hypothetical scenario where the igm is completely doubly ionised in helium by @xmath47 . assuming a gas temperature of @xmath48 k and a gas clumping factor of 30 , we estimate that the total contribution of ionising photons from he ii two - photon emission will comprise only 1% of the total number of photons produced directly from star - forming galaxies at the same redshift . given the fact that the uncertainties associated with intrinsic source intensities are larger , we ignore this additional component to simplify the calculations . once the he iii zone has been determined , the next step involves calculating the associated h ii and he ii zones around the same source . here we note that the stellar radiation field responsible for ionising the latter two species can only arise from the hottest ob stars in the galaxies . this assumption is based on the fact that only these stars have strong enough stellar winds capable of producing _ escape routes _ through the host galaxy for their ionising radiation . since the intrinsic spectra of ob stars yield a large fraction of photons with @xmath49 ev , the outer boundaries of the ionisation zones for h ii and he ii coincide , thereby allowing the tracking of both zones via a single ray tracing calculation . the approximation here will be to track h ii ionisations explicitly via ray tracing while implicitly assuming that he ii is ionised to the same degree and extent . since the relative cosmic abundance of helium is small ( @xmath50% by number density ) , the inaccuracies introduced by this approximation should be negligible . note that in carrying out the radiative transfer calculations for the ionisation of h i , we are careful to ignore h ii recombinations within the he iii regions for the reasons presented earlier . after the ionisation zones have been determined , ionisation fractions for he iii and h ii are calculated by applying the following equilibrium condition for each case , @xmath51 where @xmath52 is the volume averaged gas clumping factor within the cell , @xmath53 is the temperature dependent case a recombination coefficient , and @xmath54 , @xmath55 , and @xmath56 are the ionised , neutral , and electron number densities , respectively . in this paper , we assume a constant temperature of @xmath57 k for all regions which have undergone ionisation to he iii and @xmath58 k for regions having undergone ionisation to h ii and he ii only , ( see abel and haehnelt 1999 ) . this allows us to simplify our calculations by using constant recombination coefficients . from the fitting formulae given in hui & gnedin ( 1997 ) , we compute the relevant coefficients to be @xmath59 cm@xmath19 s@xmath61 and @xmath62 cm@xmath19 s@xmath61 for @xmath63 k and @xmath64 cm@xmath19 s@xmath61 for @xmath65 k. the photoionisation rate in an ionised cell for a given species will be approximated by @xmath66 where @xmath67 is the mean cross section for photoionisation , @xmath68 is the number of relevant ionising photons released per second from the source , and @xmath69 is the distance from the source to the cell in question . assuming that the sources have intrinsic spectral energy distributions of the form @xmath70 in the relevant frequency range of interest and that @xmath71 beyond the ionisation edge , we can approximate the mean photoionisation cross sections according to @xmath72 where @xmath73 is the cross section at the ionisation frequency . for stellar sources we choose @xmath74 ( up to 4 ryd ) based on the extrapolated far - uv synthetic spectrum for starburst galaxies derived in haehnelt et al . interestingly , this value is also consistent with estimates for the average quasar spectral index ( see the relevant discussion in sokasian et al . 2002 ) , simplifying our analysis involving reionisation by harder sources such as agns ( see 5.4 ) . in this paper we adopt @xmath75 @xmath76 and @xmath77 @xmath76 from osterbrock ( 1989 ) . to further simplify our calculations , we remove the coupling between the ionisation fractions of the two species by approximating separate expressions for the electron densities in each ionisation zone . specifically , ionisation fractions for he iii are computed first by approximating the medium within the he iii as fully ionised in he ii and h i , thus giving @xmath78 where @xmath79 and @xmath80 represent the total hydrogen and helium particle densities , and @xmath81 is the ionisation fraction for he iii . ionisation fractions for h ii are then computed using @xmath82 where we have approximated @xmath83 based on our earlier assumptions related to the nature of the sources responsible for ionising hydrogen and singly ionising helium . again , since the helium abundance is small the latter approximation should be relatively accurate . although our procedure for calculating ionisation fractions is not fully consistent , the resultant inaccuracies are relatively small and certainly fall within the broader uncertainties associated with the simulation as a whole . the approximations listed above offer computational benefits that allow us to process a large number of sources easily . since they do not compromise the main objective of this paper , which is to study the morphological evolution of reionisation , we feel justified in adopting them . the cosmological simulation we will use in our analysis is based on a sph treatment of a @xmath84cdm model with parameters @xmath85 , @xmath86 , hubble constant @xmath87 km s@xmath61 mpc@xmath61 with @xmath88 , baryon density @xmath89 , and a scale - invariant primordial power spectrum with index n=1 , normalised to the abundance of rich galaxy clusters at the present day ( @xmath90 ) . these choices are in good agreement with those inferred from the wmap observations ( bennett et al . 2003 ) . the simulation uses @xmath91 sph particles and @xmath91 dark matter particles in a @xmath92 mpc comoving box , resulting in mass resolutions of @xmath93 m@xmath1 and @xmath94 m@xmath1 in the dark matter and gas components , respectively . this particular simulation represents one of the high - resolution runs ( simulation q5 ) in the series of calculations performed by springel & hernquist ( 2003a ) . these authors used a large set of high - resolution simulations on interlocking scales and at interlocking redshifts to infer the evolution of cosmic star formation from high redshift to the present . these simulations included a novel description for star formation and feedback processes within the interstellar medium and a treatment of galactic outflows ( springel & hernquist , 2003b ) , and employed a conservative formulation of sph based on following the specific entropy as independent thermodynamic variable ( springel & hernquist , 2002).the broad range of scales encompassed by their set of simulations , together with extensive convergence tests , enabled them to obtain a converged prediction for the cosmic star formation rate , @xmath95 , within the adopted model for galaxy formation . in the following section we will discuss the motivation behind our particular choice for the simulation scale . we will also describe our methodology for selecting sources and assigning intensities in the context of the converged prediction for @xmath95 from springel & hernquist ( 2003a ) . to accurately simulate the reionisation process due to stellar sources , one requires the cosmological volume to be sufficiently large to properly sample a representative range of halo masses . at the same time , one is inhibited from adopting an excessively large volume which can compromise the technical feasibility of the simulation owing to the large number of sources which would need to be processed . furthermore , one requires sufficient mass resolution within the simulated volume in order to reliably measure star formation rates in galaxies . this is especially true at high redshifts , where the bulk of the star formation occurs in objects of relatively low mass , as expected for hierarchical growth of structure , making it important to resolve low - mass objects . from the set of simulation runs presented in springel & hernquist ( 2003a ) , we find that their q5 run provides the most appropriate compromise ( @xmath92 mpc box length ) , capable of capturing a representative sample of the halo population while maintaining a level of mass resolution that reliably matches the converged star formation rate . our approach for defining sources relies on identifying virialised halos as sites of star formation . as in springel & hernquist ( 2003a ) , halos are located by employing a `` friends - of - friends '' ( fof ) group finder algorithm to the dark matter particles . the algorithm is restricted to dark matter particles only to avoid complications associated with tracking baryonic particles whose number and type can vary with time . the two main parameters which govern the group - finding algorithm are the comoving linking length and the minimum number of particles required for a group . for the former , we choose a fixed value equal to 0.2 times the mean interparticle spacing of dark mater particles , corresponding to a group overdensity of roughly 200 . the choice for the minimum number of particles which can define a group is somewhat more arbitrary . this is due to the fact that this parameter is invariably related to the question as to what exactly constitutes a group . a reasonable definition may be that given a specific overdensity , a group is considered to be _ real _ if it can be reliablely tracked over time . more specifically , this would correspond to being able to find a given halo in a subsequent time step . below a certain threshold , one might start to identify _ transient _ groups , that dissolve away in subsequent time steps . our statistics on halo tracking indicate that this threshold is around 100 dark matter particles , with associated uncertainty growing progressively towards lower particle numbers ; i.e. below about 100 particles one starts to occasionally pick up transient halos that are probably flukes , and the fraction of these slowly grows towards smaller groups . however , as we shall show in the following section , a significant portion of the total star formation rate is predicted to come from low - mass halos with fewer than 100 dark matter particles . these objects essentially represent the star - forming mini - halos of the universe . in order to account for these sources , we adopt an aggressively low minimum value of 16 particles for our definition of a group . this will inevitably raise the level of uncertainty regarding the reality of some of the low - mass halos and their corresponding star formation rates . however , this approach provides us with a list of low - mass objects whose number density and spatial locations are statistically consistent with results from simulations conducted with higher resolution . the star formation rates in these objects can then be systematically corrected so that the total contribution from all objects matches the converged result presented in springel & hernquist ( 2003a ) . our prescription for assigning star formation rates to the groups relies first on associating each gas or star particle with its nearest dark matter particle , and the fof - group this particle resides in . we then take the total mass of each group as a `` viral '' mass , and the sum of star formation rates in all of a group s sph particles as the total star formation rate in the halo . our working assumption will then be that every halo hosts a single galaxy which will act as a source . in reality , however , there may be more than one galaxy per halo . the difference , however , is negligible in the context of the spatial resolution of the radiative transfer calculations . more specifically , in the redshift range of interest , even the most massive halos have virial radii which are less than or comparable to the width of a single cell . in figure 1 , we plot various statistical quantities related to the star formation rate from the resultant source list as a function of halo mass for redshifts z = 14.7 , 10.4 , and 7.2 . at each redshift we have binned the halos by mass in logarithmic intervals of dlog@xmath96 = 0.1 and computed the associated star formation rate density , number density , fractional star formation contribution , and cumulative star formation contribution below a given mass . from the first two rows of plots , it is evident that while the relative amplitude in the star formation rate becomes smaller towards the lower mass scales , the number of halos rises dramatically . note also the overall rise of the star formation rate with decreasing redshift . in hernquist & springel ( 2003 ) , the authors used simple analytic reasoning to identify the physical processes that govern this evolution . as part of their conclusions , they find that at early times ( @xmath97 ) , densities are sufficiently high and cooling times are sufficiently short that abundant star - forming gas exists in all dark matter halos with @xmath98 that can cool by atomic line cooling . consequently , the evolution of the star formation rate is dominated by gravitationally driven growth of the halo mass function , as is seen here . the relative contribution to the total cosmic star formation rate density as a function of halo mass is shown in the bottom two rows of figure 1 . here we see that intermediate mass scales provide the largest contribution to the overall total , with a progressively larger contribution from higher mass scales at decreasing redshifts . owing to the finite dynamic range of the cosmological simulation , we can not expect to resolve _ all _ the star forming halos that contribute to the ionising emissivity . this becomes progressively problematic at high redshifts where most of the star formation takes place in low - mass galaxies which form in abundant numbers as progenitors of more massive systems . furthermore , near the resolution limit of identifiable halos , the corresponding star formation rate becomes less reliable owing to the relatively small number of sph particles associated with the halo . the resultant star formation missed by our simulation due to these resolution effects is shown in figure 2 for the redshift range of interest . in the top panel we plot the total star formation rate from all the halos identified in our simulation ( dashed line ) and compare it to the physically motivated analytic fit to numerical converged results presented in springel & hernquist ( 2003a ) . the analytic fit , obtained from the analysis of hernquist & springel ( 2003 ) , takes the form of @xmath99 where , @xmath100 and @xmath101 , @xmath102 , and @xmath103 m@xmath1 yr@xmath61 mpc@xmath61 define the fitting parameters . the above fitting function was motivated by the bimodal behaviour in the evolution of the cosmic star formation rate . at low redshifts , the efficiency of gas cooling processes is diminished owing to the expansion of the universe and hence manifests itself through a scaling that is related to the expansion rate as quantified by the hubble constant . at high redshifts , the steep rise towards lower redshifts is related to the growth of the halo mass function which exhibits an exponential cut - off for large masses . to access the significance of the discrepancy between the numerical and the theoretical results in the context of the reionisation of the universe , we find it convenient to relate star formation rates in terms of the cumulative number of ionising photons released as a function of look - back time . in the bottom panel of figure 2 we plot this quantity in terms of a ratio between the simulation results and the theoretical prediction . here we have adopted a conversion factor of @xmath104 s@xmath61 ionising photons per star forming rate of @xmath105 m@xmath1 yr@xmath61 based in the analysis from madau et al . we find that in terms of the cumulative number of ionising photons released , the discrepancy does not become dramatic until one exceeds a redshift of @xmath106 or so . the discrepancy described above does not pose a serious obstacle in the context of our present goals ; we find that we can easily correct for the missing star formation rate through a simple algorithm by adjusting the rates in the low - mass halos . we start first by deciding upon a mass threshold for halos above which we are confident in our results for the associated star formation rate . after analysing the results from the group finding algorithm applied to our simulation , we find that independent of redshift , halos with masses @xmath107 m@xmath1 have statistically reliable values for their star formation rates . below this threshold , a halo may simply not have enough sph particles to make it possible to reliably compute the star formation rate or the halo may be unresolved altogether . the fact that our source list aggressively includes low - mass objects allows us to partly compensate for the missing objects . although the _ reality _ of some fraction of these low - mass objects may be questionable , the fact that the statistical properties regarding their number densities and locations matches those from higher resolution runs assures us that we can at least statistically capture the effect from this population of sources . by coupling the criterion regarding the _ reality _ of low - mass object with the criterion for confidence in the computed star formation rate , we can distinguish , in technical terms , two types of problematic sources : ( 1 ) objects whose _ reality _ is not questioned ( n@xmath108 ) , but whose star formation rate may be unreliable ( @xmath109 m@xmath1 ) , and ( 2 ) objects whose _ reality _ and corresponding star formation is questionable ( n@xmath110 and @xmath111 m@xmath1 ) . the net effect from both these sets of sources is to cause a deficit in the total cosmic star formation rate relative to the theoretical expectation . our method for correcting this deficit involves supplementing all objects whose masses fall below @xmath112 m@xmath1 , comprising both sets of problematic sources , with additional amounts of star formation in proportion to their mass . the total amount added to all these objects at a given redshift is normalised so that it exactly makes up the deficit relative to the theoretical prediction . in figure 3 we plot the comoving number densities and the star formation rates as a function of redshift for all the sources in the simulation . in the context of the deficit in the cosmic star formation rate , we have grouped the sources into three categories with the first column representing sources which do not require any corrections and the second and third columns including the aforementioned problematic sources . here the dashed line and solid line indicate the uncorrected and subsequently corrected contribution to the total star formation rate , respectively . it is evident from the plots that the corrections to the low - mass sources do not appear to be drastic and we are confident that the uncertainties related to these adjustments can not significantly affect our results . having adjusted star formation rates in the low - mass sources we can now estimate the magnitude of the discrepancies in the total star formation rate had we been limited to poorer mass resolutions . in figure 4 ( _ left panel _ ) we plot this discrepancy in terms of the fraction of the total star formation contributed from sources with halo masses above a certain mass . the total star formation rate is based on all halos above the resolution limit in our simulation ( @xmath113 m@xmath1 ) . in the same figure ( _ right panel _ ) we also plot source number densities as a function of redshift for the same mass ranges , with the total source density from the simulation also shown ( _ solid line _ ) . as in figure 1 , it is evident that the subsequent fraction of star formation contributed by relatively low - mass objects rises rapidly with redshift . the labeled mass inequalities were chosen to roughly match the source mass resolutions associated with recent numerical attempts to simulate igm reionisation by stellar sources ( see ciardi et al . 2003b and razoumov et al . interestingly , we find that in the context of our corrected source list , the quoted mass limits continue to significantly underestimate the relative contribution to the total star formation rate from low - mass objects even at intermediate redshifts when the ionising emissivity in the igm is expected to have reached appreciable levels . in particular , at @xmath114 the cases with @xmath115 m@xmath1 and @xmath116 m@xmath1 , which are close to the mass resolution of sources in the m3 and s3 simulations described by ciardi et al . ( 2003b ) , appear to account for only @xmath117 and @xmath118 of the total star formation rate , respectively , relative to our complete source list . it is of course possible to renormalise the star formation rates in the massive ( resolved ) sources so that the total star formation rate is corrected to agree with our converged results . however , such a simple correction fails to address the potentially significant impact the environments around sources can have on the amount of ionising radiation that can escape into the igm . in particular , by concentrating the ionising flux away from low mass system which are more likely to reside in less clumpy environments than their more massive counterparts , one runs the risk of systematically overestimating recombination rates . we will explore the ramifications from such corrections in context of our own results in 5.1 . given the redshift range of interest , @xmath119 , and our time step for updating source lists , @xmath120 yrs , the total number of sources which require processing reaches nearly @xmath121 . since our radiative transfer calculations are performed on a cartesian grid with a limiting resolution , it is logical to group together sources within cells as a method of reducing the number of sources . owing to the high resolution of our grid ( @xmath122 ) , the overall number of sources is reduced only by @xmath123 . such a marginal reduction in sources assures us that the resolution of our grid provides excellent resolving power in regards to the spatial distribution of the sources . in our analysis , we carefully follow the reionisation process from a set of models parameterised by the escape fraction . in this context , the escape fraction is defined to be the fraction of ionising photons intrinsic to each source that escape and enter into the radiative transfer calculations . it must be noted here that this parameter inevitably also carries with it any uncertainties associated with the underlying amplitude of the star formation rates computed for the sources . nevertheless , our approach allows us to predict the epoch of reionisation as a function of this parameter . by comparing these predictions with recent observational constraints on the escape fractions and neutral densities near @xmath3 , we are able to test the overall consistency of the theoretical predictions of the computed star formation rates . along the way we are also able to make general statements regarding the morphological aspects of reionisation . the evolution of the global ionisation fraction in the simulation volume provides a useful way of characterising the overall ionisation state of the universe as a function of redshift . in figures 5a and 5b we show how the ionised mass - weighted and volume - weighted fraction , respectively , evolve with redshift . the rate of the evolution to a completely ionised state is quite similar for all the models , although there is a subtle trend towards a swallower rise of ionisation fractions with decreasing escape fractions . nevertheless , in all cases the reionisation epoch , which we define as the redshift when the global volume - weighted ionisation fraction reaches @xmath124 , occurs by @xmath125 . later in this section we show how a more detailed analysis involving statistical comparisons between simulated spectra and observational results for flux transmission at @xmath7 can lead to more rigid constraints as to how early reionisation could have taken place . figures 5a and 5b also show that the ionised mass - weighted fraction is consistently larger than the volume - weighted fraction at redshifts before reionisation . this result is seen more clearly in figure 5c where we have plotted the corresponding ratio of the two fractions . the larger values for the ionised mass - weighted fraction at early redshifts is consistent with the general picture of the pre - overlap epoch where sources are preferentially ionising their relatively dense surroundings before affecting the more tenuous igm . the fact that the inequality between the mass and volume fraction is amplified with decreasing escape fraction highlights an interesting morphological feature related to the number and intensity of the sources responsible for reionisation . namely , when escape fractions are low , a relatively larger number of sources are directly involved in the process of ionising a neutral density field . since the ionisation zones associated with these sources are also smaller , the net effect is that massive regions are more likely to be ionised from the ionising radiation emanating from within rather than radiation that has first traversed the less dense igm . this predilection naturally manifests itself in relatively larger ratios for the mass - to - volume ionisation fractions during the early stages of reionisation in these cases . eventually , intervening patches of neutral material become ionised and there is a sharp rise in the intensity of the ionising background as contributions from different sources combine . as this diffuse component comes to dominate the radiation field , all regions become uniformly ionised and the ratio of mass - to - volume ionisation fraction converges to unity . a visual illustration of the reionisation process is shown in figure 6 where we plot a series of projected slices through the simulation volume . in each panel , a @xmath126mpc slice from the @xmath127 run is projected in both density and ionisation fraction . source locations in each slice are denoted by white crosses making it easier to follow how the ionised regions ( _ blue _ ) percolate to turn a neutral igm ( _ yellow _ ) into one that is highly ionised ( _ red _ ) . compared with earlier studies of reionisation , figure 6 shows that morphologically the ionised regions in our analysis trace the large scale distribution of sources . in particular , we do not find that reionisation occurs mainly from low- to high - density regions , as would be the case if photons escaped preferentially into the igm before ionising the high density regions near the sources . the panels in figure 6 graphically illustrate the large number of sources that are involved in the highly inhomogeneous reionisation process . a more dramatic illustration of this point is shown in figure 7 where we have plotted a 3d image of the same simulation showing iso - surfaces around the ionised regions at @xmath128 . this highly inhomogeneous morphology represents a departure from the general picture of reionisation proposed by gnedin ( 2000 ) where only a handful of bright sources are responsible for reionising the universe . in the following section we discuss how the morphology of the reionisation process can affect the number of ionising photons required to achieve full reionisation . the number of ionising photons per hydrogen atom required for the ionised regions around sources to completely overlap has developed into an interesting topic that is commonly addressed in studies of this kind ( ciardi et al . 2003b , razoumov et al . 2002 , haiman et al . 2001 , miralda - escud et al . 2000 , gnedin 2000 ) . to reliably estimate this quantity with simulations requires a high level of resolution capable of properly estimating gas clumping factors which govern the overall recombination rate . this requirement is especially important in light of the results presented in haiman et al . ( 2001 ) which show that mini - halos with temperatures below @xmath129k can dominate the average clumping in the early stages of reionisation and significantly increase the required photon budget necessary to achieve complete overlap . given the high resolution inherent to our hydrodynamic simulations and the corresponding radiative transfer grids , coupled with our ability to probe clumping factors on sub - cell scales based on the underlying distribution of sph particles , we are able for the first time to probe the scales needed to account for recombination effects arising from such low - mass objects . this represents a significant improvement upon previous attempts which have relied on conservative assumptions related to the escape fraction to accommodate unresolved clumping . in figure 8 , we plot the cumulative number of escaping ionising photons per hydrogen atom as a function of redshift . the diamond symbol on each curve corresponding to different values for @xmath130 represents the point at which the volume ionisation fraction exceeds @xmath124 . for @xmath131 , between 2 and 3 ionising photons are required to reionise the universe while 4 - 5 ionising photons are needed in the case where @xmath132 . the difference in the number of required ionising photons resulting from variations in the escape fraction arises from the interesting interplay between the evolution of clumping factors and the mean density of the universe . in particular , the rapid evolution of the mean density plays a slightly more important role than the evolution of clumping factors in the redshift range of interest . therefore , a delay in reionisation to lower redshifts requires fewer ionising photons since the decline in gas densities outweighs the corresponding increase in clumping factors , resulting in comparatively fewer recombinations occurring than at higher redshifts . environmental factors influence the number of photons required to produce reionisation . a related issue is whether the distribution of ionising intensities among the sources themselves can also have an important impact on this process . in particular , given the same amplitude and evolution of the cosmic star formation rate , it is interesting to explore how a redistribution of the ionising flux from dim low - mass sources to more massive sources would affect reionisation . in figure 9 , we examine this scenario by way of comparisons involving the volume - weighted ionisation fraction and the cumulative number of ionising photons that are released . specifically , we examine two situations : 1 ) the case where we have retained the full source list which includes sources down to our resolution limit of @xmath133 m@xmath1 , and 2 ) the case where we have systematically transferred ionising fluxes from sources with @xmath134 m@xmath1 to the nearest neighbouring sources with masses above this limit . the figure shows that the affect can be significant , namely that reionisation takes place later and requires more photons ( by roughly @xmath135 ) for the same escape fraction ( @xmath11 ) when low - mass sources are excluded . this result stems from the fact that low - mass sources preferentially reside in less clumpy environments than their massive counterparts . consequently , the net effect of transferring ionising flux from these sources to more massive systems amounts to increasing the overall ionising intensity in regions with higher recombination rates . this leads to recombinations playing a more dominant role and thus delaying the onset of reionisation . this finding therefore supports the requirement that dim low - mass sources be properly resolved and included in future simulations of this kind . in particular , inaccuracies related to the inability to resolve low - mass sources and their corresponding contribution to the cosmic star formation rate may be amplified due to their preferential location in less clumpy environments . recent discoveries of quasars at redshifts 5.8 and greater ( becker et al . 2001 , fan et al . 2001 , fan et al . 2000 ) are finally making possible quantitative studies of the status of the igm at high redshifts . in fan et al . ( 2000 ) the spectrum of a @xmath136 quasar ( sdss j1010 - 0125 ) was examined and revealed no evidence for a gunn - peterson trough , indicating that the igm was highly ionised near @xmath137 . in subsequent observations ( fan et . al 2001 ) the authors studied three new quasars at @xmath138 and found a significant increase in ly@xmath44 absorption from redshift 5.5 to 6.0 . in particular , they found a @xmath139 region in the ly@xmath44 forest portion of the spectrum of a @xmath4 quasar ( sdss 1030.10 + 0524 ) which had a measured flux transmittance consistent with zero , indicating a flux decrement of @xmath140 , and suggesting the possible detection of a gunn - peterson trough . to more accurately quantify these findings becker et al . ( 2001 ) obtained higher resolution spectra of the three new @xmath138 quasars , allowing them to place better constraints on the status of the high redshift igm . to relate our results to these observations , we extract artificial ly@xmath44 absorption spectra from the simulation outputs and then derive statistical probabilities for the resultant flux transmittance which can readily be compared to the observational results . we extract artificial spectra using the particle information in the sph simulation coupled with the ionisation information from our radiative transfer calculations . for each model , we generate 500 spectra along randomly selected lines of sight ( loss ) between @xmath141 . our procedure is similar to that in the tipsy software package ( katz & quinn 1995 ) , but does not require that a los be perpendicular to a box face in the simulation volume . each los has a unique and arbitrary direction relative to the box coordinate system and wraps through the simulation volume repeatedly via periodic boundaries . using the smoothing kernels of the sph particles , gas densities and temperatures are computed along a los at appropriately sampled intervals . the component of the peculiar velocity of the gas in the direction of the los is also computed at each point . once all physical quantities have been gathered , voigt profiles are fitted to each spectrum by interpolating between the corresponding line - absorption coefficients provided in harris ( 1948 ) . it is important to point out that in computing the line profiles we use a minimum gas temperature of @xmath142 k as a correction to the sph temperatures which exclude the extra heating introduced by radiative transfer effects ( see abel and haehnelt , 1999 ) . our simulations were stopped at @xmath136 , allowing us to make comparisons with the observational results from the spectrum of the @xmath4 quasar . the presence of an apparent gunn - peterson trough in the spectrum of this quasar makes it an especially interesting case . by using the results of our simulated spectra to make statistical comparisons with this region and another region with a non - zero transmittance , we can constrain the evolution of the reionisation process . our analysis is based on probability distribution functions involving transmission levels that have been averaged over the same redshift range as the observational measurements . specifically , we focus on the redshift ranges @xmath143 and @xmath144 , corresponding to the regions over which the average transmittance was measured by becker et al . ( 2001 ) in the spectrum of the @xmath4 quasar . in figure 10 , we plot the probability distribution functions of simulated transmissions averaged over the two redshift ranges and overlay the corresponding @xmath145 measurements from becker et al . ( 2001 ) ( _ vertical - dashed lines _ ) . the distributions were obtained using 500 los , although we have only plotted a restricted range of transmissions for the purposes of observational comparisons . it is clear from figure 10 that the ionising emissivities predicted in the models with @xmath5 provide the most plausible match to the observational results in terms of transmission probabilities for the two redshift ranges . for this choice , reionisation is expected to have occurred near @xmath146 , although as the analysis here demonstrates , the igm remains fairly opaque until @xmath3 . this spread in redshift between the time when the universe first became ionised and the time when opacities declined enough to allow non - zero transmission in ly@xmath44 is important for the proper interpretation of observed quasar spectra . in particular , it is incorrect to assume that a volume - weighted ionisation fraction above @xmath124 will necessarily produce non - zero transmission levels consistent with current observations of quasar spectra at @xmath7 . a much more careful analysis involving extraction of spectra from simulations which include high precision information regarding ionisation fractions is necessary in order to make meaningful comparisons with observational results . interestingly , an escape fraction of @xmath147 is reasonably close to the range of values ( @xmath148 ) that have been observationally deduced from @xmath149 starburst galaxies ( see heckman et al 2001 ; hurwitz et al . 1997 ; leitherer 1995 ) . it is important to note that these observational results may be underestimating the true values of the escape fraction owing to undetected absorption from interstellar components . also unclear , in context of our analysis , is the possible evolution of the escape fraction with redshift which may lead to significantly different values at @xmath97 . in any case , the fact that we require a reasonable escape fraction in order to match the observations lends credence to the overall consistency of the theoretical predictions of the computed star formation rates . our radiative transfer calculations have been designed to handle the simultaneous reionisation of both the hydrogen and helium components of the igm . this allows us to explore the scenario where objects with spectra harder than stellar sources were primarily responsible for the reionisation of the universe . such a scenario has been proposed by e.g. haiman & loeb ( 1998 ) . in their analysis , a theoretical extrapolation of the quasar luminosity function at fainter luminosities and higher redshifts than currently detected appears to favour reionisation by low - luminosity mini - quasars ( or agns ) . reionisation by agn type sources has also been supported on the grounds that they may be the only type of sources capable of having large enough escape fractions to allow enough ionising photons to enter the igm ( see wood & loeb 2000 ) . to explore the implications of sources with hard spectra reionising the universe , we have run a model in which the sources are capable of ionising he ii in addition to h i. specifically , he ii ionising rates are computed for the sources by assuming a single power - law form for the spectrum , @xmath150 , normalised to deliver the same number of h i ionising photons below 4 ryd predicted from soft sources with @xmath11 . this ensures that the hydrogen component of the igm will reach roughly the same level of ionisation by @xmath3 as in the @xmath11 stellar model and therefore will also provide a reasonable match to the observational results of becker et al . ( 2001 ; see 5.3 ) . extrapolated results for the far - uv synthetic spectrum of a stellar population with continuous star formation reveals a typical quasar spectrum with @xmath151 up to 4 ryd . adopting the same spectral index above 4 ryd for the hard sources allows us to relate the rate of ionising photons , @xmath68 , in he ii and h i directly with the expression @xmath152 where @xmath153 is the ratio of ionisation threshold frequencies for the two species . this leads to a conversion factor of @xmath154 he ii ionising photons for every h i ionising photon . in figure 11 , we show the resultant evolution in the volume - weighted ionisation fraction for both soft and hard sources . note how the hydrogen component in the igm becomes reionised earlier in the case where hard sources are invoked . this is due to the fact that the he iii regions now surrounding the sources act as an additional source of h i ionising photons ( see 2.1 ) . the bottom panel of figure 11 shows that the helium component becomes reionised by @xmath155 in this case . such an early reionisation epoch for helium by high redshift sources such as agns should be testable in the coming decade with the next generation space telescope . nevertheless , an interesting question we can explore currently is whether such a scenario may also leave an imprint on he ii ionisation levels at lower redshifts where observations currently exist . in particular , we would like to test whether a hypothetical population of agns which suddenly turn off at @xmath156 will have lingering effects in terms of subsequent ionisation levels when another known population of quasars turn on . for such an analysis , we re - ran our most realistic quasar model ( model 5 ) from sokasian et al . ( 2002 ) with the initial conditions that the helium component in the igm is doubly ionised everywhere in the box . we present the results showing the subsequent evolution in the volume - weighted ionisation fraction in figure 12 . here the solid line represents the scenario described above while the dotted line indicates the original simulation where quasars turn on in a medium that is only singly ionised in helium . in the case where the igm is initially fully ionised , we see that he iii recombinations quickly cause the ionisation fraction to decrease shortly after @xmath156 when the ionising emissivity from quasars is still low . eventually , the emissivity from quasars grows and the ionisation fraction starts to rise again . however by @xmath9 enough recombinations have already taken place since @xmath156 to effectively remove all traces of an early reionisation epoch , leading to ionisation levels that are governed only by the quasar population . it thus appears as if the present observations can not rule out the simple agn scenario we have considered here . the recent tentative measurement of polarisation in the cmb by the wmap satellite ( kogut et al . 2003 ) suggests that reionisation occurred in a more complex manner than indicated by the sdss quasars alone ( e.g. , spergel et al . the measurements exhibit an excess in the cmb te cross - section spectrum on large angular scales ( @xmath157 ) yielding an electron optical depth to the cmb surface of last scattering of @xmath158 . the uncertainty associated with this measurement depends on fitting all parameters concerned with the tt power spectrum and the te cross power spectrum . kogut et al . ( 2003 ) obtain a 68% confidence range of @xmath159 corresponding to an instantaneous reionisation epoch which occurred between @xmath160 . none of the models we have considered in this paper is able reionise the universe by such early redshifts . the connection between the implications of the wmap results and those of the sdss quasars for the evolution of the igm is unclear . in an attempt to determine the extent to which the sources examined in this paper would require additional ionising emissivity to provide a consistent match with the wmap measurements we consider an evolving escape fraction of the form @xmath161 . here , a positive value for the parameter @xmath162 has the effect of producing ionisation levels consistent with the observations from the @xmath4 sdss quasar , as in the model with @xmath11 , while leading to relatively larger ionising emissivities at higher redshifts . we note that with this choice for @xmath130 , the `` escape fraction '' can exceed unity . in the context of our present analysis , this simply means that the rate of production of ionising photons is higher than would be implied for our choice of the star formation history and the imf . for simplicity , we absorb these uncertainties in what follows into our definition of the `` escape fraction . '' in figure 13 we compare three different cases for this form : 1 ) a case with a constant escape fraction of @xmath11 ( @xmath163 ; _ solid - line _ ) , 2 ) a case with @xmath164 resulting in an evolving escape fraction that rises to unity at @xmath165 around when the first sources turn on ( _ dotted - line _ ) , and 3 ) a case with @xmath166 resulting in an escape fraction that exceeds unity beyond @xmath167(_dashed - line _ ) . in panels b and c of figure 13 we plot the resultant evolution of the volume weighted h ii ionisation fraction and the evolution of the electron optical depth to electron scattering , @xmath168 , given by : @xmath169 where @xmath170 @xmath76 is the thompson cross section and @xmath171 is the mean electron number density at @xmath172 . in the calculation of @xmath168 we assume that helium is singly ionised to the same degree as the hydrogen component for @xmath173 and doubly ionised everywhere at @xmath149 . it is evident from figure 13 that matching the the optical depth measurement from wmap with the sources studied here would require substantially larger ionising emissivities at high redshifts . in particular , only the case with @xmath166 , which reaches an escape fraction greater that 10 near @xmath165 , is capable of producing the measured electron optical depth . such a large increase in the stellar production rate of ionising photons would require either a greatly enhanced star formation rate at high @xmath174 relative to that obtained by hernquist & springel ( 2003 ) or an imf which becomes more top - heavy with increasing redshift , as might be expected for population iii type stars . in a recent paper ciardi et al . ( 2003a ) argue that a separate population is not necessary if one adopts a top - heavy imf for population ii type stars with an optimistically large value for the escape fraction . however , it is unclear whether the escape fraction adopted in their analysis is meaningful because of their inability to resolve either clumping on sub - grid scales or objects below @xmath175 m@xmath1 . furthermore , in their best - fit model , reionisation is complete by @xmath176 . such an early reionisation seems difficult to reconcile with observations of the gunn - peterson effect measured in the @xmath4 sdss quasar ( becker et al 2001 ) . we have utilised a fast radiative transfer code to study the reionisation of the universe by stellar sources . our method has been applied to a high resolution cosmological hydrodynamical simulation which enables us to probe scales previously unresolved in similar type of studies . by supplementing a _ one - step _ radiative transfer code specifically designed for ionisation processes with an adaptive ray - tracing algorithm , we are able to significantly speed up the calculations and are able to handle a large number sources . one of the main goals of this analysis was to study the effect of low - mass sources residing in mini - halos on the overall morphology of the reionisation process . analysis of numerically converged results for star formation rates and halo mass functions in the underlying cosmological simulation allowed us to assess the resolution effects associated with low - mass objects and apply well motivated corrections . with these corrections , we were able to reduce the effective mass resolution limit for source objects to @xmath177 m@xmath1 , which is roughly an order of magnitude smaller than previous studies of this kind . our calculations reveal that the process by which ionised regions in the igm percolate is highly inhomogeneous and especially sensitive to the inclusion of dim sources . more specifically , we find that given the same level of cosmic star formation , the number of ionising photons required to reionise the universe is overestimated by @xmath178 when only objects with masses @xmath179 m@xmath1 are used as source locations . this result stems from the fact that low - mass sources preferentially reside in less clumpy environments than their massive counterparts . consequently , their exclusion has the net effect of concentrating more of the cosmic ionising radiation in regions which have higher recombination rates . our results from the full source list that includes objects down to @xmath0 m@xmath1 appear to show a similar evolution to reionisation for various values of the escape fraction . in all cases , the epoch of reionisation , which we define as occurring when the global volume - weighted ionisation fraction reaches @xmath124 , happens by @xmath125 . this result is consistent with the absence of a gunn - peterson trough in observations of quasars below @xmath3 . to relate our results to these observations , we extract artificial ly@xmath44 absorption spectra from simulation outputs and then derive statistical probabilities for the resultant flux transmittance . our comparisons are based on probability distribution functions involving transmission levels that have been averaged over the same redshift range as the observational measurements . specifically , we focus on the redshift ranges @xmath143 and @xmath144 , corresponding to the regions over which the average transmittance was measured by becker et al . ( 2001 ) in the spectrum of their @xmath4 quasar . we find that given the amplitude and form of the underlying star formation predictions , an escape fraction near @xmath11 is most consistent with the observational results . this leads to a reionisation epoch that is complete by @xmath180 and requires roughly @xmath181 ionising photons per hydrogen atom . in an attempt to explore the implications arising from a scenario where sources with harder spectra such as agns are responsible for hydrogen reionisation , we have analysed a model in which the spectrum of sources is allowed to extend beyond 4 ryd . as a result , sources are able to reionise he ii simultaneously with h i. our analysis shows that helium reionisation occurs near @xmath182 when the spectrum is normalised to deliver the same number of h i ionising photons below 4 ryd as predicted from soft sources with @xmath11 . in a hypothetical case where these hard sources abruptly turn off before the onset of quasars below @xmath156 , we find that an early reionisation of the helium component is not necessarily in conflict with observational constraints of he ii opacities obtained at @xmath9 . this is due to the fact that by @xmath9 , enough he iii recombinations haven taken place to cause ionisation levels to converge to the values governed by quasar emissivities alone . in this paper , we have focused on reionisation by stellar sources similar to those seen in local galaxies ; i.e. population ii type stars . our results show that the star - formation history inferred by springel & hernquist ( 2003a ) based on detailed hydrodynamic simulations can account for the properties of the igm measured from quasars at @xmath183 . to the extent that these comparisons are confirmed by future theoretical and observational studies , our results support the evolutionary history of `` ordinary '' star formation derived analytically by hernquist & springel ( 2003 ) . the recent measurements by the wmap satellite ( kogut et al . 2003 ) indicate that a large fraction of the igm may have been ionised early , by @xmath184 . taken together with the observations of the sdss quasars at @xmath185 , the implications of this result for the evolution of the igm are unclear . cen ( 2002 ) has proposed a scenario in which the universe was reionised early by population iii stars , but then much of the igm recombined once these stars were no longer able to form , and the universe was reionised a second time by the next generation of stars . our results in this paper with a constant escape fraction would describe the second stage of reionisation in this model , provided that the igm mostly recombined at intermediate redshifts . our models with a constant escape fraction predict a relatively late reionisation epoch and can not account for the optical depth inferred by wmap . in an effort to study how this difference could be reconciled in the context of population ii type stars , we employed models with evolving escape fractions and found that if @xmath130 were to rise from @xmath11 near @xmath7 to @xmath12 near @xmath13 , the wmap and sdss constraints can both be satisfied . in this picture , an `` escape fraction '' larger than unity implies that the stellar production rate of ionising photons is much higher than that which would be produced according to our model assumptions . one mechanism that would boost the production rate of ionising photons would be if the imf evolved so that it became more and more top - heavy with increasing redshift . in the future , we will combine simulations of early structure formation with a detailed treatment of radiative transfer to investigate the combined effects of reionisation by both population iii and population ii stars . preliminary work on related issues by yoshida et al . ( 2003 ) indicates that dynamical heating by ongoing accretion limits the rate at which cold , dense clouds of molecular material can form in halos . in principle , this effect will severely reduce the rate of formation of massive stars through molecular cooling , perhaps rendering them unimportant as sources of ionising radiation . dynamical simulations will be required to understand these processes in detail , and to reliably calibrate semi - analytical treatments , in a manner similar to the approach described here . we thank matias zaldarriaga for discussions regarding the statistical significance of observational results in the context of our analysis . we also thank brant robertson for his services as a system administrator for the computer cluster used in this analysis . thanks daniel harvey for many constructive discussions related to this study . this work was supported in part by nsf grants aci 96 - 19019 , ast 98 - 02568 , ast 99 - 00877 , and ast 00 - 71019 and nasa atp grant nag5 - 12140 . the simulations were performed at the center for parallel astrophysical computing at the harvard - smithsonian center for astrophysics .

we present radial stellar population parameters for a subsample of 12 galaxies from the 36 isolated early - type galaxies of reda et al . using new long - slit spectra , central values and radial gradients for the stellar age , metallicity [ z / h ] and-element abundance [ e / fe ] are measured . similarly , the central stellar population parameters are derived for a further 5 isolated early - type galaxies using their lick indices from the literature . on average , the seventeen isolated galaxies have mean central [ z / h] and [ e / fe] of and respectively and span a wide range of ages from 1.7 to 15 gyrs . we find that isolated galaxies follow similar scaling relations between central stellar population parameters and galaxy velocity dispersion to their counterparts in high density environments . however , we note a tendency for isolated galaxies to have slightly younger ages , higher metallicities and lower abundance ratios . such properties are qualitatively consistent with the expectation of an extended star formation history for galaxies in lower density environments . generally we measure constant age and [ e / fe ] radial gradients . however , three galaxies show remarkable positive age gradients and two galaxies have negative age gradients . we find that the age gradients anti - correlate with the central galaxy age . thus as a young starburst evolves , the age gradient flattens from positive to almost zero . metallicity gradients range from near zero to strongly negative . for our high mass galaxies ( km / s ) metallicity gradients are shallower with increasing mass . such behaviour is not predicted in dissipational collapse models but might be expected in multiple mergers . the metallicity gradients are also found to be correlated with the central age and metallicity , as well as to the age gradients . in conclusion , our stellar population data for a sample of isolated early - type galaxies are more compatible with an extended merger / accretion history than early dissipative collapse . galaxies : elliptical and lenticular , cd - galaxies : abundances - galaxies : formation - galaxies : evolution - galaxies : kinematic and dynamics

astro-ph0703545

the stellar population properties of early - type galaxies ( e.g. age , metallicity and @xmath0-element abundance [ e / fe ] ) provide crucial clues to their evolutionary history . most work to date has concentrated on galaxy central regions . while useful , such data only sample a small fraction of the galaxy mass and do not provide any radial gradient information . radial gradients can help discriminate between different formation models . for example , dissipational collapse models ( larson 1974 ; carlberg 1984 ; kawata & gibson 2003 ) predict strong metallicity gradients that correlate with galaxy mass , whereas mergers tend to result in shallow gradients ( white 1980 ; bekki & shioya 1999 ) with little galaxy mass dependence . as [ e / fe ] may be an indicator of star formation timescale , a positive gradient indicates outside - in formation and a negative one the opposite ( e.g. ferreras & silk 2002 ) . age gradients indicate whether any young stars are confined to the central regions , and hence an indication of their mass contribution . kobayashi & arimoto ( 1999 ) compiled a list of 80 early - type galaxies with radial metallicity gradient measurements ; the average value being 0.3 dex per dex . for coma cluster ellipticals , mehlert et al . ( 2003 ) confirmed significant radial metallicity gradients but , on average , found no radial age or [ e / fe ] gradients . in contrast , snchez - blzquez et al . ( 2006a ) found significant age gradients in a sample of cluster , group and field early - type galaxies . a weak correlation between metallicity gradient and galaxy mass was found by snchez - blzquez et al . ( 2007 ) and forbes , snchez - blzquez & proctor ( 2005 ) for cluster ellipticals . the number of studies that have focused on the stellar populations of early - type galaxies in very low density environments is small . kuntschner et al . ( 2002 ) obtained spectra of three e and six s0 galaxies in low density environments ( i.e. they had less than two bright neighbours within 1.3 mpc and @xmath5350 km / s ) . five of their nine galaxies revealed emission indicating ongoing star formation . using lick absorption lines and vazdekis ( 1999 ) single stellar population ( ssp ) models they found their galaxies to be on average younger and more metal - rich than cluster galaxies . however , they noted the dominance of s0s in their sample . they also found a similar [ e / fe ] vs velocity dispersion relation as for cluster galaxies . collobert et al . ( 2006 ) extended the kuntschner et al . work with spectra of 22 galaxies , for hubble types ranging from s0/a to pure e. half came from the 2dfgrs survey ( with no bright companions within 1 mpc ) and half from the low density sample of colbert et al . ( 2001 ) ( no rc3 catalogued galaxies within 1 mpc and @xmath51000 km / s ) . after applying emission line corrections to 7 of their 22 galaxies , they applied thomas , maraston & bender ( 2003 ) models to 7 lick absorption lines with a @xmath6 fitting technique . they found a spread to younger ages than for cluster ellipticals , but no clear [ e / fe ] vs velocity dispersion relation . they speculated that isolated ellipticals _ assembled _ quite recently . both studies only probed the central regions of their sample galaxies . here we extend the previous work by examining the radial stellar population properties to around 1 effective radius in a sample of a 12 isolated galaxies . the sample presented here comes from the well - defined and highly isolated sample of early - type galaxies from reda et al . our new data are supplemented by data for 5 isolated galaxies from the literature . we also utilize the latest ssp models and fitting methods . in reda et al . ( 2004 ) , we defined a sample of 36 isolated early - type galaxies in the local universe ( @xmath7 ) . here we present new data on the stellar populations of a subsample for 12 galaxies from that sample . the basic data of these galaxies and their sources are summarised in table 1 . we supplement our data with data from the literature for other isolated galaxies of reda et al . ( 2004 ) . denicol et al . ( 2005a ) extracted 21 lick spectral indices for the central r@xmath8 region for a sample of 86 early - type galaxies . six of our 36 isolated galaxies were included in their study . denicol et al . applied an emission correction to the indices where the galaxy spectra show evidence of emission lines . comparing their extracted stellar population parameters to the literature , their measurements tend to be younger and more metal - rich ( denicol et al . 2005b ) . using their published indices of the six isolated galaxies , and applying the multi - index @xmath6 minimization technique which we are using for our observations and the same ssp model ( see sec . 2.5 ) , we have extracted the central stellar population parameters for these galaxies which are listed in table 2 . the spatially resolved stellar population of the isolated galaxy ncg 821 was previously studied by proctor et al . here we used their data to extract the central values of the age , total metallicity [ z / h ] and @xmath0-elements abundance [ e / fe ] within r@xmath8 which are also listed in table 2 . these measurements of the central parameters are consistent to our measurements using the indices from denicol et al . ( 2005a ) . we also measured the radial gradient of these parameters , considering all apertures beyond the seeing limit , to be [email protected] , @xmath90.03 , @xmath100.04 for log(age ) , [ e / fe ] and [ z / h ] respectively . llcllrll + galaxy & dist . & @xmath11 & @xmath12 & @xmath12 & @xmath13 & @xmath5 & @xmath13 + & ( mpc ) & ( mag ) & ( mag ) & source & ( pc ) & & source + ngc 682 & 73 & 24.88 & 19.9 & paper i & 3.73 & 0.10 & ps97 + ngc 1045 & 60 & 25.02 & 20.9 & paper i & 3.66 & 0.10 & paper i + ngc 1162 & 51 & 24.53 & 20.7 & paper i & 3.63 & 0.10 & 2mass + ngc 2271 & 32 & 23.84 & 20.0 & paper ii & 3.47 & 0.07 & paper ii + ngc 2865 & 35 & 24.26 & 20.7 & paper i & 3.68 & 0.01 & paper i + ngc 4240 & 26 & 22.71 & 18.8 & paper i & 3.29 & 0.10 & 2mass + eso 153-g003 & 84 & 25.04 & 20.9 & paper ii & 3.62 & 0.06 & paper ii + eso 218-g002 & 54 & 24.51 & 20.9 & paper i & 3.76 & 0.11 & paper i + eso 318-g021 & 62 & 24.43 & 20.7 & paper ii & 3.77 & 0.01 & paper ii + mcg-01 - 27 - 013 & 121 & 25.08 & 21.4 & paper i & 3.98 & 0.05 & paper i + mcg-02 - 13 - 009 & 73 & 24.58 & 21.3 & paper i & 3.67 & 0.10 & 2mass + mcg-03 - 26 - 030 & 119 & 25.79 & 21.7 & paper i & 3.92 & 0.02 & paper i + note : distances are obtained using the virgo corrected recession velocities ( from leda ) with h@xmath14 = 75 km s@xmath15 mpc@xmath15 . @xmath11 and @xmath12 are the absolute magnitudes in the @xmath16 and @xmath17-bands respectively . @xmath11 are from 2mass catalogue and @xmath12 from reda et al . ( 2004 , paper i ) and reda et al . ( 2005 , paper ii ) . effective radius r@xmath18 from the 2mass catalogue is the isophotal radius @xmath19 in the @xmath16-band ( via leda database ) converted to effective radius using the correlation in jarrett et al . ( 2003 , fig . 11 ) . for ngc 682 the effective radius is derived from the fundamental plane ( prugniel & simien 1997 ; ps97 ) [ basic ] lccccc + + galaxy & log(@xmath20 ) ( km / s ) & age@xmath1 ( gyr ) & [ e / fe]@xmath1 & [ z / h]@xmath1 + ngc 821 & 2.33@xmath5 0.06 & 4.7 @xmath5 1.9 & 0.15@xmath5 0.02 & 0.48@xmath5 0.05 + ngc 1045 & 2.36@xmath5 0.06 & 5.3 @xmath5 1.4 & 0.12@xmath5 0.02 & 0.43@xmath5 0.03 + ngc 1132 & 2.37@xmath5 0.05 & 11.2 @xmath5 4.7 & 0.36@xmath5 0.04 & 0.10@xmath5 0.07 + ngc 2128 & 2.25@xmath5 0.06 & 2.8 @xmath5 1.2 & 0.27@xmath5 0.03 & 0.48@xmath5 0.09 + ngc 6172 & 2.13@xmath5 0.05 & 2.5 @xmath5 0.5 & 0.00@xmath5 0.03 & 0.13@xmath5 0.04 + ngc 6411 & 2.25@xmath5 0.03 & 10.6 @xmath5 0.8 & 0.18@xmath5 0.04 & 0.05@xmath5 0.02 + ngc 821 & 2.29@xmath5 0.01 & 5.6 @xmath5 0.5 & [email protected] & 0.41@xmath5 0.06 + notes : parameters of the first six galaxies are obtained using the published lick indices of denicol et al . ( 2005a ) and refit to a ssp model for this work . for ngc 821 , the parameters are the average of all apertures within r@xmath21r@xmath8 from proctor et al . ( 2005 ) . [ centr1 ] our spectroscopic observations were carried out using efosc2 at the eso 3.6 m telescope on the la silla observatory in two observing runs , 2002 jan . 16 - 18 and 2004 dec . the observational set up and slit position are described in detail by hau & forbes ( 2006 ) . lick and spectrophotometric standard stars were observed at the parallactic angle . the observing nights were photometric and the seeing was generally better than @xmath22 . observations of the first run ( 2002 ) were collected using a slit of @xmath23 length and @xmath24 width which was positioned along the major axis of each galaxy , with exception for the s0 galaxy eso 153-g003 where the slit was positioned along the minor axis to avoid the disc . using the eso grism#8 of 600 line / mm provides a spectral resolution of 9.3 fwhm . the accumulated exposure time was 2 @xmath25 1200 seconds for each galaxy . in the 2004 run we used a @xmath26 slit which provides a spectral resolution of 7.8 fwhm . the accumulated exposure time was 3 @xmath25 1200 seconds for each galaxy . the multiple observations of each galaxy are combined to increase the s / n ratio . additionally , a number of spectrophotometric stars were observed for flux calibration and lick standard stars ( of type between k0 and k5 ) to be used as templates for velocity dispersion measurements and for calibration of the line - strength indices to the lick / ids system . a number of bias , dark current and dome flat were recorded each night . observations of all galaxies and stars gave a wavelength coverage of 4300 - 6300 . data reduction , including bias , dark current and flat field subtraction , is performed using tasks within iraf . wavelength calibration is done using helium - argon lamp and is good to within 0.5 , while the flux standard stars are used to calibrate the spectra . the 2-d spectra are corrected for the s - distortion and sky subtracted . for more details about the data reduction refer to hau & forbes ( 2006 ) . finally the 2-d spectra of the galaxies are spatially binned using the starlink task extract to obtain 1-d spectra of s / n ratio greater than 20 . we will refer to these 1-d spectra of each galaxy as _ apertures_. the heliocentric recession velocity of the standard stars are obtained by comparing their spectra to the solar spectrum . the _ fxcor _ task within iraf is used to fit a gaussian to the cross - correlation function to determine the maximum peak . the position of the maximum peak centre gives the required heliocentric recession velocity of the stars . once the heliocentric recession velocity of the standard stars were measured , the redshifts and velocity dispersions of the galaxies are determined by comparing their spectra to the spectra of the standard stars . the _ fxcor _ task of iraf is used to compute the cross - corelation of the galaxy spectra with the spectra of each template star . once the correlation is computed , the position of the maximum peak centre gives the recession velocity v@xmath27 and the velocity dispersion @xmath28 . the final v@xmath27 and @xmath28 are the average of the values estimated from all the template stars . the radial kinematics for the same sample of galaxies are discussed in hau & forbes ( 2006 ) . lc|cc|cc| + & & & + index & unit & mean & error of & mean & error of + & & offset & the mean & offset & the mean + fe4383 & & 0.031 & 0.329 & 0.229 & 0.120 + ca4455 & & 0.214 & 0.152 & 0.054 & 0.171 + fe4531 & & 0.163 & 0.132 & 0.342 & 0.103 + c4668 & & 0.828 & 0.097 & 0.598 & 0.187 + h@xmath29 & & 0.018 & 0.088 & 0.041 & 0.113 + fe5015 & & 0.284 & 0.079 & 0.357 & 0.185 + mg@xmath30 & mag & 0.013 & 0.002 & 0.019 & 0.007 + mg@xmath31 & mag & 0.019 & 0.003 & 0.018 & 0.006 + mg@xmath32 & & 0.083 & 0.071 & 0.039 & 0.195 + fe5270 & & 0.135 & 0.053 & 0.412 & 0.147 + fe5335 & & 0.214 & 0.093 & 0.274 & 0.215 + fe5406 & & 0.121 & 0.035 & 0.302 & 0.059 + fe5709 & & 0.006 & 0.128 & 0.031 & 0.065 + fe5782 & & 0.084 & 0.026 & 0.027 & 0.067 + na@xmath33 & & 0.058 & 0.132 & 0.239 & 0.138 + tio@xmath30 & & 0.004 & 0.005 & 0.006 & 0.003 + [ offs ] we adopted the definition of lick / ids absorption line indices from trager et al . they defined each index using a pair of pseudocontinua bracketing the line feature and the central passbands of the line itself . our observations cover a wavelength range of 4300 - 6300 which includes 16 lick / ids index . these are one balmer index ( h@xmath29 ) , three magnesium indices ( mg@xmath30 , mg@xmath31 and mg@xmath32 ) , calcium ( ca4455 ) , sodium ( nad ) , titanium oxide ( tio@xmath30 ) indices plus nine iron indices fe4383 , fe4531 , fe4668 ( referred to as c4668 ) , fe5015 , fe5270 , fe5335 , fe5406 , fe5709 and fe5782 . a number of lick standard stars were observed during each observing run and will be used in the following sections to calibrated the measured indices to the lick system . lick absorption indices are affected by a wavelength dependent instrumental resolution . this resolution ranges from minimum value of @xmath34 in the range 4900- 5400 and degrade at the blue and red sides reaching above @xmath35 ( see fig . 7 in worthey & ottaviani 1997 ) . our observed instrumental resolutions ( 9.3 and 7.8 for 2002 and 2004 runs respectively ) are different from those of the standard lick system . the galaxy velocity dispersion will also alter the observed resolution of the absorption lines . to transform to the standard lick system , two different methods are used according to the observed total resolution of the spectra ( including both instrumental and velocity dispersion effects ) . for indices with combined resolution higher than that of the lick system , spectra are degraded to the resolution of the lick / ids library by convolving the spectra with a wavelength dependent gaussian . while for indices with resolution lower than the lick / ids system , a correction factor is measured by convolving the spectra of five lick standard stars with a series of gaussians widths from 0 to 500 km / s . comparing the indices from the artificially broadened stellar spectra to the un - broadend indices , we estimate the index corrections for the broadening effect of velocity dispersion in galaxies . this factor is then used to transform the indices measured in the galaxies of low resolution spectra to the lick resolution ( proctor & sansom 2002 ; proctor et al . 2004 ) . furthermore , the original lick / ids spectra were not flux calibrated while our spectra are . to compensate for the resulting effect on the shape of the spectral continuum , the differences between the measured and the published indices values are obtained for each observed calibration star . for each index , the values from all stars are averaged and compared to the literature values to give the mean offset . the error of the mean is obtained as : @xmath36 where @xmath37 is the @xmath38 of the scatter about offset and @xmath39 is the number of the standard stars . the adopted offset of each index and its error are given in table 3 . the age , total metallicity [ z / h ] and @xmath0-element abundance [ e / fe ] properties of our sample galaxies are estimated by comparing the lick absorption lines indices to the single stellar population ( ssp ) model of thomas , maraston & korn ( 2004 , tmk04 ) . the [ e / fe ] parameter measures the ratio of the @xmath0-elements ( n , o , mg , na , si , ti ) to the fe - peak elements ( cr , mn , fe , co , ni , cu , zn ) . the multi - index @xmath6 minimization technique ( proctor & sansom 2002 ; proctor et al . 2004 ) is used to obtain the corresponding ssp model values of the age , [ z / h ] and [ e / fe ] abundance ratios for the galaxies . in this technique as many indices as possible are fit to the model . because most indices contain some information regarding each of the stellar parameters ( proctor & sansom 2002 ) , using this fitting technique has advantages of including all possible information recorded in all absorption line indices to measure these parameters . another advantage of this fitting technique is that it identifies indices that are highly deviant from model values , permitting their exclusion from the fitting process . in addition , we note that parameters measured in this way are less prone to uncertainties in data reduction such as flux calibration , stray cosmic ray and also less prone to weak emission effects ( proctor , forbes & beasley 2004 ) . therefore we start by fitting all 16 indices . the indices which deviate by more than @xmath40 are clipped and the @xmath6 fit recalculated . after clipping the highly deviated indices ( @xmath41 ) , we notice that , for some apertures within each galaxy , the fitting is still not stable and the estimated values of the age , metallictiy and [ e / fe ] are significantly affected by clipping different indices . also , after this initial fitting process , many spectral indices show a significant range of deviations from the best fitting values . considering ngc 2271 as a typical galaxy of the sample , the left panel of fig . [ chi ] shows the average deviation in units of error ( i.e. @xmath42 ) of the 16 indices from the best fit values . furthermore , we notice that some indices lie outside the model grid which can be due to residual offsets between the measured indices and the standard system , unknown dependencies of the line strength predictions on any abundance ratio or inaccuracies in ssp models . to eliminate the effects of the remaining offsets , we implemented the technique described by kelson et al . ( 2006 ) to shift the zero points of the model to our data . kelson et al . defined the reference point as a set of previously published stellar population parameters for the central massive early - type galaxy in the cluster of their study . to apply this technique to our galaxy sample we define a local reference point for each galaxy by choosing two central apertures with high s / n ratios that show higher stability during the initial fitting to the model . the initial measured age , [ z / h ] and [ e / fe ] of these two central apertures are considered as a reference zero point for that galaxy . we measure the offset between each index from the corresponding values in the model , at fixed age and metallicity as estimated by the initial fit . then we apply similar shifts to the indices of all other apertures . after applying this shift , the number of highly deviant indices from the model is reduced and consequently larger number of indices are used during the fitting process which become much faster . examining the extracted stellar population parameters before and after shifting the indices , we find that they are similar . that means applying kelson et al . method only reduces the scatter of the indices around the model . [ chi ] shows the average deviation of the indices before and after applying kelson et al . errors in the derived parameters are estimated using monte carlo type realizations of the best - fitting ssp models perturbed by the index error estimates . hence , the average estimated errors of stellar population parameters based on the observational errors are about @xmath43 dex for the log(age ) , [ z / h ] and [ e / fe ] and @xmath44 for log(@xmath28 ) . the weak [ o iii ] and/or h@xmath29 emission in the two galaxies eso318-g021 and mcg-02 - 13 - 009 ( hau & forbes 2006 ) is found to have no significant effect on our measurements of the stellar population parameters , again because of our use of all available indices and the multi - index @xmath6 minimization technique to fit the indices to the ssp model . the central stellar population parameters of each galaxy are obtained by averaging the values of all apertures within the central @xmath45 . the values are listed in table 4 with the error on the mean . in the following sections we will investigate these values and their correlations compared to those for galaxies in high density environments ( hdes ) . the stellar populations of three isolated galaxies ngc 821 , ngc 1045 and ngc 2865 were previously measured in the literature . comparing our measurements of ngc 2865 to those of snchez - blzquez et al . ( 2007 ) , we found comparable values of the central age ( 1.7 vs. 1.0 gyrs ) , [ z / h]@xmath1 ( 0.48 vs. 0.52 ) and [ e / fe]@xmath1 ( 0.07 vs. 0.13 ) . on the other hand , comparing our measurements for ngc 1045 to those obtained by using the data of denicol et al . ( 2005a ) , their data indicate a younger age ( 5 vs. 10 gyrs ) , more metal - rich ( 0.43 vs. 0.31 ) and less enhanced [ e / fe]@xmath1 ( 0.12 vs. 0.36 ) . we note that this galaxy does not show emission lines and denicol et al . did not apply any emission corrections to its indices . for ngc 821 , both measurements of proctor et al . ( 2005 ) and those we obtained by using the data of denicol et al . ( 2005a ) are comparable ( see table 2 ) . through out this paper we will consider our measurements for ngc 1045 and ngc 2865 and proctor et al . values for ngc 821 . lcccc + galaxy & log(@xmath20 ) & age@xmath1 ( gyr ) & [ e / fe]@xmath1 & [ z / h]@xmath1 + ngc 682 & 2.30 @xmath5 0.01 & 8.0 @xmath5 0.6 & 0.30 @xmath5 0.01 & 0.33 @xmath5 0.02 + ngc 1045 & 2.42 @xmath5 0.01 & 10.4 @xmath5 0.6 & 0.36 @xmath5 0.02 & 0.31 @xmath5 0.02 + ngc 1162 & 2.29 @xmath5 0.01 & 6.8 @xmath5 0.4 & 0.31 @xmath5 0.01 & 0.29 @xmath5 0.02 + ngc 2271 & 2.37 @xmath5 0.01 & 11.5 @xmath5 0.5 & 0.14 @xmath5 0.01 & 0.44 @xmath5 0.02 + ngc 2865 & 2.25 @xmath5 0.02 & 1.7 @xmath5 0.1 & 0.07 @xmath5 0.01 & 0.48 @xmath5 0.05 + ngc 4240 & 2.09 @xmath5 0.01 & 7.4 @xmath5 0.5 & 0.07 @xmath5 0.01 & 0.23 @xmath5 0.02 + eso153-g003 & 2.34 @xmath5 0.01 & 11.2 @xmath5 0.4 & 0.28 @xmath5 0.01 & 0.19 @xmath5 0.01 + eso218-g002 & 2.43 @xmath5 0.01 & 14.8 @xmath5 0.1 & 0.26 @xmath5 0.01 & 0.35 @xmath5 0.01 + eso318-g021 & 2.37 @xmath5 0.01 & 8.9 @xmath5 0.9 & 0.08 @xmath5 0.01 & 0.34 @xmath5 0.03 + mcg-01 - 27 - 013 & 2.39 @xmath5 0.01 & 8.0 @xmath5 1.6 & 0.03 @xmath5 0.01 & 0.34 @xmath5 0.04 + mcg-02 - 13 - 009 & 2.33 @xmath5 0.02 & 9.6 @xmath5 1.2 & 0.05 @xmath5 0.02 & 0.36 @xmath5 0.02 + mcg-03 - 26 - 030 & 2.47 @xmath5 0.01 & 14.5 @xmath5 0.2 & 0.15 @xmath5 0.03 & 0.27 @xmath5 0.01 + including the additional data from table 2 , the seventeen isolated galaxies show a range of central luminosity - weighted ages . roughly half ( 9 out of 17 ) are older than 9 gyr , five galaxies are between 5 and 8 gyrs , and 3 with ages younger than 3 gyrs . the sample has an average age of @xmath46 gyrs . this is younger than the average age of about 11 gyr for galaxies in the coma cluster found by mehlert et al . younger average luminosity - weighted ages for galaxies in low density environments were also found by proctor et al . ( 2004 ; see also references therein ) in comparison of galaxies in high density environments ( hdes ) such as clusters and hickson compact groups to those in loose groups and field . similar results were also found in the large sample study of terlevich & forbes ( 2002 ) . on average , our isolated galaxies have central luminosity - weighted total metallicities of [ z / h]@xmath47 . mehlert et al . ( 2003 ) quotes values of @xmath48 and @xmath49 for the coma cluster galaxies of types e and e / s0 respectively and collobert et al . ( 2006 ) found an average of [ z / h]@xmath50 for their cluster galaxies . using the published stellar population parameters in thomas et al . ( 2005 ) , we find that the early - type galaxies in their hde subsample have an average of [ z / h]@xmath51 , while their galaxies in low density environments are more metal - rich by about 0.05 - 0.1 dex . the average central luminosity - weighted [ e / fe]@xmath1 of our isolated galaxies is @xmath3 . comparing to the mean value of [ e / fe]@xmath52 reported by mehlert et al . ( 2003 ) for cluster galaxies , our isolated galaxies show lower values by order of 0.1 dex . although , if we arbitrarily divide our sample in half , then 8 out of 17 of our galaxies are significantly enhanced with @xmath53[e / fe]@xmath54 . these eight galaxies are comparable to those measured by proctor et al . ( 2004 ) for galaxies in massive and compact groups ( [ e / fe]@xmath55 ) . the other nine galaxies resemble the solar element abundance with [ e / fe]@xmath56 which is lower than what proctor et al . obtained for field galaxies ( [ e / fe]@xmath57 ) . above we have compared the central stellar population parameters of the isolated galaxies with their counterparts in hdes . however , such parameters are known to vary with galaxy velocity dispersion ( e.g. kuntschner et al . 2001 , 2002 ; mehlert et al . 2003 ; collobert et al . 2006 ; snchez - blzquez et al . 2006b ; brough et al . next we compare the correlations between central stellar population parameters of our isolated galaxies to those in hdes . bernardi et al . ( 2006 ) identified a sample of 490 early - type galaxies in hdes from the sdss at redshifts of @xmath58 . they define the hdes as those galaxy systems of total luminosity more than 10 times the luminosity of a typical early - type galaxy . to obtain a high s / n ratio for their spectra , they co - added similar objects by using narrow bins in luminosity , size , velocity dispersion , and redshift ( bernardi et al . the 490 galaxies produced 105 composite spectra of s / n@xmath59 . using their published stellar population parameters and correlations between those parameters , we have measured the @xmath60 scatter of their galaxies about these correlations . in fig . [ centr ] , solid lines are the best fit relation from bernardi et al . , and dotted lines represent the @xmath60 scatter calculated by us . similar trends to bernardi et al . were found for galaxies in hdes by previous studies ( e.g. kuntschner et al . 2002 ; mehlert et al . 2003 ; collobert et al . 2006 ; snchez - blzquez et al . 2006b ) . for our isolated galaxies , trends between the stellar population parameters in the central regions and the central velocity dispersion ( a proxy for mass ) are examined in fig . [ centr ] ( a),(b ) and ( d ) , while panels ( c ) and ( e ) show correlations of [ z / h ] and [ e / fe ] with age . isolated galaxies follow similar trends to those in hdes with more massive galaxies being older , more metal - rich and with higher [ e / fe]@xmath1 than less massive ones . isolated galaxies show slightly higher scatter towards younger ages , higher [ z / h]@xmath1 and lower [ e / fe]@xmath1 for their velocity dispersion than those in hdes . in panels ( d ) and ( e ) , the two galaxies ngc 4240 and mcg-01 - 27 - 013 are the most deviant galaxy from the correlations with very low [ e / fe]@xmath1 for their velocity dispersion and age . similarly , comparing galaxies in low density environments with those in fornax cluster , kuntschner et al . ( 2002 ) found that both samples follow similar log(@xmath20)-metallicity trends , although galaxies in low density environments show higher metallicities by @xmath61 dex . panel ( c ) shows the observed age - metallicity relation for the isolated galaxies . although the correlated errors of the age and metallicity may be partly responsible for this correlation ( e.g. kuntschner et al . 2001 ) , our isolated galaxies follow a similar correlation to hde galaxies . while the effect of the correlated errors is expected to reduce the measured metallicity of old galaxies , we note that the oldest three of our galaxies ( eso218-g002 , mcg-03 - 26 - 030 and ngc 2271 ) with age @xmath62 gyrs , tend to be more metal - rich than the average galaxy . on the other hand , the galaxy ngc 6172 tends to be younger and less metal - rich than the general trend of the correlation . the stellar population of this galaxy is extracted from the emission corrected lick absorption indices from denicol et al . ( 2005a ) , which perhaps leads to measuring a younger age . as a function of central age , panel ( e ) shows that many of the isolated galaxies have low [ e / fe]@xmath1 for their age compared to their counterparts in hdes . low values of [ e / fe]@xmath1 have been previously reported by collobert et al . ( 2006 ) for their galaxies in low density environments . in summary , isolated galaxies in our sample span a large range of ages . scaling relations between central stellar parameters are similar to those for galaxies in higher density environments albeit with a tendency to younger central ages , higher [ z / h ] and lower [ e / fe ] . [ test2 ] [ test3 ] the radial recession profile of the stellar population parameters are fit with a weighted least squares method to measure the gradient for all apertures beyond the seeing limit of @xmath63 . these fits are listed in table 5 . we note that our measurements of [ z / h ] and [ e / fe ] gradients of the galaxy ngc 2865 are comparable to those found by snchez - blzquez et al . ( 2007 ) , although they found a steeper age gradient of @xmath64 . throughout this paper we use our measurements for this galaxy which is shown in all plots as a solid triangle . in the following sections we examine the radial profile of each parameter in detail . lccc + galaxy & log(age ) & [ e / fe ] & [ z / h ] + ngc 682 & 0.03 @xmath5 0.08 & 0.01 @xmath5 0.03 & 0.25 @xmath5 0.03 + ngc 1045 & 0.19 @xmath5 0.04 & 0.05 @xmath5 0.03 & 0.24 @xmath5 0.03 + ngc 1162 & 0.29 @xmath5 0.07 & 0.02 @xmath5 0.04 & 0.44 @xmath5 0.03 + ngc 2271 & 0.03 @xmath5 0.07 & 0.07 @xmath5 0.05 & 0.05 @xmath5 0.04 + ngc 2865 & 0.63 @xmath5 0.07 & 0.03 @xmath5 0.03 & 0.47 @xmath5 0.05 + ngc 4240 & 0.16 @xmath5 0.12 & 0.01 @xmath5 0.06 & 0.30 @xmath5 0.06 + eso153-g003 & 0.02 @xmath5 0.08 & 0.03 @xmath5 0.07 & 0.22 @xmath5 0.05 + eso218-g002 & 0.04 @xmath5 0.04 & 0.04 @xmath5 0.07 & 0.28 @xmath5 0.04 + eso318-g021 & 0.16 @xmath5 0.11 & 0.06 @xmath5 0.07 & 0.01 @xmath5 0.07 + mcg-01 - 27 - 013 & 0.07 @xmath5 0.25 & 0.14 @xmath5 0.13 & 0.05 @xmath5 0.14 + mcg-02 - 13 - 009 & 0.58 @xmath5 0.27 & 0.13 @xmath5 0.10 & 0.15 @xmath5 0.15 + mcg-03 - 26 - 030 & 0.01 @xmath5 0.04 & 0.05 @xmath5 0.08 & 0.10 @xmath5 0.02 + the detailed kinematics of our sample of galaxies including the radial velocity and velocity dispersion profiles of the galaxies have been previously studied by hau & forbes ( 2006 ) . reproducing the velocity profiles ( see sec 2.3 ) , we confirm their findings of kinematic substructures in some of these galaxies . rapidly rotating cores are detected in seven galaxies , three reveal fast rotating outer discs , while three galaxies show central peaked velocity dispersion profiles ( see hau & forbes for details ) . although having a wide range of ages , figs [ test1 ] show that the majority of our sample of isolated early - type galaxies show statistically insignificant age gradients with an average of @xmath65 dex per dex . four of these galaxies ( ngc 2271 , eso153-g003 , eso218-g002 and mcg03 - 26 - 030 ) have uniform ages older than 10 gyrs indicating a formation epoch at @xmath66 . two galaxies , eso318-g021 and mcg01 - 27 - 013 , show intermediate ages of 8.9 and 8.0 gyrs which indicate an extended star formation history to @xmath67 . although two galaxies , ngc 682 and ngc 4240 , show insignificant age gradients , they are composed of both old and relatively younger populations , which suggests a secondary star formation epoch for these two galaxies . ngc 682 shows a global age of @xmath68 gyrs , with a younger population of about @xmath69 gyrs in the region around @xmath70(@xmath71)@xmath72 . the stellar kinematics at the same region and outwards indicates the presence of a rotating disk with a rotational velocity of 150 km / s combined with a low velocity dispersion which declines from about 200 km / s at @xmath73 reaching 130 km / s at the outer regions . for ngc 4240 , despite its statistically insignificant age gradient of @xmath74 , it has evidence for younger stars ( @xmath75 gyr ) in the central regions at @xmath76 . remarkable positive gradients , with the central regions being younger , are seen in two galaxies . the galaxy ngc 1162 shows a gradient of @xmath77 with the central stellar population having an intermediate age of about @xmath78 gyr for @xmath79 compared to an age @xmath80 gyr in the outer regions . ngc 2865 is the youngest member of our sample with a central stellar population as young as 1.7 gyr for @xmath81 and intermediate age stars of @xmath82 gyr outwards . this galaxy shows the steepest overall age gradient of @xmath83 . the kinematics of the central ( r@xmath84 ) regions of these two galaxies suggest rotationally supported cores ( hau & forbes 2006 ) . ngc 1045 and mcg-02 - 13 - 009 show negative age gradients of @xmath85 and @xmath86 respectively . the former galaxy has a population of intermediate age stars ( @xmath87 gyr ) beyond @xmath88 . this galaxy also shows asymmetrical rotation and a declining velocity dispersion profile outwards indicating velocity substructure ( see also hau & forbes 2006 ) . while the later galaxy shows a steady age decrease outwards , a peaked central velocity dispersion and a slowly rotating body . we note that in the hde of the coma cluster , mehlert et al . ( 2003 ) measured insignificant age gradients for 91 per cent of their galaxies . the mean metallicity gradient of the 17 isolated galaxies is @xmath89 . the two youngest galaxies of the sample ngc 1162 and ngc 2865 , both have steep positive age gradients , and show steep metallicity gradients of @xmath90 and @xmath91 respectively . ngc 821 also has a young central stellar population and a very steep [ z / h ] gradient of @xmath92 ( proctor et al . 2005 ) . on the other hand , three galaxies ( eso318-g021 , mcg -01 - 27 - 013 and mcg-02 - 13 - 009 ) which show statistically insignificant age gradients , also show negligible change of the metallicity between the centre and the outer regions with average radial gradients of @xmath93 . the flat metallicity gradient of these three galaxies , can perhaps be due to the short coverage of our spectra , i.e. only to r@xmath94 . while mehlert et al . ( 2003 ) measure on average zero age gradients for galaxies in the coma cluster , the total metallicity gradient was more pronounced ( although with a large scatter ) with their galaxies having a mean negative gradient of @xmath95 . snchez - blzquez et al . ( 2007 ) measure an average [ z / h ] gradient of @xmath96 for a sample of galaxies in mostly hdes . figs [ test1 ] show that our sample galaxies have insignificant radial gradients of [ e / fe ] with an average of @xmath97 . despite the measured flat gradient of [ e / fe ] for the galaxy ngc 2865 , visual inspection reveals fluctuations between [ e / fe]=0.1 and 0.1 which may be related to the shell structure of this galaxy . insignificant radial gradients of [ e / fe ] were also found by mehlert et al . ( 2003 ) for the early - type galaxies in the coma cluster with an average of @xmath98 . although most of our galaxies reveal statistically insignificant age gradients , fig . [ grdage1 ] shows that while galaxies older than 11 gyrs show no age gradients ( i.e. uniformally old stellar populations ) , the age gradients of young galaxies are strongly anti - correlated with the central age . no significant dependence of the age gradient on the central velocity dispersion , [ z / h]@xmath1 or [ e / fe]@xmath1 was found . this suggests that galaxy mass and star formation timescale plays little role in establishing age gradients . thus as a young starburst evolves , the age gradient flattens from positive to almost zero . [ grdz1 ] panel ( a ) shows the total metallicity gradient as a function of its central value . the dashed line represents the trend found by snchez - blzquez et al . ( 2007 ) for galaxies mostly in hdes of clusters and groups . excluding galaxies with flat metallicity gradients ( see sec . [ z_profile ] ) , our galaxies follow the trend . this trend indicates that galaxies with steeper gradients have more metal - enriched inner regions . the galaxy ngc 2271 is the most deviant from the relation with a very shallow gradient for its central metallicity . a correlation between the metallicity gradient and central age was detected in many previous studies for galaxies in different environments ( e.g. snchez - blzquez , gorgas & cardiel 2006a ; snchez - blzquez et al . 2007 ) . excluding the three galaxies with flat metallicity gradients ( see sec . [ z_profile ] ) , fig . [ grdz1 ] ( b ) shows that our galaxies follow similar trends as those found by snchez - blzquez et al . ( 2007 ) for galaxies in higher density environments . galaxies with young central ages tend to have steep metallicity gradients while galaxies with uniform old stellar populations reveal shallower metallicity gradients . since the galaxies with young central ages are those with the steepest age gradients ( fig . [ grdage1 ] ) , we expect the gradient of the total metallicity and the age gradient to be strongly correlated . [ gzgage ] shows that galaxies of steep positive age gradients also show steep negative metallicity gradients . this implies that the young central stellar populations are also more metal - rich than the old metal - poor stars in the outer regions of the galaxies . [ mass1 ] shows that galaxies of smaller mass ( lower log(@xmath20 ) ) tend to reveal steeper metallicity gradients than more massive galaxies ( larger log(@xmath20 ) ) . a hint of a similar trend was found by snchez - blzquez et al . ( 2007 ) for galaxies of log(@xmath20 ) @xmath99 ( @xmath100 km / s ) . a change in the trend direction at this velocity dispersion , has been reported in several studies ( see snchez - blzquez et al . 2007 and references therein ) . thus for galaxies with log(@xmath20 ) @xmath101 , metallicity gradients appear to get shallower for _ lower _ mass ( e.g. forbes et al . 2005 ) . we note that ngc 4240 ( log(@xmath20)@xmath102 ) is consistent with this suggestion . examining the correlation of the metallicity gradient with galaxy dynamical mass log(@xmath103 ) and the absolute magnitude in the k - band ( as other proxies of mass ) gives similar trends to that of fig . [ mass1 ] . for a sample of early - type galaxies ferreras & silk ( 2002 ) found a correlation between [ e / fe ] gradient and its central value . a steep negative gradient for low [ e / fe ] galaxies changes to a steep positive gradient for enhanced [ e / fe ] galaxies . using a simple model of star formation and a standard prescription for the rates of supernovae type ii and ia , ferreras & silk found that negative gradients imply inside - out formation which is suggestive of a dissipative collapse formation . on the other hand , a positive gradient is a sign of outside - in formation which may be a result of a past merger events which induce central star formation in the merger remnant . our isolated galaxies show statistical insignificant [ e / fe ] gradients on average . furthermore , fig . [ grdefe1 ] shows that there is no obvious correlation between [ e / fe ] gradient and its central value or the central age . several galaxies in our isolated sample reveal a number of features such as tidal tails , dust , shells , discy and boxy isophotes and highly rotating central discs ( reda et al . 2004 , 2005 ; hau & forbes 2006 ) . these structures indicate recent merger / accretion events during the evolutionary history of these galaxies . measuring the age of the stellar populations of the isolated galaxies reveals that several galaxies have central young stars which also requires a recent gaseous accretion or merger . if isolated galaxies formed purely by dissipative collapse we would expect uniformly old stellar ages . we find both young central ages and age gradients in some isolated galaxies . collapse models ( e.g. carlberg 1984 ; chiosi & carraro 2002 ; kawata & gibson 2003 ) predict steeper metallicity gradients in more massive galaxies . for the bulk of our sample , with velocity dispersions @xmath104 km / s , we find the opposite behaviour with more massive galaxies have shallower metallicity gradients . a collapse also implies a largely inside - out formation with the inner galaxy regions forming first which predicts a negative [ e / fe ] gradient ( e.g. ferreras & silk 2002 ) . we find isolated galaxies to reveal no overall [ e / fe ] gradient . thus we conclude that pure dissipative collapse can not explain our sample of relatively massive isolated early - type galaxies . the expectations for stellar population parameters in a hierarchical merger scenario are more varied depending on the gas fraction of the progenitors and their mass ratio ( e.g. kobayashi 2004 ) . gas - rich mergers are expected to induce central star formation ( e.g. mihos & hernquist 1996 ; springel 2000 ) . the gradients of age and metallicity are predicted to correlate strongly with central young stars that are also metal - rich ( bekki & shioya 1999 ) . the metallicity gradient is also predicted to be shallower for a merger remnant ( e.g. kobayashi 2004 ) . the age gradient for the majority of our isolated galaxies ( 11 out of 13 ) are found to range from flat to positive . these age gradients are also correlated with the central age which implies that a young burst dominates the luminosity in the central region where it takes place , and thus produces a steep gradient . as these young populations get older and fade , the observed gradient of the luminosity - weighted age becomes shallower . however , two galaxies ( ngc 1045 and mcg-02 - 13 - 009 ) show significant negative age gradients . the radial age profile of ngc 1045 reveals a stellar population of intermediate age at @xmath105 and outward , which implies a secondary star formation event induced in the outer regions of the galaxy . that suggests either a late gas accretion or merger with a gas - rich satellite galaxy ( kobayashi 2004 ) . such accretion may form a star forming ring . deep imaging of this galaxy reveals strong boxy isophotes and extensive extra tidal light ( reda et al . 2004 ) which argues in favour of a recent merger . while deep imaging is not available for mcg-02 - 13 - 009 , its negative age gradient and flat metallicity gradient in the central region is predicted by the simulations of kobayashi ( 2004 ) to be the result of late gas accretion rather than a major merger . most of our isolated galaxies ( 8 out of 13 ) are found to have negative metallicity gradients , as galaxies with central metal - rich stars tend to have steeper metallicity gradients . the three galaxies ngc 1162 , ngc 2865 and ngc 821 have both significant age gradients as well as metallicity gradients , with the young central stellar population also being more metal - rich than the old metal - poor stars in the outer regions . ngc 1162 has an intermediate central age of 6.8 gyrs suggesting a secondary star formation event . a major merger of two massive progenitors is expected to produce a more metal - rich stellar population and shallower metallicity gradient than observed . however , a minor merger with a satellite of mass less than 0.2 times that of the primary galaxy is more likely to explain the intermediate central metallicity and the gradient ( kobayashi 2004 ) . the galaxy ngc 2865 is the youngest ( 1.7 gyrs ) and most metal - rich ( [ z / h]@xmath106 ) galaxy of our sample . its central velocity dispersion of 178 km / s indicates an intermediate mass galaxy . deep imaging revels a surrounding shell structures ( hau , carter & balcells 1999 ; reda et al . 2004 ) and kinematics reveal a kinematically decoupled core in the central 4 arcsec or 6.4 kpc ( hau , carter & balcells 1999 ; hau & forbes 2006 ) . the kinematics and stellar population of this galaxy suggest a gas - rich accretion or merger origin for the shell and kinematics structures ( hau et al . 1999 ) . however , the steep metallicity gradient of 0.47 prefers a gas - rich accretion by a metal - rich parent galaxy . the galaxy ngc 821 has a central stellar population of 5.6 gyrs old and a high metallicity of [ z / h]@xmath107 , which indicates a secondary star burst . gas accretion or a minor merger would produce a lower central metallicity , while a major merger would have reduced the very steep metallicity gradient of 0.72 . in the detailed study of proctor et al . ( 2005 ) , ngc 821 is speculated to consume its own gas to fuel a secondary central burst of star formation . the metallicity gradient of isolated galaxies are found to be steeper for less massive galaxies . this result is consistent with a major merger history for the massive galaxies ( e.g. bekki & shioya 1999 ) . during the merger process , the high metallicity stars in the centre of the galaxy are transported to the outer regions which dilutes the metallicity gradient ( bekki & shioya 1999 ) . on the other hand , the induced star formation during the merger at the centre of the merger remnant is expected to maintain the metallicity gradient to some extent . the final metallicity gradient of the merger remnant is shallower than the progenitors and the newly formed central stars will also produce relatively steep age gradients for these galaxies ( kobayashi 2004 ) . the gas - rich merger simulation of bekki & shioya ( 1999 ) predicts a one - to - one correlation between the age gradient and metallicity gradient . however , our isolated galaxies show a correlation of slope @xmath108 . both isolated galaxies , and those in hdes , show similar age - mass relations as more massive galaxies are older than smaller ones . that suggests either the less massive galaxies have had successive star formation events ( bursts induced by mergers ) continuing until recent epochs , while more massive galaxies did not suffer such events . alternatively , all galaxies started forming their stars in a single burst ( similar to a dissipative collapse ) at the same epoch which stopped earlier in more massive galaxies , while less massive galaxies continue forming new stars for extended epochs . several mechanisms have been introduced to suppress the star formation processes in more massive galaxies such as a central active galactic nuclei ( e.g. springel , di matteo & hernquist 2005 ; de lucia et al . the dependence of the star formation efficiency on the circular velocity of the galaxy can also delay the star formation in small galaxies relative to more massive galaxies ( e.g. de lucia et al . 2004 ) . the old galaxies in our sample ( ngc 2271 , eso153-g003 , eso218-g002 and mcg-03 - 26 - 030 ) with central age @xmath109 11 gyrs show flat age gradients which indicates that all stars are formed at same time at @xmath110 . they also tend to be the most massive galaxies . this can be a result of a single burst of star formation during a rapid collapse of a single cloud ( larson 1974 , 1975 ; calberg 1984 ) . the observed metallicity gradient of these galaxies ( @xmath111 ) is much less than the predicted gradients of the dissipative collapse models . the major merger scenario can explain the shallow metallicity gradients of these galaxies but not their uniform old populations . another mechanism such as developing central active galactic nuclei ( e.g. springel , di matteo & hernquist 2005 ; de lucia et al . 2006 ) is necessary to stop any further star formation after the merger . the low metallicity of eso153-g003 ( [ z / h]@xmath112 ) suggests progenitors of relatively small size for this galaxy . the galaxy ngc 2271 has the highest metallicity among these galaxies ( [ z / h]@xmath113 ) with a flat metallicity gradient of . it also shows elongated shape ( @xmath114 ) and a solid body rotation with ( @xmath115 ( hau & forbes 2006 ) which may indicate amassive disk progenitor . the shallow metallicity gradients of the two galaxies ngc 682 and ngc 4240 indicate past major mergers . on the other hand , their uniform luminosity weighted intermediate ages suggest a very small fraction of induced star formation . the magnitude of the metallicity gradients are found to correlate to the age gradient . the suggested mechanism causing these gradients is that each star formation event will enrich the ism , followed by radial inflow , causing the subsequent stellar generations at the galaxy centre to be more metal - rich , as well as younger . a natural result of this process is the observed age - metallicity relation . the dynamical effect of supernova feedback is expected to be relatively weaker in more massive galaxies , consequently the star formation rate is more efficient in these galaxies than in less massive ones . the deep potential well of massive galaxies maintains their gas long enough to perform more complete chemical processes to produce higher metallicities ( e.g. arimoto & yoshii 1987 ; edmunds 1990 ; matteucci 1994 ) resulting in the observed mass - metallicity relation . independent of the galaxy mass , or the details of the merging process , the observed correlation between the metallicity gradient and its central value can be reproduced by the gas - rich merger simulations of bekki & shioya ( 1999 ) . the results of these models are also consistent with the observed insignificant gradients of [ e / fe ] within 1 effective radius of the isolated galaxies . the parameter [ e / fe]@xmath1 measures the abundance ratio of the @xmath0-element to the fe - peak elements which are predicted to be released to the interstellar medium by supernovae type ii and ia and on different time scales . in that sense , [ e / fe]@xmath1 is commonly used to quantify the duration of star formation ( worthey , faber & gonzalez 1992 ; matteucci 1994 ; thomas , greggio & bender 1999 ; thomas et al . 2005 ) . for the isolated galaxies as well as in hdes , the lower [ e / fe]@xmath1 of the less massive galaxies points to extended star formation for these galaxies however , the isolated early - type galaxies tend to have younger ages and lower [ e / fe]@xmath1 for their central velocity dispersion than their counterparts in hdes . furthermore , we also note that isolated galaxies of intermediate ages tend to have lower [ e / fe]@xmath1 than the galaxies in hdes . this is expected if the extended star formation is triggered with recent mergers which have stopped in hdes at higher redshifts but continue to recent epochs for galaxies in low density environments such as our isolated galaxies . isolated galaxies show a wide range of ages , [ z / h ] and [ e / fe ] for their central stellar populations . more massive galaxies tend to be older , more metal - rich and more @xmath0-element enhanced than less massive galaxies in the same sense as galaxies in high density environments . although isolated galaxies tend to be slightly younger , more metal - rich and lower [ e / fe ] for their mass . the majority of the isolated galaxies in our sample show insignificant gradients of both age and [ e / fe ] . although a correlation between the age gradient and central age indicates the tendency of the newly formed stars to locate closer to the galaxy centre . the metallicity gradient ranges from very steep to flat . metallicity gradients are found to correlate with parameters of the central stellar populations such as central metallicity , age and velocity dispersions . metallicity gradients also show a remarkable correlation with the age gradients . the formation scenario of a single dissipative collapse cloud can not explain the spatial distribution of the stellar population and kinematic properties found for isolated galaxies . mergers at different redshifts of progenitors of different mass ratios and gas fractions are needed to reproduce the observed properties of the galaxies . arimoto n. , yoshii y. , 1987 , a&a , 173 , 23 + bekki k. , shioya y. , 1999 , apj , 513 , 108 + bernardi et al . , 2003 , aj , 125 , 1882 + bernardi m. , nichol r. c. , sheth r. k. , miller c. j. , brinkmann j. , 2006 , aj , 131 , 1288 + brough et al . , 2006 , supmitted to mnras + carlberg r.g . , 1984 , apj , 286 , 403 + chiosi c. , carraro g. , 2002 , mnras , 335 , 335 + colbert j. w. , mulchaey j. s. , zabludoff a. i. , 2001 , aj , 121 , 808 + collobert m. , sarzi m. , davies r. l. , kuntschner h. , colless m. , 2006 , mnras , 370 , 1213 + de lucia g. , kauffmann g. , white s. d. m. , 2004 , mnras , 349 , 1101 + de lucia g. , springel v. , white s. d. m. , croton d. , kauffmann g. , 2006 , mnras , 366 , 499 + denicol g. , terlevich r. , terlevich e. , forbes d. a. , terlevich a. , 2005b , mnras , 358 , 813 + denicol g. , terlevich r. , terlevich e. , forbes d. a. , terlevich a. , carrasco l. , 2005a , mnras , 356 , 1440 + edmunds m. g. , 1990 , mnras , 246 , 678 + ferreras i. , silk j. , 2002 , mnras , 336 , 1181 + forbes d. a. , snchez - blzquez p. , proctor r. , 2005 , mnras , 361 , l6 + hau g. k. t. , carter d. , balcells m. , 1999 , mnras , 306 , 437 + hau g. k. t. , forbes d. a. , 2006 , mnras , 371 , 633 + jarrett t. h. , chester t. , cutri r. , schneider s. e. , huchra j. p. , 2003 , aj , 125 , 525 + kawata d. , gibson b. k. , 2003 , mnras , 346 , 135 + kelson d. d. , illingworth g. d. , franx m. , van dokkum p.g . , 2006 , astro - ph/0606642 + kobayashi c. , 2004 , mnras , 347 , 740 + kobayashi c. , arimoto n. , 1999 , apj , 527 , 573 + kuntschner h. , lucey j. r. , smith r. j. , hudson m. j. , davies r. l. , 2001 , mnras , 323 , 615 + kuntschner h. , smith r. j. , colless m. , davies r. l. , kaldare r. , vazdekis a. , 2002 , mnras , 337 , 172 + larson r. b. , 1974 , mnras , 166 , 585 + matteucci f , 1994 , a&a , 288 , 57 + mehlert d. , thomas d. , saglia r. p. , bender r. , wegner g. , 2003 , a&a , 407 , 423 + mihos j. c. , hernquist l. , 1996 , apj , 464 , 641 + proctor r. n. , forbes d. a. , beasley m. a. , 2004 , mnras , 355 , 1327 proctor r. n. , forbes d. a. , forestell a. , gebhardt k. , 2005 , mnras , 362 , 857 + proctor r. n. , forbes d. a. , hau g. k. t. , beasley m. a. , de silva g. m. , contreras r. , terlevich a. i. , 2004 , mnras , 349 , 1381 + proctor r. n. , sansom a. e. , 2002 , mnras , 333 , 517 + prugniel p. , simien f. , 1997 , a&a , 321 , 111 + reda f. m. , forbes d. a. , beasley m. a. , osullivan e. j. , goudfrooij p. , 2004 , mnras , 354 , 851 + reda f. m. , forbes d. a. , hau g. k. t. , 2005 , mnras , 360 , 693 + snchez - blzquez p. , forbes d. a. , strader j. , brodie j. , proctor r. , 2007 , submitted to mnras + snchez - blzquez p. gorgas j. , cardiel n. , 2006a , a&a , 457 , 823 + snchez - blzquez p. , gorgas j. , cardiel n. , gonzalez j. j. , 2006b , a&a , 457 , 809 + springel v. , 2000 , mnras , 312 , 859 + springel v. , di matteo t , hernquist l. , 2005 , apj , 620 , l79 + terlevich a. i. , forbes d. a. , 2002 , mnras , 330 , 547 + thomas d. , greggio l. , bender r. , 1999 , mnras , 302 , 537 + thomas d. , maraston c , bender r. , 2003 , mnras , 339 , 897 + thomas d. , maraston c , bender r. , de oliveira c. m. , 2005 , apj , 621 , 673 + thomas d. , maraston c. , korn a. , 2004 , mnras , 351 , l19 ( tmk04 ) + trager s. c. , worthey g. , faber s. m. , burstein d. , gonzalez j. j. , 1998 , apjs , 116 , 1 + vazdekis a. , 1999 , apj , 513 , 224 + white s. d. m. , 1980 , mnras , 191,1 + worthey g. , faber s. m. , gonzalez j. j. , 1992 , apj , 398 , 69 + worthey g. , ottaviani d. l. , 1997 , apjs , 111 , 377 +

the universe may harbor relics of the post - inflationary epoch in the form of a network of self - ordered scalar fields . such fossils , while consistent with current cosmological data at trace levels , may leave too weak an imprint on the cosmic microwave background and the large - scale distribution of matter to allow for direct detection . the non - gaussian statistics of the density perturbations induced by these fields , however , permit a direct means to probe for these relics . here we calculate the bispectrum that arises in models of self - ordered scalar fields . we find a compact analytic expression for the bispectrum , evaluate it numerically , and provide a simple approximation that may be useful for data analysis . the bispectrum is largest for triangles that are aligned ( have edges ) as opposed to the local - model bispectrum , which peaks for squeezed triangles ( ) , and the equilateral bispectrum , which peaks at . we estimate that this non - gaussianity should be detectable by the planck satellite if the contribution from self - ordering scalar fields to primordial perturbations is near the current upper limit .

1003.0672

a wealth of precise cosmological data are in good agreement with the predictions of the simplest single - field slow - roll ( sfsr ) inflationary models @xcite . still , no theorist considers these as anything more than toy models . realistic models must surely be more complicated , and they generically predict that there should arise , at some point , observable phenomena that depart from the predictions of sfsr inflation . some possible directions for physics beyond the sfsr approximation include multi - field models @xcite and inflaton models with non - standard kinetic terms @xcite . there has also been investigation of the consequences of topological defects @xcite produced toward the end of or after inflation @xcite . if inflation was followed by a transition associated with the breaking of a global @xmath3 symmetry , then self - ordering scalar fields ( sosfs ) are another possibly observable early - universe relic , even if there are no topological defects ( i.e. , if @xmath4 ) . here , the alignment of the scalar field as the universe expands gives rise to a scale - invariant spectrum of isocurvature perturbations , without topological defects @xcite . sample variance on the current data limit these perturbations to contribute no more than @xmath5 of large - angle cosmic - microwave - background ( cmb ) anisotropy power @xcite . sosf models are parametrized simply by the number @xmath6 of scalar fields and the vacuum expectation value @xmath7 . the cmb constraint implies @xmath8 gev , as we explain below . at this low amplitude , it is unlikely that any surviving relics leave a distinct imprint on the cmb power spectrum @xcite . in recent years , non - gaussianity has been developed as a novel tool to investigate beyond - sfsr physics @xcite . sfsr models do not predict that primordial perturbations should be gaussian , but the departures from gaussianity that they predict are unobservably small @xcite . multi - field models @xcite , such as curvaton models @xcite , string - inspired dbi @xcite models , and models with features in the inflaton potential @xcite can all produce larger , and possibly observable , deviations from non - gaussianity . for example , the detailed shape ( triangle dependence ) of the bispectrum may also help distinguish these different scenarios . the `` local - model '' bispectrum , like that which arises in curvaton and multi - field models , has a very different shape dependence than `` equilateral - model '' bispectra , like those in dbi models . non - gaussianity can be sought in the cmb @xcite , large - scale structure ( lss ) @xcite , and the abundances and properties of gravitationally - bound objects @xcite or voids @xcite . biasing may significantly amplify the effects of non - gaussianity @xcite in the galaxy distribution . the energy - density perturbations in self - ordering scalar fields are quadratic in the scalar - field perturbation , which may itself be approximated as a gaussian field . the density perturbations induced by sosfs are thus expected to be highly non - gaussian @xcite , even in the absence of topological defects . it is thus plausible that the non - gaussianity induced by sosfs might be detectable , even if they provide only a secondary contribution to primordial perturbations . in this paper , we perform the first calculation of the full shape ( triangle ) dependence of the bispectrum from sosfs . we follow the formalism for non - gaussianity developed in ref . we find considerably simplified formulas for the bispectrum , evaluate them numerically , and find a simple approximation to aid in data - analysis efforts . we estimate the current non - gaussianity constraint to the model parameter space and find it to be comparable to that from the upper limit to isocurvature perturbations from cmb fluctuations . the plan of this paper is as follows : in section [ sec : scalarfielddynamics ] , we define the model , write the scalar - field equations of motion , show that the dynamics are those of a nonlinear - sigma model , and introduce the large-@xmath6 scaling limit for the nonlinear sigma model . in section [ sec : matterperturbations ] , we write the relation between the matter - density perturbation and the scalar - field perturbation . in section [ sec : powerspectrum ] , we derive the power spectrum for density and curvature perturbations , discuss the normalization , and derive current constraints to the @xmath7-@xmath6 parameter space from upper limits to the sosf contribution to cmb fluctuations . in section [ sec : bispectrum ] , we discuss the calculation of the bispectrum , the central focus of this paper . we present a simplified version , our eq . ( [ eqn : biresult ] ) , of the matter - bispectrum expression in ref . @xcite , evaluate it numerically , and provide a simple analytic approximation for the results . we write the bispectra for matter and curvature perturbations , define a non - gaussianity parameter @xmath9 for the model , and estimate the current constraint to @xmath9 from the cmb . section [ sec : galaxysurveys ] presents the matter bispectrum for modes that entered the horizon during radiation domination , those relevant for galaxy surveys . the central results of the paper are eq . ( [ eqn : centralresult ] ) for the curvature bispectrum ; eq . ( [ eqn : fnlsigma ] ) which defines @xmath9 in terms of the sosf model parameters @xmath7 and @xmath6 ; eq . ( [ eqn : g3approx ] ) which approximates the bispectrum function @xmath10 ; and eqs . ( [ eqn : galaxybispectrum ] ) and ( [ eqn : galaxybispectrum2 ] ) which present the matter bispectrum in a form useful for galaxy surveys . we make concluding remarks in section [ sec : discussion ] . an appendix contains some calculational details and useful approximations . self - ordering scalar fields are described by an @xmath6-component scalar field with an @xmath3 symmetry that is spontaneously broken to @xmath11 . after symmetry breaking , the scalar field lies in different places in its @xmath12 vacuum manifold in different causally disconnected regions of the universe . as the universe expands and these previously causally - disconnected regions come into causal contact , field gradients tend to align the scalar field . the rate of alignment for these fields is limited only by causality , and so the fields become aligned within a few hubble times after horizon crossing . still , as the universe expands , there are continually new causally - disconnected regions , on ever larger scales , that enter the horizon . the result is thus a continual scale - invariant generation of new scalar - field perturbations . in this section , we describe the scalar - field dynamics ; the following section then describes how the gradient energy density in these scalar fields induce perturbations to the matter density . the starting point is an @xmath6-component scalar field @xmath13 , with @xmath14 real , with lagrangian density , @xmath15 where @xmath16 is the dimensionless self - coupling of @xmath17 , and @xmath7 is the magnitude of the vacuum expectation value ( vev ) in the true vacuum . at temperatures @xmath18 , the @xmath3 symmetry of the lagrangian is spontaneously broken , and the field is thereafter restricted to the @xmath12 vacuum manifold . the dynamics is thus effectively that of @xmath19 massless nambu - goldstone modes which we describe in terms of the @xmath6 fields @xmath14 with the effective lagrangian density , @xmath20 where @xmath21 is a lagrange multiplier that enforces the constraint @xmath22 . the resulting equations of motion are @xmath23\phi^a({\mathbf x},\eta ) = 0,{\nonumber}\\\end{aligned}\ ] ] where the primes denote derivatives with respect to conformal time @xmath24 , and @xmath25 in terms of the friedmann - robertson - walker scale factor @xmath26 . also , @xmath27 is here a spatial laplacian in comoving coordinates . ( [ eq : nlsm ] ) represents the non - linear sigma model ( nlsm from now on ) , that describes the evolution of the scalar field after spontaneous symmetry breaking . in the large-@xmath6 limit , the field components become independent of each other ( up to corrections of order @xmath28 ) . we thus replace the bilinear term in the equation of motion by an ensemble average , @xmath29 where there is no sum on @xmath30 in the second equality , and in the last equality we have made the usual ergodic assumption , replacing the ensemble average by a spatial average . the only timescale in the problem is that set by the ( comoving ) horizon @xmath31 , so by dimensional considerations @xmath32 , and @xmath33 , with @xmath34 . we then replace the non - linear term in the nlsm equation of motion , eq . ( [ eq : nlsm ] ) , by this expectation value and in this way linearize the equations of motion . introducing @xmath35 and fourier transforming the spatial dependence of the equations , @xmath36 we obtain @xmath37 with @xmath38 for a radiation - dominated universe and @xmath39 for a matter - dominated universe . for constant @xmath40 , the solution to eq . ( [ e : sigma2 ] ) that is finite as @xmath41 , is @xmath42 , with @xmath43 , and @xmath44 is a bessel function . here , @xmath45 is the amplitude of mode @xmath46 , and @xmath47 is fixed by @xmath48 . in the large-@xmath6 limit , the statistical distribution of each field component approaches a gaussian distribution with mean @xmath49 and variance @xmath50 . the initial field component @xmath51 takes on a random value at each point in space . we thus take the @xmath52 to be gaussian random variables with mean @xmath53 and variance , @xmath54 where @xmath55 is the dirac delta function , and @xmath56 is a normalization constant to be determined below . the power - law dependence on @xmath57 is taken since the initial conditions are scale - free . the power - law index @xmath58 in eq . ( [ eq : correlator_a ] ) is fixed by the condition that @xmath59 for all @xmath24 : @xmath60 we see that @xmath61 gives a result that is independent of time , and so we choose @xmath61 hereafter . just after symmetry breaking , at conformal time @xmath62 , the field correlation is then @xmath63 since the initial values @xmath64 are uncorrelated on scales @xmath65 , we set @xmath66 , so that the initial field is described by a white - noise power spectrum . this then fixes @xmath67 . we now return to eq . ( [ eq : beta2 ] ) to fix the normalization constant @xmath56 . from @xmath68 we find @xmath69 for @xmath39 ( matter domination ) , @xmath70 , and for @xmath38 ( radiation domination ) , @xmath71 . although the scalar field will initially take on different values in different causally - disconnected regions , the curvature perturbation is initially zero . the scalar - field gradient - energy perturbation that arises as previously causally - disconnected regions come into causal contact is then compensated by a perturbation in the matter density @xcite . in this section , we calculate the time evolution of the matter perturbation . the action of the scalar field occurs primarily within a few hubble times after a particular fourier mode @xmath57 enters the horizon . the subsequent evolution of the mode is then governed by gravitational infall as if it were a primordial perturbation ; i.e. , the perturbation amplitude grows only logarithmically during radiation domination , and then grows with the scale factor during matter domination . our strategy here will be to evaluate the matter - perturbation amplitude several hubble times after horizon crossing , a calculation that is relatively straightforward . strictly speaking , our calculation applies only to modes that enter the horizon during matter domination , but we argue below that our ultimate results for the bispectrum should also be roughly valid for the smaller - scale modes that enter the horizon during radiation domination , those relevant for galaxy surveys . as described in ref . @xcite , the scalar - field alignment involves density perturbations that then lead to gravitational - potential perturbations which in turn induce the perturbations to the matter density that are our ultimate interest . following ref . @xcite , the matter - density perturbation induced by the scalar field for modes that enter the horizon during matter domination is @xmath72 where @xmath73 is newton s constant , and @xmath74 is the @xmath75-component of the stress - energy tensor of the multicomponent scalar field . the integral in eq . ( [ eqn : matterperturbation ] ) approaches a constant for @xmath76i.e . , within a few hubble times after horizon crossing . the subsequent @xmath77 evolution in the prefactor is then simply the @xmath78 linear - theory growth of the perturbation amplitude in a matter - dominated universe . using @xmath79 , where @xmath80 gev is the planck mass , and defining @xmath81 the fourier transform of the density perturbation is @xmath82 where @xmath83 . the crucial qualitative feature is that @xmath84 is quadratic in powers of @xmath45 . and since @xmath45 is a nearly gaussian field , the density field @xmath85 will be highly non - gaussian . the power spectrum @xmath86 for matter - density perturbations induced by the scalar field is defined by @xmath87 where the angle brackets denote an average over all realizations of the random field @xmath88 . the calculation of the power spectrum is lengthy but straightforward ; details are provided in the appendix . the result , given in eq . ( [ eqn : psresult ] ) , can be re - written , @xmath89\ , , { \nonumber}\\\end{aligned}\ ] ] where @xmath90 , and @xmath91 strictly speaking , the upper limit in this integral is @xmath92 . however , here we will restrict our attention to modes that have evolved well within the horizon , @xmath93 , and so we take the upper limit of the integral in eq . ( [ eqn : mathcali ] ) to be infinity . in this case , the integral @xmath94 is antisymmetric in its arguments , and the power spectrum can be written @xmath95 where @xmath96 ^ 2\,(\hat{{\mathbf{z}}}\cdot{{\mathbf{v}}})\,\left[2(\hat{{\mathbf{z}}}\cdot{{\mathbf{v}}})-1 \right]\,,\ ] ] and @xmath97 is a unit vector . details on the evaluation of @xmath98 are given in the appendix . for @xmath39 ( matter domination ) , the integral evaluates to @xmath99 and for @xmath38 ( radiation domination ) it is @xmath100 . note that the ratios @xmath101 that appear in eq . ( [ eq : finalpowerspectrum ] ) are approximately 10 and 11 , respectively , for @xmath102 , implying that the amplitude of the matter perturbation induced by the unwinding of the scalar field is the same , to @xmath103 , for modes that enter the horizon during matter and radiation domination . we now estimate the constraints to the @xmath7-@xmath6 parameter space from the empirical constraint that the sosf provide no more than a fraction @xmath104 to @xmath105 @xcite , the cmb temperature power spectrum at multipole moment @xmath106 . on subhorizon scales during matter domination , the curvature perturbation @xmath107 is related to the gravitational potential @xmath108 by @xmath109 . the gravitational potential is related to the density perturbation through the poisson equation , @xmath110 , where @xmath111 is the mean density . in fourier space , the curvature perturbation @xmath112 is thus related to the matter - density perturbation @xmath88 by @xmath113 where we have used the friedmann equation @xmath114 , and @xmath115 is the expansion rate . the amplitude of the curvature power spectrum due to the sosf is therefore , @xmath116 ^ 2 p(k ) { \nonumber}\\ & = & 8 \left ( \frac{v}{\mpl } \right)^4 \frac{g_2 } { a^2 n } \simeq 80\ , \left ( \frac{v}{\mpl } \right)^4 \frac{1}{n}\ , , \end{aligned}\ ] ] where we have used @xmath117 during matter domination . the next step is then to determine the relation between the curvature power - spectrum amplitude @xmath118 and the temperature - fluctuation amplitude . this is a notoriously difficult calculation , but to get an estimate , we use fig . 4 in ref . @xcite , which shows that the large - angle temperature fluctuation @xmath119 in a sosf model is @xmath120 times greater than it would be in an adiabatic model with the same matter power - spectrum normalization on large scales . may be due in part to the vector and tensor perturbations that are also excited in sosf models . ] current cmb measurements indicate a curvature power spectrum @xmath121 , if primordial perturbations are adiabatic . if the sosf provides @xmath122 times more @xmath119 for fixed curvature , and if they make a fractional contribution @xmath123 to the large - angle temperature variance , then @xmath124 . we thus obtain @xmath125 where the numerical result is obtained by taking @xmath126 and @xmath127 . the numerical upper limit in eq . ( [ eqn : vnormalization ] ) is in good agreement with limits obtained from simulations @xcite . the calculation of the bispectrum proceeds analogously . the matter bispectrum @xmath128 is defined by @xmath129 although the definition of the bispectrum is nominally in terms of the vector quantities @xmath130 , the triangle constraint @xmath131 imposed by the dirac delta function , as well as statistical isotropy , imply that the bispectrum is most generally a function of the magnitudes @xmath132 of the three sides of the triangle . again , some details of the calculation are provided in the appendix . the result is @xmath133 where @xmath134 with @xmath135 we have chosen @xmath136 in eq . ( [ eqn : biresult ] ) to be in the @xmath97 direction , without loss of generality , and we have then defined @xmath137 . note that @xmath138 ; i.e. , it is a function only of the magnitudes of its arguments , and it is symmetric in its arguments . note further that @xmath139 , and thus @xmath140 . the function @xmath10 depends only on the shape of the triangle , not on its overall size i.e . , @xmath141a consequence of the scale - invariance of sosfs . we have checked that eq . ( [ eqn : biresult ] ) is equivalent to , although far simpler , than eq . ( 59 ) in ref . @xcite . given the symmetry of @xmath142 in its arguments , it is simple to check that @xmath143 , as it should ( given that the three sides of the triangle should add as @xmath144 ) . if we set the third side to have length @xmath145 , then @xmath146 . if we choose @xmath147 , then @xmath148 . we have calculated @xmath10 numerically , and the result is shown in fig . [ fig : g3 ] . we note ( prefacing the discussion below ) that the quantity , @xmath149 , with @xmath150 and @xmath151 , that we plot is the same ( up to some normalization factor ) as the quantity @xmath152 plotted in figs . 1 and 2 in ref . @xcite which show , respectively , the bispectra for the local - model and equilateral model . those figures show that the local - model bispectrum peaks sharply for `` squeezed '' triangles ( @xmath153 ) and that the equilateral - model bispectrum peaks at equilateral triangles ( @xmath154 ) . our fig . [ fig : g3 ] shows that the sosf bispectrum is , however , quite different . it is nonzero for equilateral triangles , goes to zero in the squeezed limit , and it peaks for `` aligned '' triangles , @xmath155 . to aid in data - analysis efforts , we have found that the following approximation reproduces the numerical results for @xmath10 to within a few percent : @xmath156 , { \nonumber}\\ \label{eqn : g3approx}\end{aligned}\ ] ] where we take @xmath157 in this expression . to compare with results for other models , and for comparison with cmb constraints , we next calulate the curvature bispectrum @xmath158 , defined by @xmath159 using the relations above , we can write @xmath160 and then find , @xmath161 a few observations : ( 1 ) it is only the amplitude , not the shape , that depends on the symmetry - breaking scale @xmath7 . ( 2 ) the bispectrum decreases , for fixed @xmath162 , as @xmath163 with increasing @xmath6 , again reflecting that the model should become increasingly gaussian with more fields , a consequence of the central - limit theorem . ( 3 ) the scaling with @xmath164 is @xmath165 , as opposed to the @xmath166 scaling of the local - model bispectrum . in words , the non - gaussianity is of order unity , a consequence of the fact that the density perturbation is the square of a gaussian random field [ cf . , eq . ( [ eqn : matterperturbation ] ) ] , rather than something very small , as in inflationary models . we now put the curvature bispectrum in a slightly more familiar form by defining a non - gaussianity parameter @xmath9 for the model . to do so , we recall that the local - model prediction for the curvature bispectrum is , @xmath167\ , , \label{eqn : localmodelcurvbi}\end{aligned}\ ] ] where @xmath168 is the local - model non - gaussianity parameter , defined by writing the curvature as @xmath169 $ ] in terms of a gaussian field @xmath170 . we now define the non - gaussianity parameter @xmath9 for sosfs by equating the local - model and sosf bispectra for equilateral triangles ; i.e. , equating eqs . ( [ eqn : localmodelcurvbi ] ) and ( [ eqn : almostcentralresult ] ) , we define , @xmath171 with this @xmath9 , the curvature bispectrum can then be written @xmath172 similarly , the gravitational - potential bispectrum can be obtained by multiplying this expression by @xmath173 and then replacing @xmath174 by @xmath175 . note that @xmath9 is manifestly positive , unlike @xmath176 for the local or equilateral models , which may take on either sign . as indicated in the introduction , the bispectrum can be probed with the cmb , large - scale structure , and the abundances of objects . the strongest current constraints to the local - model bispectrum come from the cmb @xcite , followed closely by galaxy - clustering constraints . given that sosfs produce a larger temperature - fluctuation amplitude for a given density - perturbation amplitude , we surmise that the cmb will provide stronger constraints to sosf non - gaussianity than galaxy clustering . we thus now estimate a constraint to @xmath9 from the cmb . before doing so , we first caution that eq . ( [ eqn : centralresult ] ) is derived for the curvature perturbation only for modes once they are well within the horizon . it is thus not , strictly speaking , appropriate for cmb modes @xmath177 . still , the shape dependence of the bispectrum , and its amplitude relative to the curvature - perturbation amplitude , arises primarily from the quadratic dependence of the density perturbation on the scalar - field perturbation as encoded in eq . ( [ eqn : delta_k ] ) . the shape dependence of the bispectrum we calculate should thus be at least roughly correct even for @xmath178 . cmb constraints to @xmath176 are typically applied assuming that the density perturbations are adiabatic , which implies a certain relation , @xmath179 , for the large - angle temperature fluctuation . in our case , though , there is roughly @xmath180 times more @xmath119 for a given @xmath181 than in adiabatic models . if so , and if all of these temperature fluctuations are due to scalar perturbations ( rather than vector and/or tensor modes ) , then the implied cmb bispectrum should be roughly @xmath182 times larger . simulations show , though , that only a fraction @xmath183 of the sosf temperature - fluctuation power is due to scalar modes , the rest coming from vector and tensor perturbations @xcite . the implied cmb bispectrum should thus scale with @xmath184 . combining these scalings with @xmath185 [ see eq . ( [ eqn : fnlsigma ] ) ] , the @xmath122 dependence of the cmb bispectrum drops out . we can therefore apply cmb constraints to @xmath9 by identifying the @xmath176 constraints obtained from the cmb for adiabatic perturbations with @xmath186 . and one final caveat : we disregard the differences in the temperature power spectra from sosfs and adiabatic perturbations . keeping these multiple caveats in mind , we proceed with our very rough estimate by noting that the wmap-7 95% c.l . limit to @xmath187 , the non - gaussianity parameter for the equilateral model , is @xmath188 @xcite ; this bispectrum is maximized for equilateral triangles . on the other hand , the sosf bispectrum is maximized for aligned triangles and is zero for squeezed triangles . we thus conclude that the constraint to @xmath189 will be stronger than that to @xmath187 , but it is not clear given the different weightings to squeezed and aligned triangles how it will compare with that to @xmath168 . applying these rough arguments to eq . ( [ eqn : fnlsigma ] ) , with @xmath190 , then we estimate a non - gaussianity parameter in excess of the predicted threshold @xmath191 for detection by planck @xcite . for now , however , we simply estimate conservatively that @xmath192 . we have carried out our calculations in the regime where analytic progress is most easily made i.e . , modes that have entered the horizon during matter domination and only after those modes have evolved well within the horizon . strictly speaking , therefore , our calculations apply only to galaxy surveys on _ very _ large scales those generally larger than extant surveys cover and possibly to mid - scale regimes in the cmb . still , our results for the bispectrum can be easily adapted to obtain roughly the bispectrum for smaller - scale modes , those that enter the horizon during radiation domination and those relevant for galaxy surveys . the calculation for the evolution of these modes is altered by three effects : ( 1 ) the index @xmath193 for the bessel functions in the scalar - field dynamics ( section [ sec : scalarfielddynamics ] ) is different . however , we have shown that this has no more than an @xmath103 effect on the power - spectrum and bispectrum . this is simply because the correlations between different fourier modes of the scalar - field energy density are imprinted , through eq . ( [ eqn : delta_k ] ) , by the dependence of those fourier modes on the scalar - field perturbations . this is the same for modes that enter the horizon during matter and radiation domination . the other two effects are ( 2 ) a slightly different amplitude for the matter perturbation , relative to the scalar - field energy - density perturbation , for modes that enter the horizon during radiation domination [ eq . ( 32 ) for rd in ref . @xcite , as opposed to eq . ( 29 ) in the same reference , our eq . ( [ eqn : matterperturbation ] ) , for md ] ; and ( 3 ) the usual linear - theory growth of primordial isocurvature perturbations through radiation domination and the transition to matter domination . these latter two effects amount to a calculation of the transfer function @xmath194 for the matter power spectrum in sosf models , which can be accomplished either with simulations [ cf . , @xcite ] or approximately with standard linear - theory calculations with primordial isocurvature fluctuations . again , however , although the calculation of the evolution of the amplitudes of the small - scale density - perturbation fourier modes will be far more complicated than the larger - scale modes we have focused upon , the _ correlations _ between those modes will be , at the @xmath103 level , the same as those we have calculated for larger - scale modes . more precisely , all we need to do is replace the density fields @xmath88 in sections [ sec : powerspectrum ] and [ sec : thecalculation ] by @xmath195 , where @xmath194 is the sosf transfer function . the matter power spectrum @xmath86 due to sosfs is then obtained from that in eq . ( [ eq : finalpowerspectrum ] ) by multiplying by @xmath196 , and the matter bispectrum is obtained by multiplying that in eq . ( [ eqn : ourbispectrum ] ) by @xmath197 . we can then write the normalization constant @xmath198 in eq . ( [ eqn : ourbispectrum ] ) in terms of the ( processed ) matter power spectra using eq . ( [ eq : finalpowerspectrum ] ) to obtain the matter bispectrum , @xmath199^{1/2}. \label{eqn : galaxybispectrum}\ ] ] valid for galaxy - survey scales . here @xmath86 is the _ processed _ power spectrum due to sosfs ; i.e. , it includes the transfer function . using eq . ( [ eqn : fnlsigma ] ) , this can be re - written in terms of @xmath9 as , @xmath200^{1/2 } { \nonumber}\\ & \simeq & 25\ , \fnl^\sigma \left ( \frac{g_{\mathrm{sw}}}{10 } \right)^3 \left ( \frac{p_\sigma}{0.1 } \right)^{-3/2 } \frac{g_3(k_1,k_2,k_3)}{g_3(1,1,1 ) } \left [ \frac { p^\sigma(k_1)p^\sigma(k_2)p^\sigma(k_3 ) } { k_1 k_2 k_3 } \right]^{1/2}. \label{eqn : galaxybispectrum2}\end{aligned}\ ] ] we leave further evaluation of this bispectrum , as well as assessment of current constraints , for future work . if some post - inflationary physics involves the spontaneous breaking of an exact @xmath3 symmetry with @xmath201 , then the ordering of these scalar fields may provide a secondary contribution to primordial perturbations . current constraints allow up to @xmath5 of the power in primordial perturbations to be due to sosfs . sosf models are appealing from the theoretical perspective because they are simple , well - defined , and parametrized only by the symmetry - breaking scale @xmath7 and number @xmath6 of fields . in this paper we have calculated the matter and curvature bispectra induced by the ordering of such scalar fields . given that the density perturbation is quadratic in the scalar - field perturbation , sosf density perturbations are expected to be highly non - gaussian , and if so , measurements of non - gaussianity may provide the means to test these models . here we have calculated analytically the bispectrum due to sosfs and presented results in a way that should be easily accessible to those doing measurements with the cmb and large - scale structure . we find that the triangle - shape dependence of the bispectrum peaks for aligned triangles , unlike the local - model bispectrum , which is largest for squeezed triangles , and the equilateral bispectrum , which is largest for equilateral triangles . we have estimated a current upper limit to the non - gaussianity parameter @xmath9 for the model and find that the implied constraints to the @xmath7-@xmath6 sosf parameter space are competitive with those from the upper limit to cmb temperature fluctuations . finally , we have already argued above , in section [ sec : currentconstraints ] , that the correlation of modes will be similar for the large - scale modes as they enter the horizon , those relevant for large - angle cmb fluctuations . we therefore believe that rough constraints to the model can be derived from cmb measurements by assuming that the curvature bispectrum we calculate is the primordial one . clearly , there is room for further numerical work to test our assumptions and to make our predictions more precise . in the meantime , though , we believe that our analytic approximation captures the essential physics and that our bispectrum can be used in the meantime as a `` working - horse '' model to derive constraints , from non - gaussianity measurements , to this interesting class of models for secondary contributions to primordial perturbations . finally , we note that the model makes a number of other predictions . given that density perturbations are actively generated as new modes come within the horizon , vector and tensor modes will be excited , and these may give rise to interesting polarization signals @xcite in the cmb and perhaps excite b modes @xcite in the cmb that might be distinguished from those due to inflation @xcite . there will also be a scale - invariant spectrum of primordial gravitational waves produced @xcite that can be sought in gravitational - wave observatories . dgf thanks caltech and cern for hospitality during the completion of this work . dgf acknowledges support from an fpu fellowship from the spanish ministry of science , with ref . ap-2005 - 1092 . this work was supported at caltech by doe de - fg03 - 92-er40701 and the gordon and betty moore foundation and at dartmouth by nsf ast-0349213 . from eq . ( [ eqn : delta_k ] ) , the two - point correlator in eq . ( [ eqn : psdefinition ] ) is expressed in terms of the correlator of four @xmath45 s . since @xmath45 s are gaussian distributed , we find via the wick theorem , @xmath202 \,{\nonumber}\\ & & \times\,|{{\mathbf{k}}}-{{\mathbf{q}}}||{{\mathbf{k}}}'-{{\mathbf{q}}}'|\,({{\mathbf{k}}}\cdot{{\mathbf{q}}})\,({{\mathbf{k}}}'\cdot{{\mathbf{q } } } ' ) \int d\tau\ , f(q\tau)f'(|{{\mathbf{k}}}-{{\mathbf{q}}}|\tau ) \int d\xi\ , f(q'\xi ) f'(|{{\mathbf{k}}}'-{{\mathbf{q}}}'|\xi)\ , . { \nonumber}\\ \label{eqn : beginning}\end{aligned}\ ] ] using eq . ( [ eq : correlator_a ] ) , we find @xmath203 @xmath204 and @xmath205 substituting these expressions into eq . ( [ eqn : beginning ] ) , we obtain @xmath206\,{\nonumber}\\ & = & \frac{c^2\eta^4}{a^2}\,\frac{k}{n}\,\int \frac{d^3v}{(2\pi)^3}\,\mathcal{i}(v,|\hat{{\mathbf{k}}}-{{\mathbf{v}}}| ) \left[(\hat{{\mathbf{k}}}\cdot{{\mathbf{v}}})^2\,\mathcal{i}(v,|\hat{{\mathbf{k}}}-{{\mathbf{v}}}| ) + ( \hat{{\mathbf{k}}}\cdot{{\mathbf{v}}})(1-\hat{{\mathbf{k}}}\cdot{{\mathbf{v}}})\mathcal{i}(|\hat{{\mathbf{k}}}-{{\mathbf{v}}}|,v ) \right]\ , , \label{eqn : psresult}\end{aligned}\ ] ] where @xmath207 is defined in eq . ( [ eqn : mathcali ] ) . from here we can then introduce @xmath208 in section [ sec : powerspectrum ] . the bispectrum is obtained by starting with , @xmath209 the expectation value of the product of six @xmath45 s can be expanded with wick contractions and , after some algebra , and using eq . ( [ eq : correlator_a ] ) , we find @xmath210\ , \big[\,((\hat{{\mathbf{k}}}+{\bf u})\cdot{\bf b } ) \,\,\mathcal{i}(b , b_2)+((\hat{{\mathbf{k}}}+{\bf u})\cdot{\bf b}_2)\,\ , \mathcal{i}(b_2,b)\,\big]\right.{\nonumber}\\ & & \hspace{.5 cm } \left.+ \big[({\bf u}\cdot{\bf b}\,\ , \mathcal{i}(b , b_{12 } ) - ( { \bf u}\cdot{\bf b}_{12})\,\ , \mathcal{i}(b_{12},b)\,\big]\ , \big[\,((\hat{{\mathbf{k}}}+{\bf u})\cdot{{\mathbf{v}}})\,\ , \mathcal{i}(v , b_{12})+((\hat{{\mathbf{k}}}+{\bf u})\cdot{\bf b}_{12})\,\ , \mathcal{i}(b_{12},v)\,\big]\right\rbrace , { \nonumber}\\\end{aligned}\ ] ] where @xmath211 , @xmath212 , @xmath213 , and we have defined @xmath214 . defining @xmath215 as in eq . ( [ eqn : ourbispectrum ] ) and performing a change of variables @xmath216 in the second term of the above integral , we then find @xmath217\big[({\bf u}\cdot{\bf v})\mathcal{i}(v , b_2 ) - ( { \bf u}\cdot{\bf b}_2)\mathcal{i}(b_2,v)\big ] { \nonumber}\\ & & \hspace{0.5 cm } \times \big[{\bf b}\cdot(\hat{{\mathbf{k}}}+{\bf u})\mathcal{i}(b , b_2 ) + { \bf b}_2\cdot(\hat{{\mathbf{k}}}+{\bf u})\mathcal{i}(b_2,b)\big ] \,.{\nonumber}\\\end{aligned}\ ] ] if the upper limit of the integral defining @xmath207 is much greater than one ( as expected for sub - horizon modes ) , such that @xmath218 , we then obtain @xmath219 finally , simple algebraic re - arrangements in eq . ( [ eq : g3compact ] ) then results in the far simpler expression , in eq . ( [ eqn : biresult ] ) for @xmath215 , @xmath220 with @xmath221 . this expression is indeed equivalent to eq . ( 59 ) in ref . @xcite , although it is written in a much more compact and simpler way . note that it is smaller by a factor @xmath222 than that of ref . @xcite , as a consequence of different conventions . once modes are well inside the horizon , the upper limit for the integral in eq . ( [ eqn : mathcali ] ) is large , and the integral can then be approximated by @xmath223 where @xmath224 , and @xmath39 for matter domination . the integral can be performed analytically ; the result is @xmath225 where @xmath226\,,\end{aligned}\ ] ] @xmath227 is the hypergeometric function , and @xmath228 is the order of the bessel function . while straightforward to evaluate numerically , this exact solution may be computationally expensive to evaluate repeatedly . we therefore use for our numerical work the approximation , @xmath229 which provides good agreement with the exact results with @xmath230 . a. h. guth and s. y. pi , phys . lett . * 49 * , 1110 ( 1982 ) ; a. a. starobinsky , phys . lett . b * 117 * , 175 ( 1982 ) ; j. m. bardeen , p. j. steinhardt and m. s. turner , phys . d * 28 * , 679 ( 1983 ) . t. j. allen , b. grinstein and m. b. wise , phys . b * 197 * , 66 ( 1987 ) ; l. a. kofman and d. y. pogosian , phys . b * 214 * , 508 ( 1988 ) ; d. s. salopek , j. r. bond and j. m. bardeen , phys . d * 40 * , 1753 ( 1989 ) ; a. d. linde and v. f. mukhanov , phys . d * 56 * , 535 ( 1997 ) [ arxiv : astro - ph/9610219 ] ; p. j. e. peebles , astrophys . j. * 510 * , 523 ( 1999 ) [ arxiv : astro - ph/9805194 ] ; p. j. e. peebles , astrophys . j. * 510 * , 531 ( 1999 ) [ arxiv : astro - ph/9805212 ] . s. mollerach , phys . rev . d * 42 * , 313 ( 1990 ) ; a. d. linde and v. f. mukhanov , phys . d * 56 * , 535 ( 1997 ) [ arxiv : astro - ph/9610219 ] ; d. h. lyth and d. wands , phys . b * 524 * , 5 ( 2002 ) [ arxiv : hep - ph/0110002 ] ; t. moroi and t. takahashi , phys . b * 522 * , 215 ( 2001 ) [ erratum - ibid . b * 539 * , 303 ( 2002 ) ] [ arxiv : hep - ph/0110096 ] ; d. h. lyth , c. ungarelli and d. wands , phys . d * 67 * , 023503 ( 2003 ) [ arxiv : astro - ph/0208055 ] ; k. ichikawa _ et al . _ , arxiv:0802.4138 [ astro - ph ] ; k. enqvist , s. nurmi , o. taanila and t. takahashi , arxiv:0912.4657 [ astro-ph.co ] ; k. enqvist and t. takahashi , jcap * 0912 * , 001 ( 2009 ) [ arxiv:0909.5362 [ astro-ph.co ] ] ; k. enqvist and t. takahashi , jcap * 0809 * , 012 ( 2008 ) [ arxiv:0807.3069 [ astro - ph ] ] ; k. enqvist and s. nurmi , jcap * 0510 * , 013 ( 2005 ) [ arxiv : astro - ph/0508573 ] ; a. l. erickcek , m. kamionkowski and s. m. carroll , arxiv:0806.0377 [ astro - ph ] ; a. l. erickcek , c. m. hirata and m. kamionkowski , phys . rev . d * 80 * , 083507 ( 2009 ) [ arxiv:0907.0705 [ astro-ph.co ] ] . g. r. dvali and s. h. h. tye , phys . b * 450 * , 72 ( 1999 ) [ arxiv : hep - ph/9812483 ] ; p. creminelli , jcap * 0310 * , 003 ( 2003 ) [ arxiv : astro - ph/0306122 ] ; m. alishahiha , e. silverstein and d. tong , phys . d * 70 * , 123505 ( 2004 ) [ arxiv : hep - th/0404084 ] . a. vilenkin and e. p. .s . shellard cambridge university press , 1994 ; r. durrer , new astron . rev . * 43 * , 111 ( 1999 ) ; r. durrer , m. kunz and a. melchiorri , phys . rept . * 364 * , 1 ( 2002 ) [ arxiv : astro - ph/0110348 ] . s. sarangi and s. h. h. tye , phys . b * 536 * , 185 ( 2002 ) [ arxiv : hep - th/0204074 ] ; n. t. jones , h. stoica and s. h. h. tye , phys . b * 563 * , 6 ( 2003 ) [ arxiv : hep - th/0303269 ] ; e. j. copeland , r. c. myers and j. polchinski , jhep * 0406 * , 013 ( 2004 ) [ arxiv : hep - th/0312067 ] ; g. dvali and a. vilenkin , jcap * 0403 * , 010 ( 2004 ) [ arxiv : hep - th/0312007 ] ; r. jeannerot , j. rocher and m. sakellariadou , phys . d * 68 * , 103514 ( 2003 ) [ arxiv : hep - ph/0308134 ] . l. kofman , a. d. linde and a. a. starobinsky , phys . rev . lett . * 76 * , 1011 ( 1996 ) [ arxiv : hep - th/9510119 ] ; g. n. felder , j. garcia - bellido , p. b. greene , l. kofman , a. d. linde and i. tkachev , phys . * 87 * , 011601 ( 2001 ) [ arxiv : hep - ph/0012142 ] ; g. n. felder , l. kofman and a. d. linde , phys . d * 64 * , 123517 ( 2001 ) [ arxiv : hep - th/0106179 ] . n. turok and d. n. spergel , phys . lett . * 66 * , 3093 ( 1991 ) ; m. kunz and r. durrer , phys . d * 55 * , 4516 ( 1997 ) [ arxiv : astro - ph/9612202 ] ; r. durrer , m. kunz and a. melchiorri , phys . rev . d * 59 * , 123005 ( 1999 ) [ arxiv : astro - ph/9811174 ] . n. bevis , m. hindmarsh and m. kunz , phys . d * 70 * , 043508 ( 2004 ) [ arxiv : astro - ph/0403029 ] . p. crotty , j. garcia - bellido , j. lesgourgues and a. riazuelo , phys . lett . * 91 * , 171301 ( 2003 ) [ arxiv : astro - ph/0306286 ] ; m. beltran , j. garcia - bellido , j. lesgourgues and m. viel , phys . d * 72 * , 103515 ( 2005 ) [ arxiv : astro - ph/0509209 ] . m. cruz , n. turok , p. vielva , e. martinez - gonzalez and m. hobson , science * 318 * , 1612 ( 2007 ) [ arxiv:0710.5737 [ astro - ph ] ] ; c. l. bennett _ et al . _ , arxiv:1001.4758 [ astro-ph.co ] . n. bartolo _ et al . _ , rept . * 402 * , 103 ( 2004 ) [ arxiv : astro - ph/0406398 ] . f. bernardeau , s. colombi , e. gaztanaga and r. scoccimarro , phys . rept . * 367 * , 1 ( 2002 ) [ arxiv : astro - ph/0112551 ] . t. falk , r. rangarajan and m. srednicki , astrophys . j. * 403 * , l1 ( 1993 ) [ arxiv : astro - ph/9208001 ] ; a. gangui _ et al . _ , astrophys . j. * 430 * , 447 ( 1994 ) [ arxiv : astro - ph/9312033 ] ; a. gangui , phys . d * 50 * , 3684 ( 1994 ) [ arxiv : astro - ph/9406014 ] . l. m. wang and m. kamionkowski , phys . d * 61 * , 063504 ( 2000 ) [ arxiv : astro - ph/9907431 ] . j. m. maldacena , jhep * 0305 * , 013 ( 2003 ) [ arxiv : astro - ph/0210603 ] . d. babich , p. creminelli and m. zaldarriaga , jcap * 0408 * , 009 ( 2004 ) [ arxiv : astro - ph/0405356 ] ; p. creminelli _ et al . _ , jcap * 0605 * , 004 ( 2006 ) [ arxiv : astro - ph/0509029 ] ; p. creminelli _ et al . _ , jcap * 0703 * , 005 ( 2007 ) [ arxiv : astro - ph/0610600 ] . s. hannestad , t. haugboelle , p. r. jarnhus and m. s. sloth , arxiv:0912.3527 [ hep - ph ] . x. c. luo , astrophys . j. * 427 * , l71 ( 1994 ) [ arxiv : astro - ph/9312004 ] ; l. verde _ et al . _ , mon . not . 313 * , l141 ( 2000 ) [ arxiv : astro - ph/9906301 ] ; e. komatsu and d. n. spergel , phys . d * 63 * , 063002 ( 2001 ) [ arxiv : astro - ph/0005036 ] . p. coles _ et al . _ , mon . not . soc . * 264 * , 749 ( 1993 ) [ arxiv : astro - ph/9302015 ] ; x. c. luo and d. n. schramm , astrophys . j. * 408 * , 33 ( 1993 ) ; e. lokas _ et al . _ , mon . not . * 274 * , 730 ( 1995 ) [ arxiv : astro - ph/9407095 ] ; m. j. chodorowski and f. r. bouchet , mon . not . soc . * 279 * , 557 ( 1996 ) [ arxiv : astro - ph/9507038 ] ; a. j. stirling and j. a. peacock , mon . not . * 283 * , 99 ( 1996 ) [ arxiv : astro - ph/9608101 ] ; r. durrer _ et al . _ , d * 62 * , 021301 ( 2000 ) [ arxiv : astro - ph/0005087 ] . l. verde and a. f. heavens , astrophys . j. * 553 * , 14 ( 2001 ) [ arxiv : astro - ph/0101143 ] ; a. buchalter and m. kamionkowski , astrophys . j. * 521 * , 1 ( 1999 ) [ arxiv : astro - ph/9903462 ] ; a. buchalter , m. kamionkowski and a. h. jaffe , astrophys . j. * 530 * , 36 ( 2000 ) [ arxiv : astro - ph/9903486 ] . w. a. chiu , j. p. ostriker and m. a. strauss , astrophys . j. * 494 * , 479 ( 1998 ) [ arxiv : astro - ph/9708250 ] ; j. robinson , e. gawiser and j. silk , astrophys . j. * 532 * , 1 ( 2000 ) [ arxiv : astro - ph/9906156 ] ; j. a. willick , astrophys . j. * 530 * , 80 ( 2000 ) [ arxiv : astro - ph/9904367 ] ; l. verde _ et al . _ , mon . not . soc . * 321 * , l7 ( 2001 ) [ arxiv : astro - ph/0007426 ] ; n. n. weinberg and m. kamionkowski , mon . not . soc . * 341 * , 251 ( 2003 ) [ arxiv : astro - ph/0210134 ] ; s. matarrese , l. verde and r. jimenez , astrophys . j. * 541 * , 10 ( 2000 ) [ arxiv : astro - ph/0001366 ] ; l. verde _ et al . _ , mon . not . 325 * , 412 ( 2001 ) [ arxiv : astro - ph/0011180 ] ; m. loverde _ et al . _ , jcap * 0804 * , 014 ( 2008 ) [ arxiv:0711.4126 [ astro - ph ] ] . m. kamionkowski , l. verde and r. jimenez , jcap * 0901 * , 010 ( 2009 ) [ arxiv:0809.0506 [ astro - ph ] ] . n. dalal _ et al . _ , d * 77 * , 123514 ( 2008 ) [ arxiv:0710.4560 [ astro - ph ] ] ; s. matarrese and l. verde , astrophys . j. * 677 * , l77 ( 2008 ) [ arxiv:0801.4826 [ astro - ph ] ] ; a. slosar _ et al . _ , arxiv:0805.3580 [ astro - ph ] ; c. carbone , l. verde and s. matarrese , arxiv:0806.1950 [ astro - ph ] . a. h. jaffe , phys . d * 49 * , 3893 ( 1994 ) [ arxiv : astro - ph/9311023 ] . a. silvestri and m. trodden , phys . lett . * 103 * , 251301 ( 2009 ) [ arxiv:0811.2176 [ astro - ph ] ] ; m. hindmarsh , c. ringeval and t. suyama , phys . d * 80 * , 083501 ( 2009 ) [ arxiv:0908.0432 [ astro-ph.co ] ] ; m. hindmarsh , c. ringeval and t. suyama , arxiv:0911.1241 [ astro-ph.co ] ; d. m. regan and e. p. s. shellard , arxiv:0911.2491 [ astro-ph.co ] . m. kamionkowski and j. march - russell , phys . b * 282 * , 137 ( 1992 ) [ arxiv : hep - th/9202003 ] ; m. kamionkowski and j. march - russell , phys . lett . * 69 * , 1485 ( 1992 ) [ arxiv : hep - th/9201063 ] ; r. holman , s. d. h. hsu , t. w. kephart , e. w. kolb , r. watkins and l. m. widrow , phys . b * 282 * , 132 ( 1992 ) [ arxiv : hep - ph/9203206 ] ; r. holman , s. d. h. hsu , e. w. kolb , r. watkins and l. m. widrow , phys . lett . * 69 * , 1489 ( 1992 ) . s. m. barr and d. seckel , phys . rev . d * 46 * , 539 ( 1992 ) . s. veeraraghavan and a. stebbins , astrophys . j. * 365 * , 37 ( 1990 ) . u. l. pen , u. seljak and n. turok , phys . * 79 * , 1611 ( 1997 ) [ arxiv : astro - ph/9704165 ] . n. turok , u. l. pen and u. seljak , phys . d * 58 * , 023506 ( 1998 ) [ arxiv : astro - ph/9706250 ] . u. l. pen , d. n. spergel and n. turok , phys . d * 49 * , 692 ( 1994 ) . p. creminelli , a. nicolis , l. senatore , m. tegmark and m. zaldarriaga , jcap * 0605 * , 004 ( 2006 ) [ arxiv : astro - ph/0509029 ] . e. komatsu _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 148 * , 119 ( 2003 ) [ arxiv : astro - ph/0302223 ] ; a. p. s. yadav and b. d. wandelt , phys . rev . lett . * 100 * , 181301 ( 2008 ) [ arxiv:0712.1148 [ astro - ph ] ] ; e. komatsu _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 180 * , 330 ( 2009 ) [ arxiv:0803.0547 [ astro - ph ] ] . e. komatsu _ et al . _ , arxiv:1001.4538 [ astro-ph.co ] . a. cooray , d. sarkar and p. serra , phys . d * 77 * , 123006 ( 2008 ) [ arxiv:0803.4194 [ astro - ph ] ] . u. seljak , u. l. pen and n. turok , phys . lett . * 79 * , 1615 ( 1997 ) [ arxiv : astro - ph/9704231 ] . m. kamionkowski , a. kosowsky and a. stebbins , phys . d * 55 * , 7368 ( 1997 ) [ arxiv : astro - ph/9611125 ] ; m. kamionkowski , a. kosowsky and a. stebbins , phys . lett . * 78 * , 2058 ( 1997 ) [ arxiv : astro - ph/9609132 ] ; m. zaldarriaga and u. seljak , phys . d * 55 * , 1830 ( 1997 ) [ arxiv : astro - ph/9609170 ] . u. seljak and m. zaldarriaga , phys . lett . * 78 * , 2054 ( 1997 ) [ arxiv : astro - ph/9609169 ] . d. baumann and m. zaldarriaga , jcap * 0906 * , 013 ( 2009 ) [ arxiv:0901.0958 [ astro-ph.co ] ] ; p. adshead and e. a. lim , arxiv:0912.1615 [ astro-ph.co ] . l. m. krauss , phys . lett . b * 284 * , 229 ( 1992 ) ; e. fenu , d. g. figueroa , r. durrer and j. garcia - bellido , jcap * 0910 * , 005 ( 2009 ) [ arxiv:0908.0425 [ astro-ph.co ] ] ; k. jones - smith , l. m. krauss and h. mathur , phys . rev . lett . * 100 * , 131302 ( 2008 ) [ arxiv:0712.0778 [ astro - ph ] ] .

we present a combined experimental and numerical study of the primary and secondary instabilities in a kolmogorov - like flow . the experiment uses electromagnetic forcing with an approximately sinusoidal spatial profile to drive a quasi - two - dimensional ( q2d ) shear flow in a thin layer of electrolyte suspended on a thin lubricating layer of a dielectric fluid . the experimental flow is modelled using a recently proposed 2d equation , derived from first principles by depth - averaging the full three - dimensional navier - stokes equations . we also introduce a realistic numerical model approximating the electromagnetic body force in the experiment , which is validated against experimental measurements . we then explore the effect of lateral boundaries by performing direct numerical simulations ( dns ) subject to different types of boundary conditions . as the strength of the forcing is increased , this q2d flow undergoes a series of bifurcations . we compare the flow regimes observed in the experiment with those obtained from the dns and find that only the simulation subject to physical no - slip boundary conditions provides close , quantitative agreement . our analysis offers additional validation of the 2d model proposed by , as well as a demonstration of the importance of properly modelling the forcing and boundary conditions .

1601.00243

fluid flows in two spatial dimensions have been the subject of substantial research efforts in recent decades . for the greater part of the twentieth century , it was generally considered that two - dimensional ( 2d ) flows were merely a theoretical idealization with limited practical relevance . this conception has changed drastically since the 1980s , when experiments in thin electrolyte layers @xcite , soap films @xcite , and liquid metals @xcite demonstrated that nearly 2d flows can indeed be realized in the laboratory . today , experimental approximations of 2d flows are widely employed as models of atmospheric and oceanic flows @xcite . being theoretically and experimentally more amenable than their three - dimensional ( 3d ) counterparts , 2d flows have also served as platforms for studying new phenomena such as turbulent cascades @xcite , coherent structures @xcite , and mixing @xcite . hence , experimental studies of 2d flows , with quantitative comparisons to theoretical predictions , are important for improving our understanding of fluid flows , in both 2d and 3d . perhaps one of the best known 2d fluid flow models is a flow introduced by andrey kolmogorov in 1959 as a mathematical problem for studying hydrodynamic stability @xcite . the kolmogorov flow represents the motion of a viscous fluid in two dimensions ( we will refer to these as @xmath0 and @xmath1 ) driven by a forcing that points along the @xmath0-direction and varies sinusoidally along the @xmath1-direction . the fluid flow is considered incompressible , @xmath2 , and is governed by the 2d navier - stokes equation , @xmath3 here , @xmath4 is the velocity field , @xmath5 is the 2d pressure field , and @xmath6 represents the driving force with amplitude @xmath7 and wavenumber @xmath8 . the parameters @xmath9 and @xmath10 are the density and the kinematic viscosity of the fluid , respectively . kolmogorov flow has served as a convenient model for understanding a wide variety of hydrodynamic phenomena in 2d , such as fluid instabilities @xcite , 2d turbulence @xcite , and coherent structures @xcite . practically realizable flows , however , are never strictly 2d . experimental approximations of kolmogorov flow have often been carried out in either shallow layers of electrolytes @xcite or in soap films @xcite , wherein geometric confinement suppresses the component of velocity along one of the spatial directions ( @xmath11 ) . the remaining two velocity components , however , generally depend on both extended and confined coordinates , making the flow `` quasi - two - dimensional '' ( q2d ) . to account for the dependence on the confined coordinate , q2d flows in shallow layers have often been modelled by adding a linear friction term to the 2d navier - stokes equation ( [ eq:2dns ] ) : @xmath12 where @xmath13 is a constant . the addition of this term was first suggested by @xcite to model a q2d flow generated in a homogeneous shallow electrolyte layer . in such a flow , the bottom of the fluid layer is constrained to be at rest because it is in contact with the solid surface of the container holding the fluid . this no - slip constraint at the bottom of the fluid layer causes a gradient in the magnitude of the horizontal velocity * u * along the confined direction @xmath11 . @xcite rationalized that the dissipation due to this shear , for sufficiently shallow fluid layers , is captured by the linear friction term . in the context of q2d flows in electrolyte layers , this term has come to be known as `` rayleigh friction . '' experimental flows in thin layers as well as models that employ equation ( [ eq:2dns_wf ] ) are now commonly referred to as `` kolmogorov - like , '' when the forcing profile is nearly sinusoidal . experimental realizations of q2d flows in recent years have employed two - fluid - layer setups : either a setup with _ miscible _ layers comprised of a heavy electrolyte fluid ( salt water ) beneath a lighter nonconducting fluid ( pure water ) @xcite , or a setup with _ immiscible _ layers comprised of a heavy dielectric fluid beneath a lighter electrolyte @xcite . the rationale behind these modifications was that in addition to confinement , density stratification and immiscibility should enhance two - dimensionality in the top layer . theoretical models of q2d experimental flows realized in stratified layers of fluids , however , have not accurately modelled the effect of the gradient in the magnitude of horizontal velocity @xmath14 along the confined direction @xmath11 . consequently , experiments were compared with simulations based on the 2d model ( [ eq:2dns_wf ] ) with empirically estimated parameters @xcite . to address this deficiency , @xcite have investigated the variation in the horizontal velocity @xmath14 along the confined direction @xmath11 for a stratified two - immiscible - layer setup . the following modified version of equation ( [ eq:2dns_wf ] ) which accounts for the inherent three - dimensionality of a q2d flow and describes its evolution quantitatively was proposed : @xmath15 where @xmath16 is a numerical prefactor to the advection term that was not accounted for previously . this prefactor reflects the decrease of the mean inertia of the fluid layer due to the variation of the horizontal velocity in the vertical direction . equation ( [ eq:2dns_mod ] ) has been derived from first principles by depth - averaging the full 3d navier - stokes equation under the q2d approximation ( cf . appendix [ sec : sensitivity ] ) . the coefficients @xmath13 , @xmath10 , and @xmath17 are estimated by including effects of both the inhomogeneity in fluid properties as well the vertical profile of @xmath14 along @xmath11 . equations ( [ eq:2dns ] ) and ( [ eq:2dns_wf ] ) can be treated as special cases of ( [ eq:2dns_mod ] ) with suitable choices of the parameters @xmath13 and @xmath17 . furthermore , equation ( [ eq:2dns_wf ] ) can also be obtained from ( [ eq:2dns_mod ] ) by rescaling the variables ( cf . appendix [ sec : nondim ] ) . in this article we study instabilities of a q2d kolmogorov - like flow realized in a setup with two immiscible fluid layers . we compare experiments with results from direct numerical simulations ( dns ) of equation ( [ eq:2dns_mod ] ) . most previous studies of kolmogorov - like flow compared experiments to theoretical analysis of a perfectly sinusoidal shear flow on unbounded or periodic domain @xcite . while some of these studies reported quantitative agreement between theory and experiment in regards to the primary instability , none where able to match simultaneously both the critical reynolds number and the critical wavenumber . even matching one of these required treating the rayleigh friction coefficient @xmath13 as an adjustable parameter . to address these difficulties , we systematically investigate the effects of lateral confinement , which was not properly addressed by the previous studies , using numerical simulations with three different sets of boundary conditions . furthermore , we investigate how the observed flow patterns and their stability are affected by the deviations in the forcing profile from perfect periodicity in the extended directions and by the variation of the forcing profile in the confined direction . finally , we compare the results of numerical simulations with the experimental observations for the secondary instability , which introduces time - dependence into the flow . the prospect of developing a 2d model that quantitatively describes the evolution of a q2d flow is particularly appealing in view of the recent theoretical advances in understanding transitional flows and weak turbulence as a dynamical process guided by nonchaotic , unstable solutions of the navier - stokes equation , often referred to as exact coherent structures ( ecs ) @xcite . the bulk of numerical studies have explored the role of ecs in 3d flows simulated on periodic domains with simple geometries , such as pipe flow , plane couette flow , and plane poiseuille flow . however , experimental evidence for the role of ecs is scarce @xcite , in part due to technical limitations in obtaining spatially and temporally resolved 3d velocity fields . q2d flows , on the other hand , can be quantified using 2d planar velocity fields which are relatively easy to measure . hence , experimental q2d flows that can be described by a 2d model serve as an ideal platform for testing and developing an ecs - based approach to understanding turbulence . recently , @xcite and @xcite have identified dozens of ecs in numerical simulations of a weakly turbulent 2d kolmogorov flow , governed by equation ( [ eq:2dns ] ) with periodic boundary conditions . we hope that this paper helps provide a rigorous foundation for future studies in the same spirit , but with a focus on experimental validation of theoretical predictions . this article is organized as follows . in [ sec : exp_setup ] , we describe the experimental setup employed to generate a q2d kolmogorov - like flow . in [ sec : model ] , we introduce a realistic model of the forcing in the experiment and discuss different types of lateral boundary conditions which are used to study the effects of confinement theoretically . in [ sec : comparison ] , we compare the flow fields obtained from experimental measurements with those from the numerical simulations for different flow regimes and characterize the bifurcations associated with increasing the forcing strength . conclusions are presented in [ sec : conclusion ] . we generate a q2d kolmogorov - like flow in the experiment using a stratified setup with two immiscible fluid layers , first introduced by @xcite . in this configuration , a lighter electrolyte is suspended on top of a denser dielectric , which serves as a lubricant between the electrolyte layer and the solid surface at the bottom of the container which holds the fluids . the fluid layers are set in motion using lorentz forces resulting from the interaction of a direct current passing through the electrolyte and a spatially varying magnetic field . we use a magnet array consisting of 14 ndfeb magnets ( grade n42 ) to generate a magnetic field that varies roughly sinusoidally along one direction , approximating the forcing in the kolmogorov flow . each magnet in the array is 15.24 cm long and 1.27 cm wide , with a thickness of @xmath18 cm . the magnetization is parallel to the thickness dimension , with a surface field strength of about 0.2 t. the magnets are positioned side - by - side along their width to form a 15.24 cm @xmath19 ( @xmath20 cm ) @xmath19 0.32 cm array such that the adjacent magnets have fields pointing in opposite directions , normal to the plane of the array . this magnet array is placed on a flat aluminum plate of dimensions 30.5 cm @xmath19 30.5 cm @xmath19 1.0 cm , and rectangular pieces of aluminum with the same thickness as the magnets ( @xmath21 cm ) are placed beside the magnet array to create a level surface . manufacturing imperfections in the individual magnets and the aluminum siding result in a surface which is not adequately smooth . hence , a thin glass plate measuring @xmath22 cm @xmath19 @xmath22 cm in area with a thickness of @xmath23 cm is placed atop the magnets and siding to provide a uniform bottom surface . a thin layer of black , adhesive contact paper ( with approximate thickness 0.005 cm ) is placed on top of the glass plate to serve as a dark background for imaging . the surface of the contact paper serves as the bottom boundary for the fluids . we place the origin of our coordinate system at this height and the lateral centre of the magnet array , with the @xmath0-coordinate aligned with the magnets longest side , the @xmath1-coordinate pointing in the direction of the magnet array periodicity , and the @xmath11-coordinate in the vertical direction . a schematic diagram is shown in figure [ expsetup ] . rectangular bars of acrylic are affixed directly onto the contact paper to create the lateral boundaries of the container that will hold the fluids . parallel to the @xmath1-direction , two bars are placed at a distance of 17.8 cm apart , centred about the origin . these solid boundaries for the fluid are henceforth referred to as the `` end walls . '' similarly , running parallel to the @xmath0-direction , two electrodes mounted on rectangular bars of acrylic are placed at a distance of 22.9 cm , symmetrically relative to the origin . these boundaries are henceforth referred to as the `` side walls '' and are used to drive the current through the electrolyte . the placement of the end walls and side walls leaves a buffer region of @xmath24 cm and @xmath25 cm , respectively , between the edge of the magnet array and these solid boundaries . the aluminum plate upon which the magnets are mounted is supported by three screws , which are adjusted to level the system . the interior of the container is filled with @xmath26 ml of a dielectric fluid and @xmath27 ml of an electrolyte to form two immiscible layers that are @xmath28 cm and @xmath29 cm thick , respectively . the dielectric fluid used is perfluorooctane , which has a viscosity of @xmath30 mpa@xmath31s and a density of @xmath32 kg / m@xmath33 at 23.0@xmath34c . the electrolyte fluid is a solution consisting of 60% 1 m copper sulfate solution and 40% glycerol by weight . the electrolyte s viscosity is @xmath35 mpa@xmath31s and the density is @xmath36 kg / m@xmath33 at 23.0@xmath34c . note that a large viscosity ratio @xmath37 has been chosen to enhance the two - dimensionality of the electrolyte , as described by @xcite . a small amount of viscosity - matched surfactant is added to the electrolyte to lower the surface tension , and a glass lid is placed on top of the container to limit evaporation . a direct current , which serves as the control parameter , is then passed through the electrolyte ; the resulting current density @xmath38 ranges from about 10 to 40 a / m@xmath39 across the different runs . the interaction of this current with the spatially alternating magnetic field @xmath40 results in a spatially alternating lorentz force @xmath41 which drives the electrolyte ( cf . figure [ expsetup ] ( a ) ) . the viscous coupling between the electrolyte and the dielectric fluids sets the dielectric fluid in motion as well . since passing a current through a resistive conductor ( the electrolyte ) results in joule heating , a calibrated thermistor is placed in the corner of the fluid domain to monitor the fluid temperature , and the aluminum plate is immersed in a temperature - controlled water bath . the water bath is regulated such that the temperature of the electrolyte is maintained to @xmath42c . by limiting the temperature fluctuations , the associated change in viscosity of the fluids is kept to a minimum . for flow visualization , we add hollow glass microspheres ( glass bubbles k15 ) manufactured by 3 m , sieved to obtain particles with mean radius @xmath43 @xmath44 m and mean density @xmath45 kg / m@xmath33 . being lighter than the electrolyte , the microspheres stay afloat at the electrolyte - air interface for the duration of the experiment . the microspheres are illuminated with white light emitting diodes placed near the end walls , outside the container holding the fluids . the flow is imaged at 15 hz with a dmk 31bu03 camera manufactured by the imaging source , placed directly above the setup . this camera has a ccd sensor with a resolution of 1024 @xmath19 768 pixels , which results in an adequate resolution of about 53 pixels per magnet width . the flow velocities are calculated using the prana particle image velocimetry ( piv ) package @xcite . this software employs a multigrid piv algorithm that deforms images to better resolve flows with high shear . the velocity field is resolved on a 169 @xmath19 126 grid , with about 9 points per magnet width . for the experimental measurements listed above , we obtain @xcite the following depth - averaged values for the parameters in equation ( [ eq:2dns_mod ] ) : @xmath46 s@xmath47 , @xmath48 , and @xmath49 m@xmath39/s . the complexity of the flow in both the experiment and simulation is characterized by the reynolds number , which we define as : @xmath50 where @xmath51 cm is the width of one magnet and @xmath52 is the measured root - mean - square ( rms ) velocity , where @xmath53 denotes spatial averaging over a sub - region with dimensions 10.16 cm @xmath19 10.16 cm at the centre of the magnet array . note that from this point on , unless noted otherwise , the characteristic length scale @xmath54 , velocity scale @xmath55 , and time scale @xmath56 will be used for nondimensionalization . in this section we present a model of the magnetic field generated by the finite array of permanent magnets in the experiment . we then introduce three types of boundary conditions used in our numerical simulations of the flow . in the discussion so far , we have not addressed an important question of how the 2d forcing function @xmath57 in equation ( [ eq:2dns_mod ] ) relates to the 3d forcing @xmath41 in the experiment . for a kolmogorov flow , the forcing @xmath57 is sinusoidal , by definition . however , for kolmogorov - like flows realized in electromagnetically driven shallow layers of electrolyte , @xmath57 needs to be computed from the 3d lorentz force * f * arising from the interaction of the magnetic field @xmath40 produced by a finite magnet array with a current density @xmath58 @xcite . the current density is easily calculated from geometrical considerations , but the magnetic field generated by the array of permanent magnets is quite complicated . for @xmath59 , the lorentz force density at any location @xmath60 within the electrolyte layer is given by @xmath61 . here , @xmath62 and @xmath63 are the @xmath0- and @xmath11-components of the magnetic field , respectively , which vary along all three coordinates @xmath0 , @xmath1 , and @xmath11 . experimental measurements show that the typical value of @xmath62 is less than 3% of the value of @xmath63 at any given location within the electrolyte . furthermore , the vertical component of the lorentz force along with the gravitational force will be balanced by the vertical gradient of the pressure . hence , the lorenz force density for all practical purposes can be approximated as @xmath64 . one can then compute @xmath57 using the expression : @xmath65 where @xmath66 and @xmath67 are the thicknesses of the electrolyte and dielectric layers , respectively , while @xmath9 is the depth - averaged density @xcite . + the black symbols in figure [ fig : mag_prof ] ( a ) show the experimental measurements of @xmath63 along the line @xmath68 , passing above the centre of the magnet array at two different heights . clearly , the magnetic field profile deviates significantly from that of a pure sinusoid . furthermore , one can not ignore the fringe fields near the edges of the array . to obtain a magnetic field profile that closely resembles the one in the experiment , one could measure the @xmath11-component of the magnetic field ( @xmath63 ) across the entire flow domain at various heights above the magnet array . using the measured field , one could then compute the depth - averaged forcing profile using equation ( [ eq : depth_averaging ] ) @xcite . however , since measuring @xmath63 on a 3d grid is an extremely tedious process , we circumvent the labour by numerically modelling the magnet array as described below . the magnets in the array are arranged such that adjacent ones have magnetization pointing in opposite directions , along @xmath69 . to obtain a magnetic field that closely resembles the one due to this array , we model each magnet as a 3d cubic lattice of identical dipoles , each with a moment @xmath70 . changing the sign of @xmath71 across adjacent magnets accounts for the alternating direction of magnetization . the magnetic field at any location @xmath60 above the array is then approximated using the linear superposition of the field contribution from all of the dipoles modelling the array . hence , we refer to this model as the `` dipole summation . '' since the strength of the dipole @xmath71 can not be measured experimentally , a single scaling parameter is calculated from a least - squares fit with the experimental measurements , taken at two heights . the rescaled dipole summation magnetic field is shown in figure [ fig : mag_prof ] ( a ) ( red lines ) , along with the experimental measurements of @xmath63 ( black symbols ) , corresponding to the line @xmath72 at heights @xmath73 and @xmath74 . figure [ fig : mag_prof ] ( b ) shows the magnetic field comparison at @xmath74 along the magnet centrelines . note that the electrolyte layer in the experiment is bounded by the planes @xmath75 and @xmath76 . hence , we compute the magnetic field @xmath77 using the dipole summation at various heights , in steps of @xmath78 , in the region @xmath79 and depth - average it using a discrete version of the expression ( [ eq : depth_averaging ] ) . in the experimental kolmogorov - like flow , vertical solid walls serve as the lateral boundaries , resulting in a no - slip boundary condition . however , for reasons of analytical and computational feasibility , kolmogorov flow has been studied almost exclusively using unbounded or periodic domains . neither an infinite lateral extent nor periodicity offer a realistic representation of the effect of boundary conditions in the experiment , as far as the flow s structure and its stability are concerned . to explore the role of boundaries , we compare the experiment to numerical simulations using computational domains with increasing degrees of confinement . the three different computational domains we study are described below . * doubly - periodic domain : this computational domain is chosen to coincide with the central @xmath80 region of the experimental domain ( @xmath81 and @xmath82 in nondimensional units ) . the simulated flow is constrained to be periodic in both the longitudinal and transverse directions , i.e. , @xmath83 and @xmath84 . along the transverse direction it spans a width equaling that of @xmath85 magnets . the 2d forcing profile @xmath86 over this doubly - periodic domain is constructed from the depth - averaged magnetic field presented in [ sec : dipole ] by retaining only the two dominant fourier modes , @xmath87 and @xmath88 , along the @xmath1-direction , where @xmath89 in dimensionless units . along the @xmath0-direction the profile is uniform : @xmath90 . * singly - periodic domain : this computational domain coincides with the region @xmath91 and @xmath82 . the longitudinal dimension is the same as that of the experiment , while the transverse one spans a width equaling that of 8 magnets , like in the doubly - periodic domain . no - slip boundary conditions are imposed at the end walls , i.e. , @xmath92 , while periodic boundary conditions are imposed along the transverse direction , i.e. , @xmath93 . the 2d forcing profile @xmath94 over this singly - periodic domain is constructed as a product of two one - dimensional profiles @xmath95 . along the @xmath1-direction the profile is once again constructed by retaining only two dominant fourier modes of the depth - averaged magnetic field , @xmath96 . along the @xmath0-direction the profile @xmath97 is chosen to be the depth - averaged magnetic field profile from the dipole summation along the magnet centreline @xmath98 . we note that the effect of transverse confinement has been studied by @xcite , and therefore is not investigated here separately . * non - periodic domain : this computational domain is identical to the experimental one in both lateral dimensions , i.e. , @xmath91 and @xmath99 , with no - slip boundary conditions imposed at both the end walls and side walls , i.e. , @xmath92 and @xmath100 . as mentioned in [ sec : dipole ] , the forcing over this domain is computed by depth - averaging the dipole summation . to compare the experimental observations with those predicted by equation ( [ eq:2dns_mod ] ) with the three types of boundary conditions described above , we have performed direct numerical simulations . the flow over the doubly - periodic domain is simulated using a pseudo - spectral method in the vorticity - stream function formulation , as described in @xcite . this simulation is henceforth referred to as the `` doubly - periodic simulation , '' abbreviated dps . for the singly - periodic and the non - periodic domains , numerical simulations have been performed using a finite - difference scheme , described in @xcite . these simulations are hereafter referred to as the `` singly - periodic simulation '' ( sps ) and the `` non - periodic simulation '' ( nps ) . details of the spatiotemporal discretization and the integration schemes employed in all the numerical simulations can be found in appendix [ sec : num_meth ] . in this section we present the results of our comparison between the experiment and the numerical simulations on the three domains described above . first , we discuss the straight uniform flow found at lower reynolds numbers , with a special emphasis on the effect of boundaries . we then perform linear stability analysis to demonstrate that equation ( [ eq:2dns_mod ] ) describes the primary instability more accurately than equation ( [ eq:2dns_wf ] ) on an unbounded domain , but still substantially underpredicts the experimentally observed critical reynolds number . next , we describe and compare the steady flow states found in the experiments and simulations above the primary instability . finally , we discuss the secondary instability which gives rise to a time - periodic flow . + + for weak driving , the flow mimics the forcing closely , with spatially alternating bands of fluid flow along the @xmath101-directions , as can be seen in figure [ fig : str_flow ] for @xmath102 . in this figure , black vectors represent the velocity field @xmath103 and the colour indicates the vorticity @xmath104 . for the experiment ( figure [ fig : str_flow ] ( d ) ) , the @xmath1-component of the velocity measured in the central region of the domain is close to zero . however , there are regions of strong recirculation near the end walls , characterized by a nonzero @xmath1-component of velocity . a closer inspection of the flow shows a slight tilt in the alignment of the flow bands . this tilt is due to the global circulation , resulting from confinement and the fluid flowing in opposite directions over the end magnets at @xmath105 . figures [ fig : str_flow ] ( a ) and ( b ) show the straight flows found in the dps and sps . it can be seen that flow fields in the dps and sps reproduce the experimental flow qualitatively away from the lateral walls . futhermore , the sps captures the turnaround flow near the end walls . however , neither the sps nor the dps displays the tilt of the flow bands observed in the experiment since the periodic flows are devoid of global circulation . in contrast , the nps generates a flow field that looks indistinguishable from the experimental one ( cf . figure [ fig : str_flow ] ( c ) ) . for a quantitative description of the straight flow profile , we have plotted in figure [ fig : vel_prof ] ( a ) the longitudinal component @xmath106 of the velocity along the line @xmath72 in the experiment . the location of this cross section is indicated by the vertical dashed line in figure [ fig : str_flow ] ( d ) . the difference in @xmath107 between the experiment and the numerical simulations along this line is shown in figure [ fig : vel_prof ] ( b ) . as can be seen , the dps and sps , which are only defined for @xmath108 , show systematic deviation from the experiment as high as 18% since they do not capture global circulation . in comparison , the nps agrees to within about 5% over the same region , with no clear systematic deviation . the disagreement between the experiment and nps in this region , we believe , is a result of the dipole summation not accounting for the variation in the strength of each individual magnet . closer to the boundaries , at @xmath109 and @xmath110 , the largest difference between the nps and the experiment is around 12% . the origin for this error is quite subtle and we shall defer its analysis to appendix [ sec : sensitivity ] . + + + the experimental longitudinal velocity component @xmath106 at @xmath111 ( along a central magnet centreline ) is shown in figure [ fig : vel_prof ] ( c ) . the very slight asymmetry in the longitudinal velocity is a result of the global circulation . in contrast , the flow in the dps is perfectly uniform and thus does not capture this asymmetry , as can be seen from the plot of its difference with the experimental profile in figure [ fig : vel_prof ] ( d ) . the sps , which is defined all the way to the end walls , also does not capture this asymmetry due to the lack of global circulation . the nps produces the closest agreement : the corresponding flow displays the asymmetry observed in the experiment , with no significant systematic deviation . in summary , the nps succeeds in capturing the effects of confinement in the experiment with good accuracy , while the dps and sps show significant systematic deviations . as the strength of the forcing increases , the flow in the experiment undergoes a qualitative change at @xmath112 , with uniform flow bands ( cf . figure [ fig : vel_prof ] ( c ) ) developing modulation that eventually gives rise to distinct stationary vortices . hence , we shall refer to this flow as the modulated flow . " several previous experimental studies have reported this transition and have characterized it using the critical reynolds number ( @xmath113 ) and wavenumber ( @xmath114 ) of the modulation @xcite . in our experiments , the wavenumber just above this transition was measured to be @xmath115 , where @xmath8 is the wavenumber associated with the forcing . in virtually all previous studies , theoretical estimates for these critical parameters have been obtained by using equation ( [ eq:2dns_wf ] ) and modelling the straight flow in experiment as a strict sinusoid @xmath116 ; the flow stability is then analyzed with respect to perturbations @xmath117 in the transverse component of the velocity . in this section , we revisit this analytical approach for equation ( [ eq:2dns_mod ] ) to provide estimates for the critical parameters . many previous studies used a different nondimensionalization , which corresponds to setting the nondimensional forcing wavenumber @xmath8 to unity . to make comparison easier , we will introduce a scaled wavenumber @xmath118 which corresponds to the convention used in those studies . the strictly sinusoidal straight flow governed by equation ( [ eq:2dns_mod ] ) on an unbounded domain becomes unstable with respect to perturbations with wavenumber @xmath119 above the reynolds number @xmath120 , which to a very good accuracy is given by : @xmath121 this expression was computed by linearizing equation ( [ eq:2dns_mod ] ) around @xmath122 and calculating its stability with respect to perturbations including three dominant modes @xcite , @xmath123 the critical reynolds number @xmath124 and the corresponding critical wavenumber @xmath125 computed using the expression ( [ eq : re_0 ] ) can then be compared with experimental observations . neutral stability curves ( [ eq : re_0 ] ) describing the primary instability . the red dashed line corresponds to @xmath46 s@xmath47 and @xmath126 , while the blue dot - dashed line corresponds to @xmath46 s@xmath47 and @xmath48 . the experimental measurement is shown as a black dot ; note that the uncertainty in @xmath113 is smaller than the size of the black circle . in all the cases , @xmath49 m@xmath39/s is held constant . , width=288 ] the neutral stability curve ( blue dot - dashed line ) which corresponds to the experimental values of parameters @xmath13 , @xmath17 , and @xmath10 is shown in figure [ fig : nsc ] . the minimum of this neutral stability curve yields a critical reynolds number @xmath127 and an associated critical wavenumber @xmath128 . the black dot on the plot indicates the critical values @xmath129 and @xmath130 , corresponding to the instability we observe in the experiment . the relative difference @xmath131 between the theoretical estimate for the critical reynolds number and that measured in experiment is about 17% . the critical wavenumber , however , is in better agreement with the experimentally measured one , with a 7% relative error . while the critical reynolds number obtained from the linear stability analysis clearly disagrees with the experimentally observed one , it is still a significant improvement over analytical estimates for a flow modelled using equation ( [ eq:2dns_wf ] ) , which corresponds to setting @xmath132 in equation ( [ eq:2dns_mod ] ) . the corresponding neutral stability curve is indicated by the red dashed line in figure [ fig : nsc ] . from equation ( [ eq : re_0 ] ) it can be seen that the entire neutral stability curve scales as @xmath133 . this implies that the critical wavenumber ( @xmath128 ) is independent of @xmath17 , while the predicted critical reynolds number for @xmath132 is @xmath134 . this is a 31% discrepancy with the experimental value , which is comparable to the 30% discrepancy reported by @xcite in a study based on equation ( [ eq:2dns_wf ] ) . as we discussed previously , the parameter @xmath17 describes the effect of the vertical variation in the magnitude of the horizontal velocity on the effective inertia and nonlinearity of the flow . equation does not account for this effect , so it is natural that its predictions are substantially less accurate . figure [ fig : mod_state ] ( a - d ) shows the modulated flow fields corresponding to the dps , sps , nps , and experiment , respectively , at @xmath135 . at this reynolds number , the modulated flow is well developed and is visually quite distinct from the straight flow . it should be noted that the size of the dps domain along @xmath0 was chosen _ a posteriori _ to be commensurate with the critical wavenumber @xmath136 . the counterclockwise global circulation in the experiment strongly affects the alignment of the vortices ( see figure [ fig : mod_state ] ( d ) ) as can be seen by comparing the modulated flows in the dps and sps with the relevant regions of the experimental flow . unlike the dps and sps , the flow field in the nps captures the features observed in the experiment remarkably well . this unambiguously demonstrates the importance of properly modelling the confinement effects in both the longitudinal and the transverse direction to reproduce the features of the flow in the experiment . + + the onset of the modulated flow is characterized by the appearance of the transverse component @xmath137 of the velocity throughout the flow domain . as the driving is increased , the magnitude of @xmath137 also increases . a bifurcation diagram characterizing the transition from the straight to the modulated flow is shown in figure [ fig : bif1 ] ( a ) . we use the spatial mean square transverse velocity , @xmath138 , as the order parameter and plot it as a function of @xmath139 . the spatial average is computed over the central region @xmath81 and @xmath108 for all simulations and experiment . in comparison to the experimental value of @xmath129 , the primary instability in the dps and sps occurs at much lower reynolds numbers @xmath140 and @xmath141 , respectively . in contrast , by imposing the correct ( no - slip ) boundary conditions in both the longitudinal and transverse directions , in addition to using a realistic model of the magnetic field , the transition can be predicted quite accurately . the laminar to modulated transition in the nps occurs at @xmath142 , which is within 5.2% of @xmath113 . finally , we note that setting @xmath143 results in a poor prediction @xmath144 even in the nps , which corresponds to a 21% error . given that the pattern of vortices in the experimental flow lacks perfect periodicity , to facilitate a quantitative comparison between the flow patterns observed in the experiment and the simulations , we use the average longitudinal wavelength @xmath145 as a measure . this wavelength is defined as a spatial average of the separation between adjacent vortex centres in the central region @xmath108 . just above onset , the vortices in the experiment form a lattice with a fairly uniform separation , @xmath146 . as the forcing is increased , the mean separation between the vortices increases , as can be seen from the plot of @xmath145 versus @xmath147 shown in figure [ fig : bif1 ] ( b ) . additionally , the vortex lattice becomes spatially irregular , as can be seen in figure [ fig : mod_state ] ( d ) . this spatial variation is quantified in the plot in figure [ fig : bif1 ] ( b ) wherein the uncertainty bars indicate one standard deviation in the spatial variation of the separation between adjacent vortices . note that immediately above onset , accurate identification of the vortex centres in the experiment is not possible because of the very weak modulation ; hence , experimental measurements are only plotted for @xmath148 . for comparison , figure [ fig : bif1 ] ( b ) also shows the average wavelength of the flow pattern in the dps , sps , and nps . finer spatial resolution in the simulation , compared to that in experiment , facilitates measuring @xmath145 closer to onset with greater accuracy . the dps does not capture the spatial variation of the wavelength or its variation with @xmath149 observed in the experiment . the sps , however , shows a qualitatively similar trend for the dependence of @xmath145 on @xmath149 . the periodicity in the transverse direction results in a uniform vortex pattern with smaller spatial variation in the separation between vortices compared to the experiment . in contrast , the nps captures both the spatial variation of the wavelength and the distortion of the lattice with increasing forcing quite satisfactorily . at @xmath150 , the discrepancy is much smaller than the uncertainty bars , but for @xmath151 , the nps overestimates the wavelength compared to what is observed in the experiment . the largest discrepancy , which is 0.46 ( a 10% relative error ) , occurs around @xmath152 . the slight difference in the flow patterns in the nps and the experiment is partially due to the deviation of the latter from being perfectly q2d . the analysis in appendix [ sec : sensitivity ] further shows that the wavelength of the pattern also sensitively depends on relatively minor changes in the forcing profile , which could also be responsible for the observed discrepancy between the numerics and experiment . while the nps provides a reasonably accurate description of the laminar to modulated transition in the experiment , as we mentioned previously , it somewhat underestimates the critical reynolds number . to resolve this discrepancy , we tested the sensitivity of the transition in the nps to changes in the forcing profile as well as variations in parameters @xmath17 , @xmath10 , and @xmath13 . in particular , we found that @xmath153 is fairly robust to spatial variation in the strength of the magnets in the array . consequently , we turned our attention to studying the sensitivity of @xmath153 to the values of the parameters @xmath17 , @xmath10 , and @xmath13 , which we discuss below . in order to match @xmath153 in the nps and experiment , we had to either decrease @xmath17 by 6% , increase @xmath10 by 7% , or increase @xmath13 by 22% . this suggests that @xmath153 is more sensitive to @xmath17 and @xmath10 than @xmath13 . figures [ fig : bif1_vary_params ] ( a ) and ( b ) show that the variation of parameters has a fairly weak effect on both the amplitude of the modulation and the wavelength of the pattern , suggesting that the disagreement between the simulation and experiment is primarily due to the deviation of the flow and/or forcing from two - dimensionality . note that for consistency , in both experiment and numerics , we use the same ( depth - averaged ) value of @xmath10 in the definition of the reynolds number , regardless of the actual value of @xmath10 used in the simulation . if experimental and numerical values of @xmath139 match according to this definition , they will also match if any other value of @xmath10 is used to define @xmath139 in _ both _ experiment and numerics . while one can also choose to adjust @xmath17 , @xmath10 , and @xmath13 simultaneously to better match the results of simulation and experiment , it is more instructive to study relation between these parameters and different physical mechanisms in the experiment they model . to begin with , @xmath17 controls the strength of the non - dissipative advection term and models the effect of the inherent three - dimensionality in suppressing the nonlinearity . a simple scaling argument ( cf . appendix [ sec : nondim ] ) shows that one can match the critical reynolds number associated with an instability by modifying @xmath17 , regardless of the values of @xmath13 and @xmath10 . in contrast , @xmath10 and @xmath13 control the dissipation primarily in the top and bottom layer , respectively . consequently , one should modify the depth - averaged parameters computed using the straight flow when the flow transitions to being modulated . the changes to the parameters can be understood , to some extent , by considering the changes in the vertical profile @xmath154 of the q2d velocity field @xmath155 near the straight to modulated flow transition . following @xcite , we approximate @xmath156.\end{aligned}\ ] ] for a sufficiently large ratio between the viscosity of electrolyte and dielectric layers ( @xmath157 ) , as is the case in our experiment , @xmath158 in the electrolyte layer and @xmath159 in the dielectric layer near the center of the domain . here , @xmath160 is the average magnitude of the local wavenumber @xmath161 of the flow pattern , and @xmath162 is the thickness of each fluid layer . for the straight flow @xmath163 , so @xmath164 . as @xmath139 increases above @xmath153 , @xmath160 should increase as the modulation of the flow increases . the corresponding changes in the parameters @xmath17 , @xmath165 , and @xmath13 can then be computed directly from their definitions : @xmath166 which yields @xmath167 , @xmath168 , and @xmath169 for @xmath170 which corresponds to the present experimental setup . since @xmath171 , this calculation predicts that , as @xmath139 increases beyond @xmath153 , @xmath17 should indeed decrease and @xmath10 should increase , compared with their depth - averaged values computed for the straight flow . the change in @xmath13 is more subtle , however . the prediction based on depth - averaging assumes that the flow is strictly horizontal , so viscosity is the only transport mechanism that can carry momentum across the dielectric layer . in fact , the vertical velocity is not vanishingly small and when the viscosity of the dielectric layer becomes small , advection due to the vertical component of the flow becomes a dominant transport mechanism . this leads to an effective increase in @xmath13 that can not be properly described using the expression which is based on the approximation . as we increase the forcing further , the modulated flow in the experiment becomes unstable at @xmath172 giving way to a time - periodic flow with a period @xmath173 ( @xmath174 s in dimensional units ) at onset . the modulated state in the nps , with the depth - averaged as well as the adjusted parameters , undergoes a hopf bifurcation as the forcing is increased . table [ table : hopf ] compares @xmath175 and @xmath176 in the experiment with those from the nps for the different parameter sets . [ cols="<,^,^,^",options="header " , ] for the depth - averaged values of parameters , @xmath175 and @xmath176 in the simulation are within @xmath177% of the experimental measurements . while separately modifying @xmath10 , @xmath13 , and @xmath17 shows some improvement in matching either @xmath175 or @xmath176 , the above results demonstrate the necessity of modifying _ all _ the model parameters simultaneously to match _ both _ @xmath175 and @xmath176 . this , as discussed in the previous section , is a consequence of the dependence of all parameters on the average wavelength of the pattern which changes between the primary and secondary instabilities . hence , a different set of parameters is required to describe the two instabilities and , more generally , there is no universal set of parameters @xmath17 , @xmath10 , and @xmath13 that correctly describes the experimental flow at all @xmath139 . in this article , we have presented a combined experimental and numerical study of bifurcations in a q2d kolmogorov - like flow . this flow is realized in the laboratory by electromagnetically driving a stratified layer of electrolyte above an immiscible layer of dielectric . this q2d flow is described using a 2d model derived from first principles by depth - averaging the 3d navier - stokes equation . in contrast , virtually all previous studies have modelled q2d flows using equation ( [ eq:2dns_wf ] ) , a semi - empirical variation of the 2d navier - stokes equation with the addition of a linear friction . also unlike previous studies of kolmogorov - like flows which have assumed a perfectly sinusoidal forcing profile , we have introduced a realistic model of the forcing which has been validated against 3d experimental measurements . to test the importance of lateral confinement , we have compared experimental measurements with numerical simulations using different boundary conditions . we have found that by incorporating realistic , no - slip boundary conditions at all lateral boundaries and a realistic forcing profile , quantitative agreement between the experiment and simulation can be achieved with no adjustable parameters . in particular , the reynolds number @xmath153 for the primary instability can be predicted to within about 5% and the critical wavenumber @xmath178 can be predicted to an accuracy higher than the measurement accuracy . these are significant improvements compared with previous studies , none of which were able to predict both @xmath153 and @xmath178 with this level of accuracy , _ despite using adjustable parameters_. furthermore , we have demonstrated that the model reasonably accurately predicts the modulated flow pattern ( both the wavenumber and the amplitude ) beyond the onset of the primary instability . moreover , this is the first study , experimental or theoretical , to provide a quantitative analysis of the secondary instability of a kolmogorov - like flow which generates a time - dependent pattern of vortices . even for the secondary instability the numerical predictions of the critical reynolds number @xmath175 and the critical period @xmath176 are in general agreement with the experiment , although the accuracy of the numerical predictions decreases with increasing @xmath139 . the discrepancy between the numerical predictions and experiments has been traced back to two factors : ( 1 ) the dependence of the parameters of the 2d model on the average wavenumber of the flow pattern , which in turn depends on the reynolds number , and ( 2 ) the variation of the forcing profile with height . both of these issues point to the limitations of the 2d model of what in reality is a 3d flow , albeit with a strongly suppressed vertical component of the velocity . nonetheless , for a select range of @xmath139 the experimental flow can be reproduced with extremely good quantitative accuracy by making fairly small adjustments to the model parameters , compared with their depth - average values computed for a simple straight flow . the ability of the model to closely reproduce the experimental flow is crucial for the utility of q2d flows for testing the geometrical description of weakly turbulent flows and studying the dynamical role of exact coherent structures . such tests will be the main focus of a follow up study . we thank rich kerswell for useful discussions and daniel borrero for his useful suggestions and insight . j. t. is grateful to samuel raben for his help with the prana piv software package . this work was supported in part by the national science foundation under grants no . cmmi-1234436 and dms-1125302 . the governing equation ( [ eq:2dns_mod ] ) was presented in dimensional form to highlight the dependence of parameters @xmath13 , @xmath17 , and @xmath10 on the properties of the fluid layers . the dimensional form also makes it easier to explore the sensitivity of the dynamics to changes in these parameters . to simplify comparison of our results with other studies it is helpful to nondimensionalize this equation . choosing the width of a magnet @xmath54 as the length scale , the rms velocity computed over the central region @xmath179 , @xmath180 as the velocity scale @xmath55 , and the ratio of these two scales as the time scale , one obtains the following nondimensional equation : @xmath181 where @xmath182 describes the relative strength of the rayleigh friction and viscous terms in ( [ eq:2dns_mod ] ) . finally , @xmath183 is the nondimensional pressure and @xmath184 is the nondimensional forcing profile . it is possible to eliminate the parameter @xmath17 from equation ( [ eq : q2dns_ndim ] ) by making the length , time , and velocity scales independent . if we again choose the width of a magnet @xmath54 as the length scale , the rms velocity as the velocity scale @xmath55 , and @xmath185 as the time scale , we instead obtain the following nondimensional equation : @xmath186 where @xmath187 , @xmath188 is the nondimensional pressure and @xmath189 is the nondimensional forcing profile . although equation does not contain @xmath17 explicitly , the reynolds number is rescaled by @xmath17 . hence , the nondimensional equations and as well as the dimensional equations and predict an identical sequence of bifurcations . however , the critical values of @xmath139 scale as @xmath133 , as we have found explicitly in equation . in this appendix , we present the details of discretization methods and numerical integration schemes employed in the nps , sps , and dps . since the nps , as well as the sps , require prescribing no - slip ( e.g. , dirichlet ) boundary conditions on the velocity field @xmath103 , numerical simulations are performed using the primitive variable ( @xmath107 , @xmath190 , and @xmath5 ) formulation by employing a semi - implicit fractional - step method detailed in @xcite . to begin with , equation ( [ eq:2dns_mod ] ) is discretized in time to obtain the following difference equation : @xmath191 in the above equation @xmath192 and @xmath193 are the velocity and kinematic pressure fields with the subscript @xmath194 indicating a discrete time instant @xmath195 , where @xmath196 is the time step for the update . for purposes of brevity we have used the notation @xmath197 and @xmath198 to represent the nonlinear and linear terms , respectively . the above equation results from a semi - implicit approximation for the temporal evolution described by ( [ eq:2dns_mod ] ) , where the linear terms in the update are treated implicitly using the crank - nicholson scheme , while the nonlinear term is handled explicitly using the adams - bashforth scheme . the velocity field @xmath199 at every instant satisfies the incompressibility condition : @xmath200 which is enforced on each update through the three - fractional - step p2 ( pressure correction ) projection method discussed in @xcite . spatial discretization of the velocity and pressure fields is carried out using the standard marker and cell ( mac ) staggered grid @xcite . the spatial derivatives in equation ( [ eq:2dns_mod_discrete ] ) are approximated using finite central differences ; the 2d laplacian operator ( @xmath201 ) uses a five - point stencil formula and the nonlinear term uses the three - point stencil formula . for both the nps and the sps , we have chosen 20 cells per magnet width @xmath54 to discretize the velocity and pressure fields . since the dimensions of the nps are identical to the lateral dimensions of the experiment , i.e. , @xmath202 , a total of @xmath203 cells were used to sample the flow domain . the sps , however , corresponds to a domain of dimensions @xmath204 , which maps to a region including the central eight magnets in the experiment . hence , a total of @xmath205 cells were used to discretize the sps domain . for both the sps and nps , a time step of @xmath206 s was used for all the numerical simulations . simulations on the doubly - periodic domain can be sped up significantly using a spectral method @xcite . since solving linear equations involving the laplacian is very cheap in the spectral method , it is convenient to use the vorticity - stream function formulation instead of the velocity - pressure formulation . taking the curl of equation ( [ eq:2dns_mod ] ) , we obtain the following equation for the @xmath11-component of vorticity @xmath207 : @xmath208 where @xmath209 . the horizontal components of the velocity field @xmath210 and @xmath211 can be computed using the stream function @xmath212 , which satisfies the poisson equation @xmath213 . the vorticity field @xmath214 is discretized in the fourier space using @xmath215 modes along each of the @xmath0- and @xmath1-directions . since the lateral dimensions of the periodic domain are @xmath80 units , the spatial resolution associated with the fourier grid corresponds to 16 grid points per magnet width @xmath54 . taking the fourier transform of equation ( [ eq:2dvor_modified ] ) , we obtain : @xmath216 + \nu \nabla^2 \omega - \alpha\omega + { \mathcal f}[w],\ ] ] where @xmath217 $ ] represents the fourier transform and @xmath218 $ ] . equation ( [ eq:2dvor_spectral ] ) is stepped forward in time ( @xmath219 ) using a 3-substep semi - implicit strang - marchuk splitting algorithm @xcite where the first and last substeps advance the vorticity field using the nonlinear term by means of a second - order explicit runge - kutta scheme ( using a time step @xmath220 ) , while the intermediate substep advances the vorticity field using the crank - nicholson scheme ( using a time step @xmath196 ) . we have used the time step @xmath221 s. in [ sec : mod_flow_sec ] we compared measurements of the average longitudinal wavelength @xmath145 of the modulated flow in the experiment and numerical simulations . the comparison between the experiment and the nps ( cf . figures [ fig : bif1 ] and [ fig : bif1_vary_params ] ) showed systematic differences in @xmath145 for both the depth - averaged and modified parameters , with the maximum difference being about 10% at @xmath222 ( cf . figures [ fig : bif1 ] ) . here we show that this deviation is likely due to the inherent three - dimensionality of the experiment , not captured by a strictly 2d model , rather than the choice of the model parameters . as we discussed in [ sec : dipole ] , the lorentz force density due to the specific arrangement of magnets employed in the experiment is to a very good approximation given by @xmath223 , where @xmath38 is the magnitude of current density and @xmath77 is the @xmath11-component of the magnetic field at any given location within the electrolyte . in deriving equation ( [ eq:2dns_mod ] ) it was assumed @xcite that @xmath63 can be decomposed as the product of a 2d horizontal profile @xmath224 , which depends exclusively on the extended coordinates @xmath225 , and a 1d vertical profile @xmath226 , which captures the variation of the magnetic field above the magnet array , i.e. , @xmath227 . this implies that , when normalized , the planar magnetic field profiles at various heights @xmath11 within the electrolyte are identical . such a magnetic field , which we call q2d , " facilitates the decomposition of the plane - parallel q2d velocity field which underpins the strictly 2d model ( [ eq:2dns_mod ] ) . a magnetic field that is truly q2d , however , can not be created using a magnet array with finite dimensions , i.e. , the _ shape _ of the magnetic field profile generated by permanent magnets in the laboratory always changes with the vertical height @xmath11 to some extent . experimental measurements of magnetic field from previous studies have shown such changes in the shape of the field profile as a function of @xmath11 @xcite . this is very much the case in our experiment as well , as can be seen from the magnetic field profiles shown in figure [ fig : mag_prof ] ( a ) . for instance , if one rescales the transverse magnetic field profiles at heights @xmath228 and @xmath229 such that they match near the centre of the array , we see that these profiles would _ not _ match near the end magnets . this is most apparent by comparing the relative heights of the peaks at @xmath230 and @xmath231 for the two profiles in figure [ fig : mag_prof ] ( a ) . a quantitative estimate of the deviation from quasi - two - dimensionality can be obtained by comparing the magnetic field profiles computed using the dipole summation at the bottom @xmath232 and the top of the electrolyte layer @xmath233 . normalizing @xmath234 and @xmath235 separately , using their respective spatial rms values computed over the entire lateral extent of the domain , we estimate the largest difference between the profiles to be approximately 12% . this difference is fairly localised towards the ends of the magnet array and is likely the reason behind the larger discrepancy in longitundinal velocity measurements over the end magnets ( cf . figure [ fig : vel_prof ] ( b ) ) . to demonstrate the impact of the @xmath11-dependence of the forcing profile on the flow , we have recomputed the straight and modulated flow fields in the nps using @xmath236 and @xmath237 , in addition to the depth - averaged magnetic field @xmath238 . the value of @xmath13 was increased by 22% relative to the depth - averaged value to reduce the influence of the uncertainty in the model parameters on the flow pattern . this choice also yields the best agreement between the average wavelengths of the flow pattern in the simulation and experiment for @xmath239 ( see figure [ fig : bif1_vary_params ] ( b ) ) . as figure [ fig : bif_4and7 ] ( a ) shows , the forcing profile strongly affects both @xmath153 and the amplitude of the modulation of the flow for @xmath240 . it also shows that @xmath238 produces substantially better agreement with experiment that either @xmath241 or @xmath242 . similarly , we find that the forcing profile strongly influences the modulation wavelength . as figure [ fig : bif_4and7 ] ( b ) shows , for both @xmath241 or @xmath238 the average wavelength agrees reasonable well with experiment , while @xmath242 produces a very poor agreement . similar results ( not shown ) are obtained if , instead of @xmath13 , either @xmath17 or @xmath10 is modified to match @xmath153 . the above analysis shows that , although the nps with the depth - averaged magnetic field profile captures the salient features of the dynamics fairly well , the flow pattern depends fairly sensitively on the details of the forcing . hence , one should expect systematic deviations between the 2d model derived for a q2d flow and the experiment where quasi - two - dimensionality is broken by the forcing . it should be mentioned that the wavelength of the modulated flow measured by either seeding the dielectric - electrolyte interface or the top surface of the electrolyte are virtually identical . this implies that viscous coupling across the fluid layers produces a q2d flow despite the fact that forcing profile driving the flow is not perfectly q2d . an important consequence of bounding the flow in the longitudinal or transverse directions is that we restrict the set of coordinate transformations that leave the governing equation ( [ eq:2dns_mod ] ) equivariant . the symmetries of the governing equation , in turn , govern the number of and relations between the distinct modulated flow solutions . in this section we discuss the symmetries implicit in the dps , sps , and nps , as well as the relations between their solutions corresponding to the modulated flow . on an unbounded or a doubly - periodic domain , equation ( [ eq:2dns_mod ] ) is equivariant under the following symmetry operations : ( a ) continuous translations in @xmath0 : @xmath243 , where @xmath244 is an arbitrary translation along the longitudinal direction , ( b ) reflection in @xmath0 combined with a discrete shift of half a period in @xmath1 : @xmath245 , and ( c ) reflections in both @xmath0 and @xmath1 : @xmath246 . the reflection @xmath247 is equivalent to a rotation by angle @xmath248 about the @xmath11-axis , while the square of the symmetry operation @xmath249 corresponds to a discrete shift @xmath250 in the @xmath1-direction , i.e. @xmath251 . these transformations that leave the governing equation equivariant , however , need not leave the flow fields invariant . when a flow field does not share a certain symmetry of the governing equation , one can generate by applying the corresponding coordinate transformation a dynamically equivalent symmetry - related copy of the flow . for example , the straight flow in figure [ fig : str_flow ] ( a ) remains unchanged when an arbitrary translation @xmath252 $ ] is applied along the @xmath0-direction , so there is a unique straight flow solution . however , the primary instability breaks the translational symmetry , so there is a continuum of distinct modulated flow solutions related by translations in the @xmath0 direction . this instability therefore corresponds an equivariant pitchfork bifurcation ( cf . figure [ fig : pitchfork ] ( a ) ) . figure [ fig : sym_copies_dps ] shows four distinct solutions @xmath253 , @xmath254 , @xmath255 , and @xmath256 of the modulated flow . they are related by discrete translations @xmath257 with @xmath258 : @xmath259 , @xmath260 , and @xmath261 . the equivariance of the governing equation under @xmath257 with arbitary @xmath244 makes the choice of the coordinate origin @xmath72 for the modulated flow arbitrary . we fix it by requiring that @xmath262 . since both the discrete symmetries @xmath247 and @xmath249 include reflection of the flow about the line @xmath72 , the choice of the origin determines whether a particular solution remains invariant under either of these discrete symmetries . in particular , @xmath253 and @xmath255 are invariant under @xmath247 , while @xmath254 and @xmath256 are invariant under @xmath249 . in summary , the equivariant pitchfork bifurcation in the dps , by virtue of the continuous translational symmetry of the governing equation , generates an infinite number of branches for the modulated flow . only four of these branches remain invariant under discrete symmetries @xmath247 or @xmath249 . the no - slip boundary condition at @xmath263 in the sps destroys the equivariance under translation @xmath257 . the governing equation , however , still remains equivariant under each of the discrete transformations @xmath249 and @xmath247 . the modulated flow solutions of the sps respect only one of these two symmetries , so we find two pairs of distinct solutions , as figure [ fig : sym_copies_sps ] illustrates . just like in the dps , @xmath253 and @xmath255 are invariant under @xmath247 , while @xmath254 and @xmath256 are invariant under @xmath249 . unlike dps , where @xmath257 relates all the distinct solutions corresponding to the modulated flow ( cf . figure [ fig : sym_copies_dps ] ) , there is no coordinate transformation that maps @xmath253 and @xmath255 to @xmath254 and @xmath256 . however , @xmath264 and @xmath265 . on an infinite domain , all four branches of the modulated flow are created at exactly the same @xmath139 ( as in dps ) , however , on a finite domain , the pitchfork bifurcations that produce the two pairs of solutions would generally happen at different @xmath139 , which depend on @xmath266 and @xmath267 ( cf . figure [ fig : pitchfork ] ( b ) ) . in particular , spatial confinement in the @xmath1 direction should suppress the instability which gives rise to @xmath254 and @xmath256 , pushing the corresponding bifurcation to higher @xmath139 . on the other hand , @xmath253 and @xmath255 should be less sensitive to the effect of confinement , and hence should appear at a lower value of @xmath139 for finite @xmath266 . the no - slip boundary condition at @xmath268 in the nps breaks the equivariance of the problem under @xmath269 , leaving the governing equation equivariant only under @xmath247 . since @xmath269 underlies the pitchfork bifurcation that gives rise to the solutions @xmath253 and @xmath255 in the sps , when this symmetry is broken , one finds an imperfect pitchfork bifurcation , as shown in figure [ fig : pitchfork ] ( c ) , with the straight flow @xmath270 at lower @xmath139 smoothly transitioning to the modulated flow @xmath253 at higher @xmath139 without an instability taking place , in perfect agreement with the results of our numerical simulations . the @xmath255 branch and the higher-@xmath139 branch of the straight flow @xmath271 are now created in a saddle - node bifurcation at @xmath272 . the states @xmath253 and @xmath255 , both of which are symmetric with respect to @xmath247 , are shown in figure [ fig : sym_copies_nps ] . while @xmath269 is no longer an exact symmetry in nps , given the large transverse extent of the domain compared with the period of the forcing ( @xmath273 = 9 ) , near the center of the domain this approximate symmetry holds and consequently @xmath274 . however , unlike @xmath253 , which remains stable up to @xmath275 in the nps , @xmath276 is unstable over the entire range of @xmath139 where it exists ( @xmath277 ) . this explains why our numerical simulations starting from randomized initial conditions have always converged to the modulated flow @xmath253 and why @xmath255 was never found in the numerical simulations or observed in the experiment . the pitchfork bifurcation which gave rise to the branches @xmath254 and @xmath256 in the sps also does not carry over into the nps . instead both @xmath271 and @xmath255 undergo hopf bifurcations , respectively , at @xmath278 and @xmath279 . this change in the nature of the bifurcation is likely caused by transverse confinement , which has a more prominent effect on @xmath254 and @xmath256 : these states are invariant under the @xmath249 symmetry in the sps , but this symmetry is broken in the nps . while we fail to observe @xmath254 and @xmath256 in the nps , the analogues of these solutions may appear for different model parameters , forcing profile , and/or the degree of confinement . 1965 example of the generation of a secondary stationary or periodic flow when there is loss of stability of the laminar flow of a viscous incompressible fluid . _ j. appl . mech . _ * 29 * ( 3 ) , 527544 .

we investigate the axisymmetric magnetosphere of an aligned rotating magnetic dipole surrounded by an ideal force - free plasma . we concentrate on the magnetic field structure around the point of intersection of the separatrix between the open and closed field - line regions and the equatorial plane . we first study the case where this intersection point is located at the light cylinder . we find that in this case the separatrix equilibrium condition implies that all the poloidal current must return to the pulsar in the open - field region , i.e. , that there should be no finite current carried by the separatrix / equator current sheet . we then perform an asymptotic analysis of the pulsar equation near the intersection point and find a unique self - similar solution ; however , a light surface inevitably emerges right outside the light cylinder . we then perform a similar analysis for the situation where the intersection point lies somewhere inside the light cylinder , in which case a finite current flowing along the separatrix and the equator is allowed . we find a very simple behavior in this case , characterized by a 90-degree angle between the separatrix and the equator and by finite vertical field in the closed - field region . finally , we discuss the implications of our results for global numerical studies of pulsar magnetospheres .

astro-ph0305288

the structure of the pulsar magnetosphere has been an active area of research for more that 30 years , starting with the pioneering work by goldreich & julian ( 1969 ) . despite the fact that real astrophysical radio - pulsars are believed to be oblique rotating magnetic dipoles , much of the theoretical effort has been devoted to a significantly simpler case of an aligned rotating magnetic dipole ( i.e. , a `` non - pulsing pulsar '' ) , in which case the problem becomes stationary and axisymmetric . the situation becomes even more tractable if one also assumes the space around the pulsar to be filled with a plasma that , on one hand , is dense enough to shorten out the longitudinal electric fields , thus providing the basis for the ideal - magnetohydrodynamic approximation , and , on the other hand , is at the same time tenuous enough for all the non - electromagnetic forces to be negligible , thus enabling one to regard the plasma as force - free . all these simplifications have lead early on to the derivation , simultaneously by several researchers , of the main equation governing the structure of the magnetosphere , the so - called pulsar equation ( scharlemann & wagoner 1973 ; michel 1973b ; okamoto 1974 ) , which is essentially a special - relativistic generalization of the well - known force - free grad shafranov equation . this equation is a quasi - linear elliptic second - order partial differential equation with a regular singular surface , the so - called light cylinder . despite the fact that this , in general non - linear , equation was first derived thirty years ago , most of the attempts to solve it , both analytical and numerical , have been limited , until very recently , to its linear special cases and to the region inside the light cylinder ( e.g. , scharlemann & wagoner 1973 ; michel 1973a ; beskin et al . 1983 ; beskin & malyshkin 1998 ) . these self - imposed restrictions can of course be attributed to the considerations of mathematical convenience and tractability ; however , they are not very well physically motivated and thus may be too restrictive to be relevant to real astrophysical systems . at the same time , by the late 1990s , the available computing resources and numerical techniques have become sufficiently powerful for the most general nonlinear problem , involving the regions on both sides of the light cylinder , to become tractable . in their pioneering work , contopoulos et al . ( 1999 , hereafter ckf ) have obtained the first ( and apparently unique ) numerical solution of the general problem , albeit with a rather poor spatial resolution . their approach has later been used by ogura & kojima ( 2003 ) who have obtained essentially the same results , but with a higher numerical resolution . both of these groups have focussed on the global structure of the solution ; consequently , they have not paid enough attention , in our opinion , to some key but subtle issues regarding the separatrix between the open and closed field - line regions and , especially , the point of intersection of this separatrix with the equatorial plane . as a result , the separatrix has not , we believe , been treated correctly , and , in particular , the separatrix equilibrium condition ( okamoto 1974 ; lyubarskii 1990 ) has not been satisfied close to the intersection point . we thus suspect that their solution is probably not quite correct . in this paper we try to clarify how the magnetic field near the separatrix and especially around the separatrix equator intersection point should be treated , as this question appears to be extremely important for setting up the _ correct _ boundary conditions for the global problem . in [ sec - basic ] we review the basic equations describing the ideal - mhd force - free magnetosphere of an aligned rotating dipole ; we pay special attention to the so - called light cylinder regularity condition and discuss how this condition could be used to find the correct form of the poloidal - current function and , simultaneously , to obtain the solution that passes smoothly across the light cylinder . then , in [ sec - y - point ] , we investigate some general properties of the magnetic field near the putative separatrix between the open and closed field regions ; we use the separatrix equilibrium condition to show that , if the separatrix intersects the equator at the light cylinder ( as has been assumed in the numerical simulations by ckf and by ogura & kojima 2003 ) , then the poloidal current and hence the toroidal magnetic field have to vanish on the last open field line above the separatrix . we then use these findings in [ sec - analysis ] to construct a unique self - similar asymptotic solution of the pulsar equation in the vicinity of such an intersection point ; in particular , we find all the power - law exponents describing the field near this point and also the angle between the separatrix and the equator . at the end of this section we show that a light surface ( a surface where the electric field becomes equal to the magnetic field and therefore where the particle drift velocity reaches the speed of light ) has to appear right outside the light cylinder ( it starts at the intersection point and extends outward at a finite angle with respect to the equator ) . we thus conclude that the only way to get a force - free solution that would be valid at least some finite distance beyond the light cylinder , is to consider the case of a separatrix equator intersection point lying some finite distance inside the light cylinder . we investigate the magnetic field structure around such a point in [ sec - t - point ] . we find that in this case the poloidal - current function may stay finite , but its derivative with respect to the poloidal flux has to go to zero on the last open field line ; we then find the asymptotic behavior of the magnetic field around the intersection point corresponding to this case . finally , we present our conclusions in [ sec - conclusions ] , where we also discuss the implications of our results for the past and future numerical studies of the axisymmetric pulsar magnetospheres . we consider the magnetosphere of an aligned rotating magnetic dipole under the assumptions of stationarity , axial symmetry with respect to the rotation axis @xmath0 , and reflection symmetry with respect to the equator @xmath1 . we shall include the effects of special relativity but will ignore general - relativistic effects ( i.e. , work in euclidean space ) . we shall assume that the dipole is surrounded by a very tenuous , but highly conducting plasma . the first of these assumptions enables us to neglect all non - electromagnetic forces ( i.e. , gravity , inertial , and pressure forces ) , and thus to conclude that the structure of the magnetosphere is governed by the relativistic force - free equation : @xmath2 here , the electric charge density @xmath3 and the electric current density @xmath4 are related to the electric and magnetic field through the steady - state maxwell equations : @xmath5 the second of our assumptions concerning the magnetospheric plasma , i.e. , the assumption of infinite conductivity , enables us to use ideal magnetohydrodynamics ( mhd ) : @xmath6 these equations ( with the appropriate boundary conditions ) should be sufficient for determining the structure of the magnetosphere . as is well known , an axisymmetric magnetic field @xmath7 can be described in terms of two functions , the poloidal magnetic flux ( per one radian in the azimuthal direction ) @xmath8 and the poloidal current @xmath9 , as @xmath10 where @xmath11 is the azimuthal ( or toroidal ) angle . in cylindrical coordinates the magnetic field components then are : @xmath13 next , by applying faraday s law @xmath14 in a steady state , we see that @xmath15 , and hence @xmath16 ; together with the assumption of axial symmetry , this gives @xmath17 that is , the electric field is purely poloidal . since , as it follows from equation ( [ eq - force - free ] ) or ( [ eq - ideal - mhd ] ) , the electric field must be perpendicular to the magnetic field , we can write it as @xmath18 where @xmath19 is the poloidal magnetic field , and @xmath20 is the unit vector in the @xmath11-direction . by substituting this relationship into @xmath15 , we find that @xmath21 , which has the meaning of the angular velocity of magnetic field lines , is constant along the field lines , @xmath22 . this is the well - known ferraro isorotation law . in the problem we are interested in , all the field lines are tied , at least at one end , to the surface of the pulsar , which is assumed to be rotating as a solid body with some uniform angular velocity @xmath23 . we also assume that our ideal - mhd assumption is valid all the way up to the pulsar surface , i.e. , that there is no appreciable gap between this surface and the ideal - mhd region . therefore , from now on , we shall assume that all the field lines rotate with the same angular velocity was considered , e.g. , by beskin et al . ( 1983 ) and by beskin & malyshkin ( 1998 ) . ] @xmath24 from equation ( [ eq - e ] ) we find @xmath25 where @xmath26 is the cylindrical radius normalized to the radius of the light cylinder ( lc ) @xmath27 . thus , we see that @xmath28 inside the lc and @xmath29 outside the lc . now let us turn to the force - free equation ( [ eq - force - free ] ) . the toroidal component of this equation , together with equation ( [ eq - e_phi=0 ] ) , gives : @xmath30 \times [ \nabla\psi\times\nabla\phi ] = 0 \quad \rightarrow \quad \tilde{i}=\tilde{i}(\psi ) \ , , \label{eq - i = i(psi)}\ ] ] and the poloidal component ( perpendicular to @xmath31 ) of equation ( [ eq - force - free ] ) can be written as @xmath32 - { { 1+x^2}\over x}\ , { \partial\psi\over{\partial x } } = - i i'(\psi ) \ , , \label{eq - pulsar}\ ] ] where we have introduced @xmath33 and where we also incorporated @xmath34 into @xmath35 on the right - hand side ( rhs ) by defining @xmath36 then the magnetic field components can be rewritten in terms of @xmath37 as @xmath38 equation ( [ eq - pulsar ] ) is the famous _ pulsar equation _ ( scharlemann & wagoner 1973 ; michel 1973b ; okamoto 1974 ) , also known in literature as the relativistic force - free grad shafranov equation ( e.g. , beskin 1997 ) . it is an elliptic second - order partial differential equation ( pde ) for the flux function @xmath37 ; the left - hand side ( lhs ) of this equation is linear , whereas the rhs is , in general , nonlinear . one very important feature of equation ( [ eq - pulsar ] ) is that it has a regular singular surface at the light cylinder @xmath39 . both indicial ( or characteristic ) exponents for this equation are equal to zero and so a general solution of equation ( [ eq - pulsar ] ) diverges logarithmically at this surface ( e.g. , bender & orszag 1978 ) . on physical grounds , one imposes an additional condition that both the function @xmath37 and its 1st and 2nd derivatives remain finite near the lc . for any given function @xmath35 , it is indeed possible to find such non - divergent solutions separately on each side of the lc . then , as can be seen from equation ( [ eq - pulsar ] ) itself , the contribution of the first term on the lhs [ the term proportional to @xmath40 vanishes as @xmath41 , and thus such non - divergent solutions on both sides of the lc have to satisfy the well - known lc regularity condition ( e.g. , scharlemann & wagoner 1973 ; okamoto 1974 ; beskin 1997 ) : @xmath42 this condition is very important and merits a few extra words of discussion . our force - free equation with a constant field - line angular velocity @xmath21 is a 2nd - order equation with one , a priori unknown , integral of motion , @xmath35 , and one singular surface , the lc . according to general theory ( beskin 1997 ) , one then needs two boundary conditions to set up the problem properly . we believe that the correct approach here is to set the boundary conditions for the function @xmath8 at both the inner boundary ( inside the lc : on the surface of the star , the symmetry axis , the separatrix between the open and closed field - line regions , and maybe the equator ) and at whatever outer boundary ( outside the lc : the equator and infinity or the light surface ) . at the same time the poloidal current function @xmath35 should not be prescribed explicitly at any boundary of the domain ; instead , it is to be determined from the matching condition at the singular surface , i.e. , the lc ( see below ) . thus , our position differs from that of beskin ( see beskin et al . 1983 , and beskin 1997 ) who proposed prescribing @xmath35 explicitly at the surface of the pulsar . indeed , if one considers the region inside the lc only , then , upon prescribing both @xmath8 and @xmath35 on the inner boundary ( the pulsar surface , etc . ) , one should be able to obtain a solution that is regular at the lc by using the regularity condition ( [ eq - regularity - lc ] ) at the other boundary of this inner region , i.e. , at the lc . thus , the solution in this inner region is then completely determined ; as such , it is totally independent of what happens outside the lc . this does not seem physical ; indeed , one should then be able to take this solution and continue it smoothly across the lc , thus prescribing both @xmath43 and @xmath44 at the outer side of the lc . but then the problem of finding a solution in the region outside the lc becomes over - determined , as we now have two conditions at @xmath39 , in addition to any conditions at the outer boundary . for example , in the very important particular case of the domain under consideration extending all the way to radial infinity , one can not actually prescribe any specific boundary conditions at infinity because equation ( [ eq - pulsar ] ) has a regular singularity there . instead , however , one imposes a regularity condition at infinity . in spherical polar coordinates ) . we hope that this does not cause any confusion . ] ( @xmath45 ) this regularity condition can be written as @xmath46 this regularity condition has a very simple physical meaning : it is the condition of the force balance in the @xmath47-direction between the toroidal magnetic field @xmath48 and the poloidal electric field @xmath49 ( which both become much larger than the poloidal magnetic field @xmath50 as one approaches infinity ) . thus , in this case the region outside the lc does not have any boundary conditions at all ( apart from the condition @xmath51 at the equator ) , but has two regularity conditions , one at the lc , and the other at infinity . these two regularity conditions are already sufficient to uniquely determine the solution @xmath52 for a given @xmath35 . therefore , if , in addition to these two regularity conditions , one also tries to impose the function @xmath43 along the lc , the system becomes over - determined . thus we come to the conclusion that prescribing @xmath35 at the inner boundary is not appropriate . one could , however , try to use the outer boundary condition ( or the regularity condition at infinity ) as the condition that fixes @xmath35 . to do this , one first sets the inner boundary conditions , picks an initial guess for @xmath35 , and uses ( [ eq - regularity - lc ] ) to obtain a regular solution inside the lc ; one then continues this solution smoothly across the lc and solves equation ( [ eq - pulsar ] ) in the outer region as an initial - value problem ( with both @xmath8 and @xmath53 specified at @xmath39 ) ; as a result , one gets a mismatch at the outer boundary between the obtained solution and the desired outer boundary condition [ or , in the case of an infinite domain , one would presumably fail to get a convergent solution satisfying the regularity condition ( [ eq - regularity - infinity ] ) ] . then one iterates with respect to @xmath35 until this outer - boundary mismatch is zero ( or until a solution regular at infinity is achieved ) . in reality , however , such an approach may not be practical as it involves solving an initial - value problem for an elliptic equation , which is not a well - posed problem . instead , we advocate the approach adopted by ckf . in this approach one considers the regions inside the lc ( @xmath54 ) and outside the lc ( @xmath55 ) separately . first , one makes an initial guess for @xmath35 and prescribes the corresponding boundary conditions for @xmath8 at the inner boundary of the the region inside the lc and at the outer boundary of the the region outside the lc [ or , if this outer boundary is at the radial infinity , imposes the regularity condition ( [ eq - regularity - infinity ] ) there ] . then , one uses , again separately in each region , the regularity condition ( [ eq - regularity - lc ] ) in lieu of a boundary condition at the lc [ in practice , one can think of ( [ eq - regularity - lc ] ) as of a mixed - type dirichlet von neumann boundary condition relating the value of @xmath8 at the surface @xmath39 to the value of its derivative normal to this surface ] . thus one obtains two solutions , one inside and the other outside the lc , that correspond to the same function @xmath35 and are both regular at @xmath39 . however , these solutions also depend on their respective boundary or regularity conditions set at the inner / outer boundaries of the domain . hence , in general , although they are both regular at the lc , these solutions are not going to coincide at @xmath39 . the mismatch @xmath56 depends on both the chosen function @xmath35 and on the inner and outer boundary conditions . one then iterates with respect to @xmath35 in order to minimize this mismatch . we believe that for a given set of boundary conditions there should be only one choice of @xmath35 for which the mismatch @xmath57 vanishes and hence the function @xmath8 becomes continuous along the entire lc , i.e. , @xmath58 . once this special function @xmath35 is found , it then follows from ( [ eq - regularity - lc ] ) , which is satisfied separately on both sides of the lc , that @xmath53 is also continuous , and so the entire solution passes smoothly across the lc . thus , one can say that the function @xmath35 is determined by the regularity condition ( [ eq - regularity - lc ] ) applied separately on both sides of the lc , plus the condition of matching of the two solutions . another , equivalent way to put it , is to say that @xmath35 is determined by the ( non - trivial ! ) requirement that the derivative @xmath59 actually _ exists _ at the lc [ whereby @xmath35 is expressed in terms of this derivative via equation ( [ eq - regularity - lc ] ) ] . this approach has been implemented successfully in the pioneering numerical work by ckf and then subsequently by ogura & kojima ( 2003 ) . both these groups have used a relaxation procedure to arrive at @xmath35 that corresponded to a unique solution that was both continuous and smooth at the lc . the main focus of this paper is the behavior of the magnetic field in the vicinity of the point of intersection of the separatrix @xmath60 between the region of closed field lines ( region i ) and the region of open field lines ( region ii ) with the equator @xmath61 ( see fig . [ fig - geometry ] ) . we shall generally call this point the _ separatrix intersection point_. our interest in this non - trivial problem is fueled by the belief that understanding the key features of this behavior is absolutely crucial to devising the proper boundary conditions and providing verification benchmarks for any future global numerical investigations of a force - free pulsar magnetosphere . at the same time , the treatment of the magnetic field around this very special point is expected to require a certain degree of subtlety and delicacy , as noted , for example , by beskin et al . ( 1983 ) and by ogura & kojima ( 2003 ) . among the questions that need to be answered are : what is the radial position @xmath62 of the intersection point ( namely , whether this point lies at the lc , @xmath63 , of inside the lc , @xmath64 ) ? and what is the angle @xmath65 at which the separatrix approaches the equator at this point , the three a priori possibilities being @xmath66 ( in which case we ll call this point the cusp - point ) , @xmath67 ( the y - point ) , and @xmath68 ( the t - point ) ? in our analysis , we shall assume that equations ( [ eq - force - free ] ) and ( [ eq - ideal - mhd ] ) that describe our system are valid almost everywhere in the vicinity of the separatrix intersection point , i.e. , everywhere with perhaps the exception of measure - zero regions . thus , we shall allow for the presence in our system of current sheets of infinitesimal thickness , across which the magnetic field can experience a finite jump . we limit our consideration to the situation where such current sheets can be present only along the separatrix between regions i and ii for @xmath69 and along the equatorial ( @xmath61 ) separatrix between the upper and lower open - field regions for @xmath70 . thus , we have one current sheet that lies on the equator at @xmath70 and at @xmath71 splits into two symmetrical current sheets ( one in each hemisphere ) lying along the separatrix @xmath72 ( see , e.g. , okamoto 1974 ) . at the same time we assume that our equations apply everywhere else at least in some region around the intersection point , including the portion of this region that lies outside the lc ( in the case @xmath63 ) . let us first assume that the separatrix between closed and open field - line regions reaches the lc ( i.e. , @xmath63 ) , and let us consider the separatrix equilibrium condition ( e.g. , okamoto 1974 ; lyubarskii 1990 ) . indeed , whether or not the separatrix contains a current sheet ( which we assume to be infinitesimally thin ) , it must satisfy the condition of force balance , which is obtained by integrating equation ( [ eq - force - free ] ) across the separatrix : ) may not be valid inside a current sheet as other forces may also be important in the pressure balance there . however , gravity and the plasma inertia are small because the separatrix current layer is assumed to be very thin ; and the plasma pressure gradient , while not necessarily small inside the current layer , gives , when integrated across the layer , just the difference between the values of the pressure on both sides outside the separatrix , both of which are small by assumption . ] @xmath73 by using equation ( [ eq-|e| ] ) and also the fact that @xmath74 in region i , we can rewrite this condition as @xmath75\ , ( 1-x^2 ) = ( b_\phi^{ii})^2 = { 1\over{r_{lc}^4}}\ , { i^2(\psi_s)\over{x^2 } } \ , , \label{eq - sepx - equil-2}\ ] ] where @xmath76 marks the distance from the intersection point ( @xmath61 ) along the separatrix . now , as the separatrix approaches the lc ( @xmath41 , @xmath77 ) , we get @xmath78 = { { i_s^2}\over{r_{lc}^4 } } \ , , \label{eq - sepx - equil - endpt}\ ] ] where we have defined @xmath79 as the value of the poloidal current of the last open field line @xmath80 ( above the separatrix current sheet ) . since @xmath81 is constant along this last open field line , and hence is independent of @xmath76 , we immediately see that in order to satisfy equation ( [ eq - sepx - equil - endpt ] ) in the limit @xmath41 , we must have either : + _ ( i ) _ @xmath82 , i.e. , all poloidal current that flows out of the pulsar must return back to the pulsar along open magnetic field lines , with no finite poloidal current flowing along the separatrix ; of the compatibility condition by beskin et al . ( 1983 ) and beskin & malyshkin ( 1998 ) . ] simultaneously , from equation ( [ eq - sepx - equil - endpt ] ) it then follows that @xmath83 , and hence no finite toroidal current flows along the separatrix either ( we make a very natural assumption that @xmath84 is in the same direction as @xmath85 ) ; thus the separatrix in this case is actually not a current sheet ; + or , + _ ( ii ) _ a finite @xmath86 but a divergent poloidal magnetic field in the closed - field region , @xmath87 , as @xmath77 , @xmath88 . note that this situation can not be dismissed automatically ; indeed , close to the lc the magnitude of the electric field becomes very close to that of the poloidal magnetic field , with the infinitely strong electric forces balancing out the infinitely strong magnetic ones to a high degree . then , the separatrix current sheet would have to carry back to the star a finite poloidal current @xmath81 ( thus closing the poloidal current circuit ) , and also a toroidal corotation current with the surface current density @xmath89 divergent near the lc : @xmath90 ; there would also have to be a divergent surface charge density on the separatrix : @xmath91 . note that these conclusions , derived from the separatrix equilibrium condition near the lc , are valid regardless of the angle @xmath65 between the separatrix and the equator at the intersection point . in our analysis in the next section we shall assume that there are no infinitely strong fields anywhere in the system , and thus shall dismiss situation _ ( ii ) _ ( @xmath63 , @xmath92 , finite @xmath86 ) as unphysical . we shall thus concentrate our attention on situation _ ( i ) _ ( @xmath82 ) . we shall try to analyze the structure of the pulsar equation ( [ eq - pulsar ] ) and obtain the asymptotic solution of this equation in the vicinity of the intersection point ( @xmath63 , @xmath93 ) . in this section we perform an asymptotic analysis of the pulsar equation ( [ eq - pulsar ] ) in the vicinity of the separatrix intersection point under the assumption that it is located at the lc , i.e. , @xmath63 , @xmath93 . our approach is actually very similar to the analysis of the magnetic field near the endpoint of a non - relativistic reconnecting current layer , performed previously by uzdensky & kulsrud ( 1997 ) . in addition to assuming @xmath63 , we shall , in this section , make use of the following two assumptions : + 1 ) the intersection point is a finite - angle y - point , i.e. , the separatrix approaches the equator at an angle @xmath65 , @xmath67 , where we measure angles from the radial vector lying on the equator and directed _ toward _ the star , as shown in figure [ fig - geometry ] ; + 2 ) the poloidal current is fully closed in the open - field region , i.e. , @xmath94 ; this assumption is motivated by the arguments presented in the previous section . furthermore , we also assume symmetry with respect to the equator and therefore consider only the upper half - space , @xmath95 . when considering the @xmath96-dependent coefficients in equation ( [ eq - pulsar ] ) in the vicinity of the y - point , we can , to lowest order in @xmath97 , replace @xmath96 by 1 everywhere except where it appears in a combination like @xmath98 ; thus we can rewrite this equation as @xmath99 + 2 \ , { { \partial\psi}\over{\partial\xi } } = - ii'(\psi ) \ , , \label{eq - pulsar - endpt}\ ] ] where we introduced a new coordinate , @xmath100 . instead of using coordinates @xmath101 , it is actually more convenient to use polar coordinates @xmath102 defined by @xmath103 when making the transition to these coordinates , we shall make use of the following expressions : @xmath104 as well as the usual expression for two - dimensional ( 2-d ) laplacian in polar coordinates : @xmath105 in these coordinates the poloidal magnetic field components are written as follows ( we do not need these expressions now but will need them later ) : @xmath106 finally , by using expressions ( [ eq - coord - derivatives - xi])([eq-2d - laplacian ] ) , we can rewrite the pulsar equation ( [ eq - pulsar - endpt ] ) in our polar coordinates ( [ eq - polar_coords-1])([eq - polar_coords-2 ] ) as @xmath107 now we need to solve this equation separately in region i ( region of closed field lines , @xmath108 ) and in region ii ( region of open field lines , @xmath109 ) and match the two solutions together at the separatrix @xmath110 . since in this paper we are dealing only with the local structure of the magnetic field near the y - point , we can , without any loss of generality , adopt the convention of counting the poloidal magnetic flux @xmath8 from the separatrix @xmath110 ( instead of the usual convention of counting @xmath8 from the rotation axis ) . thus , we shall set @xmath111 and , correspondingly , @xmath112 in region i and @xmath113 in region ii . very close to the y - point , i.e. at distances much smaller than the light cylinder radius ( @xmath114 ) , the system lacks any natural length scale . hence , we can expect the radial dependence of @xmath8 to be a power law , which enables us to make the following _ self - similar ansatz _ : @xmath115 [ we put a minus sign in equation ( [ eq - psi_ii ] ) in order to have @xmath116 . ] then , the magnetic field components are given by : + in region i : @xmath117 and in region ii : @xmath118 notice that the condition that magnetic field does not diverge near @xmath119 imposes the restriction @xmath120 whereas for the case where the poloidal magnetic flux function in region i diverges as @xmath121 [ i.e. , case _ ( ii ) _ in [ sec - y - point ] ] , we would have @xmath122 . as for the poloidal current function @xmath35 that appears on the rhs of equation ( [ eq - pulsar - polar ] ) , it takes very different forms in regions i and ii . in region i there should be no toroidal field , so @xmath123 in the open - field region ii , @xmath35 is not zero but does approach zero in the limit @xmath124 . since there is no natural a priori magnetic field scale in the vicinity of the y - point , we shall again employ a self - similar ansatz , i.e. , assume that @xmath125 scales as a power of @xmath126 near @xmath127 : @xmath128 so that the right - hand side of equation ( [ eq - pulsar - polar ] ) becomes @xmath129 where @xmath130 upon substituting relationships ( [ eq - psi_i ] ) and ( [ eq - i - regioni ] ) ( for the closed - field region ) and ( [ eq - psi_ii ] ) and ( [ eq - rhs - regionii ] ) ( for the open - field region ) into the main equation ( [ eq - pulsar - polar ] ) , we obtain two ordinary differential equations ( odes ) for the functions @xmath131 and @xmath132 , along with a relationship between the power - law indices @xmath133 and @xmath134 . in particular , in region i we get a homogeneous linear 2nd - order ode for @xmath131 : @xmath135 which is to be supplemented by two boundary conditions : @xmath136 similarly , in region ii we get a non - linear 2nd - order ode for @xmath132 : @xmath137 together with the relationship @xmath138 that follows from the requirement that the lhs of equation ( [ eq - pulsar - polar ] ) scale as the same power of @xmath139 as the rhs , i.e. , @xmath140 . [ at this point , however , we have to remark that one can not a priori exclude the possibility @xmath141 ; in such a case the poloidal current would go to zero near the separatrix so rapidly that its contribution to the pulsar equation would become completely negligible . however , as we shall discuss later in this section , it is possible to show that no continuous solutions exist in this case . ] the boundary conditions for equation ( [ eq - pulsar - ii ] ) are @xmath142 notice that the constant @xmath143 that appears on the rhs of equation ( [ eq - pulsar - ii ] ) is in fact unimportant ; it just sets the scale of variation of the function @xmath132 and , hence , the overall scale of the magnetic field strength . thus , we can rescale @xmath143 away by incorporating it into the solution ; we do this by defining a new variable @xmath144 then , equation ( [ eq - pulsar - ii ] ) becomes @xmath145 with the same boundary conditions @xmath146 thus , we see that in our problem the magnetic field structure near the y - point is completely characterized by four finite dimensionless parameters : the separatrix angle @xmath65 and the three power - law indices @xmath147 , @xmath133 , and @xmath134 . our goal is to obtain a unique solution of our problem , that is to determine the values of these parameters and , simultaneously , determine the functions @xmath131 and @xmath132 . in fact , equation ( [ eq - beta - alpha2 ] ) already gives us one relationship between the power exponents , but we still need three more relationships to fix all four parameters . therefore , we shall now discuss the conditions that will help us obtain these three additional relationships . the second condition [ the first being equation ( [ eq - beta - alpha2 ] ) ] that links our dimensionless parameters is rather obvious ; it is the condition of force - balance across the separatrix @xmath110 and it will give us a relationship between @xmath147 and @xmath133 . we have already discussed this condition in [ sec - y - point ] [ see eqs . ( [ eq - sepx - equil-1])([eq - sepx - equil - endpt ] ) ] ; with @xmath148 , @xmath149 , and @xmath82 , this condition can be written simply as @xmath150^i = [ b_{\rm pol}^2 ( r,\theta_0)]^{ii } \ , . \label{eq - sepx - equil-3}\ ] ] since in our @xmath102-coordinates the separatrix is a radial line @xmath110 , and since we expect the poloidal magnetic field on the two sides of the separatrix to be in the same direction , condition ( [ eq - sepx - equil-3 ] ) simply means that @xmath151 . then , using equations ( [ eq - br - i])([eq - btheta - ii ] ) , we immediately obtain a second relationship between the power exponents : @xmath152 and also a relationship between the normalizations of the functions @xmath131 and @xmath132 , cast in terms of their derivatives at @xmath110 : @xmath153 the remaining two relationships between the dimensionless parameters come from the properties of equations ( [ eq - pulsar - i ] ) and ( [ eq - pulsar - ii - g ] ) themselves and can not , unfortunately , be written out as explicit algebraic equations containing these parameters . this fact , however , can prevent us neither from describing and discussing the physical and mathematical conditions on which these two additional relationships are based , nor from using these conditions to obtain the actual unique values of the parameters . first , we shall show that there is a unique one - to - one relationship between the separatrix angle @xmath65 and the power exponent @xmath154 . this relationship comes from analyzing equation ( [ eq - pulsar - i ] ) for the closed - field region i. for any given @xmath65 this homogeneous linear equation with boundary conditions ( [ eq - bc - i ] ) is an eigen - value problem for the coefficient @xmath155 . it is in fact very easy to see why this must be the case . indeed , the boundary conditions ( [ eq - bc - i ] ) do not give us any scale for @xmath156 ; if some function @xmath131 is a solution of the problem ( [ eq - pulsar - i])([eq - bc - i ] ) , then @xmath157 with an arbitrary multiplier @xmath158 will also be a valid solution . thus , the normalization of @xmath131 is arbitrary : if , for given @xmath65 and @xmath154 , a non - trivial solution of the problem ( [ eq - pulsar - i ] ) ([eq - bc - i ] ) exists , then there will be infinitely many such solutions ; in particular , there will be a solution with @xmath159 . thus , we can impose an additional boundary condition @xmath159 . then , however , we have three boundary conditions for a 2nd - order differential equation , which means that for arbitrary @xmath65 and @xmath154 the system is over - determined . hence , such a solution will exist not for all values of @xmath65 and @xmath154 ; the condition that it actually does exist gives us a certain relationship between @xmath65 and @xmath154 : for a given @xmath65 we would get an infinite discrete spectrum of the allowed values of @xmath154 . of these , however , we are interested only in the lowest one because we want a solution without direction reversals of the magnetic field , i.e. , with @xmath160 everywhere . from the practical point of view , the easiest way to determine the relationship between @xmath154 and @xmath65 is the following . we scan over the values of @xmath154 ; for each given @xmath154 we use our freedom of normalization of @xmath131 to set @xmath159 . thus we now have an initial - value problem with two conditions at @xmath161 : @xmath159 and @xmath162 . with these two conditions and with the value of @xmath163 given , we can integrate equation ( [ eq - pulsar - i ] ) forward in @xmath47 ( we do it numerically using a 2nd - order runge kutta scheme ) until the point where @xmath131 becomes zero . this point is then declared the correct value of the separatrix angle @xmath65 for the given @xmath154 . the function @xmath164 we have obtained as a result of this procedure is shown in figure [ fig - theta0 ] , whereas figure [ fig - f ] shows the behavior of the function @xmath131 for several selected values of @xmath154 . we see that @xmath164 is a monotonically decreasing function with @xmath165 rad [ instead of @xmath166 that one would get if the y - point were at a finite distance inside the lc ! ] , and @xmath167 . , corresponding to the case of zero poloidal current ( see also beskin et al . 1983 ) ; this is because in his analysis he had in fact assumed that @xmath168 and then derived the corresponding value of @xmath65 . ] the asymptotic behavior of @xmath164 in the limit @xmath169 is @xmath170 , which is in fact very easy to obtain analytically . , we expect @xmath171 , and then we can neglect the @xmath172 term in equation ( [ eq - pulsar - i ] ) . as a result , we get a simple harmonic equation @xmath173 , with @xmath162 , @xmath174 ; the solution of this equation , positive everywhere in the domain @xmath175 , is @xmath176 ; it corresponds to @xmath177 . ] actually , the exact analytical fit for our numerical curve @xmath164 appears to be @xmath178 with @xmath179 . with the dependence @xmath164 thus determined , and with @xmath134 and @xmath133 related to @xmath154 via equations ( [ eq - beta - alpha2 ] ) and ( [ eq - alpha1=alpha2 ] ) , respectively , we now have everything expressed in terms of @xmath154 . the fourth ( and final ! ) condition that will help us determine @xmath154 and hence @xmath65 and the rest of the power exponents is the light cylinder regularity condition . this condition states that at the lc @xmath180 [ which is a regular singular point for equation ( [ eq - pulsar - ii - g ] ) ] the function @xmath181 should be regular , namely , it should be a continuously differentiable function , with finite 1st and 2nd derivatives . this condition can be written as @xmath182 and it is just a particular manifestation of the general lc regularity condition ( [ eq - regularity - lc ] ) . we use this condition to fix the unique value of @xmath154 ( see the general discussion of this issue at the end of [ sec - basic ] ) . here is how we do it in practice . first , we divide up region ii into two sub - regions : region ii@xmath183 ( inside the lc : @xmath184 ) and region ii@xmath185 ( outside the lc : @xmath186 ) . then , we scan over @xmath154 ; for each value of @xmath154 , we first determine the corresponding value of @xmath65 using the procedure outlined above . once @xmath65 for a given @xmath154 is found , we solve equation ( [ eq - pulsar - ii - g ] ) separately in regions ii@xmath183 and ii@xmath185 ( again , using a numerical shooting method in conjunction with the 2nd - order runge - kutta integration scheme ) . in each of these regions we use regularity condition ( [ eq - lc - regularity - g ] ) at @xmath180 and one of the boundary conditions ( [ eq - bc - ii - g ] ) at the other end of the region , i.e. , @xmath187 in region ii@xmath183 and @xmath188 in region ii@xmath185 . note that the use of the regularity condition ( [ eq - lc - regularity - g ] ) guarantees that the solutions in _ each _ of the two regions are regular . however , in general ( i.e. , for an arbitrarily chosen @xmath154 ) , the two solutions obtained in this manner do not match each other at the lc , the mismatch @xmath189 [ and hence the mismatch in @xmath190 related to @xmath191 via eq . ( [ eq - lc - regularity - g ] ) ] being @xmath154-dependent . we find that there is only one , special value of @xmath154 for which @xmath191 vanishes and the solution continues smoothly across the lc as one passes from region ii@xmath183 into region ii@xmath185 . this special value , which we call @xmath192 , is declared the correct value of @xmath154 . numerically , we find @xmath193 and , correspondingly , @xmath194 this is the way in which the unique correct solution of the problem is obtained . the functions @xmath131 and @xmath195 corresponding to this value of @xmath154 are plotted in figures [ fig - f_0 ] and [ fig - g_0 ] , respectively . now let us make a little digression and step back to discuss the possibility of finding a solution in the case @xmath196 . recall that in this case the dominant balance of the lowest - order ( in @xmath139 ) terms in the pulsar equation does not include the contribution @xmath197 due to the poloidal current . at the same time , note that relationships ( [ eq - alpha1=alpha2 ] ) and ( [ eq - sepx - f=-g ] ) , derived from the separatrix force - balance condition , still hold . hence , the solution in region ii would have to satisfy a linear equation that is identical to equation ( [ eq - pulsar - i ] ) for @xmath131 on the other side of the separatrix . this fact , combined with the condition ( [ eq - sepx - f=-g ] ) , implies that the solution just has to pass smoothly across @xmath110 , and thus this point [ which is just an ordinary point for equation ( [ eq - pulsar - i ] ) ] does not have any special significance ; it is just the point where the solution changes sign . thus , the general solution in this case takes the form @xmath198 , where @xmath131 satisfies the homogeneous linear equation ( [ eq - pulsar - i ] ) in the entire domain @xmath199 $ ] with the homogeneous boundary conditions @xmath162 and @xmath200 . in addition , the solution has to pass smoothly across the regular singular point @xmath180 ( the light cylinder ) ; the regularity condition at this point is simply @xmath201 . such a solution would start positive with a zero derivative at @xmath161 , then decrease and change sign at some angle @xmath202 , reach a minimum at the lc [ since @xmath201 ] , and finally increase and go back to zero at the equator @xmath204 . however , the extra regularity condition at @xmath180 makes the system over - determined . indeed , counting two boundary conditions at @xmath205 , an arbitrary normalization of the solution [ e.g. , @xmath159 ] , and the lc regularity condition , we now have four conditions for our second - order equation with one free parameter @xmath154 . thus , unless we are somehow extremely lucky , the solution of this problem does not exist . one can show that this is indeed the case ; it turns out that a solution of equation ( [ eq - pulsar - i ] ) in the region inside the lc ( @xmath207 $ ] ) satisfying both the boundary condition @xmath162 and the lc regularity condition @xmath201 exists only for even values of @xmath208 , whereas a solution in region ( @xmath209 $ ] ) satisfying both the boundary condition @xmath200 and the lc regularity condition exists only for odd values of @xmath210 . this demonstrates that , in the case @xmath196 , it is impossible to obtain a solution that is continuous at the lc . this leaves us with the case @xmath211 ( for which we have just found a unique suitable solution ) as the only possibility . thus we have managed to obtain a _ unique _ asymptotic solution of the ideal - mhd , force - free system ( [ eq - force - free])([eq - ideal - mhd ] ) in the vicinity of the separatrix intersection point ( @xmath63 , @xmath93 ) . this solution , characterized by dimensionless parameters ( [ eq - alpha_0 ] ) and ( [ eq - beta&theta_0 ] ) , satisfies the separatrix equilibrium condition and is continuous and smooth at the lc . let us now discuss the physical relevance of our solution to the pulsar problem . note that the obtained unique solution of our problem has a finite value of the derivative @xmath212 at the equator @xmath213 , namely , @xmath214 , and thus has a non - zero radial magnetic field just above ( and just below ) the equatorial current sheet . this actually spells bad news for the applicability of our solution . indeed , one more condition that needs to be satisfied for the solution to be physically relevant and that we have not discussed so far is the condition @xmath215 with @xmath216 related to @xmath217 via equation ( [ eq-|e| ] ) , this condition can be expressed as @xmath218 we see that this condition is satisfied trivially everywhere inside the lc , but is not automatically satisfied outside the lc . in particular , in our solution , and in fact in any solution characterized by @xmath82 , the toroidal field on the last open field line @xmath72 is zero , whereas the poloidal magnetic ( and hence electric ) field is not zero , as our explicit solution shows . this means that the inequality ( [ eq - b > e-2 ] ) is violated immediately outside the lc as one moves along the equator . more generally , this inequality is violated beyond the so - called light surface ( defined as the surface where @xmath219 ) that emanates from the y - point . it is in fact not difficult to find the location of this light surface in our solution . indeed , according to equations ( [ eq - br - ii])([eq - btheta - ii ] ) , we have @xmath220\ , , \label{eq - bpol^2}\ ] ] and , according to equations ( [ eq - b_phi-2 ] ) , ( [ eq - i - regionii ] ) , and ( [ eq - beta - alpha2 ] ) , @xmath221 using the result ( [ eq - bpol^2 ] ) , we can rewrite this as @xmath222 upon comparing this result with @xmath223 we conclude that the light surface @xmath224 emanates from the y - point at a finite angle @xmath225 which is determined from the algebraic equation [ provided that the solution @xmath181 is known ] : @xmath226\ , \cos\theta_{ls } \ , . \label{eq - theta_ls-1}\ ] ] for our solution described by the parameters ( [ eq - alpha_0])([eq - beta&theta_0 ] ) , we find @xmath227 which corresponds to the angle @xmath228 . beyond the ls ( @xmath229 ) the force - free equation ( [ eq - force - free ] ) is not applicable . the solution we have obtained in this section is valid in the region @xmath230 , but , because the process of obtaining this solution also involved the region @xmath229 [ e.g. , via the boundary condition @xmath231 , our solution may not be the correct solution of the overall problem . in fact , we can make an even more general statement : any relevant to our problem finite - field magnetic configuration with the separatrix intersecting the equator at the lc can not be almost everywhere ( that is , again , everywhere except at a finite number of infinitesimally thin current sheets ) ideal and force - free beyond the lc ! indeed , as we saw previously , if one insists on having the separatrix intersection point at the lc , then one has to contend either with having @xmath232 divergent near @xmath39 or with having @xmath233 . the latter case , however , is characterized by zero toroidal magnetic field everywhere along the last open field line @xmath72 . that the poloidal ( radial ) magnetic field does not vanish identically near the equatorial current sheet , it then follows that @xmath234 , and so the force - free equation ( [ eq - force - free ] ) becomes inapplicable . and , as our particular solution shows , the condition @xmath235 is violated not only on some singular surfaces of measure zero , such as the equatorial current sheet ( that would be acceptable to us at this stage , since we do expect our simple force - free , ideal - mhd assumptions to break down there anyway ) , but in a volume of non - zero extent in both poloidal directions , e.g. , in the region ( @xmath236,@xmath237 ) . thus we have to conclude that if one is to keep the hope of finding a force - free solution that would be valid all the way up to infinity ( in the limit of vanishing plasma density ) , or at least to some finite distance beyond the lc , then one has to make concessions regarding the anticipated location of the separatrix / equator intersection point . in other words , the requirement that @xmath238 outside the lc means that @xmath81 must be non - zero and this , in turn , forces one to conclude that the intersection point must be located at some finite distance inside the lc , i.e. , @xmath64 . this scenario has actually been the subject of several studies in the past 20 years ( e.g. , beskin et al . 1983 ; lyubarskii 1990 ; beskin & malyshkin 1998 ) . according to equation ( [ eq - sepx - equil-2 ] ) , the corresponding finite-@xmath81 configuration has to be characterized by @xmath239 staying finite in the limit @xmath240 , and , therefore , the separatrix should approach the equator at a right angle , @xmath68 , corresponding to a t - point ( see lyubarskii 1990 ) . let us now ask what other characteristic features should such a solution possess ? here we present a few simple facts that can be gleaned immediately by inspecting the force - free equation ( [ eq - pulsar ] ) and its regularity condition ( [ eq - regularity - lc ] ) . first , since beyond @xmath71 the last open field line , @xmath72 , is assumed to lie along the equator @xmath61 , the derivative @xmath241 should be zero everywhere along this line ( for @xmath70 ) . in particular , we should have @xmath242 , but then it follows from equation ( [ eq - regularity - lc ] ) that either @xmath243 or @xmath244 has to be zero . since here we are considering the case @xmath86 , we conclude that @xmath35 has to approach @xmath80 with a zero slope , @xmath245 . thus , we see that @xmath246 can not be a simple linear function , unless it is constant everywhere . next , by substituting the result @xmath247 , together with @xmath248 , into equation ( [ eq - pulsar ] ) , we find that @xmath249 for all @xmath70 . this means that the radial magnetic field also approaches the equator @xmath61 with a zero slope , @xmath250 . now let us see what one can deduce by investigating the regularity condition ( [ eq - regularity - lc ] ) just above the equatorial current sheet . according to what we have just established , above the equator the function @xmath35 can be expanded as @xmath251 where @xmath252 in order to satisfy the condition @xmath253 , as @xmath124 ( here we again set @xmath111 to simplify notation ) . then we find , to lowest order in @xmath126 , @xmath254 now , the radial magnetic field just above the equator is , in general , not zero , so we can expand @xmath255 to lowest order in @xmath256 as @xmath257 where @xmath258 is in fact just the rescaled electric field above the equator , as can be seen from equations ( [ eq - e ] ) and ( [ eq - b_r-2 ] ) : @xmath259 upon substituting equations ( [ eq - tpoint - rhs ] ) and ( [ eq - tpoint - psi - z>0 ] ) into the regularity condition ( [ eq - regularity - lc ] ) , we get @xmath260^{\beta-1 } \ , , \label{eq - tpoint - regularity - lc - z>0}\ ] ] and , therefore , @xmath261 unless @xmath262 [ and @xmath263 if @xmath264 . we can now use this result to analyze the magnetic field in the vicinity of the point @xmath265 . equation ( [ eq - pulsar ] ) in the vicinity of this point can be written as @xmath266 + { { 1+x_0 ^ 2}\over x_0}\ , { \partial\psi\over{\partial\zeta } } = -\ , ii'(\psi ) \ , , \label{eq - tpoint - pulsar}\ ] ] where @xmath267 . now , just as we did in [ sec - analysis ] , we can make a transition to the polar coordinates : @xmath268 , @xmath269 . in these polar coordinates equation ( [ eq - tpoint - pulsar ] ) becomes @xmath270 + { { 1+x_0 ^ 2}\over x_0}\ , \biggl[{\partial\psi\over{\partial r}}\cos\theta- { \partial\psi\over{r\partial\theta}}\sin\theta \biggr ] = -\ , ii'(\psi ) \ , , \label{eq - tpoint - pulsar - polar}\ ] ] again , similarly to what we did in [ sec - analysis ] [ see eq . ( [ eq - psi_ii ] ) ] , let us assume that the magnetic flux function in region ii ( the region of open field lines , @xmath271 ) is a power law of @xmath139 near the point ( @xmath272 ) : @xmath273 then , using equations ( [ eq - tpoint - rhs ] ) and ( [ eq - tpoint - beta=2 ] ) , we can express the rhs of equation ( [ eq - tpoint - pulsar - polar ] ) as @xmath274 whereas the lhs of this equation can be expressed as @xmath275 \nonumber \\ & - & \ , { { 1+x_0 ^ 2}\over x_0}\ , r^{\alpha_2 - 1}\ , \biggl [ \alpha_2 g(\theta)\cos\theta - g'(\theta ) \sin\theta \biggr ] \ , . \label{eq - tpoint - lhs - polar}\end{aligned}\ ] ] in this expression , terms of the lowest order in @xmath139 are those proportional to @xmath276 ; at the same time we see from equation ( [ eq - tpoint - rhs - polar ] ) that the rhs of equation ( [ eq - tpoint - pulsar - polar ] ) scales with @xmath139 as @xmath277 . therefore , we see that in this case of the separatrix intersecting the equator inside the lc , @xmath64 , the contribution of the rhs is completely negligible ( as @xmath278 ) . thus , the dominant balance in equation ( [ eq - tpoint - pulsar - polar ] ) dictates that the terms of order @xmath276 should balance each other and hence we get a very simple homogeneous linear ode for the function @xmath132 : @xmath279 the boundary conditions for this equation are @xmath280 and so the solution with the lowest @xmath133 ( corresponding to the simplest magnetic field topology in this region ) obviously is @xmath281 and , correspondingly , @xmath282 thus , @xmath283 . must be equal to zero according to the lc regularity condition , whereas lyubarskii assumed that @xmath284 and hence had a finite , non - vanishing @xmath285 . ] as for region i ( the region of closed field lines ) , the magnetic field there is , to lowest order in @xmath139 , purely vertical and finite at @xmath240 , so we can write @xmath286 where @xmath287 is related to @xmath81 via the separatrix equilibrium condition ( [ eq - sepx - equil-2 ] ) : @xmath288 we express hope that these simple findings will be helpful in setting up or checking the correctness of future numerical attempts to solve this nontrivial problem . in this paper we have considered the axisymmetric force - free magnetosphere of an aligned rotating magnetic dipole , under the additional assumptions of ideal mhd and of the uniformity of the field - line angular velocity [ @xmath289 . this fundamental model problem is of great importance to any attempts to understand the workings of radio - pulsars . more specifically , we have focussed most of our attention on the structure of the magnetic field in the vicinity of the point of intersection of the separatrix ( between the closed- and open - field regions ) and the equator . we call this point the separatrix intersection point . the unique singular nature of this point makes it play an extremely important role in the overall global problem ; in particular , without a thorough understanding of the subtleties of the magnetic field behavior near this point , it is impossible to prescribe the correct global boundary conditions in any sensible way . we start , however , by discussing ( in [ sec - basic ] ) the basic equations governing the global force - free pulsar magnetosphere . we give special attention to the role played by the light cylinder regularity condition in determining the poloidal electric current . after this general discussion , the rest of the paper is devoted entirely to the analysis of the separatrix intersection point . we first consider the separatrix equilibrium condition in the vicinity of this point ( see [ sec - y - point ] ) and find that if it lies at the light cylinder , then all the poloidal current @xmath35 has to return back to the pulsar in the open - field region above the equator , i.e. , there should be no singular current running in a current sheet along the equator and the separatrix . we then perform ( in [ sec - analysis ] ) an asymptotic analysis of the relativistic grad shafranov , or pulsar , equation in the vicinity of such an intersection point located at the light cylinder . we find a unique self - similar solution that can be described by the power law @xmath290 , where @xmath139 is the distance form the intersection point and @xmath291 , and by the equator separatrix angle @xmath292 rad . however , a further analysis of this solution in the region outside the light cylinder shows that a light surface ( a surface where @xmath293 ) appears just outside the light cylinder ; in particular , we find that this light surface originates right at the intersection point and extends outward at a finite angle with respect to the equator . the appearance of a light surface in this case is , of course , not surprising , taking into account the fact that ( in this case of the separatrix intersection point lying at the light cylinder ) the poloidal electric current @xmath125 and hence the toroidal magnetic field have to become zero on the last open field line . we therefore conclude that the only possibility for an ideal - mhd force - free magnetosphere above the putative equatorial current sheet to extend at least some finite distance beyond the light cylinder , is for the separatrix intersection point to be located inside , as supposed to right at , the light cylinder . ( of course we understand that the exact location of this point can only be found as a part of a global solution of the pulsar equation and that it is impossible to determine it from our local analysis ) . these findings give us the motivation to consider the case when the intersection point lies at some finite distance inside the light cylinder . in [ sec - t - point ] we examine the behavior of the function @xmath35 near the last open field line @xmath72 ( above the equatorial current sheet ) for this case ; we find that the derivative @xmath294 has to go to zero as @xmath295 , whereas the current itself can approach a finite value @xmath86 . we then perform an asymptotic analysis of the magnetic field around the intersection point ; it is essentially a somewhat simplified and trivialized analogue of our analysis in [ sec - analysis ] . we find that the separatrix approaches the equator at a right angle , @xmath68 , and that the field in the region of open field lines behaves simply as @xmath296 , while in the closed - field region the magnetic field is simply vertical and finite near the intersection point ( we call this configuration the t - point ) . finally , we would like to make several remarks regarding the numerical simulations by ckf and by ogura & kojima ( 2003 ) : \1 ) the midplane boundary conditions ( @xmath297 , @xmath298 ; @xmath51 , @xmath299 ) adopted by both groups have automatically assumed that the separatrix intersection point lies at the lc , and thus have precluded them from even considering the possibility @xmath64 . it is interesting to note , however , that from the magnetic contour plots presented by ogura & kojima ( 2003 ) it does seem that the actual intersection point lies a little bit inside the lc . \2 ) neither ckf , nor ogura & kojima ( 2003 ) have discussed or even mentioned the separatrix equilibrium condition ; thus it is not clear whether this condition has been satisfied in their simulations . in the light of our present findings and of the fact that both of these groups have found @xmath86 , we suspect that the equilibrium condition has not , in fact , been satisfied in their studies , at least close to the separatrix intersection point . this suspicion is strengthened by the fact that both groups have reported having experienced some difficulties near the separatrix . \3 ) ogura & kojima ( 2003 ) have reported that they had found @xmath300 at some finite distance outside the lc , whereas ckf have claimed that in their solution @xmath301 everywhere . the origin of this discrepancy is not clear . it may be attributed to the difference in numerical resolution , although neither of the two groups have conducted very extensive convergence studies . i am grateful to leonid malyshkin and anatoly spitkovsky for interesting and insightful discussions . this research was supported by the national science foundation under grant no . phy99 - 07949 .

we present near ir color magnitude diagrams and physical parameters for a sample of 24 galactic globular clusters toward the bulge direction . in this paper we discuss the properties of twelve new clusters ( out of 24 ) in addition to those previously studied and published by our group . the compilation includes measurements of the cluster reddening , distance , photometric metallicity , horizontal branch red clump , red giant branch morphological ( i.e. mean ridge lines ) and evolutionary ( i.e. bump and tip ) features . the compilation is available in electronic form through the worldwideweb , and it will be updated regularly .

astro-ph0612280

in the last decades galactic globular clusters ( gcs ) have proven to be extremely important astrophysical laboratories for a wide range of problematics . indeed , the study of their stellar populations address fundamental questions ranging from stellar structure , evolution and dynamics to galaxy formation and the early epoch of the universe . containing some of the oldest stars known , they are _ fossils _ from the remote and violent epoch of galaxy formation . they also serve as test particles for studying galaxy dynamics and to test stellar dynamical models . being the largest aggregates in which all post main sequence ( ms ) stars can be individually observed , they serve as fiducial templates for understanding the integrated light from distant , unresolved stellar systems . in this respect , the bulge gcs provide ideal templates to explore the high metallicity regime and thus to study the stellar content of extra galactic bulges and ellipticals . however , the high and patchy extinction , which makes optical observations difficult to impossible along most bulge lines of sight , coupled with the limited performances of the past generation of near ir instrumentation , prevented accurate determinations of the bulge gcs basic properties ( as age , metallicity , distance , etc ) , as already existing since a long time in the case of the halo clusters . the current generation of ground based ir instrumentation with high spatial resolution and wide field coverage , and with the future availability of the james webb space telescope will allow us to resolve the brightest giants in galaxies up to several mpc away . hence , a homogeneous compilation of the bulge gcs properties to be used as empirical templates of metal rich stellar populations , is thus strongly urgent . in this framework , our group started a long term project devoted to fully characterize the stellar populations in the bulge gc system , by using color magnitude diagrams ( cmds ) and luminosity functions ( lfs ) in the near ir ( see @xcite hereafter f00 ; @xcite ; @xcite hereafter vfo04a ; @xcite hereafter vfo04b ; @xcite hereafter vof05 ; @xcite hereafter o05 ) . the collected photometric database has been used to perform a detailed description of the main morphological and evolutionary features of the red giant branch ( rgb ) sequence , by means of a set of photometric indices , which have been defined and widely described in f00 , and calibrated as a function of the cluster metallicity in vfo04a and vfo04b . in this paper we present the largest , homogeneous near ir photometric database of bulge gc ever obtained . for each clusters , the compilation includes : _ i ) _ the photometric catalog ; _ ii ) _ the rgb mean ridge line , ; _ iii ) _ accurate reddening , distance , and metallicity determinations ; _ iv ) _ the luminosity of the main rgb evolutionary features ( i.e. the bump and the tip ) ; and _ v ) _ the mean magnitude of the horizontal branch red clump ( hb rc ) . moreover , in order to be easily accessible by the community , the compilation would be also provided in electronic form . the programme cluster sample is presented in 2 , while 3 is devoted to the detailed description of the observations , data reduction , and derived cmds of a sample of 12 bulge clusters , recently observed , and presented here for the first time . 4 describes the overall characteristics of the compilation and the measurement of the hb rc and the rgb bump and tip . the cluster sample presented in this study counts 24 gcs in the bulge direction observed in the near ir by our group in the last few years . the targets selection has been performed by defining _ bulge gcs _ all those located within @xmath0 and @xmath1 , where _ l _ and _ b _ are the galactic coordinates , and giving a highest priority to the metal rich population . note that this definition is mainly a working hypothesis and refers to the position of the clusters in the bulge direction , only . however , there are a number of growing evidences , based on kinematics @xcite and high resolution chemical abundances ( * ? ? ? * and references therein ) , favoring a bulge origin for the metal rich gcs within 3 kpc from the galactic center . [ map ] shows the spatial distribution in galactic coordinates of our sample ( which counts 50% of the entire bulge gc system ) , and the cobe / dirbe 3.5@xmath2 m inner bulge outline @xcite . the clusters discussed in this paper fall into three categories : * sample a * ( ngc6342 , ngc6380 , ngc6440 , ngc6441 , ngc6528 , ngc6553 , ngc6624 ) have been already published , and a detailed description of the observations , data reduction and analysis can be found in f00 , vfo04a , and vfo04b . this sample have been used to perform a detailed analysis of the main rgb morphological and evolutionary features , leading to an empirical calibration of suitable near ir photometric indices ( i.e. rgb color , magnitude , slope , bump and tip ) as a function of the cluster metallicity . * sample b * ( ngc6304 , ngc6569 , ngc6637 , ngc6638 , and ngc6539 ) have been presented in vof05 and o05 , where a detailed description of the derived ir cmds and the analysis of the rgb properties can be found . for this sample we derived the clusters metallicity by using the calibration of vfo04a , b . * sample c * ( ngc6256 , ngc6266 , ngc6273 , ngc6293 , ngc6316 , ngc6255 , ngc6388 , ngc6401 , ngc6642 , ter3 , ter5 , ter6 ) are presented and discussed here for the first time . for this sample , a detailed description of the observations , data reduction , cmds and physical properties of each cluster is provided in the following sub sections . j , h and k@xmath3 images of the clusters in _ sample c _ were obtained during three observing runs on june 2004 , july 2005 , and june 2006 using the near ir camera sofi , mounted at the eso / ntt telescope . during the observing runs two set of data were secured : _ ( i ) standard resolution set . _ a series of images in the j , h and k@xmath3 bands have been obtained by using sofi in _ large field mode _ , characterized by a pixel size of 0.288 " and a total field of view of 4.9@xmath4 4.9. on average , the images are the combination of 42 , 72 , and 99 exposures each one 3sec long in the j , h and k@xmath3 passbands , respectively . _ ( ii ) high resolution set . _ high resolution images of the inner region of each cluster were also secured using the sofi focal elongator , yielding a pixel size of 0.146 " and a total field of view 2.49 @xmath4 2.49. high resolution images are the average of 30 single exposures 1.2 sec long . all the secured images are roughly centered on the cluster center . note that the region covered by our observations allows us to sample a significant fraction of the total cluster light ( typically @xmath580 - 95% ) in all the programme clusters . during the 8 nights of observations the average seeing was always quite good ( fwhm@xmath6 ) . every image has been background subtracted by using sky fields located several arcmin away from the cluster center , and flat field corrected using halogen lamp exposures , acquired with the standard sofi calibration setup . standard crowded field photometry , including psf modeling , was carried out on each frame by using daophotii / allstar @xcite . for each cluster , two photometric catalogs ( derived from high and standard resolution images ) , listing the instrumental j , h and k@xmath3 magnitudes , were obtained by cross - correlating the single - band catalogs . the standard and high resolution catalogs have been combined by means of a proper weighted average , weighting more the high resolution measurements in the innermost region of the cluster . in principle , this strategy allows to minimize the blending effects . the internal photometric accuracy has been estimated from the _ rms _ frame to frame scatter of multiple stars measurements . over most of the rgb extension , the internal errors are quite low ( @xmath7 mag ) , increasing up to @xmath8 mag at @xmath9 . the instrumental magnitudes were then converted into the 2mass photometric system mag in the zero point calibration in all three bands has been estimated . ] , and the star positions astrometrized onto 2mass0.2 arcsec in both r.a . and dec . ] as done for the clusters in _ samples a and b_. since for terzan 6 only high resolution data were acquired , we also used j , h and k images obtained with the irac2 camera ( mounted at the eso / mpi 2.2 m telescope ) , in order to cover a larger area ( @xmath10 4@xmath44 ) . the derived calibrated catalog has been then merged with that one obtained from the sofi observations . note that , as already discussed in @xcite the calibration of the k irac2 photometry onto the k@xmath3 sofi one requires negligible color term . figure [ cmds ] presents the derived ir cmds of the global sample of 24 gcs . as shown in the figure , the photometric catalogs span the entire rgb extension , from the tip to @xmath102 - 5 magnitudes below the hb ( depending on the cluster extinction ) . the main cluster properties such as reddening and distance have been derived by using a differential method based on the comparison of cmds and lfs of clusters with similar hb morphology . however , this method allows to derive reliable estimates of reddening and distance only for intermediate high metallicity clusters , whose cmd shows a red clumpy hb morphology . in fact , the location in magnitude of the hb lf peak can be used as a reference feature and safely compared with that one of the template cluster . in 5 metal poor clusters ( i.e. ngc6256 , ngc6273 , ngc6293 , ngc6355 and ngc6401 ) the blue hb morphology coupled with the relatively high reddening and the remarkable level of field contamination prevented a safe location of the hb level , and thus the use of the differential method . hence , in these cases the cluster reddening and distance have been obtained by using the empirical method presented by ( * ? ? ? * hereafter fvo06 ) , which allows one to simultaneously get reddening , distance and metallicity of a stellar system by using a few observables , like the rgb slope , tip , and mean ridge line , of the [ k , j k ] cmd . the derived cluster reddening and distance have been used to transform the observed cmds and rgb ridge lines into the absolute plane , and to measure the following parameters : _ i ) _ ( j k)@xmath11 and ( j h)@xmath11 colors at 4 fixed absolute magnitude levels ( @xmath12 ) ; _ ii ) _ the absolute @xmath13 and @xmath14 magnitudes at constant ( j k)@xmath15(j h)@xmath16 colors ; and _ iii ) _ the slope in the [ k , j k ] and [ h , j h ] planes . then , by using the empirical calibrations from vfo04a linking this set of photometric indices to the cluster metal content , we finally derived the photometric metallicity estimates in the @xcite scale . hence , hereafter the notation [ fe / h ] refers to the @xcite scale . note that , to derive the global metallicity [ m / h ] , which takes into account the iron as well as the @xmath17element abundances , we used the calibrations presented in fvo06 within the bulge - like enrichment scenario . hence , accordingly to carney s ( 1996 ) suggestions and the more recent results on the bulge populations ( @xcite for bulge clusters ; and @xcite for giant field stars ) we adopted a [ @xmath17/fe]=0.30 constant over the entire range of metallicity up to solar [ fe / h ] . in the next paragraphs we present the cmds and rgb fiducial ridge lines for the 12 bulge clusters of _ sample c_. we briefly discuss the main cmd properties , and we provide some references for previous published photometries . ngc 6256 is a heavily reddened cluster in the outer bulge region . its extinction , whose estimates range from e(b v)=0.84 ( * ? ? ? * hereafter h96 ) to e(b v)=1.66 ( * ? ? ? * hereafter s98 ) , prevented detailed photometric studies in the optical . only two studies of the cluster can be found in the literature . the first by @xcite based on vi ground based photometry , and the second by @xcite based on hst bv observations . however , the derived cmds show an extremely scattered rgb which is barely defined . in our ir - cmd , shown in fig . [ cmds ] , the cluster rgb is rather scattered , suggesting the presence of a red component which might be due to the contamination by bulge field stars . the hb appears as a blue vertical structure at ( j k)@xmath180.5 and k@xmath1914 typical of low intermediate metallicity . the almost blue vertical sequence , at 11.5@xmath19k@xmath1915 and 0.4@xmath19(j k)@xmath191 , is likely due to the foreground disk stars . in order to investigate the nature of the observed rgb spread , we compared the sequence morphology in the cmds at different distances from the cluster center ( see fig . [ 6256rad ] ) . as shown in fig . [ 6256rad ] , the reddest rgb component becomes progressively populated as the distance to the cluster center increases , thus confirming that it is due to field stars . as previously done for other highly contaminated clusters ( see i.e. ngc 6304 in vof05 ) , the rgb fiducial ridge line has been derived considering only the innermost region of the cluster , typically only the stars lying within 30 " from the cluster center . since the absence of a prominent feature such as the red clumpy hb makes difficult to derive the cluster reddening and distance , we used the empirical method ( see fvo06 ) rather than the differential one . from the analysis of the cluster cmd we estimated the rgb slope and the observed rgb - tip to be rgb@xmath20=0.052 and k@xmath21=9.21 , respectively . using these values as input parameters , the computational routine ( assuming the _ bulge like scenario _ ) , gives a reddening e(b v)=1.2 , an intrinsic distance modulus ( m - m)@xmath22=14.79 , and metallicities [ fe / h]=1.63 dex , and [ m / h]=1.43 dex . the distance estimate agree with the value by h96 , and the derived metallicity is consistent with those published by @xcite ( [ fe / h]=1.35 dex ) , based on low medium resolution k spectra , and by @xcite ( [ fe / h]=1.01 dex ) from integrated optical spectroscopy . ngc 6266 is a high density ( [email protected] , @xcite ) , moderately reddened ( e(b v)=0.44 , h96 ; e(b v)=0.47 , s98 ) and massive ( m@xmath24=-9.19 , h96 ) cluster , which has been subject of several dynamics , kinematics and stellar variability studies . in fact , this cluster turns out to be particularly interesting because : _ i ) _ it is the fifth ( after ter 5 , 47 tuc , m 15 and m 28 ) ranking gc for wealth millisecond pulsars ( msps ) , and it is the only gc exclusively populated by msps in binary systems @xcite ; _ ii ) _ it hosts more than 200 rr lyrae variables @xcite , whose periods place this cluster in the oosterhoff type i group , though the cluster hb morphology ( a blue hb component and a very extended blue tail ) is similar to oosterhoff type ii gcs ; _ iii ) _ recent _ chandra _ x ray observations have revealed a very large number of x ray sources , suggesting the presence of a high number of cataclysmic ( and/or interacting ) binaries ; _ iv ) _ its radial velocity is smaller than the escape velocity from the bulge , indicating that , whatever its origin would be , it will never escape from the bulge @xcite . @xcite published the most extensive photometry based on a combination of multi band high resolution hst and wide field ground based observations aimed at studying the cluster dynamical state , finding that the cluster has not experienced core collapse yet . no ir photometric studies for this cluster have been published so far . the observed cmd in the [ k , j k ] plane ( see fig . [ cmds ] ) shows a well populated rgb and a blue extended hb , suggesting a low intermediate metallicity . as expected , in the near ir plane the high level of contamination by foreground disk stars affecting the optical ground based cmd of @xcite is drastically reduced , allowing us a clear definition of the rgb ridge line from the tip down to the sub giant branch ( sgb ) , at k@xmath1816 . to estimate the cluster metallicity , we have adopted the distance modulus ( m m)@xmath22=14.11 of @xcite ( based on the ( * ? ? ? * hereafter f99 ) distance scale ) , and the reddening value from h96 . the global set of photometric indices ( in both [ k , j k ] and [ h , j h ] planes ) computed for ngc 6266 have yielded a [ fe / h]=0.98 dex and [ m / h]=0.82 dex , in good agreement with the values in the literature ( see i.e. h96 ) . ngc 6273 is an intermediate low metallicity ( [ fe / h]=1.68 , see h96 ) , medium concentration ( c=1.5 * ? ? ? * ) cluster , and it is also the second most luminous ( m@xmath24=-9.50 ) in the @xcite compilation . it has been the subject of a monographic study by @xcite , based on hst b , v observations , aimed at understanding the nature of its extended hb blue tail , which shows a clear double peaked distribution and a well defined gap . from the analysis of the cmd the authors estimated an intrinsic distance modulus ( m m)@xmath22=14.77 , and an average reddening e(b v)=0.47 , though their optical cmd is strongly affected by differential reddening ( @xmath25e(b v)=0.2 mag ) . the only ir cmd available in the literature is the one published by @xcite , but it is poorly populated and quite shallow , reaching only k@xmath1815 . from the [ k , j k ] cmd , @xcite estimated a reddening e(b v)=0.38 , in excellent agreement with the value listed in literature ( i.e. e(b v)=0.38 , h96 , and e(b v)=0.31 , s98 ) . [ cmds ] shows our observed cmd in the [ k , j k ] plane along with the derived rgb ridge line ( solid line ) . the cluster rgb is well populated even in the brightest magnitude bin , and the hb appears as a vertical structure almost parallel to the rgb at ( j k)@xmath180.2 , suggesting a low intermediate metallicity . our photometry is deep enough to reach @xmath181 mag below the main sequence turn off ( ms to ) , however the sgb is quite scattered , preventing an accurate measurement of the ms to luminosity , hence a reliably cluster age estimate . because of its hb morphology , the cluster reddening and distance have been derived by using the empirical method of fvo06 . from the analysis of the observed cmd we derived the rgb slope ( rgb@xmath20=0.063 ) and the the rgb tip ( k@xmath21=8.57 ) . adopting these two values as input parameters , we found e(b v)=0.40 , ( m m)@xmath22=14.58 , [ fe / h]=1.40 , and [ m / h]=1.21 , for the cluster reddening , distance and metallicity , respectively . the derived distance is consistent ( within 0.2 mag ) with the @xcite estimate and ( within 0.1 mag ) with the value listed by h96 . our reddening measurement agree well with the values in literature ( see i.e. h96 , s98 , @xcite ) . ngc 6293 is the most metal poor cluster in the observed sample . in the literature , the best available photometries are those published by @xcite and @xcite . @xcite presented hst b , v cmd , which shows an extended blue hb , confirming the low metal content of this cluster . the near ir cmd of @xcite is poorly populated , and only reaches k@xmath515 . from the ir data , the author estimated a reddening e(b v)=0.04 , which is significantly lower than the value listed in the literature ( e(b v)= 0.39 , h96 ; e(b v)= 0.60 , s98 ) . a careful examination of the derived cmd in fig . [ cmds ] , reveals the presence of a quite scattered rgb , which can not be explained in terms of photometric errors . in order to understand the nature of the observed scatter , as already done for ngc 6256 , in fig . [ 6293rad ] we compared the radial cmds in 4 annuli at different distances from the cluster center . in the innermost 30 `` region ( fig . [ 6293rad ] , _ panel a _ ) the rgb is quite narrow , and progressively spread out with increasing the distance from the cluster center . a field component redder than the cluster rgb is present at r@xmath2660 '' . this high level of field contamination is not surprising since ngc 6293 lies in the ophiuchus complex , a dense stellar region towards the galactic center . hence , the rgb fiducial ridge line has been derived using only the stars in the innermost 30 " cluster region , where the field contamination is low . the measured reddening , distance , and metallicity estimates ( e(b v)=0.30 , ( m m)@xmath22=15.10 , [ fe / h]=1.73 , and [ m / h]=1.55 ) have been obtained by using the empirical method of fvo06 , with the following input parameters : rgb@xmath20=0.048 and k@xmath21=9.24 . the derived cluster parameters nicely agree with the corresponding values listed by h96 . in particular , the derived photometric metallicity is consistent ( within @xmath100.2 dex ) with the spectroscopic estimate published by @xcite ( i.e. [ fe / h]=1.99 dex ) . among the observed clusters , ngc 6316 is one of the most metal rich and with the largest distance to the galactic center . the [ k , v k ] cmd presented by @xcite shows a red clumpy hb and a quite well populated upper rgb , the photometry reaching only @xmath181 mag below the hb . they estimated a reddening e(b v)=0.6 , which is consistent within the values listed by h96 ( e(b v)=0.55 ) and bys98 ( e(b v)=0.98 ) . @xcite presented a hst [ v , b v ] cmd of this cluster , showing a hb morphology close to that of 47 tuc . however the moderately high reddening , and the contamination by foreground disk stars prevented a clear definition of the rgb , which appears considerably scattered even in the brightest bins . as shown in fig . [ cmds ] our cmd in the [ h , j h ] plane is deep enough to properly sample the entire rgb , from the base up to the tip , thus allowing us a safe definition of the rgb fiducial ridge line . from the cmd and the derived lf , we estimated a reddening e(b v)=0.56 , which is intermediate between the @xcite and h96 estimates , and an intrinsic distance modulus ( m m)@xmath22=15.33 , slightly longer ( @xmath180.13 mag ) than the h96 value . since no k band data were taken for this cluster , to estimate its metallicity we only measured the photometric indices in the [ h , j h ] plane . our findings , [ fe / h]=0.60 dex and [ m / h]=0.45 dex , nicely agree with the value listed by h96 , confirming that ngc 6316 has a metallicity slightly higher than 47 tuc , as previously suggested by @xcite on the basis of the rgb slope measurement . ngc 6355 lies behind a dark nebula in the eastern extension of the ophiuchus complex , and it is thus affected by a relatively high extinction ( i.e. e(b v)=0.75 , h96 ; e(b v)=1.15 , s98 ) . @xcite published a [ v , v i ] cmd showing a large field contamination and a barely detectable blue hb . from the analysis of the cmd , the authors found e(b v)=0.78 , ( m m)@xmath22=14.73 and [ fe / h]=1.3 dex . our derived cmd , shown in fig . [ cmds ] , exhibits a blue hb at 0.2@xmath19(j k)@xmath190.6 and 14.5@xmath19k@xmath1916 , and a quite scattered rgb . as expected , in the [ k , j k ] and [ h , j h ] planes , the high level of contamination by disk stars strongly affecting the @xcite cmd , is significantly reduced . however , in our cmd the main source of contamination is due to the foreground bulge field stars , which produce the observed rgb split . in fact , as shown in the radial cmds of fig . [ 6355rad ] , by increasing the distance from the cluster center , a secondary red component becomes progressively more pronounced , causing the observed scatter in the rgb . as done for ngc 6256 and ngc 6293 , the rgb ridge line has been derived by using only the stars lying within 30 " from the cluster center , and the main cluster properties have been estimated by using the empirical method of fvo06 . from the observed cmds , we measured rgb@xmath27=0.068 and k@xmath28=8.854 . adopting these two values we find e(b - v)=0.82 , ( m - m)@xmath22=14.70 , [ fe / h]=1.42 dex and [ m / h]=1.22 dex for the cluster reddening , distance and metallicity . our findings nicely agree with the @xcite results . ngc 6388 is a moderately reddened ( i.e. e(b v)=0.37 , h96 ; e(b v)=0.39 , s98 ; e(b v)=0.40 , @xcite ) , metal rich ( [ fe / h]=-0.60 , h96 , @xcite ) cluster located in the outer bulge region ( see fig . [ map ] ) . since the discovery by @xcite of an extended blue hb ( as well as in ngc 6441 ) in its hst based cmd , ngc 6388 has been subject of several optical photometric studies ( see i.e. * ? ? ? * ; * ? ? ? * ; * ? ? ? * and reference therein ) . ngc 6388 displays a quite peculiar hb morphology . in fact , besides a well populated red hb clump ( a feature normally predicted by the stellar evolution theory in the case of old , metal rich populations ) it shows an extended blue tail and a population of rr lyrae variables @xcite . ngc 6388 and ngc 6441 are thus the most metal rich examples of _ second parameter _ affecting the hb morphology . from the analysis of the cluster cmd based on washington photometry , @xcite claimed that ngc 6388 has a rgb too broad to be chemically homogeneous , suggesting a metallicity spread of @xmath29[fe / h][email protected] dex . however , very recently , no intrinsic metallicity spread has been found by @xcite , who performed high resolution ( r@xmath1040,000 ) optical spectroscopy of seven cluster members , finding an average [ fe / h]=0.44 dex with a _ rms_=0.04 dex . no ir photometric studies have been published so far for this cluster . [ cmds ] shows the first cmd in the [ k , j k ] plane , and the derived rgb ridge line . as expected , the observed cmd is characterized by a well defined red hb clump suggesting a high metallicity , while the extended blue hb is barely visible as the vertical structure at 15.5@[email protected] and 0@xmath30(j k)@xmath300.6 . the rgb appears well populated over its entire extension , from the base ( k@xmath1817 ) up to the tip , allowing us a detailed description of its morphological features . from the comparison of the cluster cmds and lfs with those of 47 tuc , we estimated the reddening and distance to be e(b v)=0.44 and ( m m)@xmath22=15.38 , respectively . the photometric indices measured along the rgb , in the [ m@xmath31 , ( j k)@xmath22 ] and [ m@xmath33 , ( j h)@xmath22 ] absolute planes , yielded the following metallicity estimates : [ fe / h]=0.61 dex and [ m / h]=0.42 dex , in good agreement with previous literature values . ngc 6401 is located in the inner bulge region , at @xmath180.8 kpc ( see table [ tab1 ] and fig . [ map ] ) from the galactic center . its physical parameters are not well determined in the literature . in particular , the extinction estimates range from e(b v)=0.53 @xcite to e(b v)=0.98 ( s98 ) , and the cluster distance values are between ( m m)@xmath22=14.39 ( h96 ) and ( m @xcite . from the analysis of the observed [ v , v i ] cmd , @xcite concluded that ngc 6401 is metal rich , with a metallicity close to 47 tuc , while from low resolution optical spectroscopy @xcite found [ fe / h]=1.1 dex . the near ir cmd here obtained is shown in fig . [ cmds ] , along with the derived rgb ridge line . the most interesting features are : _ i ) _ a blue hb tail at 0.4@xmath19(j k)@xmath190.9 and 13@xmath19k@xmath1916 ; _ ii ) _ a rather scattered rgb , which seems split in 2 separated sequences , particularly in the [ k , j k ] plane . note that , the presence of a blue extended hb is confirmed by the @xcite work based on hst bv observations . fig . [ 6401rad ] clearly shows that , in the innermost region , within 30 " from the cluster center , the cmd exhibits a blue hb and a quite narrow rgb , while by increasing the distance from the cluster center the cmd starts to be characterized by a red clumpy hb and by a rgb significantly redder than the cluster mean loci . hence , the observed rgb spread is likely due to the bulk of the bulge field population which is metal rich . in order to derive the cluster main properties we used the fvo06 empirical method , assuming the _ bulge like scenario_. we measured the cluster rgb slope considering only those stars lying in the innermost cluster region ( rgb@xmath27=-0.065 ) and we estimated the observed ( k=8.69 ) rgb tip by using the brightest stars in our catalog lying along the rgb ridge line . we used these values as input parameters for the computational routine and we obtained the following reddening , distance and metallicity estimates : e(b v)=1.1 , ( m m)@xmath22=14.43 , [ fe / h]=1.37 dex and [ m / h]=1.20 dex . the derived photometric metallicity is consistent with the spectroscopic result by @xcite ( @xmath25[fe / h]=0.25 dex ) and significantly lower than that found by @xcite ( @xmath25[fe / h]@xmath180.7 dex ) . the moderately reddened cluster ngc 6642 ( i.e. e(b v)=0.41 , h96 ; e(b v)=0.40 , s98 ) has been observed in the ir and in the optical by @xcite and @xcite , respectively . however both the published cmds show sequences which are not well populated , and with a remarkable scatter . fig . [ cmds ] shows our derived cmd in the [ h , j h ] plane , along with the mean rgb ridge line . the main features of the cmd are a blue hb and a steep rgb , typical of low intermediate metallicity clusters . in this case , since no k photometry is available , the reddening and the distance have been estimated by matching the cluster sequence loci in the cmd to those of the reference cluster ngc 6752 . by using our derived reddening ( e(b v)=0.60 ) and distance ( ( m m)@xmath22=14.68 ) , we measured the photometric indices , giving the following photometric metallicity estimates : [ fe / h]=1.20 dex and [ m / h]=1.03 dex . these values are fully consistent ( within 0.2 dex ) with the corresponding values listed by h96 ( [ fe / h]=1.35 ) and by @xcite ( [ fe / h]=1.40 ) , based on low resolution optical spectroscopy . note that , also the obtained distance nicely agree with the value listed by h96 ( i.e. ( m m)@xmath22=14.63 ) . terzan 3 is a low concentration cluster ( c=0.70 , @xcite ) projected on the outskirts of the bulge ( see fig . [ map ] ) . the only published photometry is the one by @xcite . from the analysis of the derived [ v , b v ] cmd , whose major features are a red hb and a moderately bent rgb , they derived e(b v)=0.72 , ( m m)@xmath22=14.05 and [ fe / h]@xmath180.70 dex . our cmd along with the rgb fiducial ridge line is shown in fig . this high resolution ir photometry reaches @xmath182 mag below the ms to . the presence of a red clumpy hb at [email protected] suggests a moderately high metallicity , like 47 tuc . the analysis of the cluster cmds and lfs yields a reddening e(b v)=0.73 ( in excellent agreement with the values found by @xcite and s98 ) , an intrinsic distance modulus ( m m)@xmath22=14.54 and metallicities [ fe / h]=0.82 dex and [ m / h]=0.63 dex . the significant ( @xmath180.5 mag ) discrepancy between our and @xcite distance is due to a different assumption for the selective to total absorption coefficient ( r@xmath34=3.1 , in this study ; r@xmath34=3.3 in the @xcite work ) and for the distance scale ( f99 here , and @xcite in @xcite paper ) . terzan 5 is a compact and massive cluster located in the very inner bulge region . in the last few years this extremely dense cluster has been subject of several studies aimed at investigating its stellar population ( i.e. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , as well as at understanding stellar interactions and gc dynamics ( i.e. * ? ? ? * ; * ? ? ? * ) , having the highest stellar interaction rate of any galactic gc @xcite . in fact among gcs , ter 5 ranks first in wealth of msps , and it hosts many interacting binaries ( i.e. cataclysmic variables , x ray binaries ) . being embedded in a heavily obscured zone , it is affected by a large differential reddening which makes optical observations almost impossible . in fact , the only available optical photometry is that by @xcite who published a cmd in the [ i , v i ] plane , showing largely scattered rgb and hb . they estimated e(b v)=2.49 , ( m m)@xmath22=13.74 and a metallicity close to ngc 6553 ( i.e. [ fe / h]=0.3 , @xcite ) . later on , a [ j , j h ] cmd based on hst observations has been published by @xcite and @xcite with the aim of deriving more accurate cluster parameters such as reddening , distance , and age . from the analysis of the derived cmd , @xcite found ter 5 to be coeval to ngc 6528 . @xcite obtained an ir cmd significantly deeper ( @xmath182mag ) , confirming the age found by @xcite . they , also derived e(b v)=2.16 and d@xmath35=8.7 kpc . [ cmds ] shows our observed cmd in the [ k , j k ] plane along with the derived rgb ridge line . the dominant features of the ir cmd are : _ i ) _ a red clumpy hb ; and _ ii ) _ a well populated rgb , from the base ( k@xmath1816 ) up to the tip ( [email protected] ) , which allow us a clear definition of its ridge line . from the analysis of the cmd and lf we derived e(b v)=2.38 , and ( m m)@xmath22=13.87 . the reddening value is in between the estimates provided by @xcite and @xcite , while our distance measurement is fully consistent with the one of @xcite . the photometric indices measured along the cluster rgb by adopting the reddening and distance quoted above , lead the following metallicity estimates : [ fe / h]=0.32 dex and [ m / h]=0.19 dex , thus in excellent agreement with the high resolution ir spectroscopic result obtained by @xcite ( [ fe / h]=0.21 dex ) . terzan 6 is another example of heavily reddened cluster in the inner bulge region . the only available photometric study is the one by @xcite , which provided a [ i , v the authors estimated the cluster reddening ( e(b v)=2.24 ) by comparing the derived cmd with that of ngc 6553 , the cluster distance ( ( m m)@xmath22=14.23 ) by using the absolute v magnitude of the hb level , and from the rgb morphology they concluded that ter 6 shows intermediate characteristics between 47 tuc and ngc6528/ngc6553 . our ir observations of ter 6 provided the first ir cmd ( see fig . [ cmds ] ) showing a red hb and a curved rgb which are both high metallicity indicators . from the derived cmd we found e(b v)=2.35 and ( m m)@xmath22=14.13 . both estimates nicely agree with the corresponding values found by @xcite . the set of ir photometric indices measured by using the rgb ridge lines give a cluster metallicity of [ fe / h]=0.62 dex and [ m / h]=0.43 dex . the compilation presented in this study is publicy available through the worldwideweb . the website address is : + for the entire cluster sample , two separate files are available : the first list the j , h and k photometry for all the measured stars in each cluster , together with the stars absolute position in r.a . and dec . ; the second reports the observed rgb fiducial ridge line in the [ k , j k ] and [ h , j h ] planes . table [ tab1 ] lists the derived parameters for the 24 clusters , namely the reddening , the distance and the metallicity in both the adopted scales ( columns [ 9,10,11,12 ] , respectively ) . moreover , by using our distance determinations , the _ ( l , b ) _ galactic coordinates from h96 , and assuming a distance r@xmath22=8 kpc to the galactic center @xcite , we also provided new estimates of the clusters distance from the sun ( d@xmath35 ) , from the galactic center ( r@xmath36 ) , and the distance components x , y , z , in a sun centered coordinate system ( columns [ 4,5,6,7,8 ] , respectively ) . finally for each cluster , column [ 13 ] lists the references for the ir photometry . it is worth mentioning that table [ tab1 ] lists the largest , homogeneous compilation of bulge gcs obtained so far , whose properties have been derived in a self consistent way . in fact , the derived cluster reddening , distance and metallicity are based on _ i ) _ a homogeneous photometric database analyzed by using the same data reduction procedures , and calibrated onto the 2mass photometric system ; _ ii ) _ the f99 distance scale ; _ iii ) _ a uniform and high resolution metallicity scale @xcite . the most interesting evolutionary features along the rgb are the so called bump and tip . the former flags the point ( during the post ms evolution of low mass stars ) when the narrow hydrogen burning shell reaches the discontinuity in the hydrogen distribution profile , generated by the previous innermost penetration of the convective envelope . besides providing an obvious check on the accuracy of theoretical models of stellar evolution , the identification of the rgb bump in stellar systems can be used as useful tool for providing observational constraints on a number of population parameters , since it is a sensitive function of metal content , helium abundance and stellar population age . from the observational point of view , as emphasized by @xcite and f99 , the combined use of the differential and integrated lfs is the best tool to properly detect the bump . the evolution along the rgb ends at the so called rgb tip with helium ignition in the stellar core . in gc stars , this event is moderately violent because it takes place in a electron degenerate core . because the rgb reaches its maximum extension in luminosity for stellar populations older than @xmath3712 gyr ( i.e. when stars with m@xmath38 are evolving off the ms ) and it remains approximately constant with increasing the age of the population , the tip is now widely used as a standard candle for distance determination of stellar systems . the dependence of the rgb bump and tip luminosities ( in the j , h and k bands . as well as in bolometric ) on the cluster metallicity has been investigated in vfo04b . here we have extended the same study to the new sample of bulge clusters . in doing this , we have adopted the same strategy followed in that paper , where the reader can find a detailed description of the adopted procedures . for the entire sample , the measurements of the main evolutionary features are listed in table [ tab2 ] : the observed j , h , and k magnitude of the red hb clump ( columns [ 4 , 5 , 6 ] ) , the j , h , k and bolometric bump ( columns [ 7 , 8 , 9 , 13 ] , respectively ) , and tip ( columns [ 10 , 11 , 12 , 14 ] , respectively ) magnitudes . the bolometric bump and tip magnitude have been obtained by using the bolometric corrections for population ii giants computed by @xcite . [ tip ] shows the absolute rgb tip magnitudes as a function of the metallicity , in both the adopted scales . the clusters tip determination nicely agree with the empirical relations by vfo04b ( solid lines ) , and with the theoretical prediction by @xcite ( dashed and dotted line , respectively ) . as shown in fig . [ tip_bol ] , a good agreement is also found when we compare the bolometric tip magnitudes , as a function of the clusters metallicity , with the vfo04b empirical relation ( solid lines ) , and with four different theoretical expectations by @xcite ( dashed , dotted and dotted dashed lines , respectively ) . [ bump ] plots the absolute j , h and k bump magnitudes of the bulge ( filled circles ) and halo ( open circles ) clusters ( the latter presented in vfo04b ) _ vs _ metallicity , showing that the bump becomes rapidly fainter by increasing the latter . hence , since with the present study we both enlarged the sample of high metallicity clusters and adopted the recent [ fe / h ] estimates by @xcite based on high resolution ir spectroscopy , for the two most metal rich objects in our sample ( namely , ngc 6528 and ngc 6553 ) , we calibrated new relations . as shown in fig . [ bump ] the new calibrations ( solid lines ) differ from the vfo04b ones ( dashed lines ) only in the metal rich tail , making the bump somewhat brighter . the new calibrations of the bump j , h , k absolute magnitude , in both the adopted metallicity scales , are as follows : @xmath39 + 0.28[fe / h]^2\ ] ] @xmath40 + 0.34[fe / h]^2\ ] ] @xmath41 + 0.27[fe / h]^2\ ] ] @xmath42 + 0.28[m / h]^2\ ] ] @xmath43 + 0.38[m / h]^2\ ] ] @xmath44 + 0.27[m / h]^2\ ] ] this publication makes use of data products from the two micron all sky survey , which is a joint project of the university of massachusetts and infrared processing and analysis center / california institute of technology , founded by the national aeronautics and space administration and the national science foundation . barbuy , b. , ortolani , s. & bica , e. 1997 , , 122 , 483 barbuy , b. , bica , e. & ortolani , s. 1998 , , 333 , 117 barbuy , b. , ortolani , s. & bica , e. 1998 , , 132 , 33 barbuy , b. , ortolani , s. , bica , e. , desidera , s. 1999 , , 348 , 783 beccari , g. , ferraro , f. r. , possenti , a. , valenti , e. , origlia , l. , rood , r. t. 2006 , , 131 , 2551 bica , e. , clariaa , j. j. , piatti , a. e. , bonatto , c. 1998 , , 131 , 483 caloi , v. , dantona , f. & mazzitelli , i. 1997 , , 320 , 823 carretta , e. & gratton , r. g. 1997 , , 121 , 95 carretta , e. , bragaglia , a. , gratton , r. g. , momany , y. , recio blanco , a. et al . 2006 , , submitted cassisi , s. , castellani , v. , ciarcelluti , p. , piotto , g. , zoccali , m. 2000 , , cohn , h.n . , lugger , p. h. , grindlay , j. e. & edmonds , p. d. 2002 , , 571 , 818 contreras , r . , catelan , m. , smith , h. a. ; pritzl , b. j. ; borissova , j. 2005 , , 623 , l117 cunha , k. & smith , v. v. 2006 , , astro ph/0607393 davidge , t. j. , harris , w.e . , bridges , t. j. , hanes , d.a . 1992 , , 81 , 251 315 , 679 davidge , t. j. 2000 , , 120 , 1875 dinescu , d. i. , girard , t. m. , van altena , w. f. , lopez , c. e. 2003 , , 125 , 1373 djorgovski , s. 1993 , in asp conf . ser . 50 , _ structure and dynamics of globular clusters _ , eds . s. djorgovski & g. meylan eisenhauer , f. , schodel , r. , genzel , r. , ott , t. , tecza , m. , abuter , r. , eckart , a. , alexander , t. , 2003 , , 597 , l121 ferraro , f. r. , messineo , m. , fusi pecci , f. , de paolo , m. a. , straniero , o. , chieffi , a. , limongi , m. 1999 , , 118 , 1738 ( f99 ) ferraro , f. r. , montegriffo , p. , origlia , l. , fusi pecci , f. 2000 , , 119 , 1282 ( f00 ) ferraro , f. r. , valenti , e. & origlia , l. 2006 , , 649 , 243 ( fvo06 ) fulbright , j. p. , mcwilliam , a. , & rich , m. r. 2006 , astro - ph/0609087 fusi pecci , f. , ferraro , f. r. , crocker , d. a. , rodd , t. r. & buonanno , r. 1990 , , 238,95 girardi , l. , bressan , a. , bertelli , g. & chiosi , c. 2000 , , 141 , 371 harris , w. e. 1996 , aj , 112 , 1487 ( h96)(for the 2003 updated version see http://physwww.mcmaster.ca/%7eharris/mwgc.dat ) heinke , c. o. , edmonds , p. d. , grindlay , j. e. , lloyd , d. a. , cohn , h. n. , lugger , p. m. 2003 , , 590 , 809 hughes , j. d. , wallerstein , g. , covarrubias , r. , hays , n. 2006 , aas , 208 , 6501 jones , r. v. , carney , b. w. , storm , j. , latham , d. w , 1992 , , 386 , 646 lee , j .- w . & carney , b. w. 2002 , , 124 , 1511 minniti , d. 1995 , , 303 , 468 moehler , s. , sweigart , a. v. & catelan , m. 1999 , , 351 , 519 montegriffo , p. , ferraro , f. r. , origlia , o.,fusi pecci , f. 1998 , , 297 , 872 origlia , l. , rich , r. m. , & castro , s. 2002 , , 123 , 1559 origlia , l. , & rich , r. m. 2004 , , 127 , 3422 origlia , l. , valenti , e. , rich , r. m. , ferraro , f. r. 2005 , , 363 , 897 ( o05 ) origlia , l. , valenti , e. & rich , r. m. 2005 , , 356 , 1276 ortolani , s. , barbuy , b. & bica , e. 1996 , , 308 , 733 ortolani , s. , barbuy , b. , bica , e. , renzini , a. , zoccali , m. , rich , r. m. & cassisi , s. 2001 , , 376,878 ortolani , s. , bica , e. & barbuy , b. 2003 , , 402 , 565 pritzl , b. j. , smith , h. a. , catelan , m. , sweigart , a. v. 2002 , , 124 , 949 pritzl , b. j. , venn , k. a. , ivans , i. i. , catelan , m. , layden , a. c. , kinemuchi , k. , rich , r. m. , sweigart , a. v. 2005 , aas , 206 , 3601 pietrinferni , a. , cassisi , s. , salaris , m. & castelli , f. 2004 , , 612 , 168 piotto , g. , zoccali , m. , king , i. r. , djorgovski , s. g. , sosin , d. , rich , r. m. & meylan , g. 1999 , , 118,1727 piotto , g. , king , i. r. , djorgovski , s. g. , sosin , d. , zoccali , m. , saviane , i. , de angeli , f. , riello , m. , recio blanco , a. , rich , r. m. , meylan , g. & renzini , a. 2002 , 391,945 possenti , a. , damico , n. , manchester , r. n. , camilo , f. , lyne , a. g. , sarkissian , j. & corongiu , a. 2003 , , 599 , 475 rich , r. m. et al . 1997 , , 484,25 rich , r.m . , & origlia , l. 2005 , , 634 , 1293 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 ( s98 ) salaris , m. & cassisi , s. 1997 , , 289 , 406 stephens , a. w. & frogel , j. a. 2004 , , 127 , 925 stetson , p. b. 1987 , , 99,191 straniero , o. , chieffi , a. & limongi , m. 1997 , , 490,425 ransom , s. m. , hessels , j. w. , stairs , i. h. , freire , p. c. c. , camilo , f. , kaspi , v. m. , kaplan , d. l. 2005 , science , 307 , 892 valenti , e. , ferraro , f. r. , perina , s. , origlia , l. 2004 , , 419 , 139 valenti , e. , ferraro , f. r. & origlia , l. 2004 , , 351 , 1204 ( vfo04a ) valenti , e. , ferraro , f. r. & origlia , l. 2004 , , 354 , 815 ( vfo04b ) valenti , e. , origlia , l. & ferraro , f. r. 2005 , , 361 , 272 ( vof05 ) verbunt , f. & hut , p. 1987 , iau symp . 125 , the origin and evolution of neutron stars , ed . d. j. helfand & h. h. huang ( dordrecht : reidel ) , 187 weiland , j. l. , arendt , r. g. , berriman , g. b. , dwek , e. , freudenreich , h. t. , et al . 1994 , , 425 , l81 zoccali , m. , barbuy , b. , hill , v. , ortolani , s. , renzini , a. , bica , e. , momany , y. , pasquini , l. , minniti , d. , rich , r. m. 2004 , , 423 , 507 zoccali , m. , lecureur , a. , barbuy , b. , hill , v. , renzini , a. , minniti , d. , momany , y. , gomez , a. , ortolani , s. 2006 , a&a , 457l , 1 lcccccccccccl 6256@xmath47 & -12.21 & 3.31 & 9.1 & 2.2 & 8.9 & -1.9 & 0.5 & 1.20 & 14.79 & -1.63 & -1.43&tw + 6266 & -6.42 & 7.32 & 6.6 & 1.8 & 6.5 & -0.7 & 0.8 & 0.47 & 14.11 & -0.99 & -0.80&tw + 6273@xmath47 & -3.13 & 9.38 & 8.2 & 1.4 & 8.1 & -0.4 & 1.3 & 0.40 & 14.58 & -1.40 & -1.21&tw + 6293@xmath47 & -2.38 & 7.83 & 10.5 & 2.8 & 10.4 & -0.4 & 1.3 & 0.30 & 15.10 & -1.73 & -1.55&tw + 6304 & -4.17 & 5.38 & 6.0 & 2.2 & 5.9 & -0.4 & 0.6 & 0.58 & 13.88 & -0.75 & -0.56&vof05 + 6316 & -2.82 & 5.76 & 11.6 & 3.8 & 11.6 & -0.6 & 1.2 & 0.56 & 15.33 & -0.58 & -0.38&tw + 6342 & 4.90 & 9.73 & 8.4 & 1.6 & 8.3 & 0.7 & 1.4 & 0.57 & 14.63 & -0.71 & -0.53&vfo04a + 6355@xmath47 & -0.42 & 5.43 & 9.0 & 1.1 & 8.7 & -0.1 & 0.8 & 0.81 & 14.77 & -1.42 & -1.22&tw + 6380 & -9.82 & -3.42 & 9.2 & 1.9 & 9.0 & -1.6 & -0.5 & 1.29 & 14.81 & -0.87 & -0.68&vfo04a + 6388 & -14.44&-6.74 & 11.9 & 4.8 & 11.5 & -3.0 & -1.4 & 0.44 & 15.38 & -0.61 & -0.42&tw + 6401@xmath47 & 3.45 & 3.98 & 7.7 & 0.8 & 7.7 & 0.5 & 0.5 & 1.10 & 14.43 & -1.37 & -1.20&tw + 6440 & 7.73 & 3.80 & 8.2 & 1.2 & 8.1 & 1.1 & 0.5 & 1.15 & 14.58 & -0.49 & -0.40&vfo04a + 6441 & -6.47 & -5.01 & 13.5 & 5.7 & 13.4 & -1.5 & -1.2 & 0.52 & 15.65 & -0.68 & -0.52&vfo04a + 6528 & 1.14 & -4.17 & 7.5 & 0.8 & 7.5 & 0.1 & -0.5 & 0.62 & 14.37 & -0.17 & + 0.04&f00 + 6539 & 20.80 & 6.78 & 8.4 & 3.1 & 7.8 & 3.0 & 1.0 & 1.08 & 14.63 & -0.79 & -0.60&o05 + 6553 & 5.25 & -3.02 & 4.9 & 3.2 & 4.9 & 0.4 & -0.3 & 0.84 & 13.46 & -0.30 & -0.09&f00 + 6569 & 0.48 & -6.68 & 12.0 & 4.2 & 11.9 & 0.1 & -1.4 & 0.49 & 15.40 & -0.85 & -0.66&vof05 + 6624 & 2.79 & -7.91 & 8.4 & 1.3 & 8.3 & 0.4 & -1.2 & 0.28 & 14.63 & -0.63 & -0.48&vfo04a + 6637 & 1.70 & -10.30 & 9.4 & 2.1 & 9.3 & 1.3 & -1.2 & 0.14 & 14.87 & -0.77 & -0.57&vof05 + 6638 & 7.90 & -7.15 & 10.3 & 2.9 & 10.2 & 0.3 & -1.8 & 0.43 & 15.07 & -1.00 & -0.78&vof05 + 6642 & 9.81 & -6.44 & 8.6 & 1.8 & 8.4 & 1.5 & -1.0 & 0.60 & 14.68 & -1.20 & -0.99&tw + ter 3 & -14.92&9.19 & 8.1 & 2.4 & 7.7 & -2.1 & 1.3 & 0.73 & 14.54 & -0.82 & -0.63&tw + ter 5 & 3.81 & 1.67 & 5.9 & 2.1 & 5.9 & 0.4 & 0.2 & 2.38 & 13.87 & -0.34 & -0.14&tw + ter 6 & -1.43 & -2.16 & 6.7 & 1.3 & 6.7 & -0.2 & -0.3 & 2.35 & 14.13 & -0.62 & -0.43&tw + & & & & & & & & & & & & + + + lccccccccccccc 6256 & -1.63 & -1.43 & & & & & & & 10.63 & 9.51 & 9.21 & & -3.56 + 6266 & -0.99 & -0.80 & & & & 13.50 & 12.75 & 12.65 & 9.26 & 8.10 & 7.85 & 0.44 & -3.56 + 6273 & -1.40 & -1.21 & & & & 13.65 & 13.05 & 12.85 & 9.67 & 8.83 & 8.57 & 0.10 & -3.59 + 6293 & -1.73 & -1.55 & & & & & & & 10.17 & 9.43 & 9.24 & & -3.23 + 6304 & -0.75 & -0.56 & 13.55 & 12.85 & 12.70 & 14.03 & 13.33 & 13.13 & 9.06 & 8.02 & 7.65 & 1.10 & -3.59 + 6316 & -0.58 & -0.38 & 14.93 & 14.25 & & 15.20 & 14.65 & & 10.28 & 9.17 & & & + 6342 & -0.71 & -0.53 & 14.25 & 13.60 & 13.40 & 14.65 & 13.85 & 13.75 & 9.71 & 8.67 & 8.35 & 0.98 & -3.69 + 6355 & -1.42 & -1.22 & & & & & & & 10.19 & 9.30 & 8.92 & & -3.61 + 6380 & -0.87 & -0.68 & 14.95 & 14.15 & 13.85 & 15.15 & 14.25 & 13.95 & 10.37 & 9.12 & 8.75 & 0.62 & -3.88 + 6388 & -0.61 & -0.42 & 14.90 & 14.27 & 14.17 & 15.18 & 14.47 & 14.33 & 10.24 & 9.47 & 8.81 & 0.88 & -3.76 + 6401 & -1.37 & -1.20 & & & & & & & 10.21 & 9.07 & 8.69 & & -3.42 + 6440 & -0.49 & -0.40 & 14.75 & 13.80 & 13.60 & 15.30 & 14.35 & 14.13 & 10.02 & 8.85 & 8.33 & 1.16 & -3.82 + 6441 & -0.68 & -0.52 & 15.20 & 14.55 & 14.40 & 15.70 & 14.85 & 14.77 & 10.47 & 9.49 & 9.12 & 1.10 & -3.90 + 6528 & -0.17 & + 0.04 & 14.15 & & 13.35 & 15.10 & & 14.05 & 9.16 & & 7.85 & 1.74 & -4.06 + 6539 & -0.79 & -0.60 & 14.65 & 13.85 & 13.65 & 14.90 & 14.05 & 13.83 & 10.08 & 8.92 & 8.47 & 0.71 & -3.77 + 6553 & -0.30 & -0.09 & 13.35 & & 12.40 & 14.05 & & 13.05 & 8.56 & & 6.92 & 1.28 & -3.86 + 6569 & -0.85 & -0.66 & 14.95 & 14.40 & 14.25 & 14.93 & 14.23 & 14.08 & 10.52 & 9.49 & 9.21 & 0.55 & -3.59 + 6624 & -0.63 & -0.48 & 13.95 & 13.40 & 13.25 & 14.45 & 13.60 & 13.65 & 9.30 & 8.37 & 8.08 & 1.07 & -3.85 + 6637 & -0.77 & -0.57 & 14.05 & 13.65 & 13.55 & 14.28 & 13.78 & 13.65 & 9.91 & 9.08 & 8.72 & 0.69 & -3.34 + 6638 & -1.00 & -0.78 & 14.65 & 14.05 & 13.85 & 14.45 & 13.73 & 13.58 & 9.86 & 8.89 & 8.61 & 0.49 & -3.88 + 6642 & -1.20 & -0.99 & & & & 13.85 & 13.17 & & 9.92 & 8.92 & & & + ter3 & -0.82 & -0.63 & 14.27 & 13.57 & 13.42 & 14.37 & 13.73 & 13.50 & 9.95 & 8.91 & 8.42 & 0.54 & -3.47 + ter5 & -0.34 & -0.14 & 15.15 & 13.80 & 13.35 & 15.75 & 14.50 & 13.90 & 10.25 & 8.68 & 7.99 & 1.32 & -3.96 + ter6 & -0.62 & -0.43 & 15.10 & 14.07 & 13.48 & 15.75 & 14.45 & 14.00 & 10.56 & 9.06 & 8.33 & 1.01 & -3.89 +

we investigate the behavior of weak localization , conductance fluctuations , and shot noise of a chaotic scatterer in the semiclassical limit . time resolved numerical results , obtained by truncating the time - evolution of a kicked quantum map after a certain number of iterations , are compared to semiclassical theory . considering how the appearance of quantum effects is delayed as a function of the ehrenfest time gives a new method to compare theory and numerical simulations . we find that both weak localization and shot noise agree with semiclassical theory , which predicts exponential suppression with increasing ehrenfest time . however , conductance fluctuations exhibit different behavior , with only a slight dependence on the ehrenfest time .

cond-mat0507035

according to ehrenfest s theorem , the expectation values of the position and momentum of an electron obey classical equations of motion . as long as the wavefunction of the electron is a wavepacket with minimal uncertainties in momentum and position , the expectation values are a good description of the quantum state . however , wavepackets disperse , and the ehrenfest theorem looses its relevance after a short time . in a cavity with point scatterers , which split electron wavepackets into partial waves after one scattering event , this time is simply the elastic mean free time . in a ballistic cavity with chaotic classical dynamics , this time is the so - called `` ehrenfest time '' @xmath0 , which depends on the lyapunov exponent @xmath1 of the classical motion in the cavity.@xcite for times longer than @xmath0 , a classical description no longer holds and the wave nature of the electrons becomes visible . the wave nature of electrons is the cause of some striking effects that are absent in classical systems . for transport through cavities coupled to source and drain reservoirs via point contacts , these effects are weak localization , universal conductance fluctuations , and shot noise.@xcite in the limit that transport through cavities is ergodic ( dwell time in the cavity is much longer than the time of flight through the cavity ) , the signatures of quantum transport are ` universal ' , independent of the cavity size and shape , and of the fact whether electron motion inside the cavity is ballistic and chaotic or diffusive , with repeated scattering off impurities with size smaller than the electron wavelength . random matrix theory provides a unified theoretical description of weak localization , universal conductance fluctuations , and shot noise in ballistic or diffusive cavities.@xcite if the electron motion is diffusive , the dynamics is fully quantum mechanical already at times much shorter than the time @xmath2 required for ergodic exploration of the cavity s phase space . for ballistic cavities this is true in most practical applications as well the ehrenfest time @xmath3 usually does not exceed the time of flight through the cavity but there is no fundamental reason why @xmath0 always has to be small . the case of large ehrenfest times is of theoretical interest , as it is one of very few regimes in parameter space in which one can observe differences between signatures of quantum transport in ballistic chaotic and diffusive cavities . the most prominent effects of a large ehrenfest time are found if @xmath0 is larger than the dwell time @xmath4 in the cavity . if @xmath5 , quantum transport is deterministic , and shot noise is suppressed.@xcite the suppression of shot noise has been observed experimentally by varying the dwell time @xmath4 of a chaotic cavity,@xcite and numerically , using a chaotic map as a model for a chaotic cavity.@xcite the effect of a large ehrenfest time on weak localization was first addressed by aleiner and larkin.@xcite their theory predicts a suppression of weak localization @xmath6 , if classical correlations are taken into account properly.@xcite the same suppression was found in an independent calculation by adagideli.@xcite experimental observation of the suppression of weak localization at large ehrenfest times has been reported for transport through antidot arrays.@xcite no semiclassical theory for the ehrenfest - time dependence of universal conductance fluctuations exists . however , semiclassical theories for weak localization and universal conductance fluctuations for the limit @xmath7 are essentially equal,@xcite as are diagrammatic perturbation theories for the same phenomena in diffusive cavities , supporting the expectation that the ehrenfest - time dependencies of weak localization and universal conductance fluctuations will be equal as well.@xcite direct numerical simulation of the effect of a large ehrenfest time on quantum transport through two - dimensional chaotic cavities has been problematic because of the prohibitively high computational cost of the simulations . the reason is that @xmath0 depends only logarithmically on the product of the electron wavenumber @xmath8 and the cavity size @xmath9 , @xmath10 for two - dimensional cavities , system sizes at which @xmath11 can not be simulated with present - day algorithms and processor speeds . in order to circumvent this problem , jacquod , schomerus , and beenakker proposed to replace the cavity by a quantum map.@xcite the map is opened , so that simulation of transport properties is possible . although a map has a one - dimensional phase space , a chaotic map shares many characteristics of the chaotic motion in two - dimensional chaotic cavities.@xcite the reduced dimensionality of the map s phase space made numerical simulations with larger ehrenfest times possible . for an open version of the quantum kicked rotator map , numerical simulations were reported for shot noise,@xcite weak localization,@xcite and universal conductance fluctuations.@xcite simulation results for shot noise were in good agreement with the predictions of the semiclassical theory.@xcite however , for conductance fluctuations , no dependence on @xmath0 was found , despite the fact that ehrenfest times larger than the dwell time were considered . @xcite whereas early numerical simulations of weak localization showed no ehrenfest - time dependence,@xcite we showed that there is a systematic decrease of the weak localization correction to the conductance upon increasing @xmath0 , consistent with the semiclassical theory.@xcite the main technical innovation that allowed us to detect a systematic decrease of the weak localization correction upon increasing the ehrenfest time is that we looked at time - resolved numerical simulations : the map s time evolution is truncated after a time @xmath12 , and weak localization , conductance fluctuations , and shot noise are monitored as a function of @xmath12.@xcite this procedure has two advantages . first , it allows the ensemble average over the quasienergy @xmath13 to be done analytically . ( see sec . [ numerics ] for technical details . ) this made it possible to consider significantly larger ensembles than considered previously . second , monitoring quantum corrections as a function of the ` truncation time ' @xmath12 allows us to determine the minimal time after which quantum corrections can occur . in the semiclassical theory , quantum interference requires a minimal wavepacket to be split _ and _ reunited , which takes a minimal time @xmath14 . a schematic diagram drawing relevant semiclassical trajectories for weak localization and conductance fluctuations is shown in fig . [ fig:1 ] . ( the diagram for conductance fluctuations is taken from ref . and modified to contain the effect of a finite ehrenfest time . ) not being a quantum interference effect , shot noise only requires wavepackets to be split , which happens after a time @xmath0 . comparison of the time when quantum effects appear ( the ` onset time ' ) and the rate of suppression of quantum effects as the ehrenfest time @xmath0 is increased , thus provides a quantitative test of the semiclassical theory . with such a quantitative test of the semiclassical theory , accurate simulations performed at smaller ehrenfest times can still be meaningful . = 0.8 in this paper , we present detailed results of such time - resolved numerical simulation for weak localization , conductance fluctuations , and shot noise . for all three quantum effects , we analyze their ehrenfest - time dependence ( without truncation of the time evolution ) and the onset times ( obtained from simulations with truncation of the time evolution ) . the simulation results are presented in sec.[kicked ] , together with the predictions of random matrix theory for time - resolved transport through open quantum maps . whereas we confirm our earlier conclusion that the numerical simulations for weak localization show a suppression @xmath15 , consistent with the semiclassical theory , we find that simulations for conductance fluctuations show a very small increase if @xmath0 is increased , the effect being small enough to be consistent with the simulation data reported in the literature.@xcite the onset times for conductance fluctuations are more than a factor two smaller than the onset times for weak localization , which is incompatible with the notion that conductance fluctuations arise as wavepackets are split and reunited . a second goal of this paper is to show how classical correlations are taken into account in the semiclassical theory of ref . . the importance of classical correlations no quantum diffraction takes place to or from classical trajectories with classical dwell time shorter than the ehrenfest time was pointed out in the analysis of simulation data for the quantum kicked rotator , most notably the simulations for shot noise.@xcite the original version of the semiclassical theory,@xcite which did not include these correlations , predicted a suppression of weak localization @xmath16 . it is only after accounting for the classical correlations that the proper exponential decay @xmath15 is recovered . ( classical correlations are taken into account correctly in the semiclassical theory of shot - noise suppression at large ehrenfest times.@xcite ) in addition it is demonstrated that this semiclassical theory is unitary , _ i.e. , _ that no probability is lost . our discussion of the semiclassical theory for weak localization can be found in sec . [ alcalculation ] . we conclude in sec . [ discussion ] . [ numerics ] despite their different dimensionality , chaotic quantum maps are believed to have the same phemonology as ( closed ) chaotic cavities,@xcite provided one restricts attention to the `` universal limit '' of times longer than the `` ergodic time '' . ( the `` ergodic time '' is the time required to explore the phase space . ) open kicked quantum maps have been used as a model for transport through chaotic cavities,@xcite because maps allow to simulate systems with many transport channels with relatively small computational effort . for short times , qualitative differences between maps and cavities exist , however , because the time evolution in maps is discrete , whereas the time evolution in cavities is not . quantum maps operate on a finite state vector @xmath17 of dimension @xmath18 . in the context of the quantum kicked rotator these states are discrete quasimomentum states or , alternatively , positions on a lattice with periodic boundary conditions . in the language of a cavity , the elements of @xmath17 can be thought of as points on the cavity boundary . the time evolution of such maps is discrete , and given by the floquet operator @xmath19 for the kicked rotator , this type of dynamics is relatively easy to simulate numerically , which is why it is used to numerically model quantum interference corrections at large ehrenfest times . in order to study transport , the system has to be opened . hereto two consecutive sets of @xmath20 and @xmath21 elements @xmath17 are chosen that corresponds to two ` leads'.@xcite the initial condition for a transport simulation corresponds to @xmath17 localized at one of the ` lead ' points . escape from the cavity is modeled by recording the amplitude of @xmath17 at the lead points at time @xmath22 and setting @xmath17 to zero afterwards . formally , this corresponds to the construction@xcite @xmath23^{-1 } e^{i \varepsilon } { \cal f } p^t,\ ] ] where @xmath24 is the scattering matrix corresponding to the map , @xmath13 is quasienergy , @xmath25 is a @xmath26 matrix projecting on the sites corresponding to the left ( l ) and right ( r ) contacts , and @xmath27 . the conductance coefficients @xmath28 , @xmath29 , @xmath30 , and @xmath31 are defined as @xmath32 where @xmath33 projects on the channels of the right lead , whereas @xmath34 projects on the channels of the left lead . unitarity implies @xmath35 . the dwell time corresponding to the map is @xmath36 . we note that , with this method of opening the map , the leads end abruptly . such abrupt changes lead to diffraction effects , similar to diffraction from sharp corners at the lead opening of a cavity . the semiclassical contributions of some diffracting orbits of this type were calculated for closed and open cavities , see , _ refs .. they were found to be an important contribution to shot noise at small channel numbers @xmath37 for rectangular cavities.@xcite however , to the best of our knowledge , the total contribution of such orbits to shot noise , weak localization , and conductance fluctuations in chaotic cavities at large @xmath37 are unknown . therefore , we compare our numerical results to the existing semiclassical theory , which neglects the effects of corner diffraction . in this section , we will study the transmission and reflection coefficients as a function of a truncation time @xmath38 . the truncation procedure involves writing eq . ( [ seps ] ) as a geometric series of a time - dependent scattering matrix @xmath39 , @xmath40 with @xmath41^{t-1 } { \cal f } p^t.\ ] ] instead of taking the full geometric series ( [ sexpandt ] ) , we truncate the series after time @xmath12 , @xmath42 and study transport properties as a function of @xmath12 . although such a truncation procedure does not represent a physical system , it allows us to get theoretical information on the times involved in quantum transport phenomena . in particular , we will verify at what truncation time weak localization and conductance fluctuations will first appear . the quantities of interest are the ensemble - averaged conductance coefficients @xmath28 , @xmath29 , @xmath30 , and @xmath31 , with @xmath24 replaced by @xmath43 , and the fluctuations of these coefficients . with the truncation procedure , @xmath24 is no longer unitary , so that one does not necessarily have that @xmath44 . hence , we need to consider all four conductance coefficients separately . if @xmath45 is time - reversal symmetric , one still has @xmath46 , and only three conductance coefficients need to be considered . for @xmath47 , unitarity is restored , and one conductance coefficient is sufficient . it is also of interest to study the @xmath12 dependence of shot noise . this also can be done by replacing the scattering matrix by @xmath43 , as was done for weak localization and conductance fluctuations . the resulting @xmath12-dependent fano factor is given by @xcite @xmath48.\ ] ] the ensemble average is taken by first averaging over the quasienergy @xmath13 , and then over various lead positions . the quasienergy average can be performed explicitly , see eq . ( [ eq : strunc ] ) above . for the fluctuations , we consider the variances of the transport coefficients , as well as covariances of different coefficients . the variances are taken with respect to the quasienergy @xmath13 , in order to ensure that the fluctuations are entirely of quantum origin.@xcite the variance or covariance for fluctuations with respect to @xmath13 is then averaged over different lead positions . again , having the explicit energy dependence ( [ eq : strunc ] ) at our disposal , the calculation of the variance or covariance with respect to variations of @xmath13 can be performed explicitly . as a reference for our numerical simulations , in which we take the floquet operator @xmath45 of the quantum kicked rotator , we consider averages and fluctuations of the transport coefficients for the case that @xmath45 is a random symmetric unitary matrix , taken from the circular orthogonal ensemble of random matrix theory . in the limit of large @xmath18 , @xmath20 , and @xmath21 , which is relevant for the semiclassical limit we consider throughout this paper , such averages can be calculated using the technique of ref . . we then find @xmath49 where the indices @xmath50 and @xmath51 can be taken to be l and r. the first term @xmath52 is the classical average , @xmath53 with @xmath54 , and @xmath55 is the weak localization correction , @xmath56 for the covariances we find similarly in the limit of @xmath18 , @xmath57 , @xmath58 , @xmath59 \rangle = 2 \frac{n_\alpha n_\beta n_\gamma l } + n_{\rm r}+1)^4 } { \cal r}_1 + 2 \frac{n_\alpha n_\beta}{(n_{\rm r}+n_{\rm l}+1)^3 } \left ( n_\gamma \delta_{\beta \delta } + n_\gamma \delta_{\alpha \delta } + n_\delta \delta_{\alpha \gamma } + n_\delta \delta_{\beta \gamma } \right ) { \cal r}_2 \\ + 2 \frac{n_\alpha n_\beta}{(n_{\rm r}+n_{\rm l}+1)^2 } \left ( \delta_{\alpha \gamma } \delta_{\beta \delta } + \delta_{\alpha \delta } \delta_{\beta \gamma } \right ) { \cal r}_3+o(m^{-1}),\end{gathered}\ ] ] where @xmath60 in eq . ( [ covariances ] ) the symbol @xmath61 denotes a covariance taken with respect to variations of the quasienergy only , whereas the brackets @xmath62 denote the ensemble average over @xmath45 . in our numerical simulations , we take the floquet operator @xmath45 of the quantum kicked rotator map . the map is described in detail by tworzydlo _ et al . _ , see ref .. the matrix elements of the floquet operator are @xmath63 with @xmath64 here @xmath65 is the so called stochasticity parameter that determines the classical dynamics of the map . the region @xmath66 is associated with classically chaotic dynamics . the parameter @xmath67 determines the precise quantization of the map and has no effect on the classical dynamics . in this model the size of the matrix @xmath18 is even . in the simulations we set @xmath68 . we used the floquet operator ( [ onekick ] ) to study conductance fluctuations and shot noise . the floquet operator ( [ onekick ] ) can also be used to study weak localization using the following argument : the average conductance @xmath69 consists of a classical contribution and a quantum correction , @xmath70 the classical contribution @xmath71 scales proportional to the channel number @xmath37 , whereas the quantum correction has no @xmath37-dependence ( except for a possible weak dependence on @xmath37 through the ehrenfest time ) . the quantum correction can be extracted by comparing average conductances at @xmath37 channels and @xmath72 channels , @xmath73 in order to avoid a spurious contribution to @xmath74 from classical conductance fluctuations , we made sure that the ensemble averages for @xmath75 and @xmath76 were taken for _ precisely _ the same classical dynamics ( same values of @xmath65 , same lead positions ) . this method was used previously in ref . . alternatively , one can study weak localization by considering maps with and without time reversal symmetry . the kicked rotator map ( [ onekick ] ) has time - reversal symmetry . a simple extension of eq . ( [ onekick ] ) that breaks time - reversal symmetry is the so - called three - kick model,@xcite @xmath77 where @xmath78 in this model @xmath18 is even , but not a multiple of @xmath79 . the kick potential is given by @xmath80 where @xmath81 breaks the parity symmetry of the model.@xcite the parameter @xmath82 plays the role of a magnetic field and breaks time - reversal symmetry . by comparing the transport for @xmath83 to that of @xmath84 it is possible to investigate the weak localization correction as a function of time . this is the method used in ref . . the advantage of the second method is that it does not require the cancellation of classical conductances of two different systems . the disadvantage is that it involves the additional parameter @xmath82 , which itself also affects the map s classical dynamics . however , weak localization is affected on the scale @xmath85 , so that the effect on the classical dynamics can be expected to be small in the semiclassical limit @xmath86 . for both models , the ehrenfest time is given by @xmath87 up to an @xmath37-independent constant . the dwell time reads @xmath88 the numerical algorithm used to iterate the maps ( [ onekick ] ) and ( [ 3kick ] ) can be accelerated by using a fast fourier transform.@xcite the results are computed as a function of time and not of quasienergy . this allows to truncate the series ( [ sexpandt ] ) at the maximal time studied , and to calculate the @xmath13-averages explicitly . _ weak localization_. the value of the time - reversal symmetry breaking parameter @xmath82 at which weak localization corrections to the conductance are suppressed in the three - kick model ( [ 3kick ] ) scales inversely proportional to the channel number @xmath37.@xcite hence , in order to compare weak localization corrections for different channel numbers , the average conductance is calculated as a function of the product @xmath89 . figure [ fig : wl1 ] shows the result of numerical simulations of the ensemble - averaged conductance . the ensemble average is taken over 1000 samples , choosing @xmath65 randomly in the interval @xmath90 and varying the lead positions . the figure also shows a lorentzian fit to the simulation data for small @xmath89 . the dwell time is set to be @xmath91 and the parity - symmetry breaking parameter @xmath92 . results for different @xmath37 are offset vertically . when determining the magnitude of the weak localization correction , we restrict our attention to the range @xmath93 , for which the @xmath89-dependence of @xmath94 exhibit a pronounced dip . the width of this dip seems to be ( almost ) independent on system size , in agreement with the predictions from random matrix theory.@xcite however , as can be seen from fig . [ fig : wl1 ] , the shape of this peak is not well approximated by a lorentzian at higher values of @xmath82 . for @xmath95 the average conductance typically continues to increase with @xmath89 , but at a much slower rate . figure [ fig : wl2 ] shows the difference @xmath96 for @xmath97 , as a function of the cut - off time @xmath12 . ( these results are averaged over 20 000 different realizations . ) the standard case ( without truncation ) corresponds to the limit @xmath98 . upon increasing @xmath37 , the figure shows a systematic decrease of @xmath99 as well as a shift of the onset of weak localization to larger truncation times . to study these numerical result quantitatively , we examine the dependence of the onset of weak localization and its magnitude on the number of channels . as an operational definition of the onset time @xmath100 , we define @xmath100 as that truncation time @xmath12 for which @xmath101 . in principle , onset times could also have been defined using @xmath102 . we prefer to use @xmath99 , since the latter are less impacted by the discreteness of the map s time evolution . the left panel of fig . [ fig : wl2b ] shows @xmath100 as a function of @xmath37 for the simulation curves shown in fig . [ fig : wl2 ] , as well as for similar curves calculated for @xmath103 ( time - resolved data not shown ) and for simulation data taken at dwell time @xmath104 . the right panel of fig . [ fig : wl2b ] shows the dependence of @xmath105 on @xmath37 . figure [ fig : wl2c ] depicts the same data as figure [ fig : wl2b ] , but for stochasticity parameter @xmath65 taken uniformly in the interval @xmath106 . according to the semiclassical theory , the derivative @xmath107 , where @xmath1 is the lyapunov exponent of the map . similarly , semiclassics predicts that @xmath108 . the classical map corresponding to eq . ( [ 3kick ] ) is described in ref . , and its lyapunov exponent is readily calculated using the method described in ref . . since the magnetic fields that break time reversal invariance are classically small , the lyapunov exponents should be computed for @xmath84 . numerically , we find @xmath109 for @xmath110 and @xmath111 for @xmath112 . lines with slopes corresponding these lyapunov exponents are shown in figs.[fig : wl2b ] and [ fig : wl2c ] . we conclude that the @xmath37-dependence of @xmath113 is well described by the semiclassical theory . the onset times deviate somewhat from the expected slope , however . since this deviation is more pronounced for the shorter dwell time we attribute it to fluctuations in the values of the lyapunov exponent . fluctuations of the lyapunov exponent have a stronger effect on quantum corrections at short dwell times than at large dwell times.@xcite = 0.9 = 0.9 = 0.9 = 0.9 although our data are obtained in the same way as in ref . , our conclusions are markedly different . whereas ref . finds no evidence of a systematic ehrenfest - time dependence of weak localization , we conclude that there is a systematic ehrenfest - time dependence of weak localization and that the simulation results for weak localization are consistent with the semiclassical theory . we attribute the differences to a lack of accuracy in the simulations of ref . . indeed , on average the simulations of ref do show a slight decrease upon increasing @xmath37 , but the statistical uncertainties are too large to rule whether the decrease is systematic or accidental . figure [ fig : wl6 ] shows time - resolved data for @xmath99 for @xmath114 , @xmath115 , @xmath116 , and @xmath65 taken uniformly in the interval @xmath90 . whereas the simulation data are consistent with the semiclassical theory for @xmath117 ( see the insets in fig . [ fig : wl6 ] ) , the simulation results for @xmath99 show significant deviations from the semiclassical theory for smaller channel numbers . this is no surprise , as the semiclassical theory is known to break down for small @xmath37 . despite the differences between the semiclassical theory and the magnitude of the simulated weak localization correction @xmath99 for small @xmath37 , the onset times appear to depend linearly on @xmath118 down to the smallest channel numbers considered in the simulation ( @xmath119 ) . = 0.9 the second method of computing weak localization uses the scaling of the conductance with the number of channels , see eq.([eq : scaling ] ) . time - resolved results for reflection and transmission coefficients are shown in fig . [ fig : wl3 ] and [ fig : wl4 ] , with @xmath65 taken randomly from the interval @xmath90 and dwell times @xmath91 and @xmath120 , respectively . for @xmath91 , an average was taken over @xmath121 samples , except for the data point at @xmath122 . due to the numerical cost of the calculations for @xmath122 only @xmath123 samples were considered for @xmath124 and @xmath125 samples for @xmath112 . for @xmath104 , the average was taken over @xmath126 samples for @xmath110 and over @xmath127 samples for @xmath112 . in all cases the statistical error of the weak localization correction was estimated to be @xmath128 . again , we compare the data in fig . [ fig : wl4b ] to lines with slopes which are fixed by the lyapunov exponent of the corresponding classical map . for this model the lyapunov exponents are @xmath129 when @xmath110 and @xmath130 when @xmath131 . as for the three - kick model , the results are consistent with the semiclassical theory , see the right panel of fig . [ fig : wl4b ] . figure [ fig : wl4b ] also contains results for @xmath65 taken uniformly in the interval @xmath106 ( time - resolved data not shown ) . = 0.9 = 0.9 = 0.9 the time - resolved data shown in figs . [ fig : wl3 ] and [ fig : wl4 ] show some structure at times smaller than the onset time . we can not rule out that this structure , which appears to persist despite statistical averaging , affects our quantitative conclusions for the onset times and weak localization corrections . since there is no such small - time structure in the weak localization data obtained by varying the field @xmath82 , we conclude that the small - time feature in the data of figs . [ fig : wl3 ] and [ fig : wl4 ] must be a non - magnetic - field - dependent quantum correction ( _ e.g. _ , resulting from diffraction at the contacts ) . the feature disappears quickly upon increasing the stochasticity parameter @xmath65 . the largest channel numbers in our simulations @xmath132 and @xmath122 for the scaling method , with an average over @xmath121 and @xmath127 samples ( for @xmath110 ) , respectively are smaller than the largest channel number used in the simulations ref . . the observed decrease of the weak localization correction is statistically significant and systematic , but small . we could not obtain sufficiently accurate simulation data for higher number of channels . increasing @xmath37 at fixed numerical cost implies that the number of realizations in the average has to scale @xmath133 , so that the statistical error scales @xmath134 . obtaining simulation data for @xmath135 at the same numerical cost as our @xmath136 data , would mean that only @xmath137 samples can be averaged over , increasing the error by a factor @xmath138 . at that point the statistical error becomes comparable to the expected incremental decrease of the weak localization data , and the simulation data loose statistical significance . _ shot noise . _ figure [ fig : sn1 ] shows time - resolved data for the shot noise fano factor @xmath139 , taken for the one - kick model with stochasticity parameter @xmath110 and dwell time @xmath140 . the statistical average was taken over @xmath141 realizations only , which is sufficient as the fano factor is self - averaging for large channel numbers . a more quantitative analysis of the data for @xmath110 ( circles ) , @xmath112 ( squares ) , @xmath142 ( solid lines ) , and @xmath143 ( dashed lines ) is presented in the insets . onset times , determined from the maximum of the @xmath144 graphs , are shown in the left inset of fig . [ fig : sn1 ] , together with a linear fit @xmath145 , using the lyapunov exponents obtained from the corresponding classical map . the time of the maximum was defined as onset time since it is problematic to obtain reliable onset times using the definition used previously . however , numerically obtained onset times for shot noise are significantly impacted by the discreteness of the map s time evolution , and do not show a smooth linear increase with @xmath37 over the range of channel numbers investigated in our simulations . the right inset of the figure shows values of the fano factors for the limit of large truncation times , together with fits @xmath15 , with the lyapunov exponents @xmath1 obtained from the corresponding classical map . we conclude that the rate of exponential suppression of shot noise upon increasing @xmath37 is consistent with the rate of exponential suppression of weak localization . = 0.9 = 0.85 = 0.85 = 0.85 = 0.9 _ conductance fluctuations . _ we performed time - resolved simulations for the variance of the reflection from the left contact @xmath28 , the covariance of reflection @xmath28 and transmission @xmath30 , and the covariance of reflections from the left and right contacts . we have computed the variance of the conductance coefficient @xmath28 , as well as the covariances @xmath146 and @xmath147 as a function of the truncation time @xmath12 . unitarity of the scattering matrix implies that @xmath148 if @xmath98 , but this equality does not hold for finite @xmath12 . our simulations were done for dwell times @xmath91 and @xmath149 , and for stochasticity parameters @xmath65 between @xmath149 and @xmath150 and between @xmath141 and @xmath151 . results for @xmath91 and @xmath110 , which are representative for all results obtained , are shown in fig . [ fig : ucf1 ] . the number of realizations used for the averages shown here is @xmath121 for @xmath91 , except for the @xmath135 data point , for which only @xmath127 realizations were taken . the statistical error of the variances was estimated to be @xmath152 . onset times calculated from the time - resolved covariance of @xmath28 and @xmath30 are shown in fig . [ fig : ucf3 ] , together with conductance covariances . the scatter of the data points in the right panel of fig.[fig : ucf3 ] for the short dwell time @xmath91 is larger than our statistical error and reproducable . it is probably an artefact of the discrete time evolution of the map . ( note that similar scatter exists for the weak localization data , although , in that case , it is obscured by the large systematic decrease of the weak localization correction upon increasing @xmath37 . ) as can be seen from the left panel of fig . [ fig : ucf3 ] , the onset time depend linearly on @xmath118 : @xmath153 for @xmath110 and @xmath154 for @xmath155 . however , the slopes are a factor @xmath156 smaller than for the weak localization data , both for @xmath110 and @xmath157 . further , the conductance variance ( _ i.e. , _ the large-@xmath12 limit of the data shown in fig . [ fig : ucf1 ] ) shows a slight _ increase _ with increasing @xmath37 . the total increase is less than @xmath158 over the range of channel numbers considered in our simulations and falls within the statistical uncertainty of simulations reported in refs . , where it was concluded that @xmath159 is independent of @xmath37 . simulations of conductance fluctuations for the three - kick model slow a less than 10% decrease of @xmath159 for the same range of @xmath37 , as well as onset times that increase significantly slower with @xmath37 than the onset times for weak localization in the three - kick model ( data not shown ) . clearly , the simulation data for conductance fluctuations are qualitatively different from the simulation data for weak localization ; they differ both with respect to the @xmath37-dependence of the onset times and the @xmath37-dependence of the magnitude of the quantum corrections . this contradicts the notion that the same interference processes ( diffusons and cooperons and their generalizations to ballistic systems ) underly both weak localization and conductance fluctuations , although one should note that there is no semiclassical theory of the ehrenfest - time dependence of conductance fluctuations yet . a semiclassical theory for the ehrenfest - time dependence of the weak localization correction in chaotic cavities was first formulated by aleiner and larkin.@xcite that theory predicts that the weak localization correction is suppressed @xmath16 . however , the analysis of ref . does not account for all classical correlations : it neglects the notion that no quantum diffraction takes place for electrons that spend less than a time @xmath0 in the cavity . in this section we show that accounting for all classical correlations gives a weak localization correction that still depends exponentially on the ehrenfest time , but with a different exponent : @xmath160 . while a part of our calculations has already appeared elsewhere,@xcite this section includes a calculation of the full conductance matrix @xmath161 . this allows to verify that unitarity is preserved . as the possible loss of unitarity is known to be a problem in semiclassical theories , it is important to demonstrate explicitly that the method used here preserves probability . the system under consideration is a ballistic cavity with chaotic classical dynamics . the cavity is two - dimensional , and it is coupled to two electron reservoirs via contacts of width @xmath162 and @xmath163 , see fig . ergodicity of the electron motion inside the dot is ensured by the condition @xmath164 . the cavity and the contacts are considered in the semiclassical limit @xmath165 , where @xmath8 is the electron wavenumber . . the cavity is attached to two leads , labeled `` l '' and `` r '' , and with width @xmath162 and @xmath163 , respectively . the contours @xmath166 , @xmath34 , and @xmath33 drawn in the leads are used for the calculation of transmission and reflection coefficients , see the text . [ dot1 ] ] thus , one has the separation of length scales @xmath167 or , equivalently , the separation of time scales @xmath168 , where @xmath169 is the electron velocity , @xmath170 the ergodic time , and @xmath4 the dwell time . the latter can be expressed as @xmath171 where @xmath172 is the area of the cavity . we also define @xmath173 as the probabilities that an electron at a random point in the cavity will leave through the right and left contacts , respectively . electron motion inside the cavity is assumed to be fully phase coherent and electron - electron interactions are neglected . the classical motion in the cavity is assumed to be fully chaotic ( when closed ) , with a lyapunov exponent @xmath1 . this exponent measures the rate of divergence of close phase space points . since this divergence is due to scattering from the curved walls of the cavity one can expect that @xmath174 . we neglect fluctuations in the lyapunov exponent and assume that its value is uniform . this is done to simplify the calculation as much as possible and should not affect the main physical conclusions as long as relevant times are much longer than @xmath2 . the aleiner - larkin formalism is based on a semiclassical description of transport . central objects in their theory are the non - oscillating parts of the product of an advanced and retarded green functions . the `` diffuson '' is composed of combinations of orbits with themselves ( known as the diagonal approximation in the semiclassical community ) while the `` cooperon '' is composed of orbits and their time reversal counterparts . the equation of motion for the diffuson , neglecting quantum phenomena , is the liouville equation , @xmath175 { \cal d}^0 ( \omega ; 1 , 2 ) = \delta(1,2),\ ] ] where @xmath176 denotes the phase space coordinates , limited to the energy shell : @xmath177 is the electron position while the angle @xmath67 represents is the direction of the electron s velocity @xmath178 . the operator @xmath179 is the liouville operator , taken with respect to the coordinates @xmath180 . for a hard wall cavity , @xmath181 , with appropriate boundary conditions at the walls . the symbol @xmath182 denotes a delta function on the energy shell , that is , @xmath183 . quantum effects are introduced by including a weak additional random quantum potential , treating it as a perturbation , and averaging over its realizations.@xcite this random potential leads to two different effects , which are expected in a quantum theory . the first is a diffusion in phase space , which mimics the fact that quantum dynamics can not separate small phase space structures . the second effect is quantum switching between close phase space trajectories , which results in weak localization . the strength @xmath184 of these effects is then set so that quantum corrections will appear after the correct time , the ehrenfest time @xmath0 . phase space diffusion modifies the evolution equation for the leading order diffuson as@xcite @xmath185 { \cal d}^0 ( \omega ; 1 , 2 ) = \delta(1,2).\ ] ] the strength @xmath184 of the phase space diffusion term is estimated to be @xcite @xmath186 . it is instructive to note that this term , although small , breaks the time reversal invariance which is a property of _ both _ quantum mechanics and classical mechanics ( in absence of a magnetic field ) . in the presence of quantum diffraction , the leading order cooperon , @xmath187 , also satisfies eq . ( [ diffusonw ] ) . in addition to the presence of the phase space diffusion term in the evolution equation ( [ diffusonw ] ) , quantum diffraction leads to an quantum correction to the diffuson : @xmath188 . this quantum correction is referred to as weak localization , because similar quantum corrections are the precursor of anderson localization in disordered systems . the weak localization correction for the diffuson reads@xcite @xmath189 \frac{{\cal c}^0 ( \omega ; 3,\bar{3})}{2 \pi \hbar \nu},\end{gathered}\ ] ] where @xmath190 denotes the time reversal of phase space point @xmath191 and @xmath192 is the density of states per unit area . in terms of classical trajectories , the weak localization correction @xmath193 corresponds to a configuration as shown in fig.[fig:1]a : @xmath193 is the correction to the product of a retarded and advanced green function arising from the combination of a trajectory that intersects itself at a small angle and a trajectory that avoids intersecting itself.@xcite if the intersection angle is sufficiently small , such a pair of trajectories has strongly correlated actions , resulting in a net quantum correction to the diffuson even after ensemble averaging . the pairs of trajectories before and after the ` interference region ' are represented by the leading - order diffusons @xmath194 in eq . ( [ dd ] ) , whereas the trajectories in the closed loop an orbit conjugate to its time - reversed counterpart are represented by the leading order cooperon @xmath187 . in order to calculate the weak localization correction to the conductance , the dot s conductance coefficients @xmath28 and @xmath195 are expressed in terms of the diffuson @xmath196 , @xmath197 and @xmath198.\ ] ] here @xmath34 , @xmath33 , and @xmath166 are smooth contours following a cross section of the contacts , see fig . the contour @xmath166 is taken a small amount further away from the cavity than the contour @xmath34 , see fig.[dot1 ] . the angles @xmath199 and @xmath200 denote the direction of the outward - pointing normal to @xmath34 , @xmath201 , whereas the angle @xmath202 represents the direction of the inward - pointing normal to @xmath166 . since we consider dc transport , the frequency @xmath203 has been set to zero . we refer to the appendix for a derivation . similar expressions , but with an ambiguity in the precise definitions of the integrals over the lead cross sections , were derived by takane and nakamura.@xcite the leading order `` classical '' transmission and reflection coefficients are found by substituting @xmath194 for @xmath196 . assuming that dynamics in the cavity is ergodic for all relevant time scales , one finds@xcite @xmath204 where the escape probabilities to the left and right leads were defined in eq . ( [ eq : plr ] ) above . the weak localization correction to the conductance coefficients @xmath29 and @xmath28 is obtained by substituting eq . ( [ dd ] ) , for the weak localization correction @xmath193 of the diffuson , into eqs . ( [ tott ] ) and ( [ totr ] ) . we first consider the weak localization correction to @xmath29 . the first two terms in ( [ dd ] ) contain paths which leave the dot at the point of entry , and hence these do not contribute to the transmission . hence @xmath205 { \cal c}^0 ( 3,\bar{3 } ) . \ ] ] the phase space point 3 can be viewed as the center of the interference region as depicted in fig . [ fig:1]a . the two diffusons and the cooperon head into opposite directions in phase space , so that they will sample different parts of phase space along the trajectory . thus , one can treat them as statistically uncorrelated.@xcite both the cooperon and the product of diffusons in eq . ( [ deltat ] ) exhibit non - trivial correlations since the returning path is strongly correlated with the outgoing one , cf . [ fig:1]a . in order to take into account these correlations one should examine what is meant by the phase space points @xmath79 and @xmath206 in eq . ( [ deltat ] ) . by virtue of the phase - space diffusion term in the evolution equation for the leading order diffuson , effectively , the points @xmath79 and @xmath207 do not need to be exactly time reversed phase space points , since quantum mechanics allows for a finite phase space uncertainty . an intuitive way to calculate the correlations between @xmath79 and @xmath206 was shown by vavilov and larkin.@xcite instead of using the diffusion in phase space , vavilov and larkin average over a range of initial phase space points near @xmath79 and final phase space points near @xmath206 . they show that the effect of the phase space diffusion is equivalent to such averaging ( up to logarithmic corrections ) if the ` size ' of phase space area near @xmath79 and @xmath206 scales as @xmath208 . [ to be precise , the range of coordinates @xmath209 in the average is @xmath210 , whereas the angular range is @xmath211 . ] we adopt the procedure of vavilov and larkin and consider an average over phase space points @xmath212 and @xmath213 that are within a distance of order @xmath208 of the phase space point @xmath79 . we can then follow the classical orbits which start at the points @xmath212 , @xmath213 . the phase space distance between these trajectories will diverge exponentially due to the chaotic dynamics . since the trajectories start from phase space points at a distance @xmath214 , it will take a time @xmath215 to reach a phase space distance of order @xmath9 . once the phase space distance between the trajectories is large enough they can be considered as totally uncorrelated . at this point , ergodic dynamics can be assumed . the loss of correlations happens on a time scale of @xmath216 around the time @xmath215 . since @xmath217 in the parameter regime where the ehrenfest time affects quantum transport , we can view this process as effectively instantaneous . let us first consider the cooperon in eq . ( [ deltat ] ) . as a result of the phase space correlations described above , one can not close the orbit from @xmath218 to @xmath213 for short times . in order to make this more quantitative , we denote by @xmath219 the time it takes for the classical orbit starting at phase space point @xmath191 to leave the cavity through one of the two openings . then , if @xmath220 we find @xmath221 . on the other hand , if @xmath222 we can propagate the phase space point @xmath223 for a time @xmath215 and reach phase space point @xmath138 . similarly , we can propagate @xmath213 backward in time and find , toward the phase space point @xmath224 . this leads to @xmath225 when the phase space points @xmath212 and @xmath213 are averaged over , the cooperon will have contributions from various phase space points @xmath226 , which are their distant past ( or future ) . the points @xmath138 and @xmath224 can be taken to be uncorrelated and sample the phase space with uniform probability . since the phase space point @xmath79 is eventually integrated over , there will be contributions from a sizable fraction of phase space . thus , we approximate the average contribution to the cooperon by replacing @xmath227 by its average value @xmath228 . this leads to @xmath229 where @xmath230 if @xmath231 and @xmath232 otherwise . we address the product of two diffusons in eq.([deltat ] ) by considering the integral @xmath233 again , the integration over phase space points @xmath79 and @xmath206 is replaced by an average over phase space points @xmath212 and @xmath213 within a distance of order @xmath234 from 3 . since the diffuson connects the phase space points @xmath212 and @xmath213 with points at two different contacts , the product of diffusons must be zero if @xmath235 where , as before , @xmath236 is the time it takes for the orbit at phase space point @xmath79 to leave the system . for larger @xmath236 we may , again , assume ergodicity , so that we find , after averaging over @xmath212 and @xmath213 , @xmath237 combining results , we find @xmath238 the liouville operator in eq . ( [ eq : deltat2 ] ) measures the rate of flow of probability density out of the integration range of the phase space variable @xmath79.@xcite the boundary of the integration range is composed of the lead phase space points propagated _ backward _ for a time @xmath239 . this leaves only a fraction @xmath240 of the size of the boundary at the lead . to estimate the integral we assume that each outgoing direction is equally likely to be in the system when propagated backward . however , only points which also have @xmath222 will have a non - vanishing cooperon , leading to an additional factor of @xmath240 . collecting contributions from both leads , we find @xmath241 it is of interest to compute also the weak localization correction @xmath242 . this will allow to check that probability is conserved . the calculation is similar to that of the @xmath243 , with two important differences . the first difference is that the weak localization correction of reflection is composed from two parts . in addition to the third term in eq . ( [ dd ] ) , there is a contribution from the second term in eq . ( [ dd ] ) . we write these two parts as @xmath244 and @xmath245 , respectively , and calculate them separately . substitution of the third term in eq . ( [ dd ] ) into eq . ( [ totr ] ) gives @xmath246 { \cal c}^0 ( 3,\bar{3 } ) . \label{deltar1}\end{gathered}\ ] ] the product of the two diffusons can be assumed to be uncorrelated with the cooperon . the calculation of the cooperon proceeds as for the transmission calculation . however , the behavior of the product of the diffusons contributing to reflection differ from that of the diffusons contributing to transmission . to see that , again consider the integral @xmath247 if @xmath248 both diffusons become uncorrelated before leaving the system and the average value of this integral is therefore @xmath249 . however , if @xmath250 both diffusons will exit the dot through the _ same _ lead , and one finds @xmath251 . hence , @xmath252 this is the second difference between the calculations of @xmath253 and @xmath242 . we then find @xmath254 the contribution @xmath245 is obtained by substituting the second term in the right hand side of eq . ( [ dd ] ) into eq . ( [ totr ] ) . the lead integral over @xmath255 can be calculated , resulting in @xmath256 combining eqs . ( [ eq : dr2 ] ) and ( [ eq : dr1 ] ) , one finds @xmath257 , as expected . the result ( [ eq : dtresult ] ) shows that , according to the semiclassical theory , the weak localization correction to transmission is suppressed exponentially . however , the exponent we find is different from that of ref . , where it is reported that @xmath258 . the reason for the difference with ref . is that our calculation obeys the classical correlations following from the separation of phase space into a ` classical ' and ` quantum ' part , corresponding to ( classical ) trajectories of length smaller or larger than @xmath0 , respectively . quantum diffraction does not involve the ` classical ' part of phase space . calculating the cooperon and the product of diffuson propagators assuming ergodic dynamics in the ` quantum ' part of phase space only increases the weak localization correction @xmath259 by a factor @xmath260 with respect to the calculation of ref . . the reason that @xmath259 remains exponentially small but with exponent @xmath261 , not @xmath262 is that , according to the semiclassical theory , weak localization requires a minimal path length of @xmath263 . the fraction of ` quantum ' trajectories that remain inside the cavity during the time period @xmath14 is exponentially small , @xmath264 , hence the exponentially small weak localization correction if @xmath5 . sofar we have considered the weak localization correction to the transmission . what about other quantum interference effects , such as the transmission fluctuations ? takane and nakamura have extended the semiclassical theory of aleiner and larkin to the conductance variance @xmath159 , but for the limit @xmath265 only.@xcite they found @xmath266 , in agreement with predictions from random matrix theory.@xcite just as the semiclassical theory for the weak localization correction @xmath243 was simpler than the semiclassical theory for @xmath242 ( see above ) , the semiclassical theory of transmission fluctuations takes its simplest form if applied to the covariance of reflection from the right contact @xmath31 and reflection from the left contact @xmath28 . of course , unitarity implies @xmath267 . in ref . , only one contribution to the conductance covariance @xmath147 was considered . this contribution corresponds to the four trajectories shown in fig . [ fig:1]b . for these four trajectories , a minimal dwell time @xmath14 is required : classical trajectories originating from each of the openings need to diverge and reunite , each of which takes a time @xmath0 . hence , following the same phase space arguments as for the weak localization correction , we anticipate that their contribution to the conductance fluctuations depends on the ehrenfest times as @xmath268^{1/2 } & = & [ \mbox{var}\ , g]^{1/2 } \propto e^{-\tau_{\rm e}/\tau_{\rm d } } \nonumber \\ & & \ \ \mbox{if $ \tau_{\rm e } \gg \tau_{\rm d}$}.\end{aligned}\ ] ] however , there may be contributions to the conductance fluctuations other than that of fig . [ fig:1]b . ( this possibility can not be excluded on the basis of ref . . ) for example , there may be trajectories that involve small - angle intersections of three trajectories at the same point.@xcite while it is not expected that the inclusion of such trajectories undo the suppression of the conductance fluctuations in the limit @xmath5 , they may have a non - negligible contribution for @xmath269 . a final note about the results presented in this section : in the calculation performed in this section we have replaced dynamical functions by their averages . for example , the probability that a wave packet will exit through a given lead is ( almost ) @xmath270 or @xmath232 for short enough times , depending on the classical dynamics , not @xmath271 or @xmath272 . thus , we have actually estimated the _ ensemble _ average of the weak localization correction . one can expect that if a system is close to the classical limit there will be more fluctuations in the values of diffusons and cooperons . this may lead to large deviations of the weak localization correction from that of systems with small ehrenfest times , and to large classical conductance fluctuations.@xcite in order to describe these fluctuations quantitatively , one needs a sample - specific theory for the dynamics of the diffuson , which covers the times of order @xmath0 exactly . while this is an important theoretical problem , it is beyond the scope of this paper , and should not affect the eventual exponential suppression of quantum interference phenomena for large ehrenfest times . in sec . [ numerics ] , we reported results of numerical simulations for weak localization , conductance fluctuations , and shot noise of the open quantum kicked rotator . the numerical simulations for weak localization and shot noise are consistent with an exponential suppression @xmath273 , in quantitative agreement with the semiclassical theory of sec . [ alcalculation ] . the numerical simulations for conductance fluctuations show a small increase with increasing ehrenfest time , not inconsistent with previous simulation data reported in the literature.@xcite our simulations also give information on the minimal time at which quantum effects occur . for weak localization and shot noise , these times are @xmath274 and @xmath0 , respectively , consistent with semiclassical theory . for conductance fluctuations , the onset time is less than half the onset time of weak localization . ( note that it is possible that some contribution for weak localization in _ reflection _ may appear after a time of @xmath0 . ) in order to explain the numerical simulations of shot noise , weak localization , and universal conductance fluctuations , the authors of refs . proposed a phenomenological alternative to the standard semiclassical theory , referred to as ` effective random matrix theory ' . guiding principle for the effective random matrix theory is that quantum diffraction takes place between trajectories with dwell times larger than @xmath0 only ; trajectories with dwell time shorter than @xmath0 build scattering states with transmission ( exponentially close to ) @xmath232 or @xmath270 and do not contribute to shot noise , weak localization , or universal conductance fluctuations . the importance of ` quantum ' trajectories is described by an effective number of ` quantum channels ' @xmath275 . following silvestrov _ et al._,@xcite it was then proposed that the scattering of ` quantum channels ' is described by random matrix theory as long as @xmath276 . the condition @xmath276 is generically met , even if @xmath5 . the effective random matrix theory not only correctly predicts the suppression of the ensemble - averaged shot noise at large ehrenfest times shot noise is proportional to the number of channels @xmath37 , which is replaced by @xmath277 in the effective random matrix theory , it also describes sample - specific deviations from the ensemble average that arise from the classical dynamics.@xcite the effective random matrix also has had remarkable success explaining other observables that are proportional to the channel number @xmath37 , such as the density of transmission eigenvalues,@xcite as well as the density of states in a chaotic cavity coupled to a superconductor.@xcite on the other hand , quantum - interference effects , such as weak localization and conductance fluctuations are independent of @xmath37 , and , hence , are predicted to be independent of the ehrenfest time . thus , for weak localization , the effective random matrix theory differs from the semiclassical theory , which predicts a suppression @xmath273 . the effective random matrix theory and the semiclassical description of transport not only disagree regarding the magnitude of the weak localization correction to the conductance , they also disagree with regard to the minimal time required for quantum interference effects to occur . in the semiclassical theory , quantum interference requires a minimal wavepacket to be split _ and _ reunited , which takes a minimal time @xmath14 . this is in contrast to the effective random matrix theory , where quantum interference is fully established already after a time @xmath0 . interestingly , there is no difference between semiclassics and effective random matrix theory for shot noise : not being a quantum interference effect , shot noise only requires wavepackets to be split , which happens after a time @xmath0 in both theories . these two differences between the semiclassical theory for weak localization and the effective random matrix theory are not unrelated . _ both _ theories are consistent with a fully classical description of electron dynamics for the first time interval of length @xmath0 after the electron has entered the cavity . differences between the two theories appear for electrons that escape from the cavity in the second time interval of length @xmath0 ( _ i.e. , _ for times between @xmath0 and @xmath14 ) . in the semiclassical theory , electrons that escape during this interval behave quantum mechanically but do not contribute to weak localization . as explained in sec . [ alcalculation ] , the escape of electrons between @xmath0 and @xmath14 is responsible for the suppression of weak localization with exponent @xmath261 in the semiclassical theory . in the effective random matrix theory , weak localization sets in as soon as the classical description fails , one ehrenfest time after the electron enters the cavity ; there is no time interval in which electron dynamics is neither fully classical nor described by random matrix theory . a ` microscopic theory ' supporting the effective random matrix theory and its prediction of ehrenfest - time independent weak localization and conductance fluctuations appeared recently.@xcite this theory is based on the assumption that electron dynamics in the quantum trajectories ( _ i.e. , _ trajectories with dwell times larger than @xmath0 ) is fully ergodic , with fully established quantum interference corrections . such an assumption violates the semiclassical picture of weak localization , in which the quantum correction only appears after a time @xmath14 . hence , a theory which tries to justify the effective random matrix theory must provide an entirely new semiclassical model for weak localization , in which the quantum correction appears after one ehrenfest time only . although , at present , there is no theory that explains all simulation data , we can conclude that our time - resolved simulation data for the quantum interference corrections to the conductance are not consistent with the effective random matrix theory . first , because the effective random matrix theory predicts onset time @xmath0 for weak localization , as well as an ehrenfest - time independent weak localization correction , both of which are ruled out by our simulations . second , because simulation results for weak localization and conductance fluctuations are qualitatively different , whereas the ` effective random matrix theory ' predicts equal onset times and ehrenfest - time dependences for weak localization and conductance fluctuations . of course , the question why numerical simulations for weak localization and conductance fluctuations in the open quantum kicked rotator are qualitatively different remains . clearly , this question can not receive a final answer as long as there is no semiclassical theory for the ehrenfest - time dependence of conductance fluctuations . given the difficulty of obtaining accurate numerical results in the regime @xmath5 , such a semiclassical theory needs to include the parameter range @xmath269 if a valid comparison with the numerical simulations is to be made . we would like to thank c. beenakker , s. fishman , h. schomerus , p. silvestrov , and d. ullmo for discussions . this work was supported by the nsf under grant no . dmr 0334499 and by the packard foundation . in this appendix the transmission and reflection through a quantum dot are derived in terms of lead integrals of a diffuson . similar expressions were derived previously by takane and nakamura.@xcite however , their expressions are ambiguous with respect to the exact locations of the cross sections in the leads . in our approach , all ambiguities are resolved . to calculate the total transmission and reflection in terms of diffusons , it is useful to obtain expressions for scattering matrix elements using the retarded green function of the cavity . exact expressions of this type were derived in refs .. the leads have a uniform cross section , so that the lead wavefunction can be decomposed into free waves along the lead ( denoted by @xmath278 coordinates ) and a basis of transverse wavefunctions @xmath279 . the same decomposition can also be used for the retarded green function ( at the fermi energy ) , and one defines @xmath280 for coordinates @xmath281 and @xmath282 in the leads . we use the convention that different leads are assigned different modes , so that eq . ( [ eq : g+ ] ) remains meaningful if @xmath281 and @xmath282 are in different leads . using the asymptotics of @xmath283 and of the scattering states , the green function in the leads can be written in terms of scattering matrix elements @xmath284 . one finds@xcite @xmath285 , \end{gathered}\ ] ] for @xmath286 with @xmath287 in the left lead , and to @xmath288 for @xmath289 in the right lead while @xmath290 is in the left lead . ( in the following we use units where @xmath291 . ) for the following discussion it is important to emphasize the condition @xmath286 , which appears when both @xmath289 and @xmath290 are in the same lead . one of the steps in the derivation of eq . ( [ leadg1 ] ) involves an integral over a finite domain ( bounded by the cross - section @xmath292 ) , where the integrand is proportional to @xmath293 . this integral will give different results depending whether @xmath281 is in the domain of integration ( @xmath286 ) or not ( @xmath294 ) . this leads to the inequality in eq . ( [ leadg1 ] ) . later , when we write @xmath295 in terms of cross section integrals , two different cross sections , at @xmath289 and @xmath290 , are obtained . to compute the conductance coefficients @xmath28 and @xmath30 , the absolute value squared of the matrix elements is needed . it can be expressed in terms of green function with the help of @xmath296^ { * } \right|_{\begin{subarray}{l } x_1=x_4 \\ x_2=x_3 \end{subarray } } = \\ \left\ { \begin{array}{ll } 4m^2 |s_{mn}|^2 & \mbox { if } x_1 \in r , x_2 \in l \\ 4m^2 \left ( \delta_{mn } - |s_{mn}|^2 \right ) & \mbox { if } x_1,x_2 \in l , x_1>x_2 . \end{array } \right.\end{gathered}\ ] ] the conductance coefficients @xmath28 and @xmath30 are then calculated using eq . ( [ conductances ] ) . one can use eq . ( [ derivg ] ) to simplify the summations , since the lead modes enter into @xmath297 via cross section integrals over the leads . one then uses @xmath298 replacing this approximate finite width delta function by an exact one leads , after a straightforward calculation , to @xmath299^ { * } \right|_{\begin{subarray}{l } { \bf r}_1={\bf r}_4 \\ { \bf r}_2={\bf r}_3 \end{subarray } } \\ g_{\rm ll } & = -\frac{1}{4 m^2 } \int_{c_{\rm l } } dy_1 \int_{c_{\rm l } ' } d y_2 \left ( \frac{\partial}{\partial x_1 } - \frac{\partial}{\partial x_4 } \right ) \nonumber \\ & \times \left ( \frac{\partial}{\partial x_2 } - \frac{\partial}{\partial x_3 } \right ) \left . { \cal g}^{+ } ( { \bf r}_1,{\bf r}_2)\left [ { \cal g}^{+ } ( { \bf r}_4,{\bf r}_3)\right]^ { * } \right|_{\begin{subarray}{l } { \bf r}_1={\bf r}_4 \\ { \bf r}_2={\bf r}_3 \end{subarray } } \label{leadint2}.\end{aligned}\ ] ] this expressions differ from those in ref . in that the integrals for the reflection are along separate cross sections , with the cross section @xmath34 closer to the cavity than @xmath166 . on each of the cross sections there are two points which are almost identified : @xmath300 and @xmath301 . therefore , it is natural the use the following fourier transform @xcite @xmath302 where @xmath303 and @xmath304 . note that @xmath305 and @xmath306 are on the cross sections of the lead . energy conservation can be used to simplify this probability density further . this is done by defining @xmath307 substitution of ( [ kd ] ) and ( [ defdiff ] ) in ( [ leadintforrt ] ) will lead to the expressions for @xmath29 and @xmath28 in terms of the diffuson . the calculation is straightforward . some care is needed only when one determines the boundary conditions . for @xmath29 , only diffusions that go into the cavity can get from the left lead into the right lead . these diffusons can arrive to the right lead only if they originate from the cavity . these considerations result in the theta functions in eq . ( [ tott ] ) . the calculation for @xmath28 is very similar . the only difference is that both incoming and outgoing diffusons will cross the inner ( @xmath34 ) cross section . however , _ all _ the incoming probability flux from @xmath166 must cross @xmath34 . this incoming flux just cancels the @xmath20 factor in ( [ leadint2 ] ) . the remaining contribution results in eq . ( [ totr ] ) .

appropriately normalized square random vandermonde matrices based on independent random variables with uniform distribution on the unit circle are studied . it is shown that as the matrix sizes increases without bound , with respect to the expectation of the trace there is an asymptotic-distribution , equal to that of a$]-valued r - diagonal element . research supported in part by nsf grant dms-1301604 . this work is included in the thesis of m. boedihardjo for partial fulfillment of the requirements to obtain a ph.d . degree at texas a&m university ] research supported in part by nsf grant dms-1202660 . ]

1602.02815

we consider the random vandermonde matrix @xmath2 , whose @xmath3-th entry is @xmath4 , where @xmath5 are independent with haar measure distribution on the unit circle . these have been studied in @xcite , @xcite , @xcite and @xcite and are of interest for applications in finance , signal array processing , wireless communications and biology ( see @xcite for references ) . in @xcite , ryan and debbah show that asymptotic moments of @xmath6 , ( namely , the limits @xmath7 where @xmath8 is the expectation and @xmath9 is the normalized trace on matrix algebras ) , exist and are given by sums of volumes of certain polytopes . they also compute some of these asymptotic moments . in @xcite , tucci and whiting show among other things that the asymptotic moments are given by @xmath10 for a unique measure @xmath11 on @xmath12 with unbounded support . ( this uses the stieltjes solution to the moment problem and a theorem of carleman for the former , see p. 76 of @xcite . ) further results are proved in @xcite and @xcite . g. tucci asked @xcite whether @xmath2 is asymptotically r - diagonal with respect to the expectation of the trace . in this paper , we answer tucci s question negatively , but show that @xmath2 has an asymptotic @xmath0-distribution as @xmath13 , which is in fact the @xmath0-distribution of an element that is r - diagonal over the c@xmath14-algebra @xmath1 $ ] . to be precise , we show that , for all @xmath15 and all @xmath16 , @xmath17 exists and we describe this limiting @xmath0-moment using the notion of @xmath1$]-valued r - diagonality . usual ( or scalar - valued ) r - diagonal elements are very natural in free probability theory , and have been much studied ; they were introduced by nica and speicher in @xcite . the algebra - valued version was introduced by niady and speicher in @xcite and has been further studied in @xcite . we will give the definition from @xcite , which is an easy reformulation of one of the characterizations in @xcite . the setting for algebra - valued r - diagonal elements is a @xmath18-valued @xmath0-noncommutative probability space @xmath19 , where @xmath20 is a unital inclusion of unital @xmath0-algebras and @xmath21 is a conditional expectation , namely , a @xmath18-bimodular unital projection . [ def : maxaltptn ] given @xmath15 and @xmath22 , we define the _ maximal alternating interval partition _ @xmath23 to be the interval partition of @xmath24 whose blocks are the maximal interval subsets @xmath25 of @xmath24 such that if @xmath26 and @xmath27 , then @xmath28 . for example , if @xmath29 , then @xmath30 . [ def : rdiag ] an element @xmath31 is _ @xmath18-valued r - diagonal _ if for every integer @xmath32 and every @xmath33 we have @xmath34 ( namely , odd alternating moments vanish ) and , for every integer @xmath35 , every @xmath36 and every choice of @xmath37 , we have @xmath38 where in each of the three products above , the terms are taken in order of increasing indices . note that the @xmath18-valued r - diagonality condition determines all of the @xmath18-valued @xmath0-moments @xmath39 for @xmath15 , @xmath40 and arbitrary @xmath16 , in terms of the alternating moments of even length , namely those when @xmath41 is even and @xmath28 for all @xmath42 . * contents : * the contents of the rest of the paper are as follows . in section [ sec : altmoms ] , we find asymptotics of diagonal entries of @xmath0-moments involving alternating @xmath2 and @xmath43 with certain deterministic diagonal matrices between . in section [ sec : rdiag ] , we prove our main result , characterizing arbitrary asymptotic @xmath0-moments of @xmath2 based on @xmath1$]-valued r - diagonality . in section [ sec : calcs ] , we prove results allowing the asymptotic alternating @xmath0-moments of @xmath2 found in section [ sec : altmoms ] to be computed in terms of certain integrals , we show that @xmath2 is not asymptotically scalar - valued r - diagonal , and we report the results of computations of certain @xmath1$]-valued cumulants of the asymptotic @xmath0-distribution of @xmath2 . ( details of these computations can be found in a mathematica @xcite file accompanying the arxived version of this paper . ) * notation : * on matrix algebras , @xmath44 is the normalized trace and @xmath45 is the usual trace . for partitions @xmath46 and @xmath47 of the same set , @xmath48 means their join in the lattice of all partitions of the set . we say that a set @xmath49 _ splits _ a partition @xmath50 if @xmath49 is the union of some of the blocks of @xmath50 . we write @xmath51 to mean that @xmath52 and @xmath53 are in the same block of @xmath50 . restriction _ of a partition @xmath50 to a set @xmath54 is the partition @xmath55 , and is denoted @xmath56 . if @xmath57 is a function with domain @xmath58 , then @xmath59 is the partition of @xmath58 so that @xmath60 and @xmath61 belong to the same block of @xmath59 if and only if @xmath62 . * acknowledgement : * the authors are grateful to gabriel tucci for helpful discussions about random vandermonde matrices at an early stage of this research . in this section , we investigate alternating moments in @xmath2 and @xmath43 . more specifically , we find the asymptotics of the expectations of diagonal elements of alternating moments of even length , with certain non - random diagonal matrices interspersed ( see proposition [ prop : diagexpect ] ) . let @xmath63 be the tracial state on @xmath1 $ ] given by integration with lebesgue measure . given @xmath15 , we let @xmath64 denote the lattice of all set partitions of @xmath65 . thus @xmath66 if and only if @xmath50 is a collection of disjoint , nonempty sets whose union is @xmath24 . as usual , the elements of @xmath50 are called blocks of the partition , and @xmath67 is the number of blocks in the partition . we will let @xmath68 denote the block of @xmath50 that has @xmath42 as an element . for @xmath66 and @xmath69 $ ] , let @xmath70 $ ] be defined by @xmath71 thus , @xmath72 given @xmath73 , we let @xmath74 be @xmath49 without its largest element and we let @xmath75 thus @xmath76 . naturally , we write @xmath77 for @xmath78 . for @xmath79 and @xmath73 , we write @xmath80 if and only if @xmath81 for every @xmath82 , and if this is not the case , then we write @xmath83 . then @xmath84 if @xmath85 , namely , if @xmath86 is the partition of @xmath24 into singletons , then we let @xmath87 where if @xmath88 then we let @xmath89 be the constant function @xmath90 . otherwise , if @xmath91 , then for @xmath92 $ ] we let @xmath93 and we set @xmath94 where the integral is with respect to @xmath95-dimensional lebesgue measure . [ rem : lambdapialt ] here is an alternative description of @xmath96 that is more natural . assume @xmath97 and let @xmath98 be the linear mapping given by @xmath99 then @xmath100 is injective and is onto the @xmath101-dimensional subspace @xmath102 of @xmath103 , using the convention @xmath104 . indeed , @xmath100 is injective because if @xmath105 , then @xmath106 and if @xmath107 , then @xmath108 equals @xmath109 plus @xmath110 plus some from the list @xmath111 . we see that @xmath100 maps into @xmath112 after observing @xmath113 and it is onto @xmath112 by a dimension count . note that the push - forward under @xmath100 of lebesgue measure is the @xmath101-dimensional lebesgue measure on @xmath112 corresponding to the restriction of the usual euclidean metric of @xmath103 . we consider the @xmath114-dimensional affine space @xmath115 and we let @xmath116 be the @xmath114-dimensional lebesgue measure on @xmath117 corresponding to the restriction to @xmath117 of the usual euclidean metric on @xmath103 . then @xmath100 sends @xmath118 onto @xmath119^n : s_n = t\}\ ] ] and @xmath120 is the integral of @xmath121 over @xmath122 , with respect to @xmath116 . namely , we have @xmath123 [ prop : diagexpect ] let @xmath15 and suppose @xmath124 $ ] . given @xmath125 and @xmath126 consider the deterministic @xmath127 diagonal matrix @xmath128 for @xmath92 $ ] , let @xmath129 be the least element of @xmath130 so that @xmath131 . then for all @xmath92 $ ] @xmath132 and @xmath133 furthermore , in both cases the convergence is uniform for @xmath92 $ ] , and the rate of convergence can be controlled in terms of only @xmath134 and a common modulus of continuity for @xmath135 . we have @xmath136 n^{-n}&\sum_{\{(i(1),\ldots , i(2n))\in\{1,\ldots , n\}^{2n}:i(1)=h_n(t)\}}\left(\prod_{j=1}^{2n}g_j(\frac{i(j)}n)\right ) \\ & \qquad\qquad\qquad\cdot{{\mathbb e}}(\zeta_{i(2)}^{-i(1)+i(3)}\zeta_{i(4)}^{-i(3)+i(5)}\cdots\zeta_{i(2n-2)}^{-i(2n-3)+i(2n-1)}\zeta_{i(2n)}^{-i(2n-1)+i(1 ) } ) . \end{aligned}\end{gathered}\ ] ] let us rearrange the sum by summing first over all partitions @xmath66 and then over all @xmath137 such that @xmath138 , and then over all @xmath139 such that @xmath140 , where @xmath138 means that @xmath141 if and only if @xmath142 and @xmath143 are in the same block of @xmath50 . keeping in mind that the @xmath144 are independent and @xmath145 if @xmath146 , we find that the expectation in equals @xmath147 where the summation @xmath148 is over all @xmath149 such that @xmath150 if @xmath151 , and @xmath152 is the set of all @xmath153 such that @xmath140 and for all @xmath73 , @xmath154 with the convention @xmath155 . it is straightforward from the theory of riemann integration to see @xmath156 and that the rate of convergence depends only on @xmath157 and on a common modulus of continuity of @xmath158 now , to analyze the last summation in , we have a bijection @xmath159 given by @xmath160 where @xmath161 is the set of all @xmath162 such that @xmath163 let @xmath164 be the set of all maps @xmath165 , where @xmath166 is as in , such that @xmath167 then the map @xmath168 , given by restriction , i.e. , @xmath169 , is a bijection . moreover , if @xmath170 and if @xmath171 is the corresponding element of @xmath152 , then for all @xmath172 we have @xmath173 so also @xmath174 note that @xmath175 is an evenly spaced grid of points that fills out the set @xmath176 as @xmath13 and that the above sums , when divided by @xmath177 , are riemann sums for multivariate riemann integrals . using standard estimates , we find @xmath178 and that the rate of convergence depends only on @xmath179 and on a common modulus of continuity of @xmath180 . this proves , with the desired statement on the rate of convergence . we prove similarly . we have @xmath181 n^{-n}&\sum_{\{(i(1),\ldots , i(2n))\in\{1,\ldots , n\}^{2n}:i(1)=h_n(t)\}}\left(\prod_{j=1}^{2n}g_j(\frac{i(j)}n)\right ) \\ & \qquad\qquad\qquad\cdot{{\mathbb e}}(\zeta_{i(1)}^{-i(2n)+i(2)}\zeta_{i(3)}^{-i(2)+i(4)}\cdots\zeta_{i(2n-1)}^{-i(2n-2)+i(2n ) } ) . \end{aligned}\end{gathered}\ ] ] the right - hand - side can be rewritten @xmath182 where the summation @xmath148 is over all @xmath149 such that @xmath183 and @xmath150 if @xmath151 , while @xmath184 with the convention @xmath185 . we see @xmath186 we use the bijection @xmath187 given by @xmath188 where @xmath189 is the set of all @xmath190 such that @xmath191 we let @xmath192 be the set of all maps @xmath193 such that @xmath194 and we note that the map @xmath195 given by restriction , namely , @xmath196 , is a bijection . moreover , if @xmath197 and if @xmath171 is the corresponding element of @xmath198 , then for all @xmath199 we have @xmath200 thus , in particular , @xmath201 . hence , we also have @xmath202 we have that @xmath203 is an evenly spaced grid of points that fills out the set @xmath204 as @xmath13 and that the above sums , when divided by @xmath205 , are riemann sums . using standard estimates , we find @xmath206 and that the rate of convergence depends only on @xmath207 and on a common modulus of continuity of @xmath158 . this proves , with the desired statement on the rate of convergence . in this section , we prove our main theorem ( theorem [ thm : xrdiag ] ) about asymptotic @xmath0-moments of random vandermonde matrices . it will follow from proposition [ prop : diagexpect ] above , about alternating moments , and the next proposition . [ oneprop ] let @xmath208 . let @xmath209 . let @xmath210 be the interval partition of @xmath24 defined by @xmath211 for @xmath212 . let @xmath213 be deterministic diagonal @xmath127 matrices of norm at most 1 . then @xmath214 where @xmath215 depends only on @xmath41 . we begin with some preliminaries . the following lemma can be proved using gaussian elimination , for instance . [ p3 ] let @xmath216 . let @xmath217 be a subspace of @xmath218 . let @xmath219 . then @xmath220 lemma [ p3 ] can be reformulated follows . [ p4 ] let @xmath221 . let @xmath222 . let @xmath223 . then @xmath224 [ p1 ] let @xmath225 be independent random variables uniformly distributed on the unit circle . let @xmath226 be a product of the random variables @xmath225 and their inverses , possibly with repetitions . then @xmath227 obviously if @xmath228 then @xmath229 . if @xmath230 then we write @xmath231 where @xmath232 for some @xmath233 . thus , by independence of @xmath225 , we have @xmath234 . [ p2 ] let @xmath225 be independent random variables uniformly distributed on the unit circle . let @xmath235 be products of the random variables @xmath225 and their inverses , possibly with repetition . then @xmath236 if @xmath237 for some @xmath238 then @xmath239 if @xmath240 for all @xmath238 then by lemma [ p1 ] , @xmath241 for all @xmath238 so @xmath242 [ p5 ] let @xmath225 be independent random variables uniformly distributed on the unit circle . let @xmath226 be a product of the random variables @xmath225 and their inverses , possibly with repetition . let @xmath243 . let @xmath244 be distinct . then there exists @xmath245 such that if @xmath246 satisfies @xmath247 then @xmath248 for all @xmath249 . we write @xmath226 as @xmath250 . then the result follows from lemma [ p1 ] , by taking @xmath251 for @xmath249 . combining lemma [ p4 ] and lemma [ p5 ] , we obtain [ p6 ] let @xmath221 . let @xmath222 . let @xmath225 be independent random variables uniformly distributed on the unit circle . let @xmath226 be a product of the random variables @xmath225 and their inverses , possibly with repetition . let @xmath244 be distinct . then @xmath252 equivalently , by lemma [ p1 ] , @xmath253 [ p7 ] let @xmath54 be a finite set . let @xmath254 be partitions of @xmath54 . let @xmath255 be a finite collection of vectors in a vector space @xmath217 such that whenever @xmath256 are scalars satisfying @xmath257 we have @xmath258 for all @xmath259 . then @xmath260 let @xmath261 be scalars such that @xmath262 for @xmath263 , let @xmath264 be the block in @xmath265 containing @xmath266 . then @xmath267 so by assumption , @xmath268 for all @xmath259 . hence , @xmath268 for all @xmath269 . therefore , @xmath270 thus , the result follows . [ p8 ] let @xmath271 be finite sets . let @xmath50 be a partition of @xmath58 . let @xmath272 be a partition of @xmath54 . then @xmath273 is a partition of @xmath58 and @xmath274 let @xmath275 be the union of all blocks in @xmath50 that contain an element in @xmath54 . then @xmath276 @xmath277 and @xmath278 since @xmath275 splits the partition @xmath279 , we have @xmath280 let @xmath254 be a partitions of @xmath58 . if @xmath281 then there exists a block @xmath282 that splits @xmath265 . the proof of lemma [ p9 ] is analogous to the proof of the fact that a partition of @xmath41 points with more than @xmath283 blocks must contain a singleton block . lemma [ p9 ] can be reformulated as [ p10 ] let @xmath58 be a finite set . let @xmath284 . let @xmath285 be a partition of @xmath58 . if @xmath286 then there exists a block @xmath287 such that @xmath288 and @xmath289 are disjoint . [ p11 ] let @xmath290 be a finite set in @xmath291 . for each @xmath292 , define the vector @xmath293 as @xmath294 if@xmath295 then @xmath296 for all @xmath297 . let @xmath297 . then @xmath298 since @xmath299 for all @xmath292 , if @xmath300 and @xmath301 then @xmath302 . thus , the only values of @xmath266 for which @xmath303 can possibly be nonvanishing are @xmath304 and @xmath305 . hence , @xmath306 since @xmath297 , we have : 1 . @xmath307 and @xmath308 , 2 . if @xmath309 then @xmath310 and @xmath311 , 3 . if @xmath312 then @xmath313 and @xmath311 , 4 . if @xmath314 then @xmath315 and @xmath316 , 5 . if @xmath317 then @xmath318 and @xmath316 . in all of the above cases , @xmath311 and @xmath316 . therefore , @xmath319 . so @xmath296 . [ p12 ] let @xmath290 be a finite set in @xmath291 . let @xmath320 be the equivalence relation on @xmath321 generated by @xmath322 @xmath323 . then this equivalence relation has at most @xmath324 equivalence classes . it suffices to show that every element @xmath266 of @xmath321 is related to an element in @xmath325 . if @xmath326 then @xmath327 . so replacing @xmath266 by @xmath328 , if necessary , we may assume that @xmath329 . let @xmath330 be smallest natural number for which @xmath331 . by minimality , @xmath332 for all @xmath333 . so @xmath334 so by assumption , @xmath335 for all @xmath333 . therefore , @xmath336 since @xmath337 , @xmath338 . hence , @xmath339 . [ p13 ] let @xmath208 . let @xmath209 . let @xmath210 be the interval partition of @xmath24 defined by @xmath340 for each @xmath341 , let @xmath342 then @xmath343 for all distinct @xmath344 . let @xmath345 with @xmath346 . if @xmath347 , then since @xmath348 , by the definition of @xmath210 , @xmath349 . on the other hand , if @xmath350 , then since @xmath351 and @xmath352 are disjoint blocks and are , therefore , disjoint , @xmath353 . in both cases , we have that either @xmath353 or @xmath354 . hence , @xmath355 interchanging the roles of @xmath351 and @xmath352 , we have @xmath356 since @xmath351 and @xmath352 are disjoint , @xmath357 and @xmath358 therefore , @xmath343 . [ p14 ] let @xmath209 . let @xmath210 be the interval partition defined in lemma [ p13 ] and let @xmath359 for @xmath341 be as defined there . let@xmath360 let @xmath320 be the equivalence relation on @xmath321 generated by @xmath322 for @xmath297 . then 1 . [ it : p14.1 ] every equivalence class of @xmath320 is of the form @xmath359 for some @xmath341 2 . [ it : p14.2 ] for every @xmath361 , there exists @xmath341 such that @xmath362 . we will first prove [ it : p14.1 ] . let@xmath363 we want to show that for every @xmath364 , @xmath359 is an equivalence class of @xmath320 . after proving this , we show that @xmath365 . this immediately gives the conclusion of [ it : p14.1 ] , because @xmath366 is the partition of @xmath321 that corresponds to the equivalence relation @xmath320 . we first show that @xmath367 for every @xmath364 . since @xmath364 , there exists @xmath368 such that @xmath369 and @xmath370 . since @xmath341 , by the definition of @xmath210 , we have @xmath371 . ( in particular , @xmath372 and @xmath305 are in @xmath25 . ) 1 . suppose @xmath373 and @xmath374 . we will show @xmath375 . since @xmath25 is an interval of length at least 2 , if @xmath373 then @xmath376 or @xmath377 . if @xmath376 , then by the definition of @xmath210 , @xmath378 so @xmath379 and @xmath329 . suppose @xmath380 . then either @xmath381 or @xmath382 and in either case , @xmath383 . then @xmath377 and @xmath384 . if @xmath385 , then @xmath386 and @xmath387 and @xmath326 . otherwise , if @xmath388 , then either @xmath389 or @xmath390 and we have @xmath391 . but @xmath392 , so @xmath393 , contrary to the hypothesis @xmath394 . thus , we have shown @xmath395 . 2 . suppose @xmath373 and @xmath396 . we will show @xmath397 . since @xmath341 , by the definition of @xmath210 , we must have @xmath398 unless @xmath266 is the smallest element of @xmath25 . 1 . if @xmath398 then @xmath399 so @xmath400 . 2 . if @xmath266 is the smallest element of @xmath25 , then @xmath401 . since @xmath346 and @xmath369 , @xmath402 . so @xmath403 so @xmath404 . since @xmath25 is an interval , it follows that @xmath405 . so @xmath346 , @xmath406 and @xmath407 . so @xmath408 so @xmath409 . + thus , we have shown @xmath410 . + it follows that @xmath367 . 2 . to show that @xmath359 is an equivalence class of @xmath320 , we will prove that @xmath359 is preserved by the equivalence relation @xmath320 and that all elements of @xmath359 are related . 1 . suppose @xmath411 and @xmath412 . since @xmath413 , we have @xmath414 since @xmath415 , either @xmath416 or @xmath417 . in the first case , @xmath418 and @xmath419 so @xmath420 . in the second case , @xmath421 and @xmath422 ; by , we have @xmath420 . in both cases , @xmath420 and @xmath423 , so @xmath424 . + on the other hand , if @xmath411 and @xmath424 then either @xmath425 or @xmath426 . in the first case , @xmath427 and @xmath428 . using , @xmath418 . in the second case , @xmath420 . by , @xmath418 . in both cases @xmath418 and @xmath429 , so @xmath412 . + therefore , @xmath359 is preserved by the equivalence relation @xmath320 . 2 . to prove that all elements of @xmath359 are related , note that since @xmath25 is an interval with alternating values of @xmath430 , @xmath431 is of the form @xmath432 for some @xmath433 where @xmath434 , @xmath435 , @xmath436,@xmath437 , @xmath438 , @xmath439 . thus , @xmath440 . thus , @xmath441 this means that all the elements in @xmath431 are related . using the same argument , one can show that all the elements in @xmath442 are related . just as the beginning of the first part of the proof , since @xmath364 , there exists @xmath368 such that @xmath369 and @xmath370 ( thus also @xmath371 ) . so @xmath443 and @xmath444 . since @xmath445 , @xmath446 . therefore , all elements in @xmath447 are related . + therefore , @xmath359 is an equivalence class of @xmath320 for every @xmath364 . 3 . it remains to show that @xmath365 . since @xmath367 by the first part of the proof , it suffices to show that @xmath448 . 1 . if @xmath449 then @xmath434 . let @xmath341 contain @xmath450 . then @xmath451 and @xmath364 . 2 . if @xmath452 then @xmath453 and @xmath454 . let @xmath341 contain @xmath455 . then @xmath451 and @xmath364 , since @xmath456 . this completes the proof of [ it : p14.1 ] . we now prove [ it : p14.2 ] . let @xmath361 . 1 . if @xmath369 , then since @xmath457 , either @xmath458 or @xmath459 . since @xmath460 , either @xmath461 or @xmath462 or @xmath463 . 1 . if @xmath464 and @xmath463 , then @xmath465 , which implies @xmath466 . moreover , since @xmath369 , @xmath467 . 2 . if @xmath458 and @xmath463 , then similarly , @xmath468 and we have @xmath469 and @xmath470 . 3 . if @xmath471 and @xmath472 , then @xmath462 and , since @xmath473 , we have @xmath474 and @xmath475 . 4 . if @xmath476 and if @xmath477 or @xmath478 , then similarly and @xmath474 and @xmath475 . 5 . if @xmath479 and @xmath480 , then @xmath481 and @xmath482 . if @xmath483 , then since @xmath484 we have @xmath471 . since @xmath460 , either @xmath461 or @xmath462 . in either case , we have @xmath485 and @xmath486 . this completes the proof . in the sequel , if @xmath487 is a @xmath127 random matrix and @xmath216 then @xmath488 thus , if @xmath487 is deterministic then @xmath489 is the normalized schatten @xmath330 norm . [ p15 ] let @xmath487 be a @xmath127 random matrix with integrable entries . let @xmath216 . then @xmath490 since @xmath491 is a norm on deterministic @xmath127 matrices , @xmath492 where the first inequality follows from jensen s inequality and the second inequality follows from hlder s inequality . [ p16 ] let @xmath493 and @xmath494 be @xmath127 ( deterministic ) matrices . let @xmath495 be positive real numbers such that @xmath496 . then @xmath497 applying lemma [ p16 ] repeatedly , one obtains [ p17 ] let @xmath498 be @xmath127 ( deterministic ) matrices . let @xmath499 be such that @xmath500 . then @xmath501 applying the above to random matrices , we get the following : [ p18 ] let @xmath498 be @xmath127 random matrices having finite moments of all orders . let @xmath499 be such that @xmath502 . then @xmath503 , @xmath504 taking expectations and using hlder s inequality , we obtain @xmath505 thus , holds . [ p19 ] let @xmath506 be @xmath127 random matrices having finite moments of all orders . let @xmath507 if @xmath508 and let @xmath509 if @xmath510 . then @xmath511 if @xmath510 , the result follows from the cauchy schwarz inequality . assume @xmath508 . first , @xmath512 so @xmath513 for each @xmath514 with @xmath515 , taking @xmath516 for @xmath517 , @xmath518 and @xmath519 in lemma [ p17 ] , we obtain @xmath520 & \leq\big|a_{l_{0}}^{(2)}\big|_{2}\,\prod_{j\ne l_0}\big|a_j^{({\epsilon}_j)}\big|_{2(s-1)}\le m^{s-1}\big|a_{l_{0}}^{(2)}\big|_{2},\end{aligned}\ ] ] where for the first inequality we used the trace property and the cauchy schwarz inequality , while for the second we used hlder s inequality ( lemma [ p18 ] ) . since there are @xmath521 terms in the summation in , the desired upper bound holds . we will now show that the off - diagonal entries of alternating products in @xmath2 and @xmath43 , with deterministic diagonal matrices interspersed , have expectations that are zero or are asymptotically small as the matrix size goes to infinity . let @xmath225 be independent random variables uniformly distributed on the unit circle . let @xmath244 . if @xmath522 satisfy @xmath523 then @xmath524 . let @xmath525 . by lemma [ p1 ] , @xmath526 for all @xmath73 . so @xmath527 [ odd ] let @xmath208 be an odd number . let @xmath209 be alternating . let @xmath213 be deterministic diagonal @xmath127 matrices . then @xmath528 the proof when @xmath529 and the proof when @xmath530 are similar . so we only do the case when @xmath529 . let @xmath531 . @xmath532 since @xmath529 and @xmath533 are alternating , @xmath534 when @xmath266 is odd , and @xmath346 when @xmath266 is even so @xmath535 since the sum of the exponents is @xmath536 by lemma [ odd ] , @xmath537 . thus , the result follows . using an argument similar to that in the proof of lemma [ odd ] , one obtains [ * 1 ] let @xmath538 be an even number . let @xmath209 be alternating . let @xmath213 be deterministic diagonal @xmath127 matrices . if @xmath530 then @xmath539 is a diagonal matrix . [ 1 * ] let @xmath208 be an even number . let @xmath209 be alternating . suppose that @xmath529 . let @xmath213 be deterministic diagonal @xmath127 matrices of norm at most 1 . let @xmath540 then for every integer @xmath216 , there is a constant @xmath541 such that @xmath542 let @xmath543 . by , @xmath544 since the @xmath545 have norms at most 1 , we have latexmath:[\[\label{expandineq } since @xmath534 when @xmath266 is odd and @xmath346 when @xmath266 is even , @xmath547 let @xmath548 be given by @xmath549 let @xmath550 be the restriction of @xmath57 to @xmath551 . then we have @xmath552 let @xmath50 be a partition of @xmath553 . suppose that @xmath554 . for each @xmath73 , all the @xmath555 are same for @xmath556 and we denote this value by @xmath557 . thus , @xmath558 by lemma [ p6 ] , @xmath559 if @xmath560 are scalars satisfying @xmath561 then @xmath562 . thus , by lemma [ p7 ] , @xmath563 so @xmath564 considering all the cases when @xmath565 is or is not a singleton block and @xmath566 is or is not a singleton block of @xmath50 , we see that the number of choices of @xmath567 such that @xmath554 is at most @xmath568 and , thus , @xmath569 summing over all partitions @xmath50 of @xmath553 , we have @xmath570 so by , @xmath571 so each entry in @xmath572 has absolute value at most @xmath573 . from this , the result follows easily . indeed , each entry of @xmath574 has absolute value at most @xmath575 . taken to the @xmath330-th power , every entry has absolute value at most @xmath576 , and the result follows with constant @xmath577 . using a similar argument as in the proof of lemma [ 1 * ] , ( essentially , by treating also the case @xmath578 in that proof ) one obtains the following lemma . [ bound ] for every integer @xmath216 , we have @xmath579 , where @xmath215 depends only on @xmath330 . [ offdiag ] let @xmath209 be alternating . let @xmath213 be deterministic diagonal @xmath127 matrices of norm at most 1 . let @xmath580 then for every integer @xmath216 , @xmath581 where @xmath215 depends only on @xmath41 and @xmath330 . by lemma [ bound ] , for every integer @xmath582 , @xmath583 where @xmath584 depends only on @xmath585 . thus , taking @xmath586 , @xmath587 in lemma [ p18 ] , we have @xmath588 this proves the first inequality . the other inequality follows by combining lemmas [ odd ] , [ * 1 ] and [ 1 * ] . we are now ready to prove proposition [ oneprop ] . we first prove a weaker version of it , with @xmath589 replaced by @xmath590 . the convention regarding ordering in products is described in definition [ def : rdiag ] . [ onelem ] let @xmath208 . let @xmath209 . let @xmath210 be the interval partition of @xmath24 defined by @xmath211 for @xmath212 . let @xmath213 be deterministic diagonal @xmath127 matrices of norm at most 1 . then @xmath591 where @xmath215 depends only on @xmath41 . since @xmath210 is an interval partition , we can expand @xmath592 \bigg(\mathbb{e}&\prod_{s\in\sigma}\bigg(\prod_{k\in s}(d_{k})_{i(k),i(k)}(x_{n}^{\epsilon_{k}})_{i(k),i(k+1 ) } \\ & -\mathbb{e}\left(\prod_{k\in s}(d_{k})_{i(k),i(k)}(x_{n}^{\epsilon_{k}})_{i(k),i(k+1)}\right)\bigg)\bigg ) \end{aligned } \\=&\sum_{\substack{i:\{1,\ldots , n+1\}\to\{1,\ldots , n\}\\i(n+1)=i(1)}}\prod_{k=1}^{n}(d_{k})_{i(k),i(k ) } \begin{aligned}[t ] \bigg(\mathbb{e}&\prod_{s\in\sigma}\bigg(\prod_{k\in s}(x_{n}^{\epsilon_{k}})_{i(k),i(k+1 ) } \\ & -\mathbb{e}\left(\prod_{k\in s}(x_{n}^{\epsilon_{k}})_{i(k),i(k+1)}\right)\bigg ) . \end{aligned}\end{aligned}\ ] ] since the @xmath545 have norms at most 1 , it follows that @xmath593 \bigg|\mathbb{e}&\prod_{s\in\sigma}\bigg(\prod_{k\in s}(x_{n}^{\epsilon_{k}})_{i(k),i(k+1 ) } \\ & -\mathbb{e}\left(\prod_{k\in s}(x_{n}^{\epsilon_{k}})_{i(k),i(k+1)}\right)\bigg)\bigg| . \end{aligned}\end{gathered}\ ] ] for each @xmath594 , by lemma [ p2 ] , we have @xmath595 let@xmath360 let @xmath596 . then @xmath597\\ & = \mathbb{e}\prod_{k\in u}(y_{n}^{\epsilon_{k}})_{i(k),i(k+1)}\prod_{k\in u}(y_{n}^{\epsilon_{k-1}})_{i(k-1),i(k ) } \prod_{k\in\{1,\ldots , n\}\backslash(u\cup(u-1))}(y_{n}^{\epsilon_{k}})_{i(k),i(k+1)}\displaybreak[2]\\ & = \mathbb{e}\prod_{k\in u}\zeta_{i(k+1)}^{-i(k)}\prod_{k\in u}\zeta_{i(k-1)}^{i(k ) } \prod_{k\in\{1,\ldots , n\}\backslash(u\cup(u-1))}(y_{n}^{\epsilon_{k}})_{i(k),i(k+1)}\displaybreak[2]\\ & = \mathbb{e}\prod_{k\in u+1}\zeta_{i(k)}^{-i(k-1)}\prod_{k\in u-1}\zeta_{i(k)}^{i(k+1 ) } \prod_{k\in\{1,\ldots , n\}\backslash(u\cup(u-1))}(y_{n}^{\epsilon_{k}})_{i(k),i(k+1 ) } \displaybreak[2]\\ & = \mathbb{e}\bigg ( \begin{aligned}[t ] & \prod_{k\in ( u+1)\backslash(u-1)}\zeta_{i(k)}^{-i(k-1)}\prod_{(u-1)\backslash(u+1)}\zeta_{i(k)}^{i(k+1)}\\ & \hspace{2em}\times\prod_{k\in ( u+1)\cap(u-1)}\zeta_{i(k)}^{i(k+1)-i(k-1)}\prod_{k\in\{1,\ldots , n\}\backslash(u\cup(u-1))}(y_{n}^{\epsilon_{k}})_{i(k),i(k+1)}\bigg ) . \end{aligned}\end{aligned}\ ] ] let @xmath598 . note that by the definition of @xmath290 , @xmath599 . also , if @xmath600 then @xmath601 and @xmath602 . let @xmath603 be the restriction of @xmath57 to @xmath290 . let @xmath604 be the restriction of @xmath57 to @xmath58 . with @xmath605 defined as in lemma [ p11 ] , we have @xmath606 where we think of @xmath607 as belonging to @xmath608 . let @xmath50 be a partition of @xmath58 . suppose that @xmath609 . let @xmath610 . for each block @xmath611 , all the @xmath612 are the same for @xmath556 and we denote this value by @xmath613 . it follows that @xmath614 note that the term @xmath615 is a product of the random variables @xmath616 and their inverses , possibly with repetition . thus by lemma [ p6 ] , fixing @xmath617 and summing over all @xmath607 , we have @xmath618 summing now over all @xmath617 with @xmath609 , we obtain @xmath619 let @xmath320 be the equivalence relation on @xmath321 generated by @xmath322 @xmath620 . let @xmath621 be the partition of @xmath321 that corresponds to @xmath320 . by lemma [ p11 ] and lemma [ p7 ] , @xmath622 thus , @xmath623 where the last equation follows from lemma [ p8 ] by taking @xmath624 and @xmath625 . case i : @xmath626 . in this case , by , @xmath627 thus , by , @xmath628 since @xmath629 , @xmath630 by , @xmath631 case ii : @xmath632 . by lemma [ p12 ] , @xmath633 so @xmath634 by lemma [ p10 ] , @xmath635 contains a block @xmath636 such that @xmath637 if @xmath638 then by lemma [ p14][it : p14.1 ] , @xmath639 for some @xmath640 . if @xmath641 for some @xmath642 and @xmath643 , then by lemma [ p14][it : p14.2 ] , @xmath636 is also the form . if @xmath644 or @xmath566 then since @xmath90 and @xmath645 are both in @xmath58 ( by the definition of @xmath290 ) , it follows from that @xmath646 and so @xmath647 if @xmath636 is of the form then by , @xmath648 by lemma [ p13 ] , @xmath649 . so @xmath650 note that for each @xmath651 , @xmath652 depends only on @xmath653 . thus , the random variable @xmath654 is independent of the random variables @xmath655 thus , @xmath656 so again holds . combining the conclusions of case i and case ii and summing over all partitions @xmath50 of @xmath58 , we get @xmath657 where @xmath215 is the number of partitions of @xmath58 . by , the result follows . let @xmath658 by lemma [ offdiag ] and lemma [ p15 ] , for every integer @xmath216 , @xmath659 and @xmath660 where @xmath215 depends only on @xmath41 and @xmath330 . so by lemma [ p19 ] , @xmath661 where @xmath215 depends only on @xmath41 . by lemma [ offdiag ] for @xmath662 , @xmath663 therefore , @xmath664 thus , by lemma [ onelem ] , the result follows . we are now ready to prove the main result . for a c@xmath14-algebra @xmath18 , by @xmath665 we denote the @xmath0-algebra of polynomials in noncommuting variables @xmath666 and @xmath667 with coefficients on @xmath18 ; technically this is the algebraic free product of the three algebras @xmath18 , @xmath668 $ ] and @xmath669 $ ] with amalgamation over the scalars . we endow @xmath665 with the obvious @xmath0-operation . [ thm : xrdiag ] consider the c@xmath14-algebra @xmath670 $ ] with tracial state @xmath671 obtained by integration using lebesgue measure . let @xmath672 be the linear , self - adjoint , @xmath18-bimodular map that is the identity on @xmath18 and so that with respect to @xmath673 , @xmath666 is @xmath18-valued r - diagonal with even alternating moments given by , for every @xmath15 and @xmath674 , @xmath675 then for all @xmath15 , @xmath16 and all @xmath676 , we have @xmath677 where , for @xmath678 , @xmath679 is the scalar diagonal matrix @xmath680 let @xmath15 and suppose @xmath681 and @xmath16 are arbitrary . we will prove by induction on @xmath41 . in the case of @xmath88 , the right - hand - side of is zero and , by lemma [ odd ] , so is the left - hand - side of . for the induction step , let @xmath682 be the maximal alternating interval partition of @xmath683 ( see definition [ def : maxaltptn ] ) . for @xmath341 , let @xmath684 where the product is taken in increasing order of the index @xmath42 . by @xmath18-valued r - diagonality of @xmath666 , @xmath685 where the product over @xmath341 is taken in order of increasing elements of the interval blocks @xmath25 ( since @xmath210 is an interval partition , given two distinct blocks , all the elements of one of them are less than all the elements of the other ) . expanding the above product over @xmath341 , we get a sum of @xmath686 terms that enables @xmath687 to be expressed as @xmath688 plus the sum of @xmath689 terms , each of the form @xmath690 where @xmath54 is the union of a proper subset @xmath691 of @xmath210 and for certain @xmath692 , equal to the product of @xmath693 and some of @xmath694 . we will show @xmath695 this will prove the induction step , because expansion of the left - hand - side of as a sum of @xmath686 terms will enable @xmath696 to be written as @xmath697 plus the sum of @xmath698 terms , each equal to @xmath699 for the same @xmath54 and @xmath700 as appeared in . by the inductive hypothesis , each of the terms in is equal to the corresponding term in . this shows that proof of the induction step will follow , once we have proved . in order to verify , we will use proposition [ oneprop ] , which yields @xmath701 for @xmath341 , if @xmath41 is even , then from proposition [ prop : diagexpect ] and , we have @xmath702 while if @xmath41 is odd , then by r - diagonality of @xmath666 we have @xmath703 and from lemma [ odd ] , we see that also in this case holds . we now write , for each @xmath341 , @xmath704 and , in the left - hand - side of , distribute , resulting in a sum of @xmath686 limits , each of which will be seen to equal @xmath705 . that the first of these limits is zero is precisely the import of . that each of the other limits is zero is a consquence of and hlder s inequality , ( see , lemma [ p18 ] ) . indeed , each of the other limits is of the form @xmath706 where @xmath707 is either @xmath708 or @xmath709 and for at least one @xmath341 it is the latter . now from , we conclude that , for every @xmath341 , @xmath710 remains bounded as @xmath13 . from lemma [ offdiag ] , we have that , for every @xmath341 and every integer @xmath711 , @xmath712 remains bounded as @xmath13 . consequently , for every @xmath341 , @xmath713 remains bounded as @xmath13 . of course , from we get , for every @xmath25 and @xmath330 , @xmath714 consequently , taking @xmath715 and applying hlder s inequality , we get that for every product @xmath716 of the form described at , @xmath717 now using the cauchy schwarz inequality , we conclude @xmath718 this finishes the proof of , and of the theorem . here are some results that will allow us to calculate @xmath96 for many partitions @xmath50 . the first is an easy calculation : [ lem : twointervals ] suppose @xmath722 where , for some @xmath723 , @xmath46 is a partition of @xmath724 and @xmath47 is a partition of @xmath725 . let @xmath726 be the partition obtained from @xmath47 by shifting @xmath727 to the left . then @xmath728 since @xmath729 we see @xmath730 or , with the obvious identification @xmath731 , so that the measure @xmath116 corresponds to the measure @xmath732 . thus , we find @xmath733 and @xmath734 \\ = \int_{f(\pi_1,t)}(g_1\otimes\cdots\otimes g_x)\,d\nu^{(\pi_1,t)}\ ; \int_{f({{\tilde\pi}}_2,t)}(g_{x+1}\otimes\cdots\otimes g_{n-1}\otimes 1)\,d\nu^{({{\tilde\pi}}_2,t ) } \\ = \lambda_{\pi_1}(g_1,\ldots , g_{x-1})(t)g_x(t)\lambda_{{{\tilde\pi}}_2}(g_{x+1},\ldots , g_{n-1})(t).\end{gathered}\ ] ] [ lem : internalinterval ] suppose @xmath66 and @xmath722 , where @xmath46 is a partition of @xmath735 and @xmath47 is a partition of @xmath736 , for some @xmath737 . let @xmath738 and @xmath739 be the partitions obtained from @xmath46 by applying the function @xmath740 and @xmath47 by applying the function @xmath741 , respectively , to all the points in the blocks , where @xmath740 and @xmath741 are the order - preserving bijections from @xmath742 onto @xmath743 and from @xmath744 onto and @xmath745 , respectively . then @xmath746 since @xmath747 we see @xmath748 and @xmath749 and that the measure @xmath116 decomposes as @xmath750 thus , we find @xmath751 & \int_{f({{\tilde\pi}}_1,t)}(g_1\otimes\cdots\otimes g_{x-1 } \\ & \hspace{1em}\otimes g_x(s_x)\left(\int_{f({{\tilde\pi}}_2,s_x)}(g_{x+1}\otimes\cdots\otimes g_{x+y-1})\otimes g_{x+y}(s_x)\,d\nu^{({{\tilde\pi}}_2,s_x)}\right ) \\ & \hspace{3em}\otimes g_{x+y+1}\otimes\cdots\otimes g_{n-1}\otimes1 ) d\nu^{({{\tilde\pi}}_1,t)}(s_1,\ldots , s_x , s_{x+y+1},\ldots , s_{n-1 } ) \end{aligned } \\[1ex ] & = \lambda_{{{\tilde\pi}}_1}(g_1,\ldots , g_{x-1},g_x\lambda_{{{\tilde\pi}}_2}(g_{x+1},\ldots , g_{x+y-1})g_{x+y},g_{x+y+1},\ldots , g_{n-1})(t).\end{aligned}\ ] ] [ lem : adjacent ] let @xmath66 and suppose @xmath752 for some @xmath212 . let @xmath753 be obtained from @xmath50 by gluing @xmath266 and @xmath754 together ; namely , letting @xmath755 be the order - preserving bijection , we have @xmath756 then @xmath757 where @xmath758 indicates that @xmath759 has been removed from the list of arguments , while all the others remain . in the notation of remark [ rem : lambdapialt ] , since @xmath760 , we see that in the condition for @xmath761 to belong to @xmath112 involves @xmath762 only in the summation over the block of @xmath50 that contains @xmath266 , and in this summation @xmath762 cancels with @xmath763 . thus , we easily see @xmath764 and , under the corresponding identification @xmath765 , for every @xmath92 $ ] we have the identification of measures @xmath766 where @xmath272 is lebesgue measure on @xmath767 . using , we get @xmath768}g_k\,d\lambda \\ & = \lambda_{{{\tilde\pi}}}(g_1,g_2,\ldots,\widehat{g_k},\ldots , g_{n-1})(t)\tau(g_k).\end{aligned}\ ] ] [ lem:1andn ] let @xmath719 and suppose @xmath66 has @xmath90 and @xmath41 in the same block . then for all @xmath769 $ ] , @xmath770 is a constant function . moreover , letting @xmath753 be the restriction of @xmath50 to @xmath771 , we have @xmath772 this is much like the proof of lemma [ lem : adjacent ] . recalling that we use the convention @xmath104 , since @xmath773 , in the condition for @xmath761 to belong to @xmath112 involves @xmath110 only in the summation over the block of @xmath50 that contains @xmath41 , and in this summation @xmath110 cancels with @xmath774 . thus , we have @xmath775 moreover , @xmath776^{n-1})\times\{t\}$ ] and @xmath116 corresponds to simply @xmath114-dimensional lebesgue measure on @xmath777 . using , we get @xmath778}\left(\int_{f({{\tilde\pi}},s)}(g_1\otimes\cdots\otimes g_{n-1})\,d\nu^{({{\tilde\pi}},s)}\right)\,d\lambda(s ) \\ & = \int_{[0,1]}\lambda_{{{\tilde\pi}}}(g_1,g_2,\ldots , g_{n-2})(s)g_{n-1}(s)\,d\lambda(s ) \\ & = \tau(\lambda_{{{\tilde\pi}}}(g_1,g_2,\ldots , g_{n-2})g_{n-1}).\end{aligned}\ ] ] examining remark [ rem : lambdapialt ] , we see @xmath781^n}g_1\otimes\cdots\otimes g_n\,d\nu^{(\pi)},\ ] ] where @xmath782 is the @xmath114-dimensional lebesgue measure on @xmath112 corresponding to the restriction to @xmath112 of the euclidean metric on @xmath103 . moreover , we find @xmath783 and that under this identification , @xmath784 corresponds to @xmath785 . thus , we get @xmath786^n}g_2\otimes\cdots\otimes g_n\otimes g_1\,d\nu^{(l(\pi ) ) } = \tau(\lambda_{l(\pi)}(g_2,\ldots , g_n)g_1).\ ] ] there are 14 noncrossing partitions of @xmath792 and one crossing partition , namely , @xmath793 . using remark [ rem : lambdapialt ] , we calculate @xmath794 from , we calculate @xmath795 consequently , from and , we find @xmath796 { { \mathcal{e}}}\big((x^*x)^4\big)(t)&=14+\frac12+t(1-t ) . \label{eq : ex*x4}\end{aligned}\ ] ] unlike with scalar - valued r - diagonality in the tracial setting , in the @xmath18-valued case , @xmath0-freeness is not guaranteed in a polar decomposition . this phenomenon was seen in @xcite , but is also exhibited by the asymptotic limit of the random vandermonde matrices : [ prop : polarnotfree ] the element @xmath666 does not have the same @xmath0-distribution as any element in a @xmath18-valued @xmath0-noncommutative probability space of the form @xmath797 , with @xmath290 unitary , @xmath798 and such that @xmath290 and @xmath799 are @xmath0-free over @xmath18 . if it were scalar - valued r - diagonal , then we would have @xmath806 letting @xmath678 be @xmath807 , by @xmath18-valued r - diagonality of @xmath666 and and , we have @xmath808 writing @xmath809 , expanding , distributing , using , @xmath18-valued r - diagonality again and , we get @xmath810 using we find that for @xmath811 , @xmath812 and , thus , that the quantity equals @xmath813 . but @xmath814 which shows that fails to hold . we conclude this paper with a report of calculations of some of the @xmath1$]-valued cumulant maps of the asymptotic @xmath0-distribution of random vandermonde matrices , namely , of the @xmath1$]-valued distribution @xmath673 from theorem [ thm : xrdiag ] . the details of these calculations are either straightforward to work out or can be found in the mathematica @xcite notebook that is available with this paper . let @xmath815 denote these cumulant maps , and for brevity let @xmath816 be those that need not , by virtue of r - diagonality , be zero . 1 . [ it : nosplit ] no proper subinterval @xmath817 of @xmath24 splits @xmath50 ( by proper subinterval we mean with @xmath818 and @xmath819 ) 2 . [ it : noneigh ] no block of @xmath50 contains neighbors ( modulo @xmath41 ) , namely , @xmath820 for all @xmath212 and @xmath821 . note that condition [ it : nosplit ] implies that @xmath50 has no singleton blocks . it is easy to check that @xmath822 is empty when @xmath823 , and that @xmath824 , where @xmath793 . the purely crossing projections and related quantities are studied further in @xcite .

we revisit astrophysical constraints on models with an ultra - light gravitino , in particular extending the analysis to more general models and the case of r - parity breaking . these constraints allow to restrict the value of the gravitino mass depending on the masses of scalar moduli . we perform both a frequentist and bayesian analysis and we find comparable results , even if the bayesian analysis is in part affected by volume effects . only a small window of gravitino masses can be excluded by the sn constraints in a model - independent way , while limits obtained with a definite assumption on the scalar masses result more stringent .

1410.0314

cosmology and astrophysics are among the most successful probes for very weakly interacting particles within extensions of the standard model of particle physics . indeed a substantial part of our knowledge also about the more weakly interacting standard model particles , the neutrinos , comes from cosmological and astrophysical data . particularly interesting is the case of the gravitino , the superpartner of the graviton in supergravity extensions of the standard model . for such particle the couplings are all set by the symmetry and can not be modified easily . it was therefore realized very early on , that gravitinos can be very dangerous long - lived relics in the early universe if they are not stable . this is often referred to as the _ gravitino problem _ in cosmology @xcite . of course the other possibility is for the gravitinos to be the lightest supersymmetric particle and in that case they can also successfully play the role of dark matter @xcite . we are here instead interested in a different scenario , where the gravitinos are ultra - light , with masses much smaller than the electroweak scale or even the other sm particle masses . in that case gravitinos can be in thermal equilibrium and survive as thermal relics , but they play practically no role in cosmology , since their energy density is negligible . an ultralight gravitino can be naturally obtained in supersymmetric models of the no - scale type and gives rise to a very interesting phenomenology thanks to its enhanced couplings @xcite , as we will review in sec . [ ss : gravitinolagrangian ] . in that case , thanks to the large interaction with normal matter , astrophysics provides the possibility to constrain the scenario . indeed astrophysical object like stars and supernovae are highly sensitive to any additional cooling mechanism , apart for neutrino emission , and have been exploited in the past to put limits on new particles , e.g. in the case of axions @xcite . in the past various approaches to restrict an ultra - light gravitino through astrophysical observations have been made . some early constraints on its mass were derived from cosmology , more precisely from big bang nucleosynthesis ( bbn ) . it was shown that bbn allows either a light gravitino @xmath0kev or a very heavy one @xcite . these investigations have been revisited in 1993 by moroi @xcite and four years later again by gherghetta @xcite . other studies investigated the possibility of exotic cooling of stars , red giants and white dwarfs @xcite . these limits were reviewed in @xcite , where it is shown that the limits depend as well on the masses of the scalar partners of the gravitino . our analysis is based on @xcite , wherein we revisited and generalized some of the results obtained by grifols , mohapatra and riotto during the 90s @xcite . they were able to exclude a gravitino mass range covering a few orders of magnitude in the ultra - light mass regime from the sn1987a observation . the exact mass interval however depends on the masses of the sgoldstinos , the scalar superpartner of the goldstino field , which are model - dependent . in the early works @xcite such fields were implicitly assumed to be massless . moreover most of these results were obtained assuming a specific relation between the sgoldstino s coupling parameters @xmath1 and @xmath2 and the photino and gravitino mass , namely @xmath3 , where @xmath4 are the masses of the gravitino and photino respectively . however this kind of relation does not have to occur in general models . we generalize here previous results by performing a full parameter scan in the model - dependent parameters such as the sgoldstino masses and couplings . moreover we consider also possible effects of r - parity breaking , that opens up the possibility of a single - gravitino production channel . a core collapse of stars with masses larger than a few solar masses can trigger a giant stellar explosion called a supernova . its dynamics can be explained by a `` bounce - and - shock '' model . at the end of the star evolution the iron core can no longer release energy via fusion and remains stable as long as the electron degeneracy pressure balances the gravitational pressure . however , once the core mass exceeds the chandrasekhar limit of @xmath5 it becomes unstable and starts to collapse . shortly afterwards , the gravitational binding energy @xmath6 gets released . a large fraction of this energy is emitted in the form of neutrinos . in eq . @xmath7 km is the core radius . on february 23th 1987 the blue giant sanduleak-69202 in the large magellanic cloud exploded in a supernova that was named sn1987a . it is the first , and so far the only supernova , for which the emitted neutrinos have been observed directly . they have been detected by kamiokande ii and imb @xcite . models of gravitational collapse @xcite allow to estimate the neutrino energy released by sn1987a as @xmath8 combining eqs . and , one finds the upper limit @xmath9 on the luminosity @xmath10 which can be produced by anomalous cooling mechanisms present in sn1987a . many models for physics beyond the standard model predict new light weakly interacting particles which can in principle contribute to @xmath10 . at this point we mention that constraints from anomalous sn cooling can also been found for other proposed light particles such as axions @xcite and sterile neutrinos @xcite . furthermore it is also possible to apply the energy loss argument to alternative astrophysical settings such as massive stars with @xmath11 solar masses , which has been done for axions in 2012 @xcite . in this paper we focus on supernovae , since they reach a larger core temperature compared to stars and are therefore expected to give the strongest constraints . we compute in the following the contribution to @xmath10 generated by ultralight gravitinos production via photon collisions during a supernova of type ii and we apply the constrain given in eq . . as derived in appendix a , the constraint is given as a function of the gravitino production cross - section as @xmath12 where @xmath13 are the momenta of the incoming photons , @xmath14 has been approximated by @xmath15 and the factor @xmath16 comes from the two possible photon polarizations . compared to the one used previously in @xcite . ] the described argument applies only if the gravitinos produced at sn1987a escape the supernova core and carry away energy . if the gravitino - matter coupling is sufficiently strong , the gravitino mean - free - path @xmath17 could be short enough for the gravitinos to diffuse inside the supernova core . if the gravitinos are trapped inside the core radius @xmath18 km for longer than @xmath19 s , their energy is depleted by neutrino emission and the gravitino luminosity is again compatible with @xmath20 . trapped gravitinos random - walk through the core and cover a distance @xmath21 in a time interval @xmath22 , where @xmath23 is the number of gravitino scatterings . hence , we obtain the lower bound for @xmath17 , @xmath24 the limit applies therefore for gravitino mean - free - paths larger than only . in this paper we investigate how eqs . and constrain the parameter space of the effective theory of ultra - light gravitinos that we introduce in the next section . in this section we briefly introduce the theoretical framework assumed in sec . [ sec : results ] deriving limits on the gravitino mass from supernovae observations . our general results for the cross - sections used in the analysis are given in eqs . and , obtained here for the first time . we start here with the effective lagrangian for a locally supersymmetric version of qed with broken susy and conserved r - parity , which was first derived in @xcite . it inherits a canonical khler potential , a vanishing cosmological constant and has been studied by various authors in the context of gravitino phenomenology @xcite . the matter fields @xmath25 and @xmath26 , the gauge fields @xmath27 ( of field strength @xmath28 ) and @xmath29 , and the gravitino @xmath30 , as well as their interactions are described by the following supergravity lagrangian as metric signature , fix the sign of the levi - civita symbol via @xmath31 and set @xmath32 . ] , @xmath33\lambda^{(a)}+ d_{\mu}\phi^{i}d^{\mu}\phi^{*i } - m_{\phi_i}\phi^{*i}\phi^i \nonumber\\ & + i \overline{\chi^i}_l \gamma^{\mu}d_{\mu}\chi_l^i - \frac{1}{2}m_{\chi_i } \left ( \overline{\chi^c_l}^i \chi_l^i + \text{h.c.}\right ) \nonumber\\ & - \frac{i}{\sqrt{2 } m_p}\left ( d_{\mu}\phi^{*i}\overline{\psi}_{\nu}\gamma^{\mu}\gamma^{\nu}\chi^i_l - d_{\mu}\phi^i \overline{\chi}^i_l \gamma^{\nu}\gamma^{\mu}\psi_{\nu}\right ) \nonumber\\ & - \frac{1}{4 m_p}\overline{\psi}_{\mu}\sigma^{\rho\sigma}\gamma^{\mu}\lambda \ ; f_{\rho\sigma}+\mathcal{o}(m_p^{-2})\ , , \label{eq : modell}\end{aligned}\ ] ] where @xmath34 gev is the reduced planck mass and @xmath35 is the ricci scalar . the covariant derivatives are given by @xmath36 here @xmath37 is the charge of the field on which the covariant derivative acts , and @xmath38 is the spin connection . the value of the gravitino mass @xmath39 heavily depends on the susy breaking scheme . certain models , such as no - scale models @xcite and models with gauge - mediated susy breaking ( gmsb ) @xcite allow the gravitino to be very light . as we will see below , a very small value of @xmath39 is a phenomenologically attractive possibility , since it enhances the gravitino interactions , which are otherwise suppressed by the small gravitational coupling constant @xmath40 @xcite . a massive gravitino obtains its @xmath41 helicity states by absorbing the goldstino field via the super - higgs mechanism @xcite . if the gravitino mass is very small compared to the energy scale of the relevant processes , its @xmath41 helicity states dominate in the transition amplitudes ( with some exception discussed below ) , and the gravitino effectively behaves like a massless goldstino . in this case the gravitino @xmath42 helicity states are negligible in the calculations . this result is a consequence of the susy equivalence theorem @xcite . in the limit in which the equivalence theorem applies , we can approximate the gravitino field @xmath43 as @xmath44 where @xmath45 is the spin-@xmath46 goldstino . the small @xmath39 term in the denominator of leads to an enhancement of the gravitino interactions . in certain models this enhancement of the gravitino interactions affects also the sgoldstinos , the scalar superpartners of the goldstino . in contrast to the goldstino , the sgoldstinos do not disappear from the physical spectrum . they can be very light , with masses of the order of the gravitino mass @xcite , or also much heavier . indeed , light sgoldstino are generally very long - lived and can be problematic for cosmology , so that heavy sgoldstinos are favoured from that point of view . in general though , the scenario with an ultralight gravitino could not only allow observations of gravitational effects due to the gravitino , but also due to new particles from the hidden susy breaking sector . in using eq . , we implicitly neglect terms suppressed by powers of @xmath39 higher than @xmath47 in the gravitino polarization tensor ( see eq . below ) . in the case of processes with more than one external gravitino , and involving several diagrams exhibiting different dependencies on @xmath39 , these higher order terms are not negligible . rather than using the complete spin-@xmath48 polarization tensor @xmath49 , given e.g. in @xcite , it is convenient to expand @xmath49 in powers of @xmath39 , @xmath50 performing then a careful @xmath39 power counting in the relevant amplitudes , and identifying so the leading contributions to the observables . in eq . , we denote by @xmath51 the coefficient of the @xmath52 term in this power series expansion . with this notation , @xmath53 corresponds to the polarization tensor resulting from eq . . as we have seen , the scalar partners of the goldstino remain as physical fields in the spectrum and they can be as light as the gravitino . we can write the general effective sgoldstino lagrangian as follows @xmath54 in eq . , the scalar and pseudo - scalar fields , respectively @xmath55 and @xmath56 , are the real sgoldstino components @xcite given by @xmath57 where @xmath58 is a scalar component of the @xmath59 chiral multiplet in the hidden sector , whose non - vanishing @xmath60-term breaks supersymmetry . the sgoldstino couplings constants @xmath1 and @xmath2 are model dependent , but can be related to other susy breaking parameters in specific scenarios as we will see below . they largely depend on the @xmath59-dependent contribution to the gauge kinetic function , on the khler potential and on the superpotential . also model - dependent are the sgoldstino masses . in fact , even if the sgoldstino direction is univocally determined by the susy breaking direction in the scalar field space , the scalar mass matrix is often non - diagonal and the sgoldstinos in general mix with the other moduli fields , so that they are not in general eigenstates of the scalar mass matrix ( see e.g. @xcite for constraints arising from the scalar mass matrix in minkowski and de - sitter vacua ) . we will consider the lagrangian given above to include in an effective way the presence also of those additional scalars . indeed the presence of many scalar can be approximated by an appropriate increase of the gravitino - scalar couplings . in deriving limits on the gravitino mass , we assume the lagrangian ( [ eq : lsgoldstino ] ) , with @xmath61 and @xmath62 , and the product of coupling constants @xmath63 , as free parameters . we now provide examples for how ( [ eq : lsgoldstino ] ) , and in particular @xmath39 , can be generated in specific models for supersymmetry breaking . though to specify the mechanism of supersymmetry breaking and mediation is not necessary in our model - independent study , it can allow to explicitly relate @xmath39 to the scale of supersymmetry breaking and then to the rest of the supersymmetric spectrum . the mechanism responsible for supersymmetry breaking is yet unknown , but we can consider two general classes of models , depending if the breaking is mainly due to the superpotential , @xmath60-term breaking , or the gauge sector , i.e. @xmath64-term breaking . we will here shortly review the prediction for the mass spectrum and sgoldstino couplings in the first case . in the case of @xmath64-term breaking of supersymmetry , in order to obtain a vanishing cosmological constant , also a non - vanishing superpotential @xmath65 and therefore @xmath60-terms are present at the minimum , leading to a combined @xmath64-term and @xmath60-term breaking , where often the @xmath60-term dominates @xcite . if supersymmetry is broken by non - vanishing @xmath60-terms for some of the chiral multiplets , the goldstino field is the corresponding combination of the fermionic partners , i.e. @xmath66 . very often one approximates this picture by assuming that one single chiral multiplet breaks supersymmetry and then the goldstino is simply @xmath67 . a simple computable model of this type is the polonyi model @xcite , based on a chiral multiplet @xmath59 in the hidden sector with canonical khler potential and the superpotential @xmath68 where @xmath69 is the scale of susy - breaking , since the @xmath70 f - term is @xmath71 , and the constant @xmath72 is chosen to ensure zero vacuum energy at the potential minimum . in this simple case the sgoldstinos are just the scalar and pseudoscalar components of the @xmath59 superfield as given above and their masses are of the order of the gravitino mass : @xmath73 in this scenario all the moduli fields of the theory gain similar masses via gravity mediation , giving rise to a multidimensional non - diagonal scalar field mass matrix . the goldstino direction is in general not an eigenstate of such a matrix and therefore non - trivial mixings can arise in the scalar sector . also the gauginos obtain a mass via gravity mediation of a similar order , if the gauge kinetic function has a dependence on @xmath59 . for an ultralight gravitino such a light mass spectrum is long excluded by collider bounds on the mass of the sm superpartners . another popular mechanism of mediation , providing a large mass splitting between the sm superpartners and the gravitino , is gauge mediation @xcite . this mechanism can be easily embedded in the model ( [ eq : polonyi ] ) just by adding a coupling of the polonyi field to sm charged messengers as @xmath74 with @xmath75 @xcite . then the gaugino and scalar masses are generated at one and two - loop level respectively and read @xmath76 where @xmath77 are sm representation dependent constants @xcite . we see then that the ratio between the gravitino and the photino mass is given by @xmath78 note that in the simplest model , barring cancellations , the mass of the messengers is also given by the vacuum expectation value of @xmath55 by the term @xmath79 , giving again masses of similar order . assuming though the presence of more fields @xmath70 , it is possible to disentangle the vacuum expectation value of the scalar field and the f - term and obtain also very different ratios of masses . we see that to have a large mass splitting allowing for an ultralight gravitino we are therefore obliged to consider more complicated hidden sectors and many very light messengers , just around the electroweak scale . regarding the couplings between the sgoldstinos and the qed gauge multiplet , they arise in gauge mediation via a similar one - loop diagram to that generating the gaugino masses , just substituting the f - term insertion with a physical scalar field . we therefore expect in the simplest case that the coupling @xmath1 is given by @xmath80 but in the more general case cancellations between diagrams or the presence of scalar mixing parameters can give rise to smaller values of the couplings . note that in gravity mediation the same coupling is instead @xmath81 @xcite . the couplings @xmath2 between the gravitino and the sgoldstinos are a remnant of the usual supergravity coupling between the gravitino , a scalar field and its superpartners for the goldstino multiplet . in the limit of light gravitino and single goldstino field , the dominant coupling is then of the form given above with @xmath82 while in the case of many scalar moduli the coupling can be reduced by mixing matrices . so we obtain for the relevant product of coupling , which will later appear in the cross - sections , @xmath83 we will take this value as the maximal possible value in the following and consider later also the simplified expression obtained for @xmath84 gev , i.e. @xmath85 in order to compare with previous works @xcite . note that in this case the product of couplings is not an independent parameter , but it is determined by the gravitino mass . in the general case , instead , since the scale of @xmath86 is arbitrary , @xmath87 is an independent parameter . in the model discussed above , a single scale , namely @xmath88 , determines the gravitino mass via @xmath89 , and the masses of the remaining superpartners through eqs . ( [ eq : m0m12 ] ) . in the light of existing limits on gaugino and sfermion masses , it can hence be difficult to accommodate an ultra - light gravitino within this model , even if it is possible when a very large number of messenger fields @xmath23 is considered . models where gaugino and sfermion masses are to some extent independent of the gravitino mass are of great interest for our analysis , since in that case ultra - light gravitinos can exist avoiding constraints from lhc and direct searches for new physics . within the many possibilities , we mention here a few examples of this type . a family of models where ultra - light gravitinos , and tev gauginos and sfermions can coexist is discussed in @xcite . the theory assumes a no - scale form for the khler function , and a diagonal non - canonical gauge kinetic function @xmath90 , where @xmath91 and @xmath92 are constants , and @xmath55 is the hidden sector field responsible for supersymmetry breaking . in this case the gaugino and gravitino masses are related as follows @xmath93 and properly choosing @xmath92 , @xmath39 can be ultra - light and @xmath94 at the tev scale . another way to disentangle the gaugino and gravitino masses and allow for low scale of susy breaking is also to consider dirac mass terms for the gauginos @xcite . in particular just taking simply a dirac gluino can relax strongly the lhc bounds on the neutralino sector , allowing for low majorana neutralino masses , by avoiding universality between gauginos without enhancing effects from the rges @xcite . within extra - dimensional models , even more possibilities arise . on one hand , many very light ( pseudo)goldstino states could remain physical and play a similar role to the gravitino @xcite and on the other the connections between superpartner masses become more involved . a very interesting class of models was proposed recently in @xcite and has the gravitino as a bulk field in a @xmath95-dimensional theory . the remaining particles in the mssm mass spectrum are localized on a 4-dimensional brane . supersymmetry breaking occurs in a @xmath96-brane , with @xmath97 , via an @xmath60-term denoted here by @xmath98 . after supersymmetry breaking , the mssm particles acquire masses of the order of @xmath99 tev , where @xmath35 is the size of the extra - dimensions . the gravitino kk tower instead starts at @xmath100 which can be arbitrarily small , properly choosing @xmath98 , @xmath2 , @xmath101 and @xmath102 , where @xmath102 is the fundamental gravitational scale . notably , decays of the lightest particle in the mssm mass spectrum into slightly lighter kk gravitinos are favored in this model . from the lhc perspective this model can hence resemble a 4-dimensional supersymmetric theory with a compressed mass spectrum . as a function of the sgoldstino mass @xmath103 . here we assume @xmath104 , @xmath105 100 gev , @xmath106 and @xmath107 with @xmath108 mev.,scaledwidth=45.0% ] given the general lagrangian in eqs . ( [ eq : modell ] ) and ( [ eq : lsgoldstino ] ) , we can then compute the relevant two - body processes . we discuss then possible effects of introducting r - parity breaking in the model and find that those effects are negligible in our range of parameters . ./feynman_diagrams the dominant gravitino production contribution comes from photon - photon collisions @xcite . the expression for the corresponding luminosity can be found in app . [ app : luminosity ] . the next step is to calculate @xmath110 . the contributing diagrams are given by @xmath111 & \qquad \qquad\qquad + \parbox{0.25 \textwidth } { \begin{fmfgraph*}(95,75 ) \fmfpen{0.5 thin } \fmfleft{i1,i2 } \fmfright{o1,o2 } \fmfv{label=$\alpha$,label.dist=1}{i2 } \fmfv{label=$\beta$,label.dist=1}{i1 } \fmfv{label=$\nu$,label.dist=1}{o1 } \fmfv{label=$\mu$,label.dist=1}{o2 } \fmf{dashes , tension=0.5}{v1,v2 } \fmf{photon , label=$p_1$,label.dist=1,label.side = right}{i2,v1 } \fmf{photon , label=$p_2$,label.dist=1,label.side = left}{i1,v1 } \fmf{gravitino , label=$k_2$,label.dist=1,label.side = left}{v2,o2 } \fmf{gravitino , label=$k_1$,label.dist=1,label.side = left}{o1,v2 } \fmffreeze \fmfforce{0.95w,0.9h}{r3 } \fmfforce{0.95w,0.1h}{r4 } \fmf{fermionflow , left=0.3,width=0.4}{r4,r3 } \end{fmfgraph * } } + \parbox{0.25 \textwidth } { \begin{fmfgraph*}(95,75 ) \fmfpen{0.5 thin } \fmfleft{i1,i2 } \fmfright{o1,o2 } \fmfv{label=$\alpha$,label.dist=1}{i2 } \fmfv{label=$\beta$,label.dist=1}{i1 } \fmfv{label=$\nu$,label.dist=1}{o1 } \fmfv{label=$\mu$,label.dist=1}{o2 } \fmf{dots , tension=0.5}{v1,v2 } \fmf{photon , label=$p_1$,label.dist=1,label.side = right}{i2,v1 } \fmf{photon , label=$p_2$,label.dist=1,label.side = left}{i1,v1 } \fmf{gravitino , label=$k_2$,label.dist=1,label.side = left}{v2,o2 } \fmf{gravitino , label=$k_1$,label.dist=1,label.side = left}{o1,v2 } \fmffreeze \fmfforce{0.95w,0.9h}{r3 } \fmfforce{0.95w,0.1h}{r4 } \fmf{fermionflow , left=0.3,width=0.4}{r4,r3 } \end{fmfgraph * } } \label{eq : diagrams}\ , .\end{aligned}\ ] ] this process has been calculated in @xcite in the limit of very small scalar and pseudo - scalar masses . contrary to these studies , here we do not make any assumption regarding the sgoldstino masses , and assert no relation between the couplings @xmath1 and @xmath2 . accordingly , for the amplitude of this process we obtain , @xmath112 where the four invariant amplitudes @xmath113 , @xmath114 , @xmath115 and @xmath116 are associated with the exchange of a photino , assumed here to be a mass - eigenstate , a graviton , a scalar and a pseudo - scalar , respectively . they are listed in the appendix [ app : amplitudes ] . for the calculation of @xmath117 and subsequently of the total cross - section we use the mathematica package feyncalc @xcite . we obtain the total cross - section , at leading order which may be enhanced by the couplings @xmath118 . ] @xmath119 , @xmath120 + \mathcal{o}(\sqrt{x})\ , , \label{eq : productioncs}\end{aligned}\ ] ] where @xmath121 , with @xmath122 denoting the photino mass . note that the photino t- and u - channel contribute to the cross - section only at the next order in @xmath123 and therefore the leading order result is independent of the exact value of the photino mass and the possible corrections coming by considering the full neutralino mass matrix instead than the simple photino . the parameters @xmath124 and @xmath125 , with dimension gev , describe the model dependent decay widths of the scalar and pseudo - scalar @xmath126 , respectively . note that due to the non - renormalizable nature of the interaction , the cross - section grows with energy as @xmath127 and therefore the best constraints are obtained for the highest possible energy . the cross - section is depicted in fig . [ fig : csplot ] as a function of @xmath128 for a typical average value of the energy in a sn core @xmath129 . the two limiting cases of very light or very heavy sgoldstinos , corresponding to the plateaus in fig . [ fig : csplot ] , are simply given by @xmath130 so that we can nearly recover the limit of heavy sgoldstinos also by setting @xmath131 . for @xmath132 the results are in agreement with @xcite . the maximal value of the cross - section is obtained at the resonance and depends on the value of the widths @xmath133 . so we expect the supernova environment to be most sensitive to sgoldstino with masses in the 100 mev range . we substitute the cross - section in eq . into eq . and we obtain the luminosity @xmath134 via integration . note that eq . ( [ eq : ms0 ] ) agrees with the results of @xcite and can be used to obtain an upper limit on @xmath39 straightforwardly , without relying on the complex statistical analysis of sec . [ sec : statistics ] . using eq . as explained above , we obtain @xmath135 ev , for @xmath136 gev . this result is comparable with the upper limit in @xcite , though the two expressions differ because of an extra factor of 4 in our expression for @xmath134 , and discrepancies in the numerical evaluation of the relevant formula . this finding will be refined within our likelihood analysis . we have also to consider the mean - free path of the gravitino to compare with eq . [ eq : mfpconstraint ] . the dominant gravitino scattering process in the supernova core is @xmath137 , connected to the production process by crossing symmetry . in fact the diagrams contributing to this scattering process are @xmath138 we calculate the cross - section in the same way as in sec . [ ss : luminosity ] and obtain @xmath139\right ) \nonumber\\ & + & \frac{10}{3 } \frac{m_p^{10}}{s^5 } \log\left[1+\frac{s}{m_p^2}\right ] \bigg ) \nonumber\\ & - & \xi^2 \frac{m_{3/2}^2}{s } \bigg(+\frac{2 s}{3 ( s+m_s^2)}-\frac{5}{6 } + \frac{10}{9 } \frac{m_s^2}{s } \nonumber\\ & - & \frac{5}{3 } \frac{m_s^4}{s^2 } + \frac{10}{3}\frac{m_s^6}{s^3 } \nonumber\\ & + & \frac{10}{3 } \frac{m_s^8}{s^4 } \log\left[\frac{m_s^2}{s+m_s^2}\right]\bigg)\bigg ] + \mathcal{o}(\sqrt{x})\ , . \label{eq : scatteringcs}\end{aligned}\ ] ] finally , the gravitino mean - free - path @xmath17 in the supernova core is @xmath140 with the photon number density @xmath141 given by @xmath142 the average energy of a photon is @xmath143 , corresponding to @xmath144 . for the supernova core temperature we assume @xmath145 mev @xcite . substituting and into we obtain the mean - free - path depending on the sgoldstino masses and couplings as well as on the supernova core temperature . up to now we have assumed r - parity to be conserved . this is of course well - motivated e.g. from the non - observation of proton decay . with conserved r - parity , gravitinos , or any other sparticle for that matter , can be produced in pairs only . if instead r - parity is broken , other channels giving the production of single gravitinos and containing only one supergravity vertex and less suppression by @xmath146 become possible . it is therefore important to check if such additional contributions are stronger than the rpc ones and can modify the constraints on the model . the most important channel in our context is @xmath147 , since the only neutral sm fermions , that can be produced together with the gravitino , are neutrinos . for the purpose of supernova coupling , both gravitino and neutrino escape and therefore this channel has the same effect as the rpc one . we investigate whether this process is able to compete with @xmath148 for the case of leptonic trilinear and bilinear r - parity violations ( rpv ) . + the possible trilinear rpv renormalizable couplings in the superpotential violating only the lepton number , are @xmath149 where @xmath150 and @xmath151 are generation and @xmath152 doublet indices respectively . these couplings are only between the chiral multiplets and so do not involve photons . nevertheless , at one - loop level they can generate an effective photon - photino - neutrino vertex , that has been fully calculated in @xcite . in the loop we can have fermion - sfermion pairs of any particular flavor @xmath153 , where @xmath154 . in principle we also have other types of diagrams contributing to the same channel , e.g. box - diagram , but they include an additional propagator and we therefore expect them to be more suppressed at low energy . with the one - loop effective vertex the contributing diagrams are @xmath155 here we sum over all neutrino flavors @xmath156 in the final state . the blob @xmath157 denotes an effective photon - photino - neutrino vertex , for each fermionic flavour @xmath153 . the explicit expression for the invariant amplitude is given in the appendix [ app : amplitudes ] . + since we only need to perform a rough estimate in order to see the relevance of this channel , we do not consider the full parameter dependence on the superparticle spectra and the full flavour structure of the couplings . we take instead a maximal cross - section scenario with large flavour - democratic couplings @xmath158 , maximal sfermion mixing , @xmath159 , vanishing cp phases and large mass hierarchy in the sfermion masses , i.e. we consider only the contribution of the lighter mass eigenstate sfermion of any flavour , with a common mass @xmath160 . in general the contributions of the two mass eigenstates of the same flavour tend to cancel , so this limit strongly enhances the scattering rate . with these simplifications we obtain for the total cross - section @xmath161 where @xmath162 is the qed coupling and @xmath163 the weinberg angle . note that indeed this channel is suppressed only by two powers of @xmath164 , but on the other hand also by the sixth power of the soft susy breaking masses . we compare this result to the smallest possible cross - section for gravitino pair production i.e. , where we have no contributions from the sgoldstinos , @xmath165 it becomes clear that even under exaggerated choices of parameters single gravitino production due to trilinear rpv plays no role in our context . for gravitinos heavier than @xmath166 ev the two processes do indeed become comparable , but the corresponding gravitino production would be too small to be relevant for sn cooling . alternatively we could also add bilinear r - parity violation ( rpv ) into our model by introducing the term @xmath167 to our superpotential . this term is also motivated by its ability to generate a hierarchical neutrino mass spectrum favored by observations . these models typically contain not only effective trilinear coupling as discussed above , but also neutralino - neutrino mixing since a sneutrino obtains in general a non - zero vacuum expectation value @xmath168 during electroweak symmetry breaking @xcite . in the context of supernova cooling bilinear rpv gives an additional contribution to the process @xmath169 with an intermediate neutrino . this channel is therefore not suppressed by the photino mass ( even if it does influence the neutralino - sneutrino mixing ) . contributing diagrams are @xmath170 the blob in this diagram denotes this time the coupling of the neutrinos to the photon field . since the neutrino is neutral this vertex is not present at tree level of course . however it can again be generated by radiative corrections , which lead to the real charge , magnetic dipole , electric dipole and anapole form factors of the neutrinos , which are strictly constrained by observations @xcite . it is now possible to determine whether the single gravitino production rate is of any relevance compared to the production of gravitino pairs from the same initial state . using the empirical bounds on the neutrino s electromagnetic properties and typical values for the rpv parameters , we find that the ratio of cross - sections is tiny even under the assumption of very heavy sgoldstinos , namely @xmath171 where @xmath172 and @xmath173 is the sm higgs vev . obviously the situation is the same as in the case of trilinear rpv , since the parameter @xmath174 is constrained by the physical neutrino masses to be below @xmath175 for neutralino masses at the electroweak scale . in conclusion , in the area of parameter space relevant to sn cooling the production of gravitinos via photon collisions occurs almost exclusively in pairs . neither modifications of the luminosity nor the mean - free - path can be of relevance making any further calculation or parameter scans including r - parity violations unnecessary . in sec . [ sec : results ] , we derive model independent limits on the mass of a light gravitino from supernovae observations . to this aim , we compare the general 6-dimensional effective theory of the light gravitino interactions ( sec . [ sec : theory ] ) to the bounds discussed in sec . [ sec : data ] in a global statistical analysis . in sec . [ sec : statistics ] we briefly describe the statistical methods used in this study . for a more extended introduction to statistical methods in particle physics and cosmology , we refer to the dedicated literature , e.g. refs . @xcite . .list of model parameters . for each model parameter , this table shows the type of assumed prior pdf , and the corresponding prior range . [ cols="<,^,<,^",options="header " , ] we analyze the supernovae data of sec . [ sec : data ] combining bayesian and frequentist statistics . in the bayesian approach to data analysis , we calculate the posterior probability density function ( pdf ) of the model parameters @xmath176 , where @xmath177 is the dimension of the parameter space . in the present study @xmath178 , and the six model parameters are listed in tab . [ tab : parameters ] . the parameters @xmath179 and @xmath180 vary within a range where the narrow width approximation used in eq . ( [ eq : productioncs ] ) can be applied . note that the resonance in the cross - section results smoothed by the thermal average and therefore the final results will not be strongly dependent on the sgoldstino widths . in general , the posterior pdf , @xmath181 , describes the observer s degree of belief on a certain set of model parameters , after having analyzed the available data , @xmath182 . it can be calculated applying bayes theorem to @xmath183 , the likelihood function of the data : @xmath184 in eq . ( [ eq : bayes ] ) , @xmath185 is the bayesian evidence . it is independent of the model parameters , and it hence plays the role of a normalization constant in the present analysis . @xmath186 is the prior pdf . it contains the observer s knowledge of the model parameters , before having analyzed the available data . in the analyses we assume log - priors for all model parameters , except for @xmath187 and @xmath188 this assumption allows to effectively sample the posterior pdf varying the model parameters over several order of magnitudes . tab . [ tab : parameters ] shows the range of variability that we assume for the different model parameters . the intervals chosen for @xmath61 and @xmath62 allow us to effectively explore the two extreme regimes @xmath189 and @xmath190 , where , respectively , @xmath191 and @xmath192 . we express our results in terms of 1d and 2d marginal posterior pdfs . the 2d marginal posterior pdf of the model parameters @xmath193 and @xmath194 , for instance , is defined as follows @xmath195 similar expressions hold for the other pairs of model parameters . from the 2d marginal posterior pdfs , we construct the associated 95% credible regions . they contain the 95% of the total posterior probability , and are such that @xmath196 at any point inside the region is larger than at any point outside the region . when the prior pdf contains more information than the likelihood function , the support of the posterior pdf extends over a large volume in parameter space . in this case , the integral in eq . ( [ eq : marg ] ) mostly depends on the integration volume , rather then on the information contained in the likelihood function . a similar dependence on the integration volume in eq ( [ eq : marg ] ) is called a `` volume effect '' . in the frequentist approach , from the supernovae bounds we extract the 1d and 2d profile likelihoods of the model parameters . the 2d profile likelihood of the model parameters @xmath193 and @xmath194 , for instance , is given by @xmath197 and analogously for the other pairs of model parameters . we use the profile likelihood to construct approximate frequentist confidence intervals from an effective chi - square defined as @xmath198 , where @xmath199 is the absolute maximum of the likelihood function . the bayesian approach and the frequentist approach are complementary . the former allows to obtain limits on the gravitino mass with a relatively small number of likelihood evaluations . the latter is computationally demanding , but it does not crucially depend on the choice of the prior pdf , and it is not affected by volume effects . a simple likelihood function which can be used in presence of an upper limit on the gravitino luminosity is given by the following expression @xmath200 & \qquad\qquad { \rm otherwise } , \end{array } \right . \label{eq : like1}\ ] ] where @xmath201 is an estimate of the experimental error on @xmath202 . following @xcite , in the analyses we adopt a refined version of eq . ( [ eq : like1 ] ) which includes theoretical and experimental errors for the gravitino luminosity , denoted here by @xmath203 and @xmath201 , respectively . the likelihood function used in the calculations is hence given by @xmath204 \nonumber\\ & \times \left[1-z(t_{\rm exp})\right ] + z\left ( \frac{l - l_{\rm exp}}{\tau } \right ) \label{eq : like2}\end{aligned}\ ] ] for @xmath205 m , and @xmath206 otherwise , when the supernovae constrain does not apply . the function @xmath207 in eq . ( [ eq : like2 ] ) is defined as follows @xmath208 and @xmath209 in the calculations we assume @xmath210 and @xmath211 equal to 0.05 . finally , to sample the likelihood function we use the multinest program @xcite , with parameters set to : @xmath212 and tol=@xmath213 . we use our own routines to calculate the cross - sections and the mean - free - path . figures are produced using the programs getplots @xcite and matlab . we now describe the results obtained comparing the effective theory of the light gravitino interactions to the supernovae bounds . in the analyses we use the statistical methods introduced in sec . [ sec : statistics ] . note that on general grounds we expect to exclude a window of gravitino masses , limited on the lower side by the requirement to have a sufficiently long mean - free path in the sn core and on the upper side by the sn luminosity bound . the left panel in fig . [ fig : limits ] shows the limits on the gravitino mass derived within three different realizations of the theory defined in sec . [ sec : theory ] . red lines refer to 1d profile likelihoods , whereas blue lines correspond to 1d marginal posterior pdfs . the red dashed line in the left panel of fig . [ fig : limits ] is the 1d profile likelihood of a model with light scalar and pseudo - scalar fields ( i.e. @xmath214 and @xmath215 ) with fixed sgoldstino coupling @xmath216 . from the 1d profile likelihood of this model , we are able to exclude the mass range @xmath217 at the 95% confidence level ( corresponding to @xmath218 ) . this mass window is comparable ( up to the factor 4 ) to the old exclusion obtained in @xcite with the same assumption on the value of @xmath219 . the blue dashed line in the left panel of fig . [ fig : limits ] is the 1d marginal pdf derived from the same model . compared to the 1d profile likelihood , it increases from its minimum value less steeply and it is not flat at the boundaries of the prior range . the latter effect is related to the prior pdf , which forces to zero the posterior pdf at the edge of the prior range . the red dotted and blue dotted lines in the left panel of fig . [ fig : limits ] are , respectively , the 1d profile likelihood and the 1d marginal posterior pdf of a model with heavy scalar and pseudo - scalar fields ( i.e. @xmath220 and @xmath221 ) . from the 1d profile likelihood of this case , we find the exclusion limits @xmath222 at the 95% confidence level ( corresponding to @xmath223 ) . comparing the 1d profile likelihood to the 1d marginal posterior pdf , we observe the same differences found in the case of light scalar and pseudo - scalar fields . the red solid and blue solid lines in the left panel of fig . [ fig : limits ] are , respectively , the 1d profile likelihood and the 1d marginal posterior pdf obtained fitting the full 6-dimensional effective theory of sec . [ sec : theory ] to the supernovae data , without any assumption on the sgoldstinos masses and coupling . the exclusion limits on the gravitino mass extracted from the 1d profile likelihood of this general model are one of the main results of this paper . analyzing the general 6-dimensional effective theory of the light gravitino interactions , we find the following model independent exclusion limits for the gravitino mass : @xmath224 at the 95% confidence level ( corresponding to @xmath225 ) . we see that the mass window excluded in a model - independent way is much narrower than the ranges for the two limiting cases , partially due to the resonant enhancement in the full cross - section . on the other hand these limits are very robust , since they take into account the uncertainties in the masses and couplings defining the model of sec . [ sec : theory ] in a statistically rigorous manner . the 1d marginal posterior pdf for @xmath39 , that we find marginalizing the full 6-dimensional posterior pdf over the 5 remaining model parameters , is significantly different from the 1d profile likelihood of the same 6-dimensional model . the differences between the two statistical indicators are induced by volume effects produced in the marginalization procedure , as we will see in detail below . the right panel in fig . [ fig : limits ] shows the 1d profile likelihoods and the 1d marginal posterior pdfs for the cross - section @xmath226 at @xmath227 in the three models discussed in the left panel of the same figure . from an analysis based on the general 6-dimensional effective theory of the light gravitino interactions , we extract the 95% confidence level exclusion limits @xmath228 , for @xmath226 at @xmath229 . plane . the parameters @xmath39 and @xmath219 are degenerate in the large @xmath219 limit . inspection of eqs . ( [ eq : productioncs ] ) and ( [ eq : scatteringcs ] ) shows that only the ratio @xmath230 is constrained by observations in this limit.,scaledwidth=35.0% ] we also study additional models with 2 free parameters . in fig . [ fig : limits2 ] we compare their 1d profile likelihoods . a first model is characterized by @xmath231 , @xmath232 , and @xmath219 and @xmath39 as free parameters ( red solid line ) . a second model is defined by @xmath233 , @xmath232 , and again @xmath219 and @xmath39 as free parameters ( blue solid line ) . comparing figs . [ fig : limits ] and [ fig : limits2 ] , we see that the model with @xmath219 and @xmath39 as free parameters and @xmath189 , and the full 6-dimensional model have similar 1d profile likelihoods , and confidence intervals , though the former are slightly narrower . for instance , within the 2-parameter model , we find the 95% confidence level exclusion limits for @xmath39 : @xmath234 , corresponding to @xmath235 . indeed the two parameters @xmath219 and @xmath39 capture the strongest dependence of the cross - section . let us now go back to consider the general case with 6 free parameters . fig . [ fig : correlations ] shows the 2d 95% confidence intervals ( red lines ) and the 95% credible regions ( blue lines ) in the six planes @xmath236 , @xmath237 , @xmath238 , @xmath239 , @xmath240 and @xmath241 , as obtained by comparing the general 6-dimensional theory of sec . [ sec : theory ] to the supernovae constraint . as in fig . [ fig : limits ] , @xmath242 is evaluated at the reference centre - of - mass energy @xmath229 . the bottom - left panel in fig . [ fig : correlations ] shows the results for the pair of model parameters @xmath39 and @xmath134 . in this panel , the 2d 95% confidence interval and the 95% credible region separate in two disconnected areas . one at large @xmath39 generated by the supernovae bound on the gravitino luminosity , and one at at small @xmath39 , where @xmath243 m and the supernovae bound does not apply . models with constrained @xmath244 , for finite sgoldstino masses , or large sgoldstino masses , predict @xmath245 . the diagonal structure observed in the bottom - left panel of fig . [ fig : correlations ] reflects this correlation pattern between @xmath134 and @xmath39 . however , in the general 6-dimensional theory of sec . [ sec : theory ] , the dependence of @xmath134 on @xmath39 is less obvious . for this reason we also observe `` out - of - diagonal structures '' in the @xmath236 plane , corresponding to configurations with intermediate values of @xmath246 and @xmath62 , and values of @xmath219 different from @xmath247 . the 2d credible region in the bottom - left panel of fig . [ fig : correlations ] is concentrated along the diagonal of the plane ( @xmath39,@xmath134 ) . this result is related to the volume effect also appearing in fig . [ fig : limits ] . in fact , marginalizing over the large volume in parameter space where @xmath61 , @xmath62 and @xmath219 are large and @xmath62 larger than the supernovae core temperature , and @xmath219 very large . ] leads to an artificial overweighting of configurations following the correlation pattern @xmath248 . therefore , the bayesian approach tends to favour configurations along the diagonal in the plane ( @xmath39,@xmath134 ) , because of spurious volume effects . the presence of out - of - diagonal structures in the 2d profile likelihood also induces the differences observed in fig . [ fig : limits ] between the 1d profile likelihood and the 1d marginal posterior pdf of the general 6-dimensional model . we therefore conclude that only the frequentist approach produces robust and physically relevant results , since the bayesian approach is affected by volume effects , when present supernovae data are used . the central - left and top - left panels in fig . [ fig : correlations ] refer to the pairs of model parameters @xmath237 and @xmath238 , respectively . the pair @xmath238 exhibits a statistical behavior essentially identical to the related pair @xmath236 . also the results found for the pair @xmath237 admit an analogous interpretation , with one important difference , however . in the case of the pairs @xmath237 , the two parameters are positively correlated and the correlation pattern expected for fixed @xmath244 at small scalar masses is @xmath249 . the remaining panels show the correlations between @xmath134 , @xmath242 and @xmath17 . as expected , we find that @xmath134 and @xmath242 are positively correlated , whereas @xmath134 and @xmath17 , and @xmath242 and @xmath17 are negatively correlated . finally , we briefly comment on possible degeneracies between pairs of model parameters . inspection of eqs . ( [ eq : productioncs ] ) and ( [ eq : scatteringcs ] ) shows that only the ratio @xmath230 can be constrained by observations in the large @xmath219 limit . this degeneracy is captured by the numerical analysis , as fig . [ fig : m32xi ] shows . calculating confidence intervals , we do not observe strong degeneracies between other pairs of model parameters not even between @xmath250 and @xmath251 in the heavy sgoldstino limits . we have revisited supernovae constraints on ultralight gravitinos in a more modern and model - independent way , extending them to general susy breaking models and including as well a discussion of possible effects of lepton number breaking r - parity violation . for what regards the latter , we find that the single gravitino production channels are always negligible compared to the r - parity conserving two gravitino processes for the ultralight masses relevant for the supernovae . our analysis was therefore focussed on the rpc general models , where we let the sgoldstino masses and couplings vary within reasonable expected ranges . we found that there are two very clearly distinguishable regimes of light and heavy sgoldstinos , where light / heavy is meant as compared to the sn average thermal energy scale of @xmath252 mev . in the case of light sgoldstinos , the cross - section is larger and our frequentist analysis is in agreement with the old results and gives comparable limits , for sgoldstino couplings @xmath253 . in the heavy sgoldstino region , the cross - section is slightly reduced and in order to compensate that , the excluded window is therefore slightly shifted to lower gravitino masses . note that the two different excluded ranges do overlap and are given at the 95% confidence level by @xmath254 the bayesian analysis for both cases excludes very similar ranges of the gravitino mass , but is unfortunately limited by volume effects . considering the whole range of parameters and free sgoldstino masses , we are able to set a new model - independent limit on the gravitino mass as @xmath255 at the 95% confidence level . we note that this excluded window corresponds more or less to the overlap region of the two limiting cases and is therefore reduced to be quite narrow . we also mention that a simplified analysis based on the model parameters @xmath39 and @xmath219 only , produces exclusion limits similar to the model independent limit in eq . ( [ eq : final ] ) , though slightly narrower . the gravitino mass window in eq . ( [ eq : final ] ) is within the region ruled out by bbn in the limit of massless sgoldstinos never in thermal equilibrium . for instance , one can exclude @xmath256 , demanding that the goldstino decoupling temperature is larger than about 100 mev @xcite . we however expect that for heavy sgoldstinos the same argument of @xcite would produce weaker limits on @xmath39 . if also the sgoldstinos are assumed to be in thermal equilibrium in the early universe , then the bbn lower bound on the gravitino mass becomes stronger , and of the order of 1 ev @xcite . in contrast to the bbn bounds mentioned above , our limit ( [ eq : final ] ) does not depend on assumptions made for the thermal bath at goldstino or sgoldstino decoupling . the cross - sections used in deriving the bbn bounds discussed here were first computed in @xcite . given a specific model for supersymmetry mediation , our limits ( [ eq : final ] ) can be used to exclude a range of supersymmetry breaking scales and correspondingly also superpartner masses . in the simplest models , where a single parameter , e.g. the @xmath60-term @xmath88 , determines @xmath39 , @xmath94 and @xmath257 , the resulting exclusion limits are not at all competitive with the corresponding collider limits , since they rule out supersymmetry breaking scales of the order of @xmath258 gev , which have already been probed at lep and at the lhc for a variety of models . limits on the scale of supersymmetry breaking derived from ( [ eq : final ] ) remain interesting instead for gauge mediation models with a very large number of messengers or for models which are difficult to probe at colliders , like those with compressed spectra , and especially for models where the gravitino mass @xmath39 is partially `` decoupled '' from @xmath94 and @xmath257 , as in the examples discussed in sec . [ sec : theory ] . note that the analysis at large sgoldstino couplings is unfortunately limited by the degeneracy between the gravitino mass and the sgoldstino coupling @xmath219 , so that the prior and the range chosen for such parameter influence the bayesian analysis and do not allow to pinpoint the gravitino mass completely model - independently . therefore other constraints ( or definite assumption on the susy breaking sector , such as those we tried to relax ) are needed to be able to draw definite statements . in principle our analysis could be extended to include also limits from other astrophysical objects and that would sample a different energy range and improve the constraints . for the case of light sgoldstinos , the inclusion of the process of sgoldstino production could bring additional constraining power , but at the cost of a strong dependence on the unknown coupling of the sgoldstino sector . r.c . and l.c . acknowledge partial support from the european union fp7 itn invisibles ( marie curie actions , pitn - ga-2011 - 289442 ) . t.e . acknowledges support from the dfg research training group grk1147 . the luminosity of gravitinos produced via @xmath259 is given by @xcite @xmath260 which corresponds to the gravitino energy emitted per unit time from a volume @xmath261 at a temperature @xmath262 via the collision of photons in thermal equilibrium . the temperature enters the luminosity via the photon bose - einstein distribution function @xmath263 , given by @xmath264 in terms of the cross - section , the luminosity can be expressed as @xmath265 this expression follows from the definition of total cross - section . in order to simplify the integration , we assume @xmath266 for all @xmath267 , which is a reasonable assumption in the supernova core @xcite . hence , @xmath268 where @xmath269 . in general , the cross - section depends on the photon momenta , or rather on the mandelstam variable @xmath270 . we express the photon momenta @xmath271 and @xmath272 in spherical coordinates @xmath273 and obtain @xmath274 for the sake of completeness , we list the four invariant amplitudes relevant in the calculation of gravitino pair production via photon collision , @xmath275\\ i\mathcal{m}_{\text{graviton } } & = \frac{1}{2(p_1+p_2)^2 m_p^2}\bigg ( ( \epsilon_1\cdot\epsilon_2)p_1^{\lambda}p_2^{\rho } \nonumber\\ & + \frac{1}{2}\big ( ( p_1\cdot \epsilon_2)(p_2 \cdot\epsilon_1)-(p_1\cdot p_2)(\epsilon_1\cdot\epsilon_2)\big)\eta^{\lambda\rho}\nonumber\\ & + ( p_1\cdot p_2)\epsilon_1^{\lambda}\epsilon_2^{\rho}-(p_2 \cdot \epsilon_1)\epsilon_2^{\rho}p_1^{\lambda}-(p_1\cdot \epsilon_2)p_2^{\lambda}\epsilon_1^{\rho } \nonumber\\ & + ( \rho\leftrightarrow\lambda)\bigg)\nonumber\\ & \times \overline{\psi}^{+\,\mu}(k_2 ) \bigg[\epsilon_{\mu\sigma\nu(\lambda}\gamma^5\gamma_{\rho)}(k_2-k_1)^{\sigma } \nonumber\\ & + \frac{i}{2}\epsilon_{\mu\sigma\nu(\lambda}\gamma^5 \left\ { \gamma^{\sigma},\sigma_{\rho)\tau}\right\}\nonumber\\ & -2im_{3/2}(2\eta_{\mu(\lambda}\eta_{\rho)\nu}-\eta_{\mu\nu}\eta_{\lambda\rho})\bigg]\psi^{-\,\nu}(k_1)\ , , \label{eq : amp - graviton}\displaybreak[0]\\ i\mathcal{m}_{\text{scalar } } & = \frac{i c d m_{3/2}}{m_p^2 } \frac{1}{2(p_1+p_2)^2-m_s^2}\nonumber\\ & \times \epsilon_1^{\alpha}\epsilon_2^{\beta } \;\left ( ( p_1\cdot p_2)\eta_{\alpha\beta}-p_1^{\beta}p_2^{\alpha}\right ) \nonumber\\ & \times \eta_{\mu\nu}\ ; \overline{\psi}^{+\,\mu}(k_2)\psi^{-\,\nu}(k_1)\ , , \label{eq : amp - scalar}\displaybreak[0]\\ i\mathcal{m}_{\text{pseudoscalar } } & = -\frac{i c d}{2 m_p^2}\;\frac{1}{(p_1+p_2)^2-m_p^2 } \nonumber\\ & \times \epsilon_1^{\alpha}\epsilon_2^{\beta}\,p_1^{\kappa}p_2^{\lambda}\,\epsilon_{\kappa\lambda\alpha\beta}\,(p_1+p_2)^{\zeta}\,\epsilon_{\mu\delta\nu\zeta}\ ; \nonumber\\ & \times \overline{\psi}^{+\,\mu}(k_2)\gamma^{\delta}\psi^{-\,\nu}(k_1)\label{eq : amp - pseudoscalar}\ , , \end{aligned}\ ] ] where @xmath276 . here the notation is the same of @xcite . here we give the invariant amplitude for single gravitino production via photon collision in presence of trilinear rpv . the effective photon - photino - neutrino vertex used in this calculation is given in @xcite . adopting the same notation of @xcite , we find @xmath277\label{eq : rpvamp}\end{aligned}\ ] ] where @xmath278 . here @xmath279 is the invariant amplitude for the process @xmath280 , where the index @xmath156 labels the flavor of the final state neutrino . in eq . ( [ eq : rpvamp ] ) we sum over the flavor of the fields in the loop . in addition , we denote the trilinear coupling constants by @xmath281 and introduce the gauge coupling @xmath282 , where @xmath283 is the electric charge of the fermion - sfermion pair in the loop . in the invariant amplitude , the additional gauge couplings @xmath284 and @xmath285 are given by @xmath286 we also introduce the constants @xmath287 , where @xmath288 is the number of colors . finally , the sfermion mixing angles , cp phases , and masses are donated by @xmath289 , @xmath290 , @xmath291 and @xmath292 , respectively . we conclude with the invariant amplitude for single gravitino production via photon collision in presence of bilinear rpv . using the notation of @xcite , we find @xmath293 \psi^{\mu}(k_1 ) \nonumber\\ \nonumber\\ & + & \left((p_1,\alpha ) \leftrightarrow ( p_2,\beta)\right ) \ , . \label{eq : rpvamp2}\end{aligned}\ ] ] where @xmath294 is the photon - neutrino - neutrino vertex , which at one loop is given by @xmath295 the functions @xmath296 , @xmath297 , @xmath298 and @xmath299 are the charge , magnetic dipole , electric dipole and anapole neutrino form factors , respectively . at zero momentum transfer , i.e. @xmath300 , they are strongly constrained by observations @xcite : @xmath301 evaluating eq . ( [ eq : vergleich1 ] ) , we assume @xmath302 , with @xmath303 gev . bionta r m , blewitt g , bratton c b , casper d , ciocio a , claus r , cortez b , crouch m , dye s t , errede s , foster g w , gajewski w , ganezer k s , goldhaber m , haines t j , jones t w , kielczewska d , kropp w r , learned j g , losecco j m , matthews j , miller r , mudan m s , park h s , price l r , reines f , schultz j , seidel s , shumard e , sinclair d , sobel h w , stone j l , sulak l r , svoboda r , thornton g , van der velde j c and wuest c 1987 _ phys lett . _ * 58*(14 ) 14941496 hirata k , kajita t , koshiba m , nakahata m , oyama y , sato n , suzuki a , takita m , totsuka y , kifune t , suda t , takahashi k , tanimori t , miyano k , yamada m , beier e w , feldscher l r , kim s b , mann a k , newcomer f m , van r , zhang w and cortez b g 1987 _ phys . rev . lett . _ * 58*(14 ) 14901493

we have obtained _ spitzer _ irs 5.535 m spectroscopy of the debris disk around pictoris . in addition to the 10 silicate emission feature originally observed from the ground , we also detect the crystalline silicate emission bands at 28 and 33.5 . this is the first time that the silicate bands at wavelengths longer than 10 have ever been seen in the pictoris disk . the observed dust emission is well reproduced by a dust model consisting of fluffy cometary and crystalline olivine aggregates . we searched for line emission from molecular hydrogen and atomic [ ] , , and gas but detected none . we place a 3 upper limit of on the h s(1 ) gas mass , assuming an excitation temperature of=100 . this suggests that there is less gas in this system than is required to form the envelope of jupiter . we hypothesize that some of the atomic gas observed in keplerian rotation around pictoris may be produced by photon - stimulated desorption from circumstellar dust grains .

0705.3023

the a5v star @xmath1 pictoris ( hd 39060 = hr 2020 ) at a distance @xmath9 possesses a spectacular edge - on debris disk imaged in scattered light and thermal emission that extends to radii @xmath101400@xmath11 @xcite . an age of @xmath1212 myr has been estimated for for the central star based on the properties of late - type members of the @xmath1 pic moving group @xcite . the dust in this system is believed to be replenished from a reservoir , such as collisions between parent bodies or sublimation of comets because the estimated lifetime for the dust under radiation pressure , poynting - robertson drag , and grain - grain collisions is a few orders of magnitude shorter than the age of the system @xcite . there may be evidence for the presence of large bodies in the @xmath1 pictoris disk that perturb dust grains and parent bodies and instigate collisional cascades between them . a 5@xmath13 warp at radii @xmath480 au has been observed in scattered light using stis on _ hst _ and may be generated by either a brown dwarf close to the star ( at distances @xmath43 au ) or a 0.17@xmath14 planet at larger distances from the star ( 50@xmath11 ) @xcite . the @xmath1 pictoris disk may also possess infalling , sublimating , refractory bodies . time - variable , high velocity , non - periodic , red - shifted atomic absorption features have been observed toward @xmath1 pictoris at ultraviolet ( uv ) and visual wavelengths that vary on timescales as short as hours . the velocity of the atoms , typically 100400@xmath15 , is close to the free fall velocity at a few stellar radii , suggesting that the absorption is produced as stellar photons pass through the comae of infalling bodies at distances @xmath46@xmath11 from the star @xcite . the origin of the micron - sized dust grains required to account for the observed scattered light and thermal emission is currently not well - constrained . the broad 9.7@xmath2 silicate emission feature and the narrow 11.3@xmath2 crystalline olivine emission feature observed toward @xmath1 pictoris appear grossly similar to those observed toward comets halley , bradford 1987s , and levy 1990 xx @xcite , suggesting that the grains may be cometary . models of cometary grains , idealized as fluffy aggregates of interstellar silicate cores with carbonaceous organic refractory mantles ( and additional ice mantles at distances larger than the snow - line ) , in radiative equilibrium with the central star are able to reproduce the observed 10@xmath2 silicate feature and the spectral energy distribution ( sed ) at ir through millimeter wavelengths @xcite . spatially - resolved studies of the silicate emission feature , obtained using comics on the subaru telescope , suggest that the shape of the 10@xmath2 feature changes as a function of position in the disk , with large and crystalline grains concentrated at the center and small grains located in ring - like bands at 6.4 , 16 , and 29@xmath11 from the central star @xcite . the rings of small grains may be explained by collisions between large orbiting bodies . high - resolution , mid - ir imaging of @xmath1 pictoris , obtained with trecs on the gemini south telescope , has resolved a prominent clump on the south - west side of the disk at 52@xmath11 from the central star which may also indicate that a recent collision has occured in the @xmath1 pictoris disk @xcite . spatially resolved visual spectroscopy of @xmath1 pictoris has revealed a disk of atomic gas in keplerian rotation , observed via scattered emission from , , , , , , , , and . the atomic gas possesses a ne / sw brightness asymmetry and an inner warp similar to that observed in the dust disk although the gas extends to larger heights than the dust @xcite . estimates of the radiation pressure acting on fe and na atoms suggest that these species should be not be in keplerian rotation but should be accelerated to terminal outflow velocities @xmath12100s1000s km / sec @xcite . @xcite showed that a ring of neutral hydrogen at distance 0.5 au could slow down the radial flow of gas . however , @xcite have recently proposed that the gas will not be radially accelerated if the carbon is overabundant ; their model does not require a population of undetected atomic hydrogen . since carbon does not experience strong radiation pressure ( @xmath16 = @xmath17 @xmath18 0 ) and also has a large ionization fraction ( @xmath120.5 ) , they suggest that coulomb interactions between and other ions reduce the effective radiation pressure on the bulk gas . in this case , the ions and neutral atoms in the disk couple together into a fluid , with an effective radiation pressure coefficient , that is bound to the system and that brakes the gas if @xmath19 @xmath4 0.5 . in particular , they suggest that if the carbon abundance is @xmath1010@xmath20 solar , then all the atomic gas will be retained . measurements of the column density of the stable component of atomic carbon ( with zero velocity relative to the star ) via absorption from ( @xmath21p ) @xmath221613 @xcite and absorption from @xmath221036 and @xmath23 @xmath221037 superimposed on chromospheric @xmath221038 , suggest that the bulk of the atomic gas is composed of carbon with a c / fe abundance ratio that is 16 times the solar value and an ionization fraction of 0.5 @xcite . we report the results of a _ spitzer _ irs study of the dust and gas around @xmath1 pictoris , building on the model for the composition and spatial distribution of the grains by @xcite and placing 3@xmath3 upper limits on the line emission from h@xmath6 s(2 ) , s(1 ) , s(0 ) and [ ] , , and gas . based upon the similarity in the spatial distribution of the dust and gas observed in scattered light , we hypothesize that the dust and gas in this system are physically associated and that the observed gas is secondary ; it has been produced from circumstellar material since the star formed . we quantitatively show that the observed , seen in keplerian rotation around the star , can be produced via photon - stimulated desorption in a time that is shorter than the residence time of the gas in the disk . we obtained ir spectrograph ( irs ) @xcite spectral mapping mode observations of @xmath1 pictoris using the short - low ( sl ; 5.214 @xmath2 ; @xmath24 ) , short - high ( sh ; 9.919.6 @xmath2 ; @xmath25 ) , and long - high ( lh ; 18.737.2 @xmath2 ; @xmath25 ) modules on the _ spitzer space telescope _ @xcite . the sl2 slit has a size 3.6@xmath2657@xmath27 ; the sl1 slit has a size 3.7@xmath2657@xmath27 ; the sh slit has a size [email protected]@xmath27 ; the lh slit has a size [email protected]@xmath27 . both the low and high resolution observations were made with the spectrograph long slits aligned along the position angle of @xmath1 pictoris disk to within 5@xmath13 . the sl2 ( aor key : 8972288 ) and sl1 ( aor key : 8972544 ) spectra were obtained on 16 november 2004 by stepping the slit in 7 and 11 1.8@xmath27 intervals perpendicular to the disk , respectively , with the center position centered on the disk . each sl2 and sl1 map position was observed using 9 and 14 cycles , respectively , of 6 second ramps . the sh ( aor key : 4879616 ) and lh ( aor key : 4876800 ) spectra were obtained on 15 december 2003 and 4 march 2004 , respectively , by stepping the slit in 3 2.4@xmath27 and 4.8@xmath27 intervals perpendicular to the disk , respectively , with the center position centered on the disk . each sh and lh map position was observed using 25 and 24 cycles , respectively , of 30 and 14 second ramps . high accuracy peak - up , with the blue ( 13.318.7@xmath2 ) array , on the nearby star hd 38891 ( located 12@xmath28 north of @xmath1 pictoris ) , was used to center the disk in the sl2 , sl1 , and sh slits ; high accuracy peak - up , with the red array ( 18.526@xmath2 ) , on hd 38891 was used to center the disk in the lh slit . the raw data were processed into calibrated ( flat - fielded , stray light corrected ) 2-dimensional spectra with version s11.0.2 of the ssc pipeline . we removed masked and rogue pixels from the basic calibrated data by interpolating between their nearest neighbors , coadded the resulting images , and extracted 1-dimensional spectra using the irs team s smart program @xcite . since separate nod observations of the sky were not made when @xmath1 pictoris was observed , we approximated the sl sky background as a constant at each wavelength and subtracted this value from each pixel in the spatial direction across the source . since the sl slits are long enough to sample the psf in the spatial direction , we fit 1-dimensional gaussians to the sl spectra in the spatial direction to determine the positions of the source in the slits . we extracted our low resolution spectra by summing over a window whose width varied as a function of wavelength to account for the changing size of the point spread function and was centered on the source position . finally , we multiplied the resulting spectra with an @xmath29 lac relative spectral response function ( rsrf ) , obtained by dividing an @xmath29 lac template spectrum with an observed @xmath29 lac sl spectrum @xcite . since point sources fill the high resolution slits , we extracted our high resolution spectra ( without background subtraction ) by summing over all of the pixels in the sh and lh slits . the extracted high resolution spectra were multiplied by a @xmath30 dra high resolution rsrf . we constructed a spectrum that is consistent with the lh slit , the largest slit we used to observe @xmath1 pictoris , centered on the disk . since the source is extended in our irs observations , we added multiple sl measurements together to construct an aperture equal to the lh aperture . since our sh map was not large enough to cover the lh central slit position , we scaled our sh spectrum by a factor of 1.068 to match the sl and lh spectra , using the ll spectrum at 5.1939 @xmath0 m to 9.9232 @xmath0 m , the sh spectrum at 9.9461 @xmath0 m to 19.3179 @xmath0 m , and the lh spectrum at 19.2636 @xmath0 m to 35.9892 @xmath0 m . the final spectrum of @xmath1 pic ( shown in figure [ fig : sedobs ] ) is that of a 11@xmath27@xmath2022@xmath27 region centered on the star and oriented along the disk . the signal : noise in our spectrum is very high . over spans of many spectral resolution elements , the sensitivity of our spectrum is limited by flat - fielding errors to 1 - 2% of the continuum flux density . over smaller spans , @xmath31 10 spectral resolution elements or so , the sensitivity is limited only by photon noise from the background and target , and is significantly better : 0.1 - 0.5% of the continuum flux density in sh and lh , and 0.05 - 0.2% in sl , consistent with the point - to - point scatter in the spectrum over such small spans , and ( perforce ) consistent with the _ spitzer_-irs facility sensitivity tool , spec - pet . thus , every spectral feature visible in our @xmath1 pic spectrum that is at least as wide as the spectral resolution is real , and the upper limits on weaker spectral lines are significantly smaller than 1% of the continuum . for simplicity we have calculated such upper limits directly from the point - to - point variation of the continuum - subtracted spectrum . figure [ fig : sedobs ] plots the 5.236@xmath2 irs spectrum and the phoenix model stellar atmospheric spectrum with effective temperature @xmath32 and surface gravity log@xmath33 = 4.2 . figure [ fig : sedobs ] clearly shows that the dust excess emission above the stellar atmospheric radiation is even seen at wavelengths as short as @xmath3458@xmath2 , indicating that there may exist a population of hot dust ( @xmath12400600@xmath8 ) at a distance of @xmath410@xmath11 from the central star [ see figure 1 of @xcite ] . mostly prominent in figure [ fig : sedobs ] are the broad 10@xmath2 feature ( which reveals the presence of both amorphous silicates and crystalline silicates in the @xmath1 pictoris disk ) , and the 28@xmath2 and 33.5@xmath2 crystalline olivine emission features ( see the inset in figure [ fig : sedobs]a ) which closely resemble the @xcite model spectrum ( see their figure 6c).fe@xmath35sio@xmath36 grains also have a rich set of features in the mid - ir . but their peak wavelengths and relative feature strengths are inconsistent with the irs spectrum ( e.g. see chihara et al . olivine mg@xmath37fe@xmath38sio@xmath39 is a mixture of forsterite mg@xmath40sio@xmath39 and fayalite fe@xmath40sio@xmath39 with a mixing proportion @xmath41 . in this work as well as in li & greenberg ( 1998 ) we adopt the refractive indices of mg@xmath42fe@xmath43sio@xmath39 ( mukai & koike 1990 ) . although in principle it is possible to infer the mg : fe ratio from the locations and strengths of the ir features ( e.g. see koike et al . 1993 , 2003 , fabian et al . 2001 ) , neither the irs spectrum reported here reveals a large number of crystalline silicate features nor does it cover a sufficiently broad wavelength range for constraining the exact crystal composition . as shown in figure 2 , the dust model consisting of crystalline mg@xmath42fe@xmath43sio@xmath39 olivine reproduces the ir features reasonably well . ] this is the first time that the silicate bands at wavelengths longer than 10@xmath2 have ever been detected in the @xmath1 pictoris disk . for comparison , we also show in figure [ fig : sedobs ] the iras broadband photometry of @xcite , the photometry of @xcite at 8.8 , 10.3 , 11.7 and 12.5@xmath2 obtained using the _ big mac _ spectral filters at the nasa _ infrared telescope facility _ ( irtf ) , the irtf @xmath44 band ( 10.1@xmath2 ) and @xmath45 band ( 20@xmath2 ) photometry of @xcite , the irtf 2.613.5@xmath2 intermediate resolution ( @xmath46 ) spectrometry of @xcite , and the _ kuiper airborne observatory _ ( kao ) 47 and 95@xmath2 photometry of @xcite . while the photometry of @xcite and the irtf @xmath44 band photometry of @xcite agree with the spectrometry of @xcite very well , they are below the irs spectrum by a factor of @xmath121.94 ( see figure [ fig : sedobs]b and figure [ fig : sedmod]b ) . this difference is much larger than the calibration uncertainty of irs and is attributed to the fact that the spitzer apertures ( @xmath47 for sl , @xmath48 for sh , @xmath49 for lh ) are much larger than the apertures of ground - based observations [ e.g. @xmath50 diameter of @xcite , @xmath51 of @xcite , @xmath52 diameter of @xcite ] , and therefore spitzer detects more flux from extended material than in previous observations.3.6@xmath53 , approximately the same as @xcite s spectroscopic observation ) is close to that of @xcite . ] the nature ( e.g. size , composition , and morphology ) of the dust and its spatial distribution in protoplanetary and debris disks is mostly revealed through the interaction of the dust with the electromagnetic radiation of the central star : absorbing , scattering and polarizing the stellar radiation and re - radiating the absorbed uv / visible photons at longer wavelengths , ranging from the ir to submillimeter and millimeter . since the disk around @xmath1 pictoris subtends more than 100@xmath27 , it has been studied in great detail . there exists a vast variety of observational data for the dust in this disk ( see @xcite ) , including ( 1 ) imaging observations of scattered light @xcite ; ( 2 ) imaging observations of dust thermal emission in the mid - ir @xcite and submillimeter @xcite ; ( 3 ) mid - r narrow band photometry @xcite , as well as broadband photometry in the mid - ir @xcite and millimeter @xcite ; ( 4 ) optical @xcite and near - ir @xcite imaging of polarized light ; and ( 5 ) mid - ir spectroscopy @xcite . to accurately describe the dust grains in the @xmath1 pictoris disk , all of the aforementioned observational data must be modeled _ simultaneously_. while recognizing the importance of a comprehensive , simultaneous modeling of all of the @xmath1 pictoris observations , in this paper we present a simple model focusing on our _ spitzer _ irs spectrum . we defer the comprehensive modeling of the dust around @xmath1 pictoris that attempts to simultaneously fit all of the observed data to a separate paper ( a. li et al . 2007 , in preparation ) . we model the _ spitzer _ irs spectrum of @xmath1 pictoris building on the work of @xcite which includes ( 1 ) a population of cometary dust , composed of low - density porous aggregates of amorphous silicate core - carbonaceous mantle grains , based on the cosmic abundance consideration ( see appendix a of li & lunine 2003 ) . ] and ( 2 ) a population of crystalline olivine aggregates with the same size and spatial distributions as the amorphous cometary grains ( but actually we do not require crystalline dust outside @xmath1260@xmath11 from the star since at distances @xmath1060@xmath11 from the star silicate grains are too cold to emit at the characteristic mid - ir bands ) . for the amorphous cometary and crystalline olivine fluffy aggregates , we assume a vacuum volume fraction @xmath54=0.90 ( see 2 of li , lunine , & bendo 2003 for justification ) and approximate the porous grains as spheres with radii @xmath55 . we assume a power - law dust size distribution @xmath56 , with a minimum grain radius @xmath57 , and a maximum grain radius @xmath58 ( see li , lunine , & bendo 2003 for justification ) . similar to @xcite , we assume a modified power - law spatial distribution for the dust , @xmath59 $ ] . this functional form , peaking at @xmath60 , behaves like a power - law @xmath61 at large distances ( @xmath62 ) . at @xmath63 , the @xmath64 term dominates , approximating an increasing profile in the inner , dust - relatively - depleted region . we take @xmath65 and @xmath66 as derived from scattered light modeling @xcite . the inner boundary @xmath67 of the disk is taken to be the location inside which silicate dust sublimates . for the @xmath1 pictoris disk , micron - sized silicate grains possess an equilibrium temperature @xmath68 at @xmath69 ; therefore , we take @xmath70 . we take @xmath71=2000@xmath11 . in modeling the dust ir emission , we with all other parameters ( i.e. @xmath72 , @xmath73 , @xmath67 , @xmath71 ) pre - chosen are therefore left with only 2 free parameters : @xmath29 the dust size distribution power - index , and @xmath74 which determines the amount of dust in the inner disk region @xmath63 [ we note that @xcite took @xmath74=1 ] . it is found that , with @xmath75 and @xmath76 , we are able to obtain a reasonably good fit to the overall sed and the spitzer irs spectrum including the silicate emission features at 10 , 28 and 33.5@xmath2 ( see figure [ fig : sedmod ] ) . ) is inverse proportional to @xmath74 . with @xmath74=0.15 and @xmath73=2.7 , the dust spatial distribution peaks at @xmath77 . at a first glance , this appears inconsistent with the scattered light images which suggest a dust spatial distribution slope change at @xmath1280120au ( e.g. see @xcite ) . however , the dust spatial distribution from 46au ( @xmath78 ) to @xmath72 is rather flat : @xmath79 only changes by a factor of @xmath182.3 from @xmath80 to @xmath72 , while it drops much more steeply from @xmath72 to 2@xmath72 ( by a factor of @xmath185.8 ) . in a subsequent paper ( a. li et al . 2007 , in preparation ) , we will investigate in detail whether the present dust spatial distribution is able to reproduce the scattered light images . ] the required dust masses are approximately @xmath81 and @xmath82 for the amorphous and crystalline components , respectively . as shown in figure [ fig : sedmod]b , the model is somewhat deficient at @xmath3458@xmath2 and @xmath341724@xmath2 . this is due to the oversimplified dust spatial distribution function adopted above . by including 2 rings or clumps of dust at @xmath83 and at @xmath842030@xmath11 as implied by @xcite and @xcite , it is expected that the model will account for the excess emission at 58@xmath2 and 1724@xmath2 ( see a. li et al . 2007 , in preparation ) . the oversimplification of the assumed dust spatial distribution is also reflected by the model - predicted ir emission from the dust within a 3.6@xmath53 diameter ( corresponding to @xmath1235@xmath11 ; see figure 2a ) : the model - predicted emission from the inner 35@xmath11 region accounts for almost all the observed @xmath22@xmath415@xmath2 emission for the entire disk . although this is consistent with the previous spatially resolved mid - ir spectroscopy which show that 10@xmath2 silicate emission originates from the inner @xmath1220@xmath11 @xcite , it appears to contradict the irs detection of an appreciable amount of silicate emission from the outer @xmath1235@xmath11 ( see figure 1a ) . this will be investigated in detail in a subsequent paper ( see a. li et al . 2007 , in preparation ) . core - accretion models suggest that giant planets accrete their gaseous envelopes on timescales between 1 and 20myr . therefore the @xmath1 pictoris disk , with an age of @xmath8512myr , is an excellent source to search for bulk gas . if the gas has a solar composition , then the bulk gas is expected to be hydrogen . _ iso _ observations indicated line emission from h@xmath6 s(0 ) at 28.2@xmath2 and s(1 ) at 17.0@xmath2 , suggesting that the disk possesses 54@xmath5 warm h@xmath6 with an excitation temperature @xmath7=110@xmath8 @xcite . however , apparently conflicting _ fuse _ observations constrain the circumstellar h@xmath6 lyman series absorption and place a 3@xmath3 upper limit of @xmath86 on the h@xmath6 column density , significantly lower than ( 5500)@xmath87 expected if the _ iso_-detected h@xmath6 were uniformly distributed in an edge - on disk in the beam @xcite , an assumed geometry that is consistent with observations of the atomic gas . detailed studies of gas drag on the dust dynamics suggest that the gas : dust ratio is less than 1 ( or @xmath40.4 @xmath88 molecular hydrogen exists in the disk ) . for example , if the disk possessed 40 m@xmath89 gas , then small grains would collect at distances @xmath10200 au and would increase the scattered light surface brightness by more than a factor of 10 @xcite . searches for 21 cm emission constrain the mass of atomic hydrogen @xmath90 @xcite ; recent chemical models constrain the total mass of hydrogen @xmath9115 m@xmath89 including the molecular component based on these observations @xcite . we searched for emission from h@xmath6 s(2 ) , s(1 ) , s(0 ) and [ ] , , and but did not detect any of these species . we place 3@xmath3 upper limits on their line fluxes toward @xmath1 pictoris ( see table 1 ) . our 3@xmath3 upper limit on the h@xmath6 s(1 ) line flux is @xmath41.2@xmath92 , a factor of @xmath856.4 times lower than the reported _ iso _ detection ( see figure [ fig : h2_spec ] ) . although the _ spitzer _ irs sh slit is a factor of two shorter than the _ iso _ sws slit ( 14@xmath9327@xmath27 ) , our upper limits effectively constrain the h@xmath6 line emission because the warm , bulk h@xmath6 is expected to be located at radii @xmath4100@xmath11 ( @xmath46@xmath27 ) . converting 3@xmath3 upper limits on the line fluxes from any species into gas masses depends sensitively on the assumed gas temperature . detailed models of the heating and cooling of molecular and atomic gas via gas - grain collisions , cosmic rays , line emission , etc . have been used to infer the temperature and chemical structure of the @xmath1 pictoris disk . however , these models depend on the initial gas : dust ratio or gas mass assumed . for example , a 2@xmath5 disk ( with an interstellar gas : dust ratio of @xmath12100 ) heated primarily by gas - grain collisions may possess gas as warm as 100150@xmath8 at intermediate heights at distances of 300500@xmath11 where [ ] fine structure emission at 63.2@xmath2 is the dominant coolant ; while , a 0.2@xmath5 disk may possess gas as warm as 300@xmath8 at lower heights and similar radii in the disk where [ ] and h@xmath6 rotational / vibrational emission are the dominant coolants @xcite . estimates of the total gass mass , inferred from the measured column densities of atomic species and the scattered light gas density profile of , suggest that the @xmath1 pictoris disk contains @xmath127.4@xmath2010@xmath94 @xmath88 measured gas , corresponding to a gas : dust ratio @xmath120.019 , significantly less than assumed in many detailed chemical models . even if the gas has a solar hydrogen abundance relative to the heavy elements , then the disk possesses @xmath123.7@xmath2010@xmath95 @xmath88 gas , corresponding to a gas : dust ratio @xmath120.093 ( a. roberge , private communication ) , still significantly less than the gas : dust ratios of 100 or 10 assumed in these models . to estimate an upper limit on the mass of h@xmath6 based on our s(1 ) line flux upper limit , we must assume a gas excitation temperature . gas temperatures can be inferred from ( 1 ) detailed models of gas in thermal balance that calculate the composition and density structure of disks in addition to the temperature structure ( as described above ) or ( 2 ) observations if multiple electronic transitions are observed . far - uv observations of @xmath1 pictoris have constrained the bulk circumstellar gas temperature . analysis of the ( @xmath21p ) @xmath221613 and @xmath221561 multiplets suggest that the excitation temperature of the stable component , @xmath7= 50100@xmath8 @xcite . since bulk hydrogen with a gas : dust ratio of 100 has not been directly detected ( as assumed for the kamp & van zadelhoff models ) , we rely on the measured excitation temperature to act as a guide for the bulk gas kinetic temperature @xmath96=50100@xmath8 . therefore , we place 3@xmath3 upper limits on the mass in each of the listed species assuming gas temperatures @xmath7=50 and 100@xmath8 ( see table 1 ) . the total flux produced by @xmath44 atoms or molecules @xmath97 where @xmath98 is the distance to the star , @xmath99 is the energy of the radiated photons , @xmath100 is the fraction of atoms or molecules in level @xmath101 , and @xmath102 is the transition probability . since the temperature of any h@xmath6 is uncertain , we plot the 3@xmath3 upper limits on the h@xmath6 mass obtained from our s(1 ) and s(0 ) line fluxes as a function of temperature ( see figure [ fig : h2_mass ] ) . if @xmath103=100k , we estimate that @xmath417@xmath5 h@xmath6 hydrogen remains in the disk from our h@xmath6 s(1 ) upper limit . the nondetection of h@xmath6 , [ ] , , and emission toward @xmath1 pictoris is consistent with _ observations of other 1030myr old disks . _ iso _ observations of the 20myr old dusty a1v star 49 ceti have indicated line emission from h@xmath6 s(0 ) at 28.2@xmath2 , suggesting that the 49 cet disk possesses @xmath10110@xmath5 warm h@xmath6 with an excitation temperature @xmath104 @xcite . however , high resolution _ spitzer _ irs observations place 3@xmath3 upper limits on the h@xmath6 s(0 ) line flux that are a factor @xmath129 lower than the reported detection @xcite . a recent _ spitzer _ irs search for h@xmath6 emission from warm gas and millimeter search for @xmath105co from cool gas around 8 sun - like systems with ages @xmath430 myr measured upper limits @xmath40.04 @xmath106 gas within a few au of the disk inner radius ( 140 au ) and @xmath40.04 @xmath88 gas at 1040 au @xcite . _ spitzer _ irs observations place 5@xmath3 upper limits on the line flux from h@xmath6 s(1 ) and s(0 ) toward hd 105 , a g0v member of the tucana - horolgium association at 40@xmath107 from the sun , suggesting that less than 1500 , 12 , and 0.95@xmath5 , respectively , at @xmath103=50 , 100 , or 200@xmath8 remains in this disk at an age of 30myr @xcite . together , these observations suggest that warm molecular hydrogen , with @xmath108 = 100 k , dissipates from circumstellar disks on timescales @xmath410 myr . the origin of the observed stable atomic gas in keplerian rotation around @xmath1 pictoris ( described in the introduction ) , like the micron - sized dust grains , is currently not well - constrained . there have been suggestions in the literature that the stable gas component is produced by infalling refractory bodies @xcite or by collisions between dust grains @xcite . we consider these possibilities and hypothesize that some of the observed atomic sodium is generated by photon - stimulated desorption from sio@xmath6-like surfaces . infalling refractory bodies undoubtably contribute at least some portion of the observed gas but whether all of the gas is generated by the evaporation of parent bodies is uncertain . the spatial distribution of the gas may provide a clue to its origin . the atomic gas is detected to distances of 150200@xmath11 on the southwest side and distances of 300350@xmath11 on the northeast side of the disk . these distances correspond to the inner regions of the dusty disk that has been detected in thermal emission and scattered light to distances @xmath101400 au . the sw / ne dust brightness asymmetry observed in scattered light is also seen in the resonantly scattered gas emission lines ; however , the asymmetry in the emission is more pronounced . finally , the gas and dust populations share an inner disk that is tilted @xmath125@xmath13 with respect to the outer disk @xcite . if infalling refractory bodies produce all of the observed gas , why are the spatial distributions of the gas and dust so similar ? to explain the similarity in the spatial distributions of the dust and gas around @xmath1 pictoris , @xcite recently proposed that the atomic gas is produced via collisions between dust grains that vaporize at least some portion of the dust . vaporization of dust grains in energetic collisions is believed to produce atomic gas in supernovae shocks and may generate a non - trivial fraction of the gas observed in the exospheres of mercury and the moon @xcite . in the @xcite model , radiation pressure accelerates sub - blowout sized grains ( @xmath1 meteoroids ) to large radial velocities ; these small particles then collide with bound , orbiting grains with relative velocities as high as 90 km / sec ( for carbonaceous grains ) . the observed abundances of the @xmath1 pictoris dust and atomic gas may provide a clue to the origin of the stable atomic gas . if the gas is liberated in vaporizing collisions and no atomic species are selectively removed ( e.g. via outgassing or differentiation ) , then the gas composition should be the same as the grain composition . our models for the dust grains around @xmath1 pictoris reproduce the _ spitzer _ irs spectrum assuming a carbon relative to silicon abundance ratio of @xmath858.9 , significantly smaller than the measured ratio of the column densities of atomic carbon and silicon gas , @xmath85500 @xcite . photon - stimulated desorption ( psd ) may produce a substantial fraction of the sodium observed in the tenuous exospheres of mercury and the moon @xcite and may produce some of the spatially - extended sodium gas observed via resonant scattering toward @xmath1 pictoris . since psd produces gas directly from the dust , it might naturally explain the similarity in the gas and dust density distributions . the cross - section for psd jumps dramatically at uv wavelengths , becoming most efficient at @xmath109 @xcite . @xmath1 pictoris has been observed extensively in the far and near ultra violet using _ fuse _ @xcite , _ hst _ ghrs and stis @xcite , with a continuum flux at @xmath110 measured by _ iue _ , @xmath111 @xcite . @xcite model the observed stellar ultra violet spectrum of @xmath1 pic using a rotationally broadened ( @xmath112 = 130@xmath1134kmsec@xmath114 ) phoenix model with effective temperature @xmath115 , and surface gravity , log@xmath33=4.2 , suggesting a stellar uv photon production rate , @xmath116 = 2.0@xmath2010@xmath117 s@xmath114 . the efficiency with which photons desorb sodium , averaged over the surface of a sphere , is @xmath118 @xcite where @xmath45 is the photon - stimulated desorption cross section and @xmath3 is the atomic surface coverage . laboratory experiments measure @xmath119 @xmath12 ( 3@xmath1131 ) @xmath20 10@xmath120 @xmath121 for lunar samples with temperatures of 250 k and using incident photons with @xmath22 @xmath4 2500 , suggesting an efficiency of @xmath122 = 2.3 @xmath20 10@xmath123 for materials with a lunar temperatures and compositions , @xmath124 = 3 @xmath20 10@xmath105 @xmath125 @xcite . in addition , the photon - stimulated desoprtion efficiency of sodium is also function of substrate temperature , @xmath126 . we estimate @xmath127 from measurements of sodium desorption yields in the laboratory @xcite , normalized to the sodium desorption efficiency at the moon as discussed above . the rate at which atomic sodium is produced in the @xmath1 pictoris disk is @xmath128\,\pi a^{2}\,n_{\rm dust}(r , z)\ ] ] where @xmath129 and @xmath130 are the minimum and maximum radii , and @xmath131 and @xmath132 are the minimum and maximum heights at which the atomic sodium gas is observed . since the gas in the @xmath1 pic disk is located at distances of up 350 au , well beyond our irs slit , we use a dust density distribution inferred based on observations of the disk on large scales ( rather than the dust density profile that we derive in section 3 ) . artymowicz ( private communication ) has inferred the quantity @xmath133 , summed over all grain sizes , from _ hst _ stis observations @xcite assuming a grain albedo 0.5 : @xmath134^{-1/2 } \exp\left[-\left(\frac{z}{w}\right)^{0.7}\right]\ ] ] where @xmath135=120au , @xmath136=6.6@xmath137 , and @xmath138=2@xmath139 . we assume that the bulk of the surface area is contained in 1 @xmath0 m amorphous olivine grains and estimate the grain temperature as a function of distance using mie theory ( the grains are spherical ) and using laboratory - measured indices of refraction @xcite . if @xmath140 and @xmath141 , we estimate a sodium production rate @xmath142 from the grains via photon - stimulated desorption . we do not use the same dust spatial distribution as adopted in 3 for modeling the ir emission because the desorption properties of porous aggregates invoked in 3 are poorly known , while the desorption properties of compact grains ( assumed in eq.[4])are better quantified ( e.g. experimental data are available ) . moreover , the dust spatial distribution adopted in 3 is already oversimplified . we estimate the number of sodium atoms in the @xmath1 pictoris disk from observations of resonantly scattered and infer the time required to produce the circumstellar sodium . since the ionization potential of sodium is 5.1 ev and @xmath1 pic possesses a high uv luminosity , the majority of the circumstellar sodium is expected to be ionized . @xcite have written a photoionization code that calculates the densities of neutral and ionized elements from h to ni , assuming that the @xmath1 pic disk is optically thin and has a solar composition ; they estimate a sodium neutral fraction of 3@xmath2010@xmath94 at the disk midplane at 100@xmath11 from the central star . since the sodium neutral fraction is a function of distance from the stellar ionizing source , we have written a simple photoionization code which calculates the densities of neutral and ionized carbon and sodium in the disk as a function of position assuming that ( 1 ) the gas is optically thin , ( 2 ) the disk gas contains only carbon and sodium , ( 3 ) ionization of carbon produces the electron density in the disk , ( 4 ) the gas temperature , @xmath7 = @xmath126 , and ( 5 ) the circumstellar possesses a density distribution @xmath143^{-1/2 } \exp\left[-\left(\frac{z}{\alpha r}\right)^{2}\right]\ ] ] @xcite where @xmath144 , @xmath145 , @xmath146 , @xmath147 , and @xmath148 . we use published carbon and sodium photoionization cross sections @xcite and radiative recombination coefficients @xcite to estimate the carbon and sodium densities assuming that the atoms are in ionization equilibrium at each point in the disk . if the gas has a solar carbon : sodium abundance ratio , then the @xmath1 pictoris disk possesses 4.6@xmath2010@xmath117 sodium atoms and a time of @xmath1213000 years is required to produce the sodium gas . models of the breaking of atomic gas in the @xmath1 pic disk suggest that the lifetime of an atom in the disk is @xmath1210@xmath149 to 10@xmath150 years @xcite , suggesting that the sodium gas is in a steady state if it is produced by psd and braked by coulomb interactions . we can estimate the radial dependence of the sodium number density and compare it to that inferred from resonantly scattered sodium observations @xcite . if atomic sodium produced in the disk does not migrate radially , then sodium gas generated by photon - stimulated desorption will have a number density distribution @xmath151\,\pi a^{2}\,n_{\rm dust}(r , z)\ ] ] where @xmath152 is the distance to the central star and @xmath153 is the surface area contained in the dust grains . the predicted psd - produced sodium density distribution at the disk midplane falls too quickly as a function of radius compared to that inferred from observations ( see figure 5 ) . the rapid fall off in predicted sodium density is the result of the 1/@xmath154 dilution of stellar photons . the sodium desorption efficiency is able to partially overcome this effect because @xmath155 $ ] @xmath156 @xmath152 if the grains are 1 @xmath0 m olivine spheres . however , even taking into account the temperature dependence of the desorption efficiency , our predicted sodium number density falls off too quickly with radius by a factor of @xmath152 . if psd produces the bulk of the observed sodium , sodium atoms must migrate to larger distances via radiation pressure . our model for the origin of circumstellar sodium could be extended to other atomic species in the @xmath1 pictoris disk . at the current time , few laboratory measurements have been made for the photon - stimulated desorption rates of other atoms . potassium is the only species , other than sodium , for which psd rates have been measured . @xcite measure cross sections for postassium desorption from cr@xmath6o@xmath157 surfaces , @xmath158 = 2 @xmath20 10@xmath120 @xmath121 at 2500 - 3500 and 2 @xmath20 10@xmath159 @xmath121 at 1900 . if photon - stimulated desorption of potassium has the same temperature dependence as observed for sodium and the @xmath1 pic circumstellar grains have a solar potassium to sodium abundance ratio , @xmath160 = 2.1 @xmath20 10@xmath161 @xmath125 , then we estimate a potassium production rate in the @xmath1 pic disk of 3.3 @xmath20 10@xmath162 k atoms s@xmath114 , @xmath123000 times less than the sodium production rate . atomic potassium has not yet been detected in the @xmath1 pictoris disk and 3@xmath3 upper limits on the circumstellar potassium mass have not yet been reported in the literature . the simple dust model presented in 3 , consisting of cometary grains with a power - law size distribution and a modified power - law spatial distribution , provides a reasonably good fit to the overall sed and the observed _ spitzer _ irs spectrum . at a first glance , the model has many free parameters : ( 1 ) @xmath163 , @xmath164 and @xmath29 for the dust size distribution @xmath165 ; and ( 2 ) @xmath67 , @xmath71 , @xmath74 , @xmath72 , @xmath73 for the dust spatial distribution @xmath166 . with general constraints from the disk structure and the dust absorption and emission properties , we actually have only 2 free parameters @xmath29 and @xmath74 left ( see 3 for details ) . in so doing , 2 assumptions have been made : ( 1 ) all grains have the same spatial distribution ; and ( 2 ) the grains producing the irs emission spectrum are also responsible for the optical scattered light or their spatial distribution follows that of the dust responsible for the optical scattered light . admittedly , the former assumption is oversimplified since the response of the grains to the stellar radiation pressure and poynting - robertson drag varies with grain size and therefore the simple radial power law is probably not all that representative of the actual grain distribution uniformly with size . to justify the latter assumption , a simultaneous modeling of both the scattered light and ir emission is required . nevertheless , the general conclusion for the dust ( the presence of crystalline silicates and an inner warm dust component in the disk ) remains valid . we have presented a model for the production of circumstellar sodium gas around @xmath1 pictoris from dust via photon - stimulated desorption . our model is able to generate the inferred circumstellar sodium within the residence time of the gas but is unable to reproduce the radial dependence of the number density distribution if the sodium atoms do not migrate to larger distances via radiation pressure . our estimate for the time required to desorb the observed sodium is a function of ( 1 ) the circumstellar grain temperature , ( 2 ) the circumstellar gas temperature , and ( 3 ) the gas carbon : sodium abundance ratio . the photon - stimulated desorption rate of sodium from orbiting dust grains is dependent on the grain temperature with sodium more efficiently desorbed from warmer surfaces . since the bulk of the surface area of the dust is contained in the smallest grains , we estimated the grain temperature assuming that the grains have radii of @xmath55 = 1 @xmath0 m , are composed of amorphous silicates , and are spherical . however , if the grains are very small ( 2@xmath167 ) then the grain temperature , @xmath126 = @xmath168^{0.2 } t_\star$ ] , and the sodium desorption rate , @xmath169 = @xmath170 , and 4000 years are required to produce the sodium gas . or , if the grains are very large ( black bodies ) , then the grain temperature , @xmath126 = @xmath171^{0.5 } t_\star$ ] , and the sodium desorption rate , @xmath169 = @xmath172 , and 58000 years are required to produce the sodium gas . in our simple model , we have assumed that the gas and dust have the same temperatures and we have explored the possible range in dust temperature profiles . we also consider how changes in the gas temperature profile affect our estimates for the sodium neutral fraction and therefore the total number of sodium atoms in the disk . the estimated neutral fraction of sodium is weakly dependent on the gas temperature and composition . the radiative recombination coefficients for carbon and sodium decrease by factors of @xmath125 if the gas temperature increases from 10 k to 100 k or 100 k to 1000 k. if the gas has a temperature profile similar to small grains or black bodies ( instead of 1 @xmath0 m olivine spheres ) , then we estimate that the @xmath1 pictoris disk possesses either 7.0@xmath2010@xmath117 or 5.4@xmath2010@xmath117 sodium atoms , respectively , similar to the 4.6@xmath2010@xmath117 sodium atoms expected if the gas has the temperature profile expected from 1 @xmath0 m amorphous silicate grains . the assumed abundance of sodium relative to carbon also affects the total number of sodium atoms inferred from the neutral sodium observations . in our model , we have assumed that the gas has a solar composition ; however , a recent inventory of the circumstellar atomic gas around @xmath1 pictoris suggests that carbon is enriched compared to iron and oxygen by factors of 16 and 18 , respectively @xcite . if the @xmath1 pictoris disk is enriched in carbon relative to sodium by a factor of 16 , then the disk possesses 1.7@xmath2010@xmath117 sodium atoms , somewhat less than the number inferred if the gas has a solar composition . we have obtained _ spitzer space telescope _ irs spectra of the 12myr old debris disk around @xmath1 pictoris . we find that : \1 . in addition to the 10@xmath2 silicate emission feature originally detected in ground - based observations , we , for the first time , also observe weak crystalline silicate emission features at 28@xmath2 and 33.5@xmath2 . \3 . no h@xmath6 , [ ] , , or emission is detected . our 3@xmath3 upper limits suggest that @xmath417@xmath5 h@xmath6 remains in the disk significantly less than the previously reported detections of h@xmath6 s(1 ) and s(0 ) emission . the circumstellar disk around @xmath1 pictoris has too little gas to support the formation of giant planets . some of the observed resonantly scattered , observed at visual wavelengths , may be produced via photon - stimulated desorption ; the timescale to generate all of the inferred circumstellar sodium is 13000 yr , approximately the residence time of the gas in the disk if the gas is braked by coulomb forces . we would like to thank a. brandeker for providing the rotationally broadened phoenix stellar atmosphere model for @xmath1 pictoris and i. mann for providing us a preprint of their manuscript on their model for a collisional origin for the @xmath1 pic atomic gas prior to publication . we would also like to thank our two anonymous referees and e. chiang , d. hollenbach , m. jura , j. najita , a. roberge , n. samarasinha , w. sherry , a. weinberger , k. willacy , and y. wu for their helpful comments and suggestions . support for this work at noao was provided by nasa through the spitzer space telescope fellowship program , through a contract issued by the jet propulsion laboratory , california institute of technology under a contract with nasa . support for this work at the university of missouri was provided in part by the university of missouri summer research fellowship , the university of missouri research board , a nasa / hst theory program grant , and a nasa / spitzer theory program grant . support for this work at the university of arizona was provided by nasa through contract number 1255094 issued by jpl / caltech . , d. k. , moore , t. j. t , roche , p. f. , smith , c. h. , & wright , c. m. 1993 , mnras 265 , l41 , p. , burrows , c. , & paresce , f. 1989 , apj , 337 , 494 , p. , & clampin , m. 1997 , apj , 490 , 863 , j .- c . , nelson , r. p. , lagrange , a. m. , papaloizou , j. c. b. , & mouillet , d. 2001 , a&a , 370 , 447 , d. e. , gillett , f. c. , & witteborn , f. c. 1992 , apj , 385 , 670 , d. e. , & paresce , f. 1993 , in protostars and planets iii , eds . e. levy and j. i. lunine ( tuscon : university of arizona press ) , 1253 , h. , & valiron , p. 2007 , a&a , 466 , 201 , a. , liseau , r. , olofsson , g. , & fridlund , m. 2004 , a&a , 413 , 681 , c. h. , sargent , b. , bohac , c. , kim , k. h. , leibensperger , e. , jura , m. , najita , j. , forrest , w. j. et al . 2006 , apjs , 166 , 351 , h. , koike , c. , & tsuchiyama , a. 2001 , , 53 , 243 , r. , krugel , e. , shustov , b. , tutukov , a. , & kreysa , e. 1991 a&a , 252 , 220 , a. , & mann , i. 2007 , apj , 660 , 1541 , w. r. f. , greaves , j. s. , & coulson , i. m. 2005 , mnras , 359 , 663 , j. , begemann , b. , henning , t. , jaeger , c. , & mutschke , h. 1995 , a&a , 300 , 500 , d. , henning , t. , jger , c. , mutschke , h. , dorschner , j. , & wehrhan , o. 2001 , , 378 , 228 , r. , brandeker , a. & wu , y. 2006 , apj , 643 , 509 , w. , lagrange , a .- m . , vidal - madjar , a. , ferlet , r. , & forveille , t. 1995 , a&a , 301 , 231 , f. c. 1986 , in light on dark matter , proc . first infra - red astronomical satellite conf . ( dordrecht : reidel ) , 61 , t. m. , scarrott , s. m. , & wolstencroft , r. d. 1991 , mnras , 252 , 50 , d. a. , durrance , s. t. , & clampin , m. 1993 , apj , 411 , l41 , d. a. , ardila , d. r. , krist , j. e. , clampin , m. , ford , h. c. , illingworth , g. d. , bartko , f. , bentez , n. et al . 2006 , aj , 131 , 3109 , u. , & hollenbach , d. 2004 , apj , 613 , 424 , m. r. , hollenbach , d. , & erickson , e. f. 1991 , apj , 374 , 555 , p. m. , smith , b. j. , difrancesco , j. , colome , c. , & low , f. j. 1996 , apj , 471 , 973 , s. r. , lindler , d. j. , lanz , t. m. , cornett , r. h. , hubeny , i. , maran , s. p. , & woodgate , b. 2000 , apj , 539 , 435 , i. , walker , h. j. , klauss , u. , sylvester , r. j. , & lemke , d. 1999 , mnras , 304 , 589 , s. j. u. , devost , d. , higdon , j. l. , brandl , b. r. , houck , j. r. , hall , p. , barry , d. , charmandaris , v. et al . 2004 , pasp , 116 975 , w. s. , greaves , j. s. , zuckerman , b. , webb , r. a. , mccarthy , c. , coulson , i. m. , walther , d. m. , dent , w. r. f. et al . 1998 , nature , 392 , 788 , d. , gorti , u. , meyer , m. , kim , j. s. , morris , p. , najita , j. , pascucci , i. , carpenter , j. et al . 2005 , apj , 631 , 1180 , j. r. , roellig , t. l. , van cleve , j. , forrest , w. j. , herter , t. , lawrence , c. r. , matthews , k. , reitsema , h. j. 2004 , apjs , 154 , 18 , d. e. , weber , a. , & brault , j. w. 1987 , j. mol . , 126 , 19 , m. , chen , c. h. , furlan , e. , green , j. , sargent , b. , forrest , w. j. , watson , d. m. et al . 2004 , apjs , 154 , 453 , p. , & jewitt , d. 1995 , aj , 110 , 794 , p. , larwood , j. , smith , b. a. , schultz , a. 2000 , apj , 530 , l133 , i. , & van zadelhoff , g .- j . 2001 , a&a , 373 , 641 , i. , freudling , w. , & chengalur , j. n. 2007 , apj , 660 , 469 , r. m. , potter , a. e. , reiff , p. h. , sarantos , m. , jackson , b. v. , hick , p. , & giles , b. 2001 , j. geophys . res . , 106 , 20509 , r. f. , fajardo - acosta , s. b. , telesco , c. m. , hackwell , j. a. , lynch , d. k. , russell , r. w. 1993 , apj , 418 , 440 , c. , shibai , h. , & tuchiyama , a.1993 , mnras , 264 , 654 , c. , chihara , h. , tsuchiyama , a. , suto , h. , sogawa , h. , & okuda , h. 2003 , a&a , 399 , 1101 , n. a. , krivova , a. v. , & mann , i. 2000 , apj , 539 , 424 , p. o. , & pantin , e. 1994 , nature , 369 , 628 , a .- m . , beust , h. , mouillet , d. , deleuil , m. , feldman , p. d. , ferlet , r. , hobbs , l. , lecavelier des etangs , a. et al . 1998 a&a , 330 , 1091 , h. , wurz , p. , patel , m. r. , killen , r. , kolb , c. , massetti , s. , orsini , s. et al . 2003 , icarus , 166 , 238 , t. , heap , s. r. , & hubeny , i. 1995 , apj , 447 , l41 , f. , barbieri , c. , cremonese , g. , verani , s. , cosentino , r. , mendillo , m. , sprague , a. , & hunten , d. 2006 , icarus , 185 , 395 , a. , perrin , g. , ferlet , r. , vidal - madjar , a. , colas , a. , colas , f. , buil , c. , sevre , f. 1993 , a&a , 274 , 877 , a. , vidal - madjar , a. , roberge , a. , feldman , p. d. , deleuil , m. , andr , m. , blair , w. p. , bouret , j. c. , et al . 2001 , nature , 412 , 706 , a. , & greenberg , j. m. 1998 , a&a , 331 , 291 , a. , & lunine , j. i. 2003 , apj , 590 , 368 , a. , lunine , j. i. , & bendo , g. j. 2003 , apj , 598 , l51 , r. , brandeker , a. , fridlund , m. , olofsson , g. , takeuchi , t. , & artymowicz , p. 2003 , a&a , 402 , 183 , t. e. , yakshinskiy , b. v. , ageev , v. n. , & johnson , r. e. 1998 j. geophys . res . , 103 , 5873 , i. , khler , m. , kimura , h. , cechowski , a. , & minato , t. 2006 , , 13 , 159 , d. , larwood , j. d. , papaloizou , j. c. b. , & lagrange , a .- m . 1997 , mnras , 292 , 896 , t. , & koike , c. 1990 , icarus , 87 , 180 , s. n. 1995 , apj , 101 , 423 , h. , & storey , p. j. 1980 , a&a , 308 , 1980 , y. k. , kataza , h. , honda , m. , yamashita , t. , onaka , t. , watanabe , j .- i . , miyata , t. , sako , s. , et al . 2004 , nature , 431 , 660 , e. , lagage , p. o. , & artymowicz , p. 1997 , a&a , 327 , 1123 , f. , & burrows , c. 1987 , apj , 319 , l23 , i. , gorti , u. , hollenbach , d. , najita , j. , meyer , m. r. , carpenter , j. m. , hillenbrand , l. a. , herczeg , g. j. et al . apj , 2006 , 651 , 1177 , a. , feldman , p. d. , lagrange , a .- m . , vidal - madjar , a. , ferlet , r. , jolly , a. , lemaire , j. l. , & rostas , f. 2000 , apj , 538 , 904 , a. , feldman , p. d. , weinberger , a. j. , deleuil , m. , & bouret , j .- c . 2006 , nature , 441 , 724 , i. , dent , w. r. f. , & wyatt , m. c. 2004 , mnras , 348 , 1282 , b. a. , & terrile , r. j. 1984 , science , 226 , 1421 , m. , fukagawa , m. , kimura , h. , yamamoto , t. , suto , h. , & abe , l. 2006 , apj , 641 , 1172 , c. m. , & knacke , r. f. 1991 , apj , 372 , l29 , c. m. , fisher , r. s. , wyatt , m. c. , dermott , s. f. , kehoe , t. j. j. , novotny , s. , marinas , n. , radomski , j. t. , packham , c. , de buizer , j. , & hayward t. l. 2005 , nature , 433 , 133 , p. , & augereau , j .- c . 2005 , a&a , 437 , 141 , w. f. , blake , g. a. , van dishoeck , e. f. , van zadelhoff , g. j. , horn , j. m. m. , becklin , e. e. , mannings , v. , sargent , a. i. , van den ancker , m. e. , & natta , a. 2001 , nature , 409 , 60 , d. a. , & ferland , g. j. 1996 , apjs , 103 , 467 , d. a. , ferland , g. j. , korista , k. t. , & yakovlev , d. g. 1996 , apj , 465 , 487 , a. , lecavelier des etangs , a. , & ferlet , r. 1998 , planet space sci . 46 , 629 , n. v. , & krgel , e. 1999 , a&a , 352 , 508 , z. , koerner , d. w. , ressler , m. e. , werner , m. w. , backman , d. e. , & sargent , a. i. 2003 , apj , 584 , l27 , a. j. , becklin , e. e. , & zuckerman , b. 2003 , apj , 584 , l33 , m. w. , roellig , t. , low , f. , rieke , g. , rieke , m. , hoffmann , w. , young , e. , houck , j. et al . 2004 , apjs , 154 , 1 , l. , simbotin , i. , & dalgarno , a. 1998 , apjs , 115 , 293 , r. d. , scarrott , s. m. , & gledhill , t. m. 1995 , ap&ss , 224 , 395 , b. v. , & madey , t. e. 1999 , nature , 400 , 642 , b. v. , & madey , t. e. 2004 , icarus , 168 , 53 , b. , forveille , t. , & kastner , j. 1995 , nature , 373 , 494 , b. , song , i. , bessell , m. s. , & webb , r. a. 2001 , apj , 562 , l87 lccccc h@xmath6 s(2 ) & 12.279 & @xmath45.0@xmath2010@xmath173 & @xmath41.6@xmath2010@xmath161 & @xmath41.7@xmath2010@xmath149 & 2,4 + h@xmath6 s(1 ) & 17.035 & @xmath41.2@xmath2010@xmath173 & @xmath42.1@xmath2010@xmath150 & @xmath417 & 2,4 + @xmath174@xmath175 $ ] & 25.249 & @xmath48.6@xmath2010@xmath173 & @xmath40.047 & @xmath40.00016 & 1 + & 25.988 & @xmath46.1@xmath2010@xmath173 & @xmath40.29 & @xmath40.0011 & 3 + h@xmath6 s(0 ) & 28.221 & @xmath44.1@xmath2010@xmath173 & @xmath43400 & @xmath443 & 2,4 + & 34.814 & @xmath44.9@xmath2010@xmath176 & @xmath40.028 & @xmath40.00046 & 1 +

using self - consistent three - dimensional ( 3d )-body simulations , we investigate the physical properties of non - axisymmetric features in a disk galaxy created by a tidal interaction with its companion . the primary galaxy consists of a stellar disk , a bugle , and a live halo , corresponding to milky - way type galaxies , while the companion is represented by a halo alone . we vary the companion mass and the pericenter distance to explore situations with differing tidal strength parameterized by either the relative tidal force or the relative imparted momentum . we find that the formation of a tidal tail in the outer parts requires or . a stronger interaction results in a stronger , less wound tail that forms earlier . similarly , a stronger tidal forcing produces stronger , more loosely wound spiral arms in the inner parts . the arms are approximately logarithmic in shape , with both amplitude and pitch angle decaying with time . the derived pattern speed decreases with radius and is close to the curve at late time , with and denoting the angular and epicycle frequencies , respectively . this suggests that the tidally - induced spiral arms are most likely kinematic density waves weakly modified by self - gravity . compared to the razor - thin counterparts , arms in the 3d models are weaker , have a smaller pitch angle , and wind and decay more rapidly . the 3d density structure of the arms is well described by the concentrated and sinusoidal models when the arms are in the nonlinear and linear regimes , respectively . we demonstrate that dynamical friction between interacting galaxies transfers the orbital angular momentum of one galaxy to the spin angular momentum of the companion halo .

1505.04857

spiral arms play an important role in galactic evolution in diverse ways ( e.g. , @xcite and references therein ) . they exert non - axisymmetric torque to stars and gas clouds in galaxy rotation and cause their radial migrations , leading to secular density changes of the disks ( e.g. , @xcite ) . they also provide sites for active star formation , which is either triggered or organized by the stellar spiral potentials ( e.g. , @xcite ; see also @xcite and @xcite for review ) . therefore , understanding the nature and properties of spiral arms is crucial to understand secular , chemical , and dynamical evolution of disk galaxies . in terms of lifetime of spiral arms , the theory of spiral structure has forked into two branches : long - lived quasi - stationary density waves @xcite and short - lived transient features @xcite . the first picture requires that self - gravity plays a key role in organizing density waves into a self - sustained global pattern that rotates almost rigidly about the galaxy center @xcite . it successfully predicts systematic offsets between narrow dust lanes and star - forming regions associated with the arms ( e.g. , @xcite ; see also section 6.4.3 of @xcite ) , although it does not address the origin of the density waves . in the second picture , on the other hand , spiral arms are transient waves that wind up over time as they propagate inward in the radial direction . in this case , spiral arms last only for a few rotation periods . numerical simulations of isolated disk galaxies show that noises inherent in a stellar disk ( e.g. , @xcite ) or perturbations provided by giant molecular clouds ( e.g. , @xcite ) are amplified as they swing from leading to trailing configurations ( e.g. , @xcite ) , and form transient but recurrent spiral arms . such swing - amplified arms in isolated galaxies are usually ragged with multiple arms rather than being grand - design spirals with prominent two arms . observations indicate that among samples that include both non - barred and barred spiral galaxies , the probability of having grand - design arms is higher for galaxies with companions ( e.g. , @xcite ) .. recently , @xcite analyzed spiral structures in a sample of galaxies from the _ spitzer _ infrared nearby galaxies survey and found that the strength of grand - design arms is rather tightly correlated with tidal forcing from nearby companion galaxies , while the arm morphologies depend very weakly on the galaxy parameters such as stellar mass , gas fraction , disk / bulge ratio , rotational velocity , etc . this indicates that a large fraction of grand - design arms , especially for very strong arms , are driven most likely by tidal interactions . @xcite pioneered a numerical study for tidal interaction of disk galaxies using noninteracting test particles , and found that extended structures such as tidal tails and bridges develop in the course of galaxy encounters . subsequent @xmath0-body simulations focused on very strong encounters that lead to galaxy mergers or significant transformation of disk morphologies ( e.g. , @xcite ) . some studies concentrated on the orbital parameters required to reproduce the observed morphologies and kinematic features of interacting galaxies ( e.g. , @xcite ) , while others investigated properties of tidally - induced spiral arms ( e.g. , @xcite ) . in particular , @xcite found that the spiral arms generated in a cold disk by weak tidal interaction wind up from sc to sa appearances . @xcite showed that inner spiral arms are created if the tidal strength parameter @xmath1 ( see below for definition ) is larger than @xmath8 , suggesting that even a low - mass companion can tidally excite grand - design arms if the pericenter distance is small enough . to characterize the tidal strength , @xcite instead used the @xmath2 parameter ( see also below for definition ) that takes into account the interaction duration , finding that a tidal encounter with @xmath9 deforms the outer disk into an `` ocular '' shape . numerical simulations often show that tidally - induced arms are transient and posses the characteristics of kinematic density waves . for example , @xcite found that spiral arms driven by tidal forces resemble kinematic density waves when self - gravity is weak . in modeling tidal interactions of the m51/ngc 5195 system , @xcite found that spiral arms in m51 can not be described by a single pattern speed , indicative of kinematic density waves . @xcite showed that spiral arms in an isolated galaxy generated by swing amplification of random perturbations have a pattern speed that decreases with the galactocentric radius @xmath10 ( see also @xcite ) . @xcite also found that the pattern speed of @xmath11 arms driven by a satellite on a bound eccentric orbit decreases with radius . to quantify how the physical properties of the tidally - driven arms depend on @xmath2 , ( * ? ? ? * hereafter paper i ) ran a series of @xmath0-body simulations by considering a stellar disk inside a fixed halo interacting with a point mass companion on a prescribed parabolic orbit . paper i found that a tidal bridge consists of the disk particles pulled out by the tidal perturbations , with their epicycle phases locked to the companion , while a tidal tail forms as strongly perturbed near - side particles overtake mildly perturbed far - side particles . paper i also found that a stronger encounter produces stronger , more open arms that start to develop earlier at smaller @xmath10 . the arm pattern speed turned out to be a decreasing function of @xmath10 even when the arms are strongest , and converges to the @xmath12 curve as they decay . this suggests that tidally - driven arms are unlikely to be quasi - stationary density waves . similar results were obtained by @xcite who included a gaseous disk as well in simulating the m51/ngc 5195 system , although gravity of the gaseous component tends to increase the arm pattern speed . @xcite suggested that these are caustic waves maintained by coherent epicycle oscillations triggered by tidal forcing . while the results of the paper i are informative in assessing the quantitative effects of tidal perturbations on the arm properties , they were based on highly idealized galaxy models . first of all , paper i considered an infinitesimally - thin , two - dimensional ( 2d ) stellar disk . this not only neglects nonplanar motions but also overestimates self - gravity of the disk , making the spiral arms stronger than in disks with finite thickness . in addition , paper i employed fixed gravitational potentials for the halo / bulge and the perturbing companion . this precludes the possibility of dynamical friction occurring due to their gravitational reactions to the tidal perturbations . moreover , the galaxy and the companion in paper i were set to follow the prescribed parabolic orbits , which ignores their orbital decay caused by angular momentum loss . by evolving the system in the frame in which the galaxy remains stationary , paper i also ignored indirect forces arising from the orbital motion of the galaxy relative to the center of mass of the system , potentially suppressing the growth of lopsided spiral modes in the stellar disk ( e.g. , @xcite ) . to overcome the caveats mentioned above , we in this paper extend paper i by considering a self - consistent three - dimensional ( 3d ) galaxy model in which the stellar disk has a finite thickness and a central bugle and a dark halo are represented by live particles rather than by fixed potentials . the perturbing companion is modeled by a live halo alone for simplicity . as in paper i , we control the strength of a tidal encounter by varying two parameters : the pericenter distance and the galaxy - to - companion mass ratio . our objective is four - fold . first , we wish to measure the physical properties of tidally - induced spiral arms in more realistic 3d models and compare them with those in the razor - thin counterpart . we will also find the quantitative dependence of the arm properties on both @xmath2 and @xmath1 . second , @xcite and @xcite showed that the formation of tidal tails depends on the shape of the galactic potential . we will show that the tail formation depend not only on the potential shape but also on the tidal stregnth . third , we want to explore 3d density structures inside spiral arms and compare them with the analytic formulae proposed by @xcite . finally , we will show that dynamical friction of the companion in the course of a tidal interaction can be a source of the spin angular momentum of the primary halo . this paper is organized as follows . in section [ sec : method_model ] , we describe our galaxy model and introduce the parameters for tidal strength . in section [ sec : properties ] , we present the properties of tidal tails formed in the outer parts and spiral arms in the inner parts . in section [ sec : armdensity ] , we analyze the 3d density structures of spiral arm in the radial and vertical directions and compare them with the prediction of @xcite . in section [ sec : orbit_decay ] , we explore the orbital decay of the galaxy caused by dynamical friction , which results in spin - up of the initially non - rotating halo . finally , we summarize our results and discuss their astronomical implications in section [ sec : summary ] . ( a ) radial profile of the circular velocity of the initial disk and the contribution of each component . ( b ) radial distributions of the characteristic angular frequencies @xmath6 ( solid ) , @xmath5 ( dotted ) , and @xmath13 ( dashed).,width=302 ] in this paper we use 3d @xmath0-body simulations to study the tidal interaction of a disk galaxy with its companion . the galaxy consists of a stellar disk , a spherical bulge , and a dark matter halo ; the effect of a gaseous component is not considered in the present study . for the halo , we adopt a truncated @xcite model @xmath14 with the mass @xmath15 , the scale radius @xmath16 , and the truncation radius @xmath17 , within which @xmath18 of @xmath19 is enclosed . for the bulge , we take a plummer sphere @xmath20 with @xmath21 and @xmath22 . the stellar disk is initially axisymmetric with density distribution @xmath23 where @xmath24 is the total disk mass , @xmath25 is the radial scale length , and @xmath26 is the vertical scale height . the corresponding stellar surface density is @xmath27 at @xmath28 . the mass ratio of each component of the galaxy is @xmath29 , with the total mass @xmath30 within the truncation radius . lcccccccc ta1 & 0.44 & 25 & 0.410 & 0.248 & 0.12 & 0.69 & 30.9 & 0.173 + ta2 & 0.44 & 35 & 0.180 & 0.166 & 0.16 & 0.61 & 23.1 & 0.108 + ta2h & 0.44 & 35 & 0.175 & 0.163 & 0.15 & 0.61 & 25.5 & 0.113 + ta3 & 0.44 & 45 & 0.096 & 0.121 & 0.21 & 0.55 & 13.6 & 0.066 + tb1 & 0.22 & 25 & 0.202 & 0.134 & 0.19 & 0.45 & 14.2 & 0.111 + tb2 & 0.22 & 35 & 0.089 & 0.091 & 0.25 & 0.45 & 10.2 & 0.061 + tb3 & 0.22 & 45 & 0.046 & 0.065 & 0.35 & 0.39 & 7.6 & 0.039 figure [ fig : rotprofile ] plots the radial distribution of the circular velocity and the contribution of each component as well as the characteristic angular frequencies @xmath6 and @xmath31 in the initial disk . the presence of a strong bulge makes the @xmath5 curve rise steeply toward the center , which suppresses the formation of a bar in our tidal encounter models ( e.g. , @xcite ) . to find the equilibrium velocity distribution of the disk particles under the total gravitational potential , we solve the axisymmetric jeans equations using the method described in @xcite and @xcite . we ensure that the dark matter halo and bulge do not rotate initially . for the disk , the radial profile of the radial velocity dispersion is chosen such that the toomre @xmath32 stability parameter of the 3d disk is similar to that of the 2d counterpart considered in paper i ; we set @xmath33 over @xmath34 , which is large enough to suppress a spontaneous growth of spiral structures when evolved in isolation . we model the perturbing companion using a dark matter halo alone , which follows the truncated hernquist profile @xmath35 with the scale radius of @xmath36 , and @xmath37 for @xmath38 . we vary the companion mass @xmath39 to study the dependence of tidal features on the strength of tidal force ( see section [ sec : param ] ) . compared with the point - mass model considered in paper i , the current extended - halo model realizes a more realistic , smooth variation of the perturbing gravitational potential especially when the companion is close to the pericenter . most of the simulations presented in this paper use the galaxy model constructed by distributing @xmath40 , @xmath41 , and @xmath42 particles for the halo , bulge , and disk of the primary galaxy , respectively , and @xmath43 particles for the companion . we evolve our 3d galaxy model in isolation for 2 gyr , finding that the disk remains almost axisymmetic , with very weak ( @xmath44 density variations ) nonaxisymmetric features forming at its outskirts . this suggests that our initial disk is globally stable , due to a strong bulge as well as quite large value @xmath32 @xmath45 , in the absence of tidal perturbations . to check the dependency of simulation outcomes on the particle numbers , we also run a high - resolution model using @xmath46 , @xmath47 , @xmath48 , and @xmath49 particles . we find that resolution does not make significant differences in the properties of tidal features : the spiral arms in the low - resolution model are @xmath50 weaker and have a @xmath50 smaller pitch angle than in the high - resolution counterpart . initially , we place the two galaxies at large separation , and make them move on mutual parabolic orbits with a pericenter distance @xmath51 that is calculated under the assumption that the total mass of each galaxy is concentrated at its own center of mass . the galaxy orbits are prograde and located at the same plane as the disk of the primary galaxy . to study tidal interactions with differing strength , we consider six models that differ in @xmath39 and @xmath51 . the responses of the stellar disk to the tidal field of the companion can be parameterized by either @xmath52 or @xmath53 where @xmath54 , taken equal to @xmath55 in this work , is the characteristic size of the primary galaxy ( e.g. , @xcite ) , @xmath56 is the galaxy mass within @xmath54 , @xmath57 is time elapsed for the companion to orbit one radian near the pericenter , and @xmath58 is time taken by the disk particles at @xmath59 to rotate one radian about the galaxy center . the @xmath1 parameter in equation directly measures the tidal force exerted by the companion relative to the gravitational force of the primary at @xmath54 ( e.g. , @xcite ) . on the other hand , the @xmath2 parameter in equation takes allowance for the duration of tidal interaction and thus corresponds to the ratio of the linear momentum imparted to a disk particle by the companion with respect to its original linear momentum of galaxy rotation ( e.g. , @xcite ) . note that the current definition of @xmath2 differs slightly from that given in paper i , since the companion now has an extended density distribution rather than being treated as a point mass . when the tidal perturbations are applied _ impulsively _ over the time scale of @xmath57 , @xmath2 is a good measure of tidal strength ( e.g. , section 8.2 of @xcite ) . one can thus expect that @xmath2 is a better measure of tidal forcing for tidal tails formed in the outer disk where the rotation period is longer than @xmath57 . table [ tab : model ] lists the parameters of each model and some simulation outcomes . column ( 1 ) labels each run . we choose model ta2 with @xmath60 and @xmath61 as our fiducial model , which is the 3d counterpart of model a2 of paper i. model ta2h is identical to model ta2 except that the former employs 20 times more particles than the latter . column ( 2 ) gives the companion mass relative to the primary galaxy . column ( 3 ) lists the pericenter distance of two galaxies under the keplerian orbits . columns ( 4 ) and ( 5 ) give the dimensionless interaction strengths @xmath1 and @xmath2 , respectively . columns ( 6 ) , ( 7 ) , and ( 8) list the time @xmath62 when the tidal tail develops strongest , and its pitch angle @xmath63 and surface density @xmath64 at @xmath65 , respectively . here , @xmath66 is the surface density of the initial disk at @xmath67 . finally , column ( 9 ) gives the peak strength of the spiral arms that are induced . we take @xmath68 , @xmath69 , and @xmath70 as the units of mass , length , and velocity , respectively . in these units , the gravitational constant is @xmath71 , and the corresponding unit of time is @xmath72 . the simulation time is set such that @xmath73 corresponds to the closest approach of the galaxies under the prescribed parabolic orbits , but the tidal deformation of the galaxies makes the actual time of the closest approach delayed to @xmath740.04 , with a smaller value corresponding to weaker tidal forcing . all the simulations start from @xmath75 , when the two galaxies are separated more than @xmath76 . for model ta2 , the initial positions and velocities of the galaxies are @xmath77 and @xmath78 for the primary , and @xmath79 and @xmath80 for the companion . the simulations are performed using the gadget code that employs the barnes - hut hierarchical tree algorithm @xcite to solve the poisson equation @xcite . we adopt the force error tolerances of @xmath81 and @xmath82 . the gravitational softening parameter in the low - resolution models is set to @xmath83 for the disk , the halo , and the companion , and @xmath84 for the bulge , corresponding to 2.8 times the mean particle distance in each system ( e.g. @xcite ) . tidal perturbations disturb not only the otherwise axisymmetric disk , producing a tidal tail and a bridge in the outer parts ( @xmath85 ) and spiral waves in the inner parts ( @xmath86 ) , but also the primary halo and companion galaxy . in this section , we explore the correlations between the physical properties of the tidal features and the interaction strength parameters @xmath1 and @xmath2 . the deformation of the halo and companion will be presented in section [ sec : orbit_decay ] . figure [ fig : snapshots ] shows morphological changes of the stellar disk of the primary galaxy in our fiducial model ta2 projected on to the orbital plane . the grey circle in each frame indicates the center of mass of the companion at a given epoch , with its size corresponding to the region encompassing a half of the companion mass . the solid curves represent the actual trajectories of the two galaxies . the dashed lines in the @xmath87 frame draw the parabolic orbits they would follow if their extended mass distributions were undisturbed over the course of the encounter . the difference between the solid and dashed orbits are due to the angular momentum loss caused by dynamical friction , as will be discussed in section [ sec : orbit_decay ] in detail . at early time when the galaxies are separated widely ( @xmath88 ) , tidal perturbations are so weak that the orbits are parabolic and the disk of the primary remains almost axisymmetric . as the perturbing galaxy approaches the pericenter , the tidal force begins to excite the epicycle motions of stars in the outer disk , first shaping them into a bridge at the near side to the companion ( @xmath89 ) and then a tail at the opposite side ( @xmath90 ) . in the bridge , the epicycle phases of the constituting particles are forced to be aligned to the companion in such a manner that the radial velocities and the gradient of the azimuthal velocities are always maximized along the line connecting the galaxies ( paper i ) . on the other hand , the tidal tail develops as strongly perturbed particles at the near side catch up and overlap with mildly perturbed particles at the far side ( paper i ; @xcite ) . the tail is approximately logarithmic in shape , indicating that its pitch angle is independent of @xmath10 . the tails and bridges produced in our models are transient and decay after @xmath91 . as the companion recedes , the tidal force becomes weaker and is thus unable to keep the coherency of the forced epicycle phases of the bridge particles . composed of particles gathered from a wide range of radii in the unperturbed disk , the tails have a large velocity dispersion , tending to disperse with time . they disappear almost completely at @xmath92 . a small fraction ( @xmath93 ) of the disk particles that achieve velocities larger than the escape velocity become unbound , either captured by the companion or escaping from the system . with no further perturbations from the companion at @xmath94 , on the other hand , the bound particles once in the bridge or tail spread out , following galaxy rotation with large eccentricities . the disk becomes increasingly featureless over time . paper i found that the formation time , shape , and strength of the tails in 2d models are well correlated with @xmath2 . to study how the correlations change in the current self - consistent 3d models , we define the tail formation epoch , @xmath62 , as the time when it becomes densest at @xmath67 , and measure its pitch angle @xmath95 and surface density @xmath96 at that time . figure [ fig : sp - tail ] plot these values as filled symbols against @xmath1 and @xmath2 ; these are also tabulated in columns ( 6)(8 ) of table [ tab : model ] . with weak tidal interaction , model tb3 does not produce a readily identifiable tail , suggesting that the formation of tidal tail requires @xmath3 or @xmath97 . this threshold is probably a lower limit since all of our models consider in - plane , prograde encounters , the most favorable condition for the tail development . a stronger tail tends to form earlier and has a larger pitch angle . the dashed lines are the best power - law fits to the 3d results : @xmath98 , @xmath99 , and @xmath100 in terms of @xmath1 ; @xmath101 , @xmath102 , and @xmath103 in terms of @xmath2 . as expected , the tail properties are better correlated with @xmath2 than @xmath1 : the linear fitting coefficients are @xmath104 and @xmath105 against @xmath1 and @xmath2 , respectively . for comparison , we also plot the 2d results taken from paper i as as open circles . overall , the 2d and 3d results agree very well , although the 3d tails are slightly weaker , which is probably caused by weaker disk gravity due to finite disk thickness . figure [ fig : sp - tail ] shows that the tail in model ta2h with high resolution has almost the same properties as in model ta2 , suggesting that our results do not depend sensitively on resolution . dependence on @xmath2 of ( a ) @xmath106 , the maximum fraction of the disk particles at @xmath107 , and ( b ) @xmath108 , the fraction of the disk particles in the tail at @xmath107 . filled circles are from our 3d simulations . the results of @xcite for models with @xmath109 ( regular triangles ) , @xmath110 ( inverted triangles ) , and @xmath111 ( squares ) are compared . the solid lines are the best power - law fits : @xmath112 and @xmath113.,width=302 ] another way to quantify the tail strength is to use the tidal response introduced by @xcite who ran a number of @xmath0-body simulations for galaxy mergers with differing halo mass and interaction strength . they defined @xmath106 as the maximum fraction of the disk particles in each model that reach distances beyond @xmath114 ( @xmath115 in our models ) from the disk center in the course of interaction . figure [ fig : spteff](a ) plots @xmath106 from our tail - forming models ( filled circles ) as well as the models ( open symbols ) from @xcite as a function of @xmath2 : regular triangles are for models with @xmath116 ; inverted triangles for @xmath110 ; squares for @xmath111 . all of our models have @xmath117 . despite differences in the galaxy models such as potential depth and rotation curve , etc . , our results are overall in good agreement with the extrapolation of @xcite for @xmath118 . note that in very strong encounter models with @xmath119 in @xcite represented by open squares , the companion directly passes through the inner parts of the primary disk , for which the @xmath2 parameter relying on a distant tide approximation is not useful in characterizing the interaction strength . except for these very strong - encounter models , there is a good correlation between @xmath106 and @xmath2 with some scatters . the best fit for models with @xmath120 is @xmath112 , which is shown as the solid line in figure [ fig : spteff](a ) . since @xmath106 includes the particles not only in a tail but also escaped or captured to the companion , to measure the tail strength we similarly define @xmath108 as the fraction of the disk particles consisting only of a tail at @xmath107 . figure [ fig : spteff](b ) plots @xmath108 for our models as a function of @xmath2 . the solid line draws the best fit @xmath113 , which has a tighter correlation than the @xmath106@xmath2 relation . the @xmath108@xmath2 relationship can be explained as follows . let us assume that the tidal perturbations are applied impulsively near the pericenter during the time interval @xmath57 . the radial velocity increment at @xmath54 is then given by @xmath121 . this enhances the epicycle amplitudes of the disk particles to @xmath122 . with the epicycle frequency @xmath123 at @xmath54 for flat rotation , it follows that @xmath124 . since the perturbed mass that goes into the tail is proportional to @xmath125 , one obtains @xmath126 , similar to our numerical results . we have seen in the previous section that the tidal force from the companion provides strong perturbations to epicycle orbits of particles in the outer disk , resulting in a tidal tail and a bridge . tidal disturbances in the inner part of the disk are not as strong as in the outer parts , but can nevertheless induce spiral arms there ( e.g. , @xcite ) . figure [ fig : surf_xy ] plots close - up snapshots of the stellar surface density in logarithmic scale in model ta2h projected onto the orbital plane , with the center of mass of the disk shifted to the origin ( @xmath127 ) . at @xmath128 , the outer edges of the arms are quite sharp , which is a common feature of tidally - induced arms ( e.g. , @xcite ) . to delineates the spiral structure , figure [ fig : surf_phir ] plots the perturbed surface density , @xmath129 , relative to the initial surface density @xmath130 in the @xmath131@xmath132 plane . before the companion passes by the pericenter ( @xmath133 ) , the disk appears almost axisymmetric , without forming a bar or spiral structures , indicating again that the disk is globally stable in isolation . tidal disturbances excite epicycle motions of the particles in the inner disk , which are coherently organized to build up a well - defined two - armed global spiral pattern . with @xmath134 , the effect of swing amplification on the growth of the spiral arms is not significant ( paper i ) . it is apparent that the arms are approximately logarithmic . in model ta2h , they become strongest at @xmath135 and decay as they keep winding out over time . in these single encounter experiments , no spiral arms are apparent after @xmath136 ( see also fig . [ fig : surf_xy ] ) . ( a ) radial dependence of the arm strength @xmath137 averaged over the time interval @xmath138 centered at the time of the peak strength . ( b ) temporal dependence of @xmath137 averaged over @xmath139 for models ta1 , ta2 , and tb1 , and over @xmath140 for models ta3 , tb2 , and tb3 . after the peak , @xmath137 decays as @xmath141 . in both panels , the results of the 2d model a2 from paper i are compared as dashed lines.,width=302 ] one of the most important parameters that control the responses of gas flows across a stellar spiral arm is the arm strength ( e.g. , @xcite ) . it has often been customary to quantify the arm strength using the dimensionless parameter @xmath142 where @xmath143 denotes the amplitude of the @xmath11 fourier mode of @xmath144 . note that @xmath137 effectively measures the maximum gravitational force due to the spiral arms as a fraction of the axisymmetric gravitational force in the unperturbed disk @xcite .- armed tightly - wound spirals with pitch angle @xmath145 and radial wavenumber @xmath146 , the corresponding gravitational potential at @xmath147 of a razor - thin disk is @xmath148 with @xmath149 ( e.g. , eq.(6.31 ) of @xcite ) . therefore , @xmath150 . ] figure [ fig : f - r](a ) plots the radial variation of @xmath137 averaged over the time interval @xmath151 centered at the epoch when the arms attain the maximum strength in each model , while figure [ fig : f - r](b ) gives the temporal change of @xmath137 averaged over a range of radii where the arms are strong . definitely , a weaker tidal interaction produces weaker arms that grow more slowly at larger galactocentric radii : they are maximized at @xmath152 for the intermediate encounter models ta1 , ta2 , and tb1 with @xmath153 ( or @xmath154 ) , and at @xmath155@xmath156 for the weak encounter models ta3 , tb2 , and tb3 with @xmath157 ( or @xmath158 ) . the difference between the results of models ta2 and ta2h is insignificant , indicating again that resolution does not affect the simulation outcomes much . for comparison , figure [ fig : f - r ] also plots the results of the 2d model a2 taken from paper i as dashed lines . note that the arms in model a2 have @xmath137 about 70% stronger than in the 3d counterpart , which is primarily due to the overestimated self - gravity in the 2d , razor - thin geometry . dependence of the relative amplitude @xmath159 of the arms on the @xmath1 parameter . the solid circles mark the averaged over @xmath160 and @xmath161 , with the errorbars indicating the standard deviations . the solid line is the best fit : @xmath162.,width=302 ] figure [ fig : sp - f ] plots the peak arm strength @xmath163 as functions of @xmath1 and @xmath2 . the solid lines represent the best power - law fits @xmath164 and @xmath165 to the 3d results , with the linear correlation coefficients of @xmath166 and @xmath167 relative to @xmath1 and @xmath2 , respectively . these can be compared to the 2d results of @xmath168 from paper i shown as dotted lines . observations usually measure the arm strength in terms of the relative fourier amplitude @xmath169 ( e.g. , @xcite ) . in order to facilitate comparisons with observations , figure [ fig : m2m0_p ] presents the dependence on @xmath1 of @xmath170 calculated from our 3d results . the filled circles plot the averages over @xmath171 and @xmath172 , with the errorbars corresponding to the standard deviations . the solid line is the best fit : @xmath162 . the spiral arms achieve the full strength at @xmath173@xmath174 with smaller values corresponding to stronger tidal interaction , after which they begin to decay . this is mainly caused by the increase in the velocity dispersions of the disk particles . first of all , the growth of spiral arms requires the particles at different radii to gather into the arms , which increases the velocity dispersions ( paper i ) . also , heating of the disk particles due to gravitational scattering off the arms becomes efficient once the arms are strong enough , inhibiting further growth of the arms ( e.g. , @xcite ) . the increased velocity dispersions make the epicycle orbits in the arms kinematically less coherent . figure [ fig : f - r](b ) shows that the amplitudes of the arms in the current 3d models decrease with time almost exponentially , with a characteristic timescale of @xmath175gyr , about twice faster than in the 2d models of paper i. the faster decay of the arms in the 3d models is presumably due to weaker disk gravity in a stratified disk , which makes the arms wind out at a faster rate as well . temporal variations of the pitch angle @xmath176 of the spiral arms located at @xmath177 for the intermediate forcing models ta1 , ta2 , and tb1 , and at @xmath178 for the weak forcing models ta3 , tb2 , and tb3 . the result of the 2d , razor - thin model a2 is compared as a thick grey line . the arm pitch angle in the 3d models decays as @xmath179 after the peak , which is steeper than @xmath180 in the 2d model.,width=321 ] figure [ fig : surf_phir ] shows that the tidally - induced spiral arms are approximately logarithmic in shape over a wide range of radii . this appears consistent with the observational results that grand design spirals are close to be logarithmic in optical and near - infrared images ( e.g. , @xcite ) . as in paper i , we calculate the arm pitch angle at a given time by first finding the fourier coefficients in @xmath131 and @xmath132 defined as @xmath181,\ ] ] where @xmath182 is the number of particles located at @xmath183@xmath184 for the intermediate tidal - forcing models with @xmath153 ( or @xmath185 ) , and at @xmath186@xmath156 for the weak forcing models with @xmath157 ( or @xmath158 ) , @xmath187 are the coordinates of the @xmath188-th particle , and @xmath189 is a real number corresponding to the slope of a two - armed spiral in the @xmath190 plane . we then calculate @xmath191 that maximizes @xmath192 and calculate the pitch angle of the arms through @xmath193 at each time ( e.g. , @xcite ) . figure [ fig : winding ] plots the temporal changes of the arm pitch angles for all 3d models . the 2d result of model a2 from paper i is also compared . when the arms stand out initially , they have moderate pitch angles amounting to @xmath194 . after attaining the peak amplitude , the arms in the 3d models wind out over time as @xmath195 , which is steeper than @xmath196 in the 2d models . the faster winding rates in 3d models are again because of reduced self - gravity in vertically - extended disks . note that the pitch angle of purely kinematic density waves with no self - gravity decays as @xmath197 ( e.g. , section 6.2 of @xcite ) . the pattern speed of spiral arms is also an important parameter , although it is not well constrained observationally . to calculate the arm pattern speed induced in our models , we use the normalized cross - correlation of the perturbed surface densities at two different epoches separated by @xmath198 as @xmath199 by taking a sufficiently small value of @xmath200 , we find @xmath201 that maximizes @xmath202 at given radius and time . the instantaneous pattern speed is then given by @xmath203 . figure [ fig : omegap ] plots as contours the amplitudes of the @xmath204 in the @xmath10@xmath205 plane for some selected epoches of models ta2 ( left ) and tb2 ( right ) . the angular frequencies @xmath6 and @xmath206 from the initial rotation curve are overplotted as a solid line and two dotted lines , respectively . the locus of @xmath201 in each panel draws the arm pattern speed as a function of @xmath10 . shortly after the pericenter passage of the companion ( @xmath207 ) , the perturbed surface density is dominated by the tidal bridge rather than the arms . the bridge has an almost fixed pattern speed at @xmath208 km s@xmath209 kpc@xmath209 , which corresponds to the angular speed of the companion near the pericenter . at this time , the particles inside the bridge are tidally locked to the companion . as the companion moves away from the primary , the bridge becomes weaker and @xmath204 becomes progressively dominated by the spiral arms . at @xmath210 when the arms have substantial amplitudes , @xmath211 is larger than @xmath212 , but only slightly . since @xmath211 is not constant over radius , the arms produced in our models are not exactly a pattern in a strict sense . they are rather similar to gravity - modified kinematic density waves . although self - gravity tends to increase @xmath211 , its effect is not significant in our 3d models . at late time ( @xmath213 ) , @xmath211 converges to the @xmath12 curve . to summarize this section , spiral arms induced by a tidal interaction in our models are not quasi - stationary density waves envisaged by @xcite . they are rather kinematic density waves slightly modified by gravity . due to reduced self - gravity , spiral arms in the 3d models are weaker and decay faster than those in the razor - thin counterparts . the arms in vertically - stratified disks have a smaller pitch angle and wind up more rapidly than those in the razor - thin , 2d disks . while there are numerous studies on the generation of stellar spiral arms , they concentrate mostly on arm morphologies and longevity without focusing on the arm density structures ( e.g. , @xcite ) . without much information on the vertical structure of stellar spiral arms , most previous works that studied galactic spiral shocks across the arms employed the arm potentials that are independent of , or varying weakly with , the vertical height ( e.g. , @xcite ) . @xcite suggested an analytic expression for 3d spiral density perturbations as described below . in this section , we analyze the density structures of spiral arms produced in our models and compare them with the analytic suggestion of @xcite . starting from a physically - motivated trial function for the spiral gravitational potential that drops off exponentially in the radial direction , @xcite obtained a simple formula for the @xmath11 spiral density perturbation @xmath214 with @xmath215,\ ] ] where @xmath216 is the normalization coefficient , @xmath217 is the fiducial radius , @xmath218 is the radial scale length of the perturbed density , @xmath219 is the vertical scale height , @xmath220 is the fourier amplitude of an azimuthal mode @xmath221 , and @xmath222 denotes the azimuthal phase of the arm at @xmath223 . the summation over @xmath221 is to allow for linear superpositions of various modes along the azimuthal direction for two - armed , logarithmic spirals . as examples , @xcite considered two different cases : _ sinusoidal _ arms having @xmath224 with @xmath225 ( e.g. @xcite ) and _ concentrated _ arms having @xmath226 with @xmath227 , @xmath228 , and @xmath229 . the concentrated arms have flatter interarm regions than the sinusoidal arms . the four - armed concentrated spirals were used by @xcite and @xcite to study gas dynamics driven by spiral arms . the thick spiral potential considered by @xcite corresponds to the sinusoidal arms . figure [ fig : rhoz_arm ] plots the vertical density profiles in the arm regions at @xmath230 and @xmath231 of model ta2h at @xmath232 when the arm strength is nearly peaked ( left ) , and at @xmath233 when the arms are in the decaying phase ( right ) . the black solid , dotted , and dot - dashed lines draw the total density @xmath234 , the azimuthally - averaged density @xmath235 , and the perturbed density @xmath236 , respectively . the three density distributions are well described by @xmath237 , shown as thin red lines , with @xmath238 , @xmath239 , @xmath240 at @xmath232 for the total , azimuthally - averaged , perturbed densities , respectively , and with @xmath241 , @xmath239 , @xmath242 at @xmath233 . the temporal increase of the scale heights between @xmath232 and @xmath243 is due to disk heating . but , these agree within 10% , validating the vertical dependence of @xmath244 in equation , with a scale height similar to that of the background averaged disk . torques on the disk / bulge exerted by its own halo ( dotted ) and the companion ( dashed ) in model ta2h as functions of time . the solid line gives the total torque . while the halo torque is always negative , the companion torque is positive over @xmath245 , caused by the displacement of the disk / bulge from the center of mass of the halo ( see figure [ fig : bulgedisp ] ) . with the positive net torque , the disk / bulge gains angular momentum during @xmath246 . , width=321 ] figure [ fig : rhor_arm ] plots as solid lines @xmath244 as a function of @xmath10 along the cuts with @xmath247 for model ta2h in the midplane . the top and second rows are at @xmath232 , while the third and bottom rows are at @xmath233 . the thin solid lines give the simulation results , while the thick grey lines draw the best fits using equation based on the sinusoidal arms ( left panels ) and the concentrated arms ( right panels ) . the fact that equation describes the arm positions represented by the over - densities fairly well implies that the spiral arms are closely logarithmic in shape . overall , the concentrated arms fit the radial arm structure better at @xmath232 when the arms are strong with @xmath248 , whereas the sinusoidal arms provide a better fit at @xmath233 when the arms are weak with @xmath249 . we note however that these fits are reasonably good only at @xmath250 and fail at smaller @xmath10 . this is expected from the fact that the tidally - driven arms do not decline monotonically with @xmath10 but are peaked at @xmath251 ( see fig . [ fig : f - r ] ) , while equation requires arms to decay exponentially in the radial direction over all radii . as figure [ fig : snapshots ] shows , the actual trajectories of the primary and companion galaxies during the tidal encounter deviate from those under the assumption of the rigid halos . the discrepancy is due to dynamical friction occurring when two galaxies overlap at least partly . as the companion enters the primary halo , the former creates a density wake in the latter that in turn exerts gravitational drag force on the former . the drag force removes the orbital angular momentum from the companion , decreasing the pericenter distance . the orbital angular momentum extracted from the companion is transferred to the spin angular momentum of the primary halo . a similar process also occurs when the primary enters the companion halo , which results in the angular momentum transfer from the orbit of the primary to the spin of the companion . to quantify the amount of the angular momentum transfer , figure [ fig : angmom ] plots the temporal variations of the @xmath252-component of the angular momenta of all the components as functions of time . in each panel , the solid line draws the net ( orbital plus spin ) angular momentum , while the dotted and dashed lines give the orbital and spin angular momenta defined , respectively , by @xmath253 and @xmath254 where @xmath255 is the total mass , @xmath256 is the position vector of the center of mass , and @xmath257 is the mean velocity of each component , measured in the center - of - mass frame of the whole system . the dot - dashed line in figure [ fig : angmom](a ) plots the total angular momentum of the whole system , showing that it is conserved within 3% throughout the entire evolution . initially , all the components except the disk do not spin , and their angular momenta are dominated by the orbital motions . as the primary and companion approach the pericenter , they start to experience dynamical friction , and the primary halo and the companion begin to continuously acquire the spin angular momentum at the expense of the orbital angular momentum . in order to check if the decrease in the orbital angular momentum of the companion is really caused by dynamical friction due to the primary halo , we use the @xcite formula for the drag force @xmath258 \mathrm{\mathbf{v}}_{\rm ptb},\end{aligned}\ ] ] where @xmath259 is the average density of the halo , @xmath260 is the relative speed of the companion , @xmath261 , with @xmath262 denoting the mean velocity dispersion of halo particles , and @xmath263 is the coulomb logarithm introduced to avoid a singularity in the force evaluation , with @xmath264 and @xmath265 representing the maximum and minimum impact parameters of the background particles , respectively . for orbital decay of a satellite galaxy in a dark matter halo , equation ( [ eq : df_chandra ] ) matches the numerical results provided @xmath264 is taken to be of order of the orbital radius of the halo , and @xmath266 with @xmath267 being the half - mass radius ( e.g. , @xcite ) . to apply equation ( [ eq : df_chandra ] ) to model ta2h at @xmath73 , we calculate the mean density of the halo as @xmath268 , where @xmath269 and @xmath270 denote the position vectors of the halo and the companion , respectively . the velocity of the primary relative to the halo is @xmath271 . with @xmath272 and @xmath273 , @xmath274 . the resulting torque on the companion is given by @xmath275 which is plotted as a thick solid line in figure [ fig : angmom](a ) , in fairly good agreement with the instantaneous decreasing rate of the orbital angular momentum of the companion . this validates that dynamical friction is indeed a primary cause of the angular momentum transfer in our numerical simulations . a close inspection of figure [ fig : angmom ] reveals that while the orbital angular momenta of the halo and companion decrease monotonically with time , those of the disk and bulge stay almost constant at @xmath276 , increase slightly for @xmath277 , and subsequently decrease slowly . this indicates that the disk and bulge experience positive torque temporarily near the pericenter , while the halo and companion always receive negative torque . to analyze what causes this unexpected behavior of the angular momenta of the disk and bulge , figure [ fig : torque ] plots as dashed and dotted lines the torques exerted on the disk / bulge due to particles in the companion and the halo , respectively it is apparent that the torque by the companion is overall positive , while the halo gives a negative torque to the disk / bulge . the resultant net torque , shown as the solid line , is positive for @xmath278 , which boosts the orbital motion of the disk / bulge system . to understand why the companion torque is positive , figure [ fig : bulgedisp ] plots the density distributions of the companion ( gray scale with black contours ) and the disk and bulge of the primary ( white contours ) at @xmath73 ( left ) and @xmath279 ( right ) of model ta2h in the orbital plane . the contour levels decrease by a factor of @xmath280 starting from the innermost level of @xmath281 for both companion and disk / bulge . note a clear density wake in the companion formed at @xmath73 near @xmath282 due to the primary . the filled and open star symbols in cyan mark the centers of mass of the halo and disk / bulge of the primary , respectively , while the arrows in green indicate the velocity vectors of the disk / bulge in the center of mass frame of the whole system . the dynamical friction of the primary due to the companion is stronger for the ( outer ) halo than the disk / bulge owing to proximity to the companion . this causes the halo to move slower than the rest of the primary galaxy , gradually making them displaced from each other . at @xmath73 , the center of mass of the disk / bulge is entering the third quadrant and moving in the negative-@xmath283 direction , while that of the halo is still in the second quadrant . at this time , the position vector of the bulge , indicated by the blue arrow , is inclined to the line connecting the centers of mass of the disk / bulge and the companion ( the red dotted arrow ) . in such a geometrical configuration , the companion pulls the disk / bulge forward by providing a positive torque , while the halo tends to slow them down . it turns out that the positive torque is stronger than the gravitation pull by the halo at this time , as figure [ fig : torque ] shows . the net effect is that the separation between the centers of mass of the disk / bulge and the halo increases at a faster rate . the positive torque becomes strongest @xmath284 when the companion is closest to the disk / bulge , and then decreases as the companion moves away . at @xmath279 , the positive torque by the companion is smaller than the negative torque by the halo . the disk / bulge is subsequently pulled back to , and moves in phase with , the halo after @xmath285 . does the dynamical friction between two galaxies really transfer the orbital angular momentum of one galaxy to the spin angular momentum of the other ? figure [ fig : vtiso_halo ] plots the temporal changes of the axially - averaged rotation velocities @xmath286 of the halo ( left ) and companion ( right ) . the vertical arrow in each panel marks the position of the companion and the primary in the left and right panels , respectively . initially ( @xmath75 ) , they are non - rotating , spherical , and about @xmath76 apart . as the center of the companion ( halo ) moves through the outer parts of the primary halo ( companion ) , it creates a density wake in the halo ( companion ) that pulls it backward . at the same time , the wake in the halo ( companion ) is pulled _ forward _ by the companion ( halo ) , acquiring positive rotational velocities as shown in figure [ fig : vtiso_halo ] . the mass - weighted mean rotational velocities of the companion and the halo at @xmath287 is @xmath288 and @xmath289 , respectively . this suggests that tidal interactions can be a source of the spin angular momentum of a dark matter halo that might be non - rotating when it first formed . using self - consistent , 3d @xmath0-body simulations , we have investigated the physical properties of tidal features induced in a disk galaxy interacting with its companion . the disk galaxy consists of a halo , a bulge , and a stellar disk , corresponding to milky - way type galaxies , all of which are represented by live particles . the perturbing companion is simplified by a live halo alone . this work is a straightforward extension of paper i that considered a 2d , razor - thin disk residing within a fixed halo . we do not consider the effect of the gaseous component in the present work . by varying the companion mass and the pericenter distance , we explore the intermediate and weak interaction regimes with the strength parameter of @xmath290 or @xmath291 ( see eqs.and [ [ eq : s ] ] for the definitions of @xmath1 and @xmath2 ) . we analyze the properties of tidal tails formed in outer regions , and spiral arms in the inner regions , and study their dependence on @xmath1 and @xmath2 . we also study the 3d density structures of spiral arms as well as the orbital decay of the galaxies caused by dynamical friction occurring during a tidal encounter . the main results of this work can be summarized as follows . 1 . the tidal force from the companion excites epicycle motions in the outer regions of the primary , forming a bridge at the near side and an extended tail at the far side for interactions with @xmath292 or @xmath293 : tidal forcing weaker than this is unable to produce a tidal tail . the tail formation time , pitch angle , and surface density measured at @xmath67 scale as @xmath294 , @xmath295 , and @xmath296 in terms of @xmath1 , and @xmath297 , @xmath298 , and @xmath299 in terms of @xmath2 . this is in good agreement with the 2d results of paper i , although the 2d tails are slightly stronger than the 3d counterparts , due to stronger gravity in the disk midplane . we also find that the tail response @xmath108 defined by the fraction of the disk particles consisting of a tail at @xmath300 has an excellent correlation with @xmath2 as @xmath113 . the tidal force also excites two - armed spiral density waves over a range of radii in the inner parts of the disk . we calculate the spatial and temporal dependence of the arm strength @xmath137 defined by the ratio of the maximum gravitational force due to the @xmath11 modes to the centrifugal force of disk rotation ( eq . [ [ eq : f ] ] ) . for intermediate tidal interactions with @xmath153 ( or @xmath185 ) , the arms are strongest at @xmath301@xmath184 , while weak interaction models with @xmath157 ( or @xmath158 ) have the arms induced at @xmath155@xmath156 . the spiral arms are stronger in models with larger @xmath1 or @xmath2 , with the peak arm strength given by @xmath164 or @xmath165 . the spiral arms are approximately logarithmic in shape , with a pitch angle @xmath194 when the arms are strongest . after the peak , the arms decay with a timescale of @xmath302 , and wind out as @xmath303 . the derived arm pattern speed @xmath211 is a decreasing function of @xmath10 and becomes close to the @xmath5 curve at late time . this suggests that the induced spiral arms are kinematic density waves modified weakly by self - gravity . compared to the 2d counterparts , arms in 3d models are weaker , have a smaller pitch angle , and wind and decay more rapidly with time , due to weaker gravity . when compared with the analytic expression of @xcite , the 3d density structure of tidally - induced , @xmath11 arms can be reasonably well described by the concentrated arms when the arms are in the nonlinear regime with @xmath248 . on the other hand , arms in the linear regime with @xmath249 are better fitted by the sinusoidal model . the perturbed density in the arm regions follows the characteristic profile @xmath304 along the vertical direction , with the scale height @xmath305 not much different from that of the initial disk . 4 . the halos of the primary and companion become partially overlap in the course of a tidal encounter . this creates gravitational density wakes in the halos , resulting in dynamical friction . the dynamical friction has a few notable effects . first , it transports the orbital angular momentum of one galaxy to the spin angular momentum of the halo of its companion , making the pericenter distance smaller than that under the rigid halos . second , the initially rotation - free halos achieve the mass - weighted rotational velocities amounting to @xmath306 and @xmath307 for the primary and companion , respectively . third , the dynamical friction is stronger for the halo , causing offsets between the centers of mass of the halo and disk / bulge of the primary . thus , the disk / bulge can temporarily gain angular momentum by a positive torque from the companion near the pericenter , although they are later pulled back to the center of mass of the halo as the companion moves away . tidal interaction of a galaxy with its companion in a prograde orbit is studied using fully consistent @xmath0-body models . we find that the physical properties of the tidally - induced nonaxisymmetric features are tightly correlated with the interaction strength parameters @xmath1 and @xmath2 . these correlations are in qualitative agreement with those from the 2d results presented in paper i , although the structures in the 3d models are generally weaker and decay more rapidly . these are also qualitatively consistent with the results of @xcite who described in terms of @xmath2 the formation criteria of the outer disk features such as double arms corresponding to spiral arms decoupling from the tidal tail . @xcite similarly found that the tidal strength @xmath1 determines the strength of grand - design spiral arms in interacting galaxies . overall , @xmath2 is a better parameter for characterizing tidal tails formed in the outer regions , while spiral arms induced in the inner regions are better correlated with @xmath1 . strength and shapes of tidal tails have been used to constrain the masses of dark halos ( e.g. , @xcite ) . using numerical simulations , @xcite found that a tidal tail becomes shorter and less massive as the mass and size of a dark halo increase due to a shorter interaction time as well as a deeper potential well ( e.g. , @xcite ) . more quantitatively , @xcite suggested @xmath308 as an indicator of the susceptibility of a disk galaxy to the tail formation , where @xmath309 and @xmath310 refer to the escape and circular velocities at @xmath311 , respectively . note that @xmath312 measures the depth of the halo potential relative to the specific kinetic energy of the disk . using galaxy models with @xmath313 , @xcite showed that the maximum tidal response @xmath106 described in section [ sec : tail ] is inversely proportional to @xmath312 . they argued that tidal forcing produces a tail with a substantial amplitude only when @xmath314 in their models . @xcite conducted an extensive survey of galaxy collisions , and confirmed that @xmath312 is really a governing parameter and must be less than @xmath315 for the tail formation . our primary galaxy has @xmath316 , corresponding to the lower end in the models of @xcite ( see their figure 10 ) . based on the results of @xcite and @xcite , therefore , one can naturally expect strong tidal tails produced in all of our simulations . however , it turns out that a tail is absent in model tb3 with @xmath317 , while all the other models with @xmath318 do produce a tail with strength depending on @xmath2 . model tb3 has a very weak tidal response at @xmath319 from particles captured by the companion , even if it is more prone , in terms of @xmath312 , to the tail formation than any model of @xcite . this suggests that the tail formation should depend not only on @xmath312 but also on the interaction strength . our numerical results imply that a criterion for producing a tidal tail is @xmath318 ( or @xmath320 ) for @xmath316 . the critical values of @xmath2 and @xmath1 would increase with increasing @xmath312 , since it would then become increasingly difficult to make disk particles climb out of the halo potential well . we found that spiral arms induced by a tidal encounter are approximately logarithmic , consistent with many of observed grand - design spirals including m51 ( e.g. , @xcite ) . in our models , a logarithmic shape results from the fact that the spiral arms in the present simulations are close to kinematic density waves , slightly modified by self - gravity , for which the pitch angle varies with time and space as @xmath321 ( e.g. , eq . ( 6.26 ) of @xcite ) . the rotation curve of our galaxy model shown in figure [ fig : rotprofile ] gives @xmath322 at @xmath323 where arms are induced , making @xmath324 roughly constant over @xmath10 . this predicts that kinematic density waves wind out with time as @xmath325 when self - gravity is neglected . since @xmath326 for galactic disks , this also predicts that @xmath324 is smaller for larger @xmath327 , consistent with the observational results that the arm pitch angle has a negative correlation with the shear rate ( e.g. , @xcite ) . our results show that self - gravity reduces the arm - winding rate to @xmath328 , although it is not strong enough to maintain the arms quasi - steady . this holds even in 2d models with overestimated self - gravity at the disk midplane ( paper i ) . another prediction of kinetic density waves is that the arm pattern speed is not strictly constant over the radius but a decreasing function of @xmath10 . indeed , @xcite found that the pattern speed of the arms in m81 interacting with ngc 3077 is best described by @xmath329@xmath330 based on the @xcite method , after taking allowance for the radial variation of @xmath211 . @xcite applied a similar method to m51 in obvious interaction with ngc 5195 , and found that the arms consist of multiple patterns with larger pattern speeds at smaller radii ( see also @xcite ) , although it is questionable whether they are really multiple patterns with different pattern speeds or a single pattern with radially - varying @xmath211 . realistic interaction models of @xcite for the m51/ngc 5195 pair also predict @xmath211 close to the @xmath5 curve over a range of radii where the arms are strong . it is interesting to note the spiral arms in the isolated galaxy ngc 1068 have a pattern speed that decreases steeply with @xmath10 , indicative of a short lifetime of order of @xmath331 ( @xcite ; see also @xcite ) . all of these suggest that arms in real spiral galaxies are more likely transient rather than possessing the characteristics of quasi - stationary density waves unless the disks are strongly self - gravitating . while we employ very simple galaxy models for tidal interactions , our models can well be applied to the m51/ngc 5195 system that have the mass ratio of @xmath332@xmath333 ( e.g. , @xcite ) and a pericenter distance of @xmath334@xmath335 ( e.g. , @xcite ) , corresponding to models ta1 and ta2 . as figure [ fig : f - r ] shows , the arm strength is maximized @xmath336 after the pericenter passage . @xcite found observationally that the age distribution of star clusters in the arms of m51 has a narrow peak at @xmath337@xmath338 and a broad peak at @xmath339@xmath340 . if the enhanced star formation is really triggered by tidal effects , the m51/ngc 5195 pair might have undergone double interactions , the first one @xmath341@xmath342 ago and the second one @xmath343@xmath344 ago @xcite , considering the time delay between the arm formation and the closest approach . the peak arm strength in models ta1 and ta2 is @xmath345 , consistent with the radially averaged value of @xmath346 from @xmath347-band observations ( e.g. , @xcite ) . the arm pitch angle at peak is @xmath348 in models ta1 and ta2 ( figure [ fig : winding ] ) , which is also consistent with the observed pitch angle of @xmath349 ( e.g. , @xcite ) . based on our numerical results , the stellar spiral arms in m51 are in the nonlinear regime and their 3d density structure may be described by the concentrated arm model of @xcite . by analyzing the properties of @xmath11 spiral arms in the _ spitzer _ infrared nearby galaxies survey , @xcite very recently found that spiral morphology depends only weakly on the galaxy properties , while the arm strength is tightly correlated with tidal forcing from nearby companions . their figure 17 shows that @xmath350 is an increasing function of @xmath1 , which agrees with our numerical results qualitatively , but seemingly not quantitatively . that is , @xmath170 of the arms in the _ spitzer _ samples varies in the range of @xmath351@xmath352 for @xmath353@xmath354 , while our results predict @xmath355@xmath356 only for @xmath357 and no arm induced for @xmath358 ( see also @xcite ) . as @xcite noted , this apparent quantitative discrepancy may arise because the observed @xmath1 values in the _ spitzer _ samples correspond to the projected distance of a nearest neighbor at the current epoch , while @xmath1 in our models measures the tidal force using the 3d distance at the epoch of the pericenter passage . the projection effect overestimates the tidal forcing , while the distance at the current epoch is likely to underestimate the true @xmath1 significantly . for example , a comparison between figure 17 of @xcite and our figure [ fig : m2m0_p ] suggests that ngc 1566 with @xmath355 and the current forcing estimate of @xmath359 might probably have undergone an interaction with its companion ngc 1581 with @xmath360 at the pericenter passage . finally , we remark on the absence of a bar in our numerical simulations . as figures [ fig : surf_xy ] and [ fig : surf_phir ] show , tidal forcing in our models induces spiral structures at @xmath361 , while the inner regions with @xmath362 remain almost unaffected . this is in contrast to the results of @xcite , @xcite , and @xcite who showed that disks subject to tidal interaction can be unstable to form a bar . the major difference between the galaxy models used by the present paper and their work is that our models possess a relatively strong bulge , while their models have no or very weak bulge . the strong bulge in our models puts an inner lindbland resonance which suppresses feedback to swing amplification that would otherwise produce a bar near the center @xcite . this suggests that a strong bulge may be responsible for the absence of a bar in non - barred grand design spiral galaxies . we are grateful to the referee for a thoughtful and constructive report . this work was supported by the national research foundation of korea ( nrf ) grant , no . 2008 - 0060544 , funded by the korea government ( msip ) . the computation of this work was supported by the supercomputing center / korea institute of science and technology information with supercomputing resources including technical support ( ksc-2014-c3 - 003 ) .

we evaluate differential distributions for exclusive scalar meson ( glueball candidate ) production for ( fair@gsi ) and ( j - parc@tokai ) . both qcd diffractive , pion - pion meson exchange current ( mec ) components as well as double - diffractive mechanism with intermediate pionic loop are calculated for the first time in the literature . the pion - pion component , which can be reliably calculated , dominates close to the threshold while the diffractive component may take over only for larger energies . at the moment only upper limit for the qcd - diffractive component can be obtained . the diffractive component is calculated based on two - gluon impact factors as well as in the framework of khoze - martin - ryskin approach proposed for diffractive higgs boson production . different unintegrated gluon distribution functions ( ugdfs ) from the literature are used . rather large cross sections due to pion - pion fusion are predicted for panda energies , where the gluonic mechanism is shown to be negligible . the production of close to threshold could limit the so - called form factor in the region of larger pion virtualities . we discuss in detail the two - pion background to the production of the meson .

0906.0286

many theoretical calculations , including lattice qcd , predicted existence of glueballs ( particles dominantly made of gluons ) with masses @xmath4 1.5 gev . no one of them was up to now unambiguously identified . the nature of scalar mesons below 2 gev is also not well understood . lattice qcd approach with quenched quarks find a scalar gluonium ( glueball ) at approximately 1.6 gev @xcite . states @xcite . ] also the analyses in the framework of chiral lagrangians @xcite indicate that @xmath0 is dominantly gluonium state . the qcd sum rules @xcite suggest that the states at approximately 1 gev and 1.5 - 1.6 gev are admixtures of gluonium and @xmath5 states . a recent analysis in the framework of gaussian qcd sum rules @xcite , which is well suited to @xmath5 - gluonium mixing , find that the states at about 1 gev ( @xmath6 ) and at about 1.4 gev are strongly mixed with the preference of the higher - mass state to have slightly larger gluonium admixture . summarizing this discussion , it may be very difficult to find a clear signal of gluonium . further studies of the scalar meson production in several processes may shed more light on the quite complicated problem . the lowest mass meson considered as a glueball candidate is a scalar @xmath0 @xcite discovered by the crystall barrel collaboration in proton - antiproton annihilation @xcite . the branching fractions are consistent with the dominant glueball component @xcite . it was next observed by the wa102 collaboration in central production in proton - proton collisions in two - pion @xcite and four - pion @xcite decay channels at @xmath7 30 gev . close and kirk @xcite proposed a phenomenological model of central exclusive @xmath0 production . in their language the pomerons ( transverse and longitudinal ) are the effective ( phenomenological ) degrees of freedom @xcite . the close - kirk amplitude was parameterized as dependence applies in the meson rest frame ( current - current c.m . ) ] @xmath8 in their approach there is no explicit @xmath0-rapidity dependence of the corresponding amplitude . since the parameters were rather fitted to the not - normalized wa102 data @xcite no absolute normalization can be obtained within this approach . furthermore the parameterization is not giving energy dependence of the cross section , so predictions for other ( not - measured ) energies are not possible . in the present paper we will investigate rather a qcd - inspired approach . it provides absolute normalization , energy dependence and dependence on meson rapidity ( or equivalently on @xmath9 of the meson ) . the nature of the @xmath0 meson still remains rather unclear . new large - scale devices being completed ( j - parc at tokai ) or planned in the future ( fair at gsi ) may open a new possibility to study the production of @xmath0 in more details . in the present analysis we shall concentrate on exclusive production of scalar @xmath0 in the following reactions : @xmath10 while the first process can be measured at j - parc , the latter two reactions could be measured by the panda collaboration at the new complex fair planned in gsi darmstadt . the combination of these processes could shed more light on the mechanism of @xmath0 production as well as on its nature . if @xmath0 is a glueball ( or has a strong glueball component @xcite ) then the mechanism shown in fig . [ fig : qcddiff ] may be important , at least in the high - energy regime . this mechanism is often considered as the dominant mechanism of exclusive higgs boson @xcite and @xmath11 meson @xcite production at high energies . there is a hope to measure these processes at lhc in some future when forward detectors will be completed . at intermediate energies the same mechanism is , however , not able to explain large cross section for exclusive @xmath12 production @xcite as measured by the wa102 collaboration . explanation of this fact is not clear to us in the moment . the sketch of the bare qcd mechanism for diffractive production of the glueball . the kinematical variables are shown in addition.,scaledwidth=40.0% ] at lower energies ( @xmath13 20 gev ) other processes may become important as well . since the two - pion channel is one of the dominant decay channels of @xmath0 ( 34.9 @xmath14 2.3 % ) @xcite one may expect the two - pion fusion ( see fig.[fig : pion_pion ] ) to be one of the dominant mechanisms of exclusive @xmath0 production at the fair energies . the two - pion fusion can be also relative reliably calculated in the framework of meson exchange theory . the pion coupling to the nucleon is well known @xcite . the @xmath3 form factor for larger pion virtualities is somewhat less known . this may limit our predictions close to the threshold , where rather large virtualities are involved due to specific kinematics . at largest hesr ( antiproton ring ) energy , as will be discussed in the present paper , this is no longer a limiting factor as average pion virtualities are rather small . the sketch of the pion - pion mec mechanism . form factors appearing in different vertices and kinematical variables are shown explicitly.,scaledwidth=30.0% ] the sketch of the double - diffractive mechanism with pionic loop for exclusive production of the glueball candidate @xmath0 . the stars attached to @xmath15 mesons denote the fact that they are off - mass - shell.,scaledwidth=40.0% ] in this paper we concentrate on the mechanism of the reaction . our aim here is to explore a possibility of studying exclusive @xmath0 meson production in the fair and j - parc energy range and explore the potential of these facilities . while their are some ideas about the reaction mechanism at higher energies , the mechanism at lower energies was never studied . we shall investigate new mechanisms of pion - pion fusion shown in fig.[fig : pion_pion ] , the qcd mechanism shown in fig.[fig : qcddiff ] and a mechanism with intermediate pionic loop shown in fig.[fig : diffractive_pion_triangle ] . the second ( qcd ) mechanism is typical for high energies but here we wish to investigate its role at intermediate energies and in particular its vanishing at low energies and the interplay with the pion - pion fusion mechanism . the cross section for a general 3-body reaction @xmath16 can be written as @xmath17 above @xmath18 is the mass of the nucleon . the three - body phase space volume element reads @xmath19 at high energies and small momentum transfers the phase space volume element can be written as @xcite @xmath20 where @xmath21 , @xmath22 are longitudinal momentum fractions carried by outgoing protons with respect to their parent protons and the relative angle between outgoing protons @xmath23 . changing variables @xmath24 one gets @xmath25 the high - energy formulas ( [ dps_element_he1 ] ) and ( [ dps_element_he2 ] ) break close to the meson production threshold . then exact phase space formula ( [ dps_element ] ) must be taken and another choice of variables is more appropriate . we choose transverse momenta of the outgoing nucleons ( @xmath26 ) , azimuthal angle between outgoing nucleons ( @xmath27 ) and rapidity of the meson ( @xmath28 ) as independent kinematically complete variables . then the cross section can be calculated as : @xmath29 where @xmath30 denotes symbolically discrete solutions of the set of equations for @xmath31 and @xmath32 : @xmath33 where @xmath34 and @xmath35 are transverse masses of outgoing nucleons . the solutions of eq.([energy_momentum_conservation ] ) depend on the values of integration variables : @xmath36 and @xmath37 . the extra jacobian reads : @xmath38 in the limit of high energies and central production , i.e. @xmath39 0 ( very forward nucleon1 ) , @xmath40 0 ( very backward nucleon2 ) the jacobian becomes a constant @xmath41 . the matrix element depends on the process and is a function of kinematical variables . the mechanism of the exclusive production of @xmath0 close to the threshold is not known . we shall address this issue here . therefore different mechanisms will be considered and the corresponding cross sections will be calculated . according to khoze - martin - ryskin approach ( kmr ) @xcite , the amplitude of exclusive double diffractive colour singlet production @xmath16 can be written as @xmath42 the normalization of this amplitude differs from the kmr one @xcite by the factor @xmath43 and coincides with the normalization in our previous work on exclusive @xmath12-production @xcite . the amplitude is averaged over the colour indices and over two transverse polarisations of the incoming gluons @xcite . the bare amplitude above is subjected to absorption corrections which depend on collision energy ( the bigger the energy , the bigger the absorption corrections ) . we shall discuss this issue shortly when presenting our results . the vertex factor @xmath44 in expression ( [ ampl ] ) describes the coupling of two virtual gluons to @xmath0 meson . recently the vertex was obtained for off - shell values of @xmath45 and @xmath46 in the case of @xmath47 exclusive production @xcite . an almost alternative way to describe the vertex is to express it via partial decay width @xmath48 .. this will give us an upper estimate . as a consequence this will allow us to show that the gluonic component is negligible for future experiments with the panda detector . ] the latter ( approximate ) method can be used also for the @xmath0 meson production . in the original khoze - martin - ryskin ( kmr ) approach @xcite the amplitude is written as @xmath49}{q_{0,t}^2q_{1,t}^2q_{2,t}^2 } f_g^{kmr}(x_1,x'_1,q_{1,t}^2,\mu^2;t_1)f_g^{kmr}(x_2,x'_2,q_{2,t}^2,\mu^2;t_2 ) \ ; , \label{kmr_amplitude}\ ] ] where only one transverse momentum is taken into account somewhat arbitrarily as @xmath50 and the normalization factor @xmath51 can be written in terms of the @xmath52 decay width ( see below ) . in the kmr approach the large meson mass approximation @xmath53 is adopted , so the gluon virtualities are neglected in the vertex factor @xmath54\simeq(q_{1,t}q_{2,t})=(q_{0,t}+p'_{1,t})(q_{0,t}-p'_{2,t } ) . \label{kmr_vert}\ ] ] the kmr ugdfs are written in the factorized form : @xmath55 with @xmath56 gev@xmath57 @xcite . in our approach we use somewhat different parameterization of the @xmath58-dependent isoscalar form factors . please note that the kmr and our ( general ) skewed ugdfs have different number of arguments . in the kmr approach there is only one effective gluon transverse momentum ( see eq.([glue_momenta ] ) ) compared to two independent transverse momenta in general case ( see eq.([skewed_ugdfs ] ) ) . the kmr skewed distributions are given in terms of conventional integrated densities @xmath59 and the so - called sudakov form factor @xmath60 as follows : @xmath61 \ ; . \label{kmr_ugdf}\ ] ] the square root here was taken using arguments that only survival probability for hard gluons is relevant . it is not so - obvious if this approximation is reliable for light meson production . the factor @xmath62 in the kmr approach approximately accounts for the single @xmath63 skewed effect @xcite . please note also that in contrast to our approach the skewed kmr ugdf does not explicitly depend on @xmath64 ( assuming @xmath65 ) . usually this factor is estimated to be 1.31.5 . in our evaluations here we take it to be equal 1 to avoid further uncertainties . following now the kmr notations we write the total amplitude ( [ ampl ] ) ( averaged over colour and polarisation states of incoming gluons ) in the limit @xmath66 as @xmath67\frac{f^{off}_{g,1}(x_1,x_1',q_{0,t}^2 , q_{1,t}^2,t_1)f^{off}_{g,2}(x_2,x_2',q_{0,t}^2,q_{2,t}^2,t_2 ) } { q_{0,t}^2\,q_{1,t}^2\ , q_{2,t}^2}\ , , \label{ampl - kmr}\end{aligned}\ ] ] where the normalization constant is @xmath68 in addition to the standard kmr approach we could use other off - diagonal distributions ( for details and a discussion see @xcite ) . in the present work we shall use a few sets of unintegrated gluon distributions which aim at the description of phenomena where small gluon transverse momenta are involved . some details concerning the distributions can be found in ref . we shall follow the notation there . in the general case we do not know off - diagonal ugdfs very well . in @xcite we have proposed a prescription how to calculate the off - diagonal ugdfs : @xmath69 where @xmath70 and @xmath71 are isoscalar nucleon form factors . they can be parameterized as ( @xcite ) @xmath72 in the following for brevity we shall use notation @xmath73 which means @xmath74 or @xmath75 . above @xmath74 and @xmath75 are total four - momentum transfers in the first and second proton line , respectively . while in the emission line the choice of the scale is rather natural , there is no so - clear situation for the second screening - gluon exchange @xcite . even at intermediate energies ( @xmath76 = 10 - 50 gev ) typical @xmath77 are relatively small ( @xmath78 0.01 ) however , characteristic @xmath79 are not too small ( typically @xmath80 10@xmath81 ) . therefore here we can not use the small-@xmath82 models of ugdfs . in the latter case a gaussian smearing of the collinear distribution seems a reasonable solution : @xmath83 where @xmath84 are standard collinear ( integrated ) gluon distribution and @xmath85 is a gaussian two - dimensional function @xmath86 above @xmath87 is a free parameter which one can expect to be of the order of 1 gev . based on our experience in @xcite we expect strong sensitivity to the actual value of the parameter @xmath87 . summarizing , a following prescription for the off - diagonal ugdf seems reasonable : @xmath88 where @xmath89 is one of the typical small-@xmath82 ugdfs ( see e.g.@xcite ) . so exemplary combinations are : kl @xmath90 gauss , bfkl @xmath90 gauss , gbw @xmath90 gauss ( for notation see @xcite ) . the natural choice of the scale is @xmath91 . this relatively low scale is possible with the grv - type of pdf parameterization @xcite . we shall call ( [ mixed_prescription ] ) a `` mixed prescription '' for brevity . the amplitude in the previous section , written in terms of off - diagonal ugdfs , was constructed for rather large energies . the smaller the energy the shorter the qcd ladder . it is not obvious how to extrapolate the diffractive amplitude down to lower ( close - to - threshold ) energies . here we present slightly different method which seems more adequate at lower energies . at not too large energies the amplitude of elastic scattering can be written as amplitude for two - gluon exchange @xcite @xmath92 in analogy to dipole - dipole or pion - pion scattering ( see e.g. @xcite ) the impact factor can be parameterized as : @xmath93 at high energy the net four - momentum transfer : @xmath94 . @xmath95 in eq.([impact_factor ] ) is a free parameter which can be adjusted to elastic scattering . for our rough estimate we take @xmath96 . the sketch of the two - gluon impact factor approach . some kinematical variables are shown explicitly.,title="fig:",scaledwidth=35.5% ] the sketch of the two - gluon impact factor approach . some kinematical variables are shown explicitly.,title="fig:",scaledwidth=34.0% ] generalizing , the amplitude for exclusive @xmath0 production can be written as the amplitude for three - gluon exchange shown in fig.[fig:2g_impact_factor ] : @xmath97 at high energy and @xmath98 0 the four - momentum transfers can be calculated as : + @xmath99 , @xmath100 . + at low energy and/or @xmath101 0 the kinematics is slightly more complicated . let us define effective four - vector transfers : @xmath102 then @xmath103 and @xmath104 . close to threshold the longitudinal components @xmath105 0 and @xmath106 0 . then the amplitude ( [ 3g_exclusive_amplitude ] ) must be corrected . then also four - vectors of exchanged gluons ( @xmath107 , @xmath108 and @xmath109 ) can not be purely transverse and longitudinal components must be included as well . to estimate the effect we use formula ( [ 3g_exclusive_amplitude ] ) but modify the transferred four momenta of gluons entering the @xmath110 production vertex : @xmath111 and leave @xmath107 purely transverse . this procedure is a bit arbitrary but comparing results obtained with formula ( [ 3g_exclusive_amplitude ] ) with that from the formula with modified four - momenta would allow to estimate related uncertainties . we write the vertex function @xmath112 in the following tensorial form :- factorization approach we work in the on - shell approximation . in the on - shell approximation ( or infinitely heavy meson approximation ) the vertex is expressed through decay width and all vertices should be equivalent . even if the off - shell effects are included we do not expect very different energy dependence of the cross section for different tensorial forms as due to kinematics only small virtualities of gluons enter into game . ] @xmath113 the normalization factor is obtained from the decay of @xmath0 into two soft gluons : @xmath114 of course the partial decay width is limited from above : @xmath115 the amplitudes discussed here involve transverse momenta in the infra - red region . then a prescription how to extend the perturbative @xmath116 dependence to a nonperturbative region of small gluon virtualities is unavoidable . in the following @xmath116 is obtained from an analytic freezing proposed by shirkov and solovtsev @xcite . the @xmath117 is the second most probable decay channel @xcite . as a consequence the mechanism shown in fig.[fig : pion_triangle_kinematics ] may play important role in the exclusive production of @xmath0 @xcite . it is relatively easy to estimate the contribution of this mechanism at high energies @xcite . in this paper we shall make an estimate of the corresponding cross section not far from the threshold , where the situation is slightly more complicated . a sketch of the double - diffractive mechanism with pionic loop for exclusive production of the glueball candidate @xmath0 . some kinematical variables are shown explicitly.,scaledwidth=40.0% ] the amplitude of the process @xmath118 sketched in fig.[fig : diffractive_pion_triangle ] can be written in a simplified form @xcite as : @xmath119 the delta functions are related to helicity conservation in hadronic processes . while the pomeron ( sub)amplitudes are dominantly imaginary , the reggeon ( sub)amplitudes have both real and imaginary parts . the factor @xmath120 . in the formula above @xmath121 and @xmath122 are a so - called reggeon trajectories . for brevity we use notation @xmath73 which means @xmath74 or @xmath75 . we take from the phenomenology : @xmath123 = 1.0808 , @xmath124 = 0.25 gev@xmath57 , @xmath125 = 0.5475 and @xmath126 = 0.93 gev@xmath57 @xcite . the strength parameters for the @xmath127 scattering fitted to the corresponding total cross sections @xcite : @xmath128 = 13.63 mb and @xmath129 mb . and @xmath130 scattering . ] at not too high energies the slope parameter @xmath131 6 gev@xmath57 . the subchannel mandelstam variable @xmath132 and @xmath133 are related to center - of - mass energies of relevant pion - nucleon subsystem . in principle , they are functions of pion - four momenta in the triangle : @xmath134 and @xmath135 and in general should be put inside of the triangle function @xmath136 which depend on the 3-body kinematics , i.e. on four - momenta of the two exchanged pomerons ( in general pomeron - reggeon , reggeon - pomeron or reggeon - reggeon ) . in order to simplify the calculation we take instead average values calculated as : @xmath137 the factors @xmath138 are added to cut off the region of small @xmath139 , where the naive regge parametrization does not apply . we parametrize them in terms of the smooth function : @xmath140 where @xmath141 . the parameter @xmath142 gives the position of the cut and parameter @xmath143 describes how sharp is the cut off . the latter parameter can have significant influence on the numerics . we shall take @xmath144 gev and @xmath145 gev . for large energies @xmath146 1 and close to kinematical threshold @xmath147 @xmath146 0 . this means that we limit to double - diffractive contribution only . the effective regge parametrizations of @xmath127 interactions @xcite are for both colliding particles being on - mass - shell . in our case the triangle pions are off - mass - shell . we correct the regge strength parameters by multiplying by two vertex form factors @xmath148 ( see fig.[fig : pion_triangle_kinematics ] ) . we take them in the following factorized form : @xmath149 @xmath150 is in principle a free parameter . in the calculation shown in the result section we shall take @xmath150 = 1 gev . the dependence on triangle four - momenta forces us to merge the form factors inside the triangle integration which leads to a modified pion - triangle function : @xmath151 where standard triangle integrand @xmath152 reads : @xmath153 in addition to three pion propagators we have written three vertex form factors which are functions of four momenta of corresponding legs . in principle , these functions are relatively well known for space - like pions . we parametrize the triangle - vertex form factors in the following factorized exponential form : @xmath154 in this factorized form each exponent is associated with individual leg in the vertex . such form factors ( exponents ) are normalized to unity when pions in the loop are on - mass shell . please note that we symmetrically ( modulus in ( [ triangle_formfactors ] ) ) damp configurations above and below pion - mass shell . @xmath155 is related to the size of the pions in the triangle . it is natural to expect : @xmath156 . in the calculation presented here we shall take @xmath157 = 0.5 gev . since the configurations close to the mass shells give the biggest contributions the sensitivity to the actual value of the form factor @xmath158 ( see eqs.([triangle_function_hat ] ) and ( [ triangle_formfactors ] ) ) is not substantial . the @xmath159 coupling constant can be calculated from the corresponding partial decay width @xcite . calculating the triangle function for running kinematics of the @xmath160 process ( each point of the phase space ) is in practice impossible . we calculate numerically the triangle function for : @xmath161 transverse components are on average small and are neglected in the present approximation . close to threshold @xmath162 and @xmath163 . it is straightforward to evaluate the pion - pion meson exchange current contribution shown in fig.[fig : pion_pion ] . if we assume the @xmath164 type coupling of the pion to the nucleon then the born amplitude squared and averaged over initial and summed over final spin polarizations reads : @xmath165 \times 2 \nonumber \\ \times & \dfrac{g^2_{{\pi nn } } t_k } { ( t_1 - m_{\pi}^2)^2 } f_{\pi nn}^2(t_1 ) \ ; \times \ ; |c_{f_0(1500 ) \to \pi\pi}|^2 v_{\pi \pi \to f_0(1500)}^2(t_1,t_2 ) \ ; \times \ ; \dfrac{g^2_{{\pi nn } } t_k } { ( t_2 - m_{\pi}^2)^2 } f_{\pi nn}^2(t_2 ) \nonumber \\ \times & \left [ \left ( e_2 + m \right)\left ( e_2 ' + m \right ) \left(\dfrac{{\bf p}_2 ^ 2}{(e_2 + m)^2 } + \dfrac{{\bf p}_2^{'2}}{(e_2^ { ' } + m)^2 } - \dfrac { 2 { \bf p}_2 \cdot { \bf p}^{'}_2}{(e_2 + m)(e_2^ { ' } + m)}\right)\right ] \times 2 \ ; . \nonumber \\ \label{pion_pion_amplitude}\end{aligned}\ ] ] in the formula above @xmath18 is the mass of the nucleon , @xmath166 and @xmath167 are energies of initial and outgoing nucleons , @xmath168 and @xmath169 are the corresponding three - momenta and @xmath170 is the pion mass . the factor @xmath171 is the familiar pion - nucleon coupling constant and is relatively well known @xcite ( @xmath172 = 13.5 14.6 ) . in our calculations we take @xmath172 = 13.5 . the isospin factor @xmath173 equals 1 for the @xmath174 fusion and equals 2 for the @xmath175 fusion . limiting to nucleons in the final state , in the case of proton - proton collisions only @xmath176 final state channel is possible and therefore the @xmath174 fusion is allowed while in the case of proton - antiproton collisions both @xmath177 and @xmath178 final state channels are possible , i.e. both @xmath174 and @xmath175 mec are allowed . in the case of central heavy meson production rather large transferred four - momenta squared @xmath74 and @xmath75 are involved and one has to include extended nature of the particles involved in corresponding vertices . this is incorporated via @xmath179 or @xmath180 vertex form factors . the influence of the t - dependence of the form factors will be discussed in the result section . in the meson exchange approach @xcite they are parameterized in the monopole form as @xmath181 typical values of the form factor parameter are @xmath182 = 1.21.4 gev @xcite , however the gottfried sum rule violation prefers smaller @xmath183 0.8 gev @xcite . the normalization constant @xmath184 in ( [ pion_pion_amplitude ] ) can be calculated from the partial decay width as @xmath185 where @xmath186 gev . the branching ratio is @xmath187 = 0.349 @xcite . the off - shellness of pions is also included for the @xmath188 transition through the extra @xmath189 form factor which we take in the factorized form : @xmath190 it is normalized to unity when both pions are on mass shell @xmath191 in the present calculation we shall take @xmath192 = 1.0 gev . let us start with the qcd mechanism relevant at higher energies . we wish to present differential distributions in @xmath9 , @xmath74 or @xmath75 and relative azimuthal angle @xmath27 . in the following we shall assume : @xmath193 . this assumption means that our differential distributions mean upper limit of the cross section . if the fractional branching ratio is known , our results should be multiplied by its value . there are almost no absolutely normalized experimental data on exclusive @xmath0 production in the literature , except of ref.@xcite . the absolutely normalized data of the abcdhw collaboration @xcite put emphasis rather on @xmath194 production . in principle , some ( model - dependent ) information on glueball wave function could be obtained from radiative decays @xmath195 and @xmath196 @xcite . the present data are not good enough to provide a detailed information on coupling of gluons to @xmath0 . in fig.[fig : dsig_dxf_kl ] we show as example distribution in feynman @xmath9 for kharzeev - levin ugdf ( solid ) and the mixed distribution kl @xmath90 gaussian ( dashed ) for several values of collision energy in the interval @xmath76 = 10 50 gev . in general , the higher collision energy the larger cross section . with the rise of the initial energy the cross section becomes peaked more and more at @xmath197 0 . the mixed ugdf produces slightly broader distribution in @xmath9 . the distribution of @xmath0 in feynman @xmath9 for w = 10 , 20 , 30 , 40 , 50 gev ( from bottom to top ) . in this calculation the kharzeev - levin ugdf ( solid line ) and the mixed distribution kl@xmath90 gauss ( dashed line ) were used . , scaledwidth=50.0% ] in fig.[fig : dsig_dt_kl ] we present corresponding distributions in @xmath198 . the slope depends on ugdf used , but for a given ugdf is almost energy independent . distribution in @xmath198 for kharzeev - levin ugdf for @xmath199 10 , 20 , 30 , 40 , 50 gev ( from bottom to top ) . the notation here is the same as in fig.[fig : dsig_dxf_kl ] . , scaledwidth=50.0% ] finally we present corresponding distributions in relative azimuthal angle between outgoing protons or proton and antiproton . these distributions have maximum when outgoing nucleons are back - to - back . again the shape seems to be only weekly energy dependent . distribution in relative azimuthal angle for kharzeev - levin ugdf for @xmath76 = 10 , 20 , 30 , 40 , 50 gev ( from bottom to top ) . the notation here is the same as in fig.[fig : dsig_dxf_kl ] . , scaledwidth=50.0% ] what about the pion - pion fusion mechanism ? can it dominate over the gluonic mechanism discussed in the previous subsection ? in fig.[fig : sigma_w ] we show the integrated cross section for the exclusive @xmath0 elastic production @xmath200 and for double charge exchange reaction @xmath201 the thick solid line represents the pion - pion component calculated with monopole vertex form factors ( [ f_pinn_formfactor ] ) with @xmath182 = 0.8 gev ( lower ) and @xmath182 = 1.2 gev ( upper ) . the difference between the lower and upper curves represents uncertainties on the pion - pion component . the pion - pion contribution grows quickly from the threshold , takes maximum at @xmath202 6 - 7 gev and then slowly drops with increasing energy . the gluonic contribution calculated with unintegrated gluon distributions drops with decreasing energy towards the kinematical threshold and seems to be about order of magnitude smaller than the pion - pion component at w = 10 gev . we show the result with kharzeev - levin ugdf ( dashed line ) which includes gluon saturation effects relevant for small - x , khoze - martin - ryskin ugdf ( dotted line ) used for the exclusive production of the higgs boson and the result with the `` mixed prescription '' ( kl @xmath90 gaussian ) for different values of the @xmath87 parameter : 0.5 gev ( upper thin solid line ) , 1.0 gev ( lower thin solid line ) . in the latter case results rather strongly depend on the value of the smearing parameter . the thick dash - dotted line corresponds to the second diffractive mechanism with pionic triangle . it is above the wa102 experimental data point . this is probably because of absorption effects not included in the present calculation . this contribution stays below the pion - pion fusion contribution at the gsi hesr energies . for comparison we show also experimental data point of the wa102 collaboration from ref.@xcite which lies between the results obtained with `` kl '' and `` mixed '' off - diagonal ugdfs . the thick long - dashed line corresponds to the second diffractive mechanism with pionic triangle . it is above the wa102 experimental data point . this is probably because of absorption effects not included in the present calculation . this contribution stays below the pion - pion fusion contribution at the gsi hesr energies . we calculate the gluonic contribution down to w = 10 gev . extrapolating the gluonic component to even lower energies in terms of ugdfs seems rather unsure . at lower energies the two - gluon impact factor approach seems more relevant the impact factor approach result is even order of magnitude smaller than that calculated in the kmr approach ( see lowest dash - dotted ( red ) line in fig . [ fig : sigma_w ] ) , so it seems that the diffractive contribution is completely negligible at the fair energies . ( color online . ) the integrated cross section as a function of the center of mass energy for @xmath203 ( left panel ) and @xmath204 ( right panel ) reactions . the thick solid lines are for pion - pion mec contribution ( @xmath182 = 0.8 , 1.2 gev ) , the dashed line is for qcd diffractive contribution obtained with the kharzeev - levin ugdf , the dotted line for the kmr approach and the thin solid lines ( blue ) are for `` mixed '' ugdf ( kl @xmath90 gaussian ) with @xmath87 = 0.5 , 1 gev . the dash - dotted line represents the two - gluon impact factor result . the thick long - dashed line corresponds to the second diffractive mechanism with intermediate pionic triangle . the experimental data point at w = 29.1 gev is from ref.@xcite . , title="fig:",scaledwidth=49.0% ] ( color online . ) the integrated cross section as a function of the center of mass energy for @xmath203 ( left panel ) and @xmath204 ( right panel ) reactions . the thick solid lines are for pion - pion mec contribution ( @xmath182 = 0.8 , 1.2 gev ) , the dashed line is for qcd diffractive contribution obtained with the kharzeev - levin ugdf , the dotted line for the kmr approach and the thin solid lines ( blue ) are for `` mixed '' ugdf ( kl @xmath90 gaussian ) with @xmath87 = 0.5 , 1 gev . the dash - dotted line represents the two - gluon impact factor result . the thick long - dashed line corresponds to the second diffractive mechanism with intermediate pionic triangle . the experimental data point at w = 29.1 gev is from ref.@xcite . , title="fig:",scaledwidth=49.0% ] our calculation suggests that quite different energy dependence of the cross section may be expected in elastic and charge - exchange channels . experimental studies at fair and j - parc could shed more light on the glueball production mechanism . the smaller energies the larger values of @xmath205 and @xmath206 are involved . many of unintegrated gluon distributions in the literature are formulated in the region of very small @xmath82 . extrapolation of the method down to small energies automatically means going to the region of large @xmath82 . below we wish to demonstrate this fact . in fig.[fig : large_x ] we show the ratio of the cross sections @xmath207 as a function of center - of - mass energy . above @xmath208 was introduced to define the region of small / large @xmath82 . the solid line corresponds to @xmath208 = 0.1 and the dashed line to @xmath208 = 0.2 . down to largest hesr energies one stays in the region of @xmath209 0.2 . the ratio of the cross sections ( see eq . ( [ ratio_large_x ] ) ) as a function of center - of - mass energy . the solid line corresponds to @xmath208 = 0.1 and the dashed line to @xmath208 = 0.2 . , scaledwidth=50.0% ] let us concentrate now on @xmath210 collisions at energies relevant for future experiments at hesr at the fair facility in gsi . here the pion - pion mec ( see fig.[fig : pion_pion ] ) seems to be the dominant mechanism , especially for the charge exchange reaction @xmath211 . as discussed in the previous section the gluonic component can be there safely neglected . in fig.[fig : average_t1andt2_w ] we show average values of @xmath74 ( or @xmath75 ) for the two - pion mec as a function of the center of mass energy . close to threshold @xmath212 the transferred four - momenta squared are the biggest , of the order of about 1.5 gev@xmath213 . the bigger energy the smaller the transferred four - momenta squared . therefore experiments close to threshold open a unique possibility to study physics of large transferred four - momenta squared at relatively small energies . this is a quite new region , which was not studied so far in the literature . below we shall present cross sections for the @xmath204 reaction . the cross section for the @xmath203 reaction can be obtained by rescaling by the factor of 1/4 . ( color online . ) average value of @xmath214 as a function of the center - of - mass collision energy for the two - pion exchange mechanism . in this calculation @xmath182 = 0.8 gev . , scaledwidth=50.0% ] the maximal energy planned for hesr is @xmath215 = 5.5 gev . at this energy the phase space is still very limited . in fig.[fig : dsig_dy_pipi_f0_1500 ] we show rapidity distribution of @xmath0 calculated including pion - pion fusion only . for comparison the rapidity of incoming antiproton and proton is 1.74 and -1.74 , respectively . this means that in the center - of - mass system the glueball is produced at midrapidities , on average between rapidities of outgoing nucleons . rapidity distribution of @xmath0 ( @xmath175 fusion only ) produced in the reaction @xmath211 for w = 3.5 , 4.0 , 4.5 , 5.0 , 5.5 gev ( from bottom to top ) . in this calculation @xmath182 = 0.8 gev . , scaledwidth=50.0% ] in fig.[fig : dsig_dpt_pipi_f0_1500 ] we show transverse momentum distribution of neutrons or antineutrons produced in the reaction @xmath211 . the distribution depends on the @xmath216 form factors @xmath179 and @xmath180 in formula ( [ pion_pion_amplitude ] ) . transverse momentum distribution of neutrons or antineutrons produced in the reaction @xmath211 ( @xmath175 fusion only ) for w = 3.5 , 4.0 , 4.5 , 5.0 , 5.5 gev ( from bottom to top ) . in this calculation @xmath182 = 0.8 gev . , scaledwidth=50.0% ] in fig.[fig : dsig_dphi_pipi_f0_1500 ] we show azimuthal angle correlation between outgoing hadrons ( in this case neutron and antineutron ) . the preference for back - to - back configurations is caused merely by the limitations of the phase space close to the threshold ( the matrix element for pion - pion fusion is @xmath27-independent ) . this correlation vanishes in the limit of infinite energy . at high energy , where the phase space limitations are small , the distributions are isotropic , there is no dependence on azimuthal angle . in practice far from the threshold the distribution becomes almost constant in azimuth . this has to be contrasted with similar distributions for pomeron - pomeron fusion shown in fig.[fig : dsig_dphi_kl ] which are clearly peaked for the back - to - back configurations . therefore a deviation from the constant distribution in relative azimuthal angle for the highest hesr energy of w = 5.5 gev for @xmath217 can be a signal of the gluon induced processes and/or the presence of subleading reggeon exchanges , e.g. @xmath218 . it is not well understood what happens with the gluon induced diffractive processes when going down to intermediate ( w = 5 - 10 gev ) energies . our calculations shows , however , that the diffractive component is negligible compared to the pion - pion fusion at @xmath219 10 gev . possible future experiments performed at j - parc could bring some new insights into this issue by studying distortions ( probably very small ) from the pion - pion fusion mechanism . azimuthal angle correlations between neutron and antineutron produced in the reaction @xmath211 ( @xmath175 fusion only ) for w = 3.5 , 4.0 , 4.5 , 5.0 , 5.5 gev ( from bottom to top ) . in this calculation @xmath182 = 0.8 gev . , scaledwidth=50.0% ] up to now we have neglected interference between pion - pion and pomeron - pomeron contributions ( for the same final channel ) . this effect may be potentially important when both components are of the same order of magnitude . at j - parc energies there could be , in principle , some small interference effect . while the pomeron - pomeron contribution is dominantly nucleon helicity preserving the situation for pion - pion fusion is more complicated . in the latter case we define 4 classes of contributions with respect to the nucleon helicities : @xmath220 ( both helicities conserved ) , @xmath221 ( first conserved , second flipped ) , @xmath222 ( first flipped , second conserved ) and @xmath223 ( both helicities flipped ) . the corresponding ratios of individual contributions to the sum of all contributions are shown in fig.[fig : map_pt1pt2_pipi_f0_1500_rat ] . in practice , only the @xmath220 @xmath224 contribution may potentially interfere with the gluonic one . from the figure one can conclude that this can happen only when both transverse momenta of the final nucleons are small . we shall leave numerical studies of the interference effect for future investigations , when experimental details of such measurements will be better known ; but already now one can expect them to be rather small . helicity decomposition of the cross section on the ( @xmath225 ) plane for w = 10 gev . @xmath226 ( upper left ) , @xmath227 ( upper right ) , @xmath228 ( lower left ) , @xmath229 ( lower right ) . the standard nucleon dipole form factor was used in this calculation . , title="fig:",scaledwidth=40.0% ] helicity decomposition of the cross section on the ( @xmath225 ) plane for w = 10 gev . @xmath226 ( upper left ) , @xmath227 ( upper right ) , @xmath228 ( lower left ) , @xmath229 ( lower right ) . the standard nucleon dipole form factor was used in this calculation . , title="fig:",scaledwidth=40.0% ] helicity decomposition of the cross section on the ( @xmath225 ) plane for w = 10 gev . @xmath226 ( upper left ) , @xmath227 ( upper right ) , @xmath228 ( lower left ) , @xmath229 ( lower right ) . the standard nucleon dipole form factor was used in this calculation . , title="fig:",scaledwidth=40.0% ] helicity decomposition of the cross section on the ( @xmath225 ) plane for w = 10 gev . @xmath226 ( upper left ) , @xmath227 ( upper right ) , @xmath228 ( lower left ) , @xmath229 ( lower right ) . the standard nucleon dipole form factor was used in this calculation . , title="fig:",scaledwidth=40.0% ] now we wish to show the size of the triangle - double - diffractive ( tdd ) component at the gsi hesr energy range . in fig.[fig : pipifusion_triangle ] we compare it with the pion - pion fusion component . the tdd component vanishes quickly with decreasing energy and stays below the pion - pion fusion component for the hesr energy range . the quick decrease of the cross section is caused mainly by the @xmath230 and @xmath231 factors in eq.([diffractive_pion_triangle_amplitude ] ) and ( [ wpin_correction_factor ] ) and reflects smallness of @xmath127 subchannel energies . comparison of the pion - pion fusion component ( solid lines ) and the double - diffractive one with pionic triangle ( dashed line ) . the details concerning the double - diffractive component are explained in section ii - d . the vertical gray band shows the range of the center - of - mass energy available by the panda experiment . , scaledwidth=50.0% ] in the previous section we have shown that in the panda energy range the pion - pion fusion is the dominant reaction mechanism for the production of the glueball candidate @xmath0 . up to now we have calculated the cross section for production of @xmath0 meson a process with three particles ( @xmath233 , @xmath234 and @xmath0 ) in the final state . the 2 @xmath235 4 amplitude for the reaction through the glueball candidate @xmath0 . the stars attached to @xmath236 mesons denote the fact that they are off - mass - shell.,scaledwidth=45.0% ] in practice one must select a given decay channel of @xmath0 . there are a few options : ( a ) a two - pion decay ( @xmath175 or @xmath174 ) , ( b ) a four - pion decay ( c ) a two - kaon decay . the first one is attractive due to its simplicity but may have a large background . the second requires more complicated analysis but may have smaller background . the branching fraction for the last option is smaller by a factor of about 5 than for the two - pion channel . the 2 @xmath235 4 amplitudes with the intermediate @xmath237-meson ( reggeon ) exchange as an example of the background.,title="fig:",scaledwidth=30.0% ] the 2 @xmath235 4 amplitudes with the intermediate @xmath237-meson ( reggeon ) exchange as an example of the background.,title="fig:",scaledwidth=30.0% ] let us consider now an estimate of the background to the @xmath238 reaction . in fig.[fig : diagram_2to4_f0_1500 ] we present our reaction of interest the reaction which proceeds through the scalar resonance @xmath0 . this reaction is viewed now as a process with four particles ( @xmath239 ) in the final state . unavoidably there exists a nonreduceable background to this process . in fig.[fig : diagram_2to4_background ] we show an example of the background . we shall call these two complex diagrams as @xmath237-meson(reggeon ) exchanges for brevity . the region of @xmath240 1.5 gev is slightly above the region of application of the standard meson - exchange formalism and slightly below the region of application of high - energy regge approach . in principle , one should consider both approaches . in the @xmath237 meson - exchange formalism the reduced amplitude for the @xmath241 process can be written as : @xmath242 above @xmath243 the quantities @xmath244 in ( [ amplitude_rho_exchange ] ) describe couplings of extended objects : pions and the exchanged @xmath237-meson . we parameterize them in the exponential form : @xmath245 consistent with the definition of the coupling constant the form factors are normalized to unity when @xmath237 meson is on - mass - shell . we take @xmath246 = 2.6 , which reproduces the @xmath237 meson decay width @xcite , and @xmath182 = 1 gev ( @xmath247 = 4 gev@xmath57 ) . in the case of @xmath237-reggeon exchange the amplitude can be written as @xmath248 we parameterize the vertex form factors in the standard exponential form used usually in the regge phenomenology @xmath249 in practical calculation we take @xmath247 = 6 gev@xmath57 . in the formula ( [ amplitude_reggeon_exchange ] ) @xmath250 is a signature factor which we take here @xmath251 and @xmath252 is a so - called reggeon trajectory . we take from the phenomenology : @xmath253 = 0.5475 @xcite and @xmath254 = 0.9 gev@xmath57 . the strength parameter @xmath255 can be obtained assuming regge factorization ( see e.g.@xcite ) and using the known strength parameters for the @xmath256 and @xmath127 scattering fitted to the corresponding total cross sections @xcite . the simple regge parameterizations apply for energies @xmath257 2 gev ( see e.g.@xcite ) . in our case of pion - pion scattering energies of @xmath258 1.5 gev are of interest . here a small modification of the regge formula ( [ amplitude_reggeon_exchange ] ) may be in order . consistent with meson - exchange formalism ( spin-1 exchange ) one may expect saturation of the @xmath241 cross section at lower energies . the following freezing of the energy factor in ( [ amplitude_reggeon_exchange ] ) seems a reasonable correction : @xmath259 where @xmath260 . one may expect @xmath261 = 1.5 2.0 gev . the compatibility of the regge formalism with low - energy approaches for pion - pion scattering was discussed in ref.@xcite . the 2 @xmath235 2 amplitudes ( [ amplitude_rho_exchange ] ) and ( [ amplitude_reggeon_exchange ] ) may be inserted into the 2 @xmath235 4 amplitude of fig.[fig : diagram_2to4_background ] . when doing so we include in addition the correction for off - shellness of incoming pions as was done for the @xmath0 meson using exponential form factors ( as in [ off - shell_form_factors ] ) . now we can perform a genuine 2 @xmath235 4 calculation including four - body phase space . pion - pion invariant mass distribution for @xmath0 ( solid line ) and the @xmath237-reggeon exchange background ( dashed lines ) for naive ( upper line ) and corrected ( two lower lines with @xmath261 = 1.5 , 2 gev ) extrapolations to low energies . here the full phase space has been included . the calculation was performed for the highest panda center - of - mass energy @xmath76 = 5.5 gev.,scaledwidth=50.0% ] here we discuss the results with @xmath237-reggeon exchange only . we have checked that the @xmath237-meson exchange formalism discussed in this section provides approximately the same results at @xmath262 1.2 gev as the modified reggeon exchange with @xmath261 = 1.5 2.0 gev ( see formula ( [ freezing_reggeon ] ) ) . therefore the modified reggeon - exchange calculation provides a realistic predictions in the broad range of pion - pion energies , both above and below the @xmath0 resonance . in fig.[fig : dsigma_dmpipi_full ] we show two - pion invariant mass distribution . the solid line corresponds to our resonance contribution . the dashed lines correspond to the @xmath237-reggeon exchange contribution . here the resonance contribution is much lower than the @xmath237-exchange background . in this calculation the integration over whole phase space was done . rapidity distribution of pions from the decay of the glueball candidate @xmath0 ( solid line ) and from the @xmath237-reggeon exchange background ( dashed lines ) for naive ( upper line ) and corrected ( lower lines ) extrapolations to low energies . for the background we impose in addition : 1.4 gev @xmath263 1.6 gev . the calculation was performed for the highest panda center - of - mass energy @xmath76 = 5.5 gev.,scaledwidth=50.0% ] in fig.[fig : dsigma_dypi ] we show corresponding distributions in rapidity of @xmath264 or @xmath265 ( identical ) . in order to better see the overlap of the signal and background for the @xmath237-reggeon exchange we impose in addition 1.4 gev @xmath263 1.6 gev ( the region of the @xmath0 resonance ) . limiting to very small center - of - mass rapidities one can further improve the signal - to - background ratio . in fig.[fig : dsigma_dmpipi_cuty3y4 ] we show two - pion invariant mass distribution with extra cuts : -0.5 @xmath266 0.5 . while the @xmath0 contribution is only slightly modified , the @xmath237-reggeon background contribution is reduced by more than order of magnitude . one can clearly see the signal over background in this case . especially the high - energy side of the @xmath0 meson is now free of the @xmath237-exchange background . a better separation can be done by using pion - pion partial wave analysis . pion - pion invariant mass distribution for @xmath0 ( solid line ) and the @xmath237-reggeon exchange background ( dashed lines ) for naive ( upper line ) and corrected ( lower lines with @xmath261 = 1.5 , 2.0 gev ) extrapolations to low energies . here an additional condition on center - of - mass rapidities -0.5 @xmath266 0.5 has been imposed . the calculation was performed for the highest panda center - of - mass energy @xmath76 = 5.5 gev.,scaledwidth=50.0% ] close to the two - pions production threshold the roper resonance excitation and its subsequent decay ( @xmath267 ) is known to give the dominant contribution to the @xmath268 reaction @xcite . the same may be expected also for the @xmath238 reaction . the roper resonance produces the two - pions in dominantly the @xmath269=0 and @xmath270=0 state is angular momentum between pions and @xmath270 is the total isospin of the pion pair . ] ( the tail of the @xmath271 meson ) , i.e. the strength is concentrated at @xmath272 much lower than @xmath0 . the kinematical constraint gives @xmath273 0.5 gev . in addition , this contribution could be eliminated by extra cuts on invariant masses @xmath274 and @xmath275 . the same method can , at least in principle , be used to eliminate the double @xmath276 , @xmath277 excitations followed by their decays @xmath278 and @xmath279 . = 5.5 gev reduces the signal by less than about 5% . ] in this sense the last two contributions ( roper and double isobar excitations ) are reduceable . to which extend precision of the real apparatus will allow such a reduction is a matter of further investigations . certainly complete analysis requires including more processes and an analysis of cuts allowing for improving the signal - to - background ratio . this certainly goes beyond the scope of the present paper where we only signal a huge increase of the @xmath0 meson production cross section in the panda energy range due to pion - pion fusion , the mechanism never discussed before in the literature . the @xmath280 reaction may be more favorable as far as the signal - to - background ratio is considered . unfortunately theoretical calculation of background are not feasible in this case . it is not clear to us at present if the 6-body channel can be measured by the panda detector at fair . in the case of the @xmath281 reaction the relevant branching fraction is smaller , but not negligible ( @xmath282 = 8.6 % @xcite ) . on the other hand the contribution from nucleon resonances is probably considerably smaller . there is , however , unreduceable contribution from @xmath283 exchange in the @xmath284 subprocess . the parameters for the latter reaction are much less known than those for the @xmath241 subprocess . at high energies another two - pion continuum may be of interest . pairs . [ fig:2pi_mechanisms ] , title="fig:",width=207 ] pairs . [ fig:2pi_mechanisms ] , title="fig:",width=207 ] the underlying mechanism was proposed long ago in ref.@xcite . the general situation is sketched in fig.[fig:2pi_mechanisms ] . the corresponding amplitude for the @xmath268 process ( with four - momenta @xmath285 ) can be written as @xmath286 where @xmath287 denotes `` interaction '' between nucleon @xmath288=1 ( forward nucleon ) or @xmath288=2 ( backward nucleon ) and one of the two pions @xmath30=3 @xmath289 , @xmath30=4 @xmath290 . in the regge phenomenology they can be written as : @xmath291 above @xmath292 , where @xmath293 is the center - of - mass energy in the @xmath294 subsystem . the first terms describe the subleading reggeon exchanges while the second terms describe exchange of the leading ( pomeron ) trajectory . we have neglected real parts of the reggeon exchanges amplitudes for simplicity . the strength parameters of the @xmath127 interaction are taken from ref.@xcite . the extra form factors @xmath295 and @xmath296 `` correct '' for off - shellness of the intermediate pion in the middle of the diagrams shown in fig.[fig:2pi_mechanisms ] . in the fallowing they are parametrized as @xmath297 i.e. normalized to unity on the pion - mass - shell . we take @xmath150 = 1 gev . more details of the calculation will be presented elsewhere @xcite . the @xmath298 amplitude ( [ regge_amplitude ] ) is used to calculate the corresponding cross section including limitations of the four - body phase - space . to excludes resonance regions we `` correct '' the regge parametrization ( [ regge_propagators ] ) by multiplying byfactors @xmath299 ( as in [ wpin_correction_factor ] ) . in fig.[fig : dsigma_dmpipi_full_space_b ] we show the two - pion invariant mass distribution of the double - diffractive background together with the @xmath0 signal ( @xmath224 fusion ) . we show three curves corresponding to different cuts on both @xmath300 : @xmath301 = 2.0 ( solid ) , 1.9 ( dashed ) , 1.8 ( dotted ) gev . the figure suggests that the double - diffractive background should not disturb observing the @xmath0 signal at the panda experiment energies . in this case , unlike for the two - pion involved @xmath237 exchange discussed in the previous subsection , imposing cuts on pion rapidities would not be helpful as the double - diffractive contribution is concentrated at midrapidities as shown in fig.[fig : dsigma_dypi_b ] . at lower hesr energies the situation is better , the double - diffractive two - pion background is relatively smaller . pion - pion invariant mass distribution for @xmath0 ( solid line ) and the double - diffractive background ( dashed lines ) for different ( sharp ) cut - off for @xmath300 . here the full phase space has been included . the calculation was performed for the highest panda center - of - mass energy @xmath76 = 5.5 gev.,scaledwidth=50.0% ] rapidity distribution of pions from the decay of the glueball candidate @xmath0 ( solid line ) and from the double - diffractive background ( dashed lines ) for different ( sharp ) cut - off for @xmath300 . for the background we impose in addition : 1.4 gev @xmath263 1.6 gev . the calculation was performed for the highest panda center - of - mass energy @xmath76 = 5.5 gev.,scaledwidth=50.0% ] for the first time in the literature we have estimated the cross section for exclusive @xmath0 meson ( glueball candidate ) production not far from the threshold . we have included both gluon induced diffractive and triangle - double - diffractive mechanisms as well as the pion - pion exchange contributions . the qcd diffractive component was obtained by extrapolating down the cross section in the khoze - martin - ryskin approach with unintegrated gluon distributions from the literature as well as using two - gluon impact factor approach . a rather large uncertainties are associated with the qcd diffractive component . at present only upper limit can be obtained for the diffractive component as the @xmath302 decay coupling constant remains unknown . the coupling constant could be extracted only in high - energy exclusive production of @xmath0 where other mechanisms are negligible . we have found rather large contribution of pionic - triangle - double - diffractive component at higher energies ( @xmath303 10 gev ) . however , at the gsi hesr energies this contribution is strongly damped because of the phase space limitations on the @xmath127 subchannel energies . the calculation of the mec contribution requires introducing extra vertex form factors . at largest panda energies they are relatively well known and the pion - pion fusion can be reliably calculated . the situation becomes more complicated very close to the threshold where rather large @xmath304 and @xmath305 are involved . the cross section for energies close to the threshold is very sensitive to the functional form and parameters of vertex form factor . therefore a measurement of @xmath0 close to its production threshold could limit the so - called @xmath216 form factors in the region of exchanged four - momenta never tested before . we predict the dominance of the pion - pion contribution close to the threshold . our calculation shows that the diffractive components ( in fact its upper limit for the qcd mechanism ) are by more than order of magnitude smaller than the pion - pion fusion component in the energy region of future panda experiment . the diffractive components may dominate over the pion - pion component only for center - of - mass energies @xmath257 15 gev . taking into account rather large uncertainties the predictions of this component should be taken with some grain of salt . clearly an experimental program is required to disentangle the reaction mechanism at energies @xmath257 15 gev . disentangling the mechanism of the exclusive @xmath0 production not far from the meson production threshold would require study of the @xmath203 , @xmath204 processes with the panda detector at fair and @xmath118 reaction at j - parc . in the case the pion exchange mechanism is a dominant process one expects : @xmath306 . at high energies , when the gluonic , or double - diffractive with intermediate triangle , components dominate over mec components @xmath307 . at intermediate energies one can not exclude a priori subleading reggeon exchanges like @xmath218 for instance . however , we do not know how to reliably calculate them from first principles . we believe that the distortions from the pion - pion at low energies and/or distortions from the qcd gluonic mechanism at high energy may tell us more and allow for a phenomenological analysis taking into account the @xmath218 component explicitly . we leave this problem for a future analysis when experimental data will be available . only a careful studies of different final channels in the broad range of energies could help to shed light on coupling of ( nonperturbative ) gluons to @xmath0 and therefore would give a new hint on its nature . the experimental studies of exclusive production of @xmath0 are not easy at all as in the @xmath224 decay channel one expects a large continuum . we have performed an involved calculation of the four - body @xmath308 background . our calculation shows that imposing extra cuts should allow to extract the signal of the glueball @xmath0 candidate at the highest panda energy . a partial wave @xmath224 analysis should be helpful in this context . the two - pion continuum will be studied in more detail in our future work . a smaller continuum may be expected in the @xmath309 or four - pion @xmath0 decay channel . this requires , however , a good geometrical ( full solid angle ) coverage and high registration efficiencies . panda detector seems to fulfill these requirements , but planning real experiment requires a dedicated monte carlo simulation of the apparatus . it is a central problem of our field if @xmath0 is a @xmath5 or glueball type . unfortunately , our analysis does not allow to give a definite answer to this important question . some information on baryon - baryon correlation may be helpful but certainly not decisive . if the cross section at high energies ( where the contribution of subleading reggeon exchanges may be neglected ) is much smaller than predicted based on the kmr method it means that gluons only weakly couple to @xmath0 . this could provide some indirect information on the @xmath0 structure . a direct comparison of the shape of differential distributions at high energies may provide a valuable test of the kmr method originally proposed for exclusive higgs production ( the latter experiment is very difficult as very small statistics is predicted ) . a possible disagreement with the prediction for exclusive @xmath0 production at high energies could put into question the kmr approach , at present state of art in the field . experiments at rhic could be useful in this context and could shed light on the nonperturbative coupling of gluons to @xmath0 . s. narison , g. veneziano , int . j. mod * a4 * ( 1989 ) 2751 ; + s. narison , nucl . b509 * ( 1998 ) 312 ; + s. narison , phys . * d73 * ( 2006 ) 114024 . tao huang , hong ying jin and ai - lin zhang , phys . * d59 * ( 1998 ) 034026 . kisslinger , j. gardner and c. vanderstraeten , phys . lett . * b410 * ( 1997 ) 1 . d. harnett , t.g . steele , nucl . * a695 * ( 2001 ) 205 . g. orlandini , t.g . steele , d. harnett , nucl.phys . * a686 * ( 2001 ) 261 . steele , d. harnett , g. orlandini , aip conf . * 688 * ( 2004 ) 128 , [ arxiv : hep - ph/0308074 ] . c. amsler et al . ( crystal barrel collaboration ) , phys . b327 * ( 1994 ) 425 ; + c. amsler et al . ( crystal barrel collaboration ) , phys . b333 * ( 1994 ) 277 ; + c. amsler et al . ( crystal barrel collaboration ) , phys . * b340 * ( 1994 ) 259 . khoze , a.d . martin and m.g . ryskin , phys . lett . * b401 * ( 1997 ) 330 ; + v.a . khoze , a.d . martin and m.g . ryskin , eur . j. * c23 * ( 2002 ) 311 ; + a.b . kaidalov , v.a . khoze , a.d . martin and m.g . ryskin , eur . j. * c31 * ( 2003 ) 387 , [ arxiv : hep - ph/0307064 ] ; + a.b . kaidalov , v.a . khoze , a.d . martin and m.g . ryskin , eur . j. * c33 * ( 2004 ) 261 ; + v.a . khoze , a.d . martin , m.g . ryskin and w.j . stirling , eur . j. * c35 * ( 2004 ) 211 .

ic 1848 is one of the young open clusters in the giant star forming cas ob6 association . several interesting aspects relating to star formation processes in giant star forming regions attracted us to study the initial mass function ( imf ) , star formation mode , and properties of pre - main sequence stars ( pms ) . a and h photometric study of the young open cluster ic 1848 was conducted as part of the sejong open cluster survey " ( sos ) . we have selected 105 early - type members from photometric diagrams . their mean reddening is mag . using the published photometric data with near- and mid - infrared archival data we confirmed the normal reddening law ( ) toward the cluster ( ic 1848 ) . a careful zero - age main sequence fitting gives a distance modulus of mag , equivalent to kpc . h photometry and the list of young stellar objects identified by koenig et al . permitted us to select a large number of pms stars comprising 196 h emission stars , 35 h emission candidates , 5 class i , 368 class ii , and 24 transition disk candidates . from the hertzsprung - russell diagram using stellar evolution models , we estimate an age of 5 myr from several evolved stars and 3 myr from the pms stars . the imf was derived from stars with mass larger than , and the slope is slightly steeper ( ) than the salpeter / kroupa imf . finally , we estimated the mass accretion rate of pms stars with a uv excess . the mean mass accretion rate is about in the mass range of to , whereas intermediate - mass stars ( ) exhibit a much higher accretion rate of . [ firstpage ] open clusters and associations : individual ( ic 1848 ) circumstellar matter stars : luminosity function , mass function accretion , accretion disks

1311.6553

there are three well - known large - scale star forming regions ( w3/w4/w5 @xcite ) in the cas ob6 association . w5 ( also called the ic 1848 h ii region ) is the eastern part of the molecular cloud complex and consists of two components w5 west and w5 east @xcite . there are four known o stars , hd 17505 , hd 17520 , hd 237019 and bd + 60 586 in w5 west , while one visible o star hd 18326 is at the center of w5 east . all but hd 237019 are surrounded by many low - mass stars in clusters . these o stars are thought to be the ionizing sources of the h ii region . the simple morphology of each h ii region helps us to study the influence of strong uv radiation from massive stars on the surrounding materials and the triggering mechanisms for star formation @xcite . hence , w5 is an ideal laboratory to study star formation processes with feed back from high - mass stars . the initial mass function ( imf ) of this star forming region is also an interesting scientific issue . in this context , our target of interest is the ic 1848 cluster ( hereafter ic 1848 ) , also known as ocl 364 , situated in the center of w5 west , containing the brightest star hd 17505 , which is a multiple system consisting of an o6.5iii , two o7.5v , and an o8.5v stars @xcite . many investigators ( @xcite and therein ) provided radio maps at various frequencies to investigate the large - scale structures and the physical properties of the w3/w4/w5 complex . since active star formation takes place in these regions , investigators have a continuing interest in the processes involved . @xcite and @xcite studied star formation in w5 a ( or ic 1848 a ) , which is a bright - rimmed molecular cloud to the eastern end of w5 . @xcite discussed many interesting observational aspects associated with the triggered star formation in w5 . a census of embedded stellar sources in the w3/w4/w5 region have been published by @xcite . the authors identified 19 clusters , with about half the stars in 5 rich clusters . @xcite identified several young stellar objects ( ysos ) using the iras point source catalog and found that the young stars are mainly distributed around the edge of the h ii region . the authors argued from a comparison of evolutionary time scales that there is a distinct difference in the star formation history of the ionizing stars and the young stars on the border of expanding h ii regions . because of the presence of clumps evaporating away from the ionising sources and the shorter time scale for star formation under the assumed expanding velocities , they also suggested that radiatively driven implosion - star formation triggered by the strong uv radiation from young ob stars @xcite - may be the dominant triggering mechanism in w5 . _ spitzer_/irac and mips imaging data , @xcite provided photometric data for more than 17,000 point sources and yso classifications . the identified ysos made it possible to study the clustering properties of these ysos and to discuss plausible triggering mechanisms required to explain the distinct generations within w5 . the high resolution irac and mips images revealed several isolated cometary globules and elephant trunk structures with polycyclic aromatic hydrocarbon ( pah ) emission . later @xcite studied the disk evolution of intermediate - mass stars . cometary globules , bright - rimmed clouds , and h ii region outflows are suggested as evidence of triggered star formation , and many studies have investigated the physical properties of these objects @xcite . several photometric studies @xcite involving ic 1848 have been made in the optical bands . these studies provided useful photometric data and fundamental parameters , such as reddening , distance , and age for ic 1848 . however , most analyses assumed a normal reddening law ( @xmath13 ) , and the multiplicity of early - type stars was not fully taken into account . furthermore , the limiting magnitudes were not deep enough to study the pre - main sequence ( pms ) stars fainter than @xmath14 mag in the young clusters or associations in the perseus arm . recently , a deep @xmath15 photometric study for ic 1848 w5 east has been made by @xcite with archival data . the authors presented the imf for stars in the mass range from @xmath16 to @xmath17 , as well as their fundamental parameters . however , no deep optical photometric study has been made for ic 1848-w5 west . since the results of @xcite provided very good membership criteria for the pms stars with accretion disks , the age distribution and mass accretion rates of pms stars , the star formation history within the region , and the imf down to the @xmath18 regime can be studied . in addition , the two micron all sky survey ( 2mass ) @xcite and _ spitzer_/irac data @xcite make it possible to test carefully the reddening law in a wide wavelength range . the previously published distance to ic 1848 , derived using various methods , is in the range 1.7 to 2.4 kpc . the photometric surveys mentioned above provided distances of 1.7 kpc @xcite , @xmath19 kpc @xcite , @xmath5 kpc @xcite , and 2.3 kpc @xcite . some studies have implicitly assumed that the distance to ic 1848 is the same as that to w3 or w4 ( or ic 1805 ) ( 1.9 2.4 kpc ) . @xcite independently derived a distance of 2.3 kpc using their galactic rotation model . early studies have relied on this distance . recently , @xcite derived a distance of @xmath20 kpc for a w3oh maser using its vlbi parallax . @xcite preferred to use a distance of 1.89 kpc to match the observed and theoretical stellar radii . since distance is a critical parameter in converting observational quantities to reliable physical parameters , we need to revisit its determination . the sejong open cluster survey ( sos ) project is dedicated to provide homogeneous photometric data down to @xmath21 mag for many open clusters . the overview of the project can be found in @xcite ( paper 0 ) . the young open cluster ngc 2353 was studied as part of this project @xcite . this paper on ic 1848 is the third in the series . the observations and comparisons with previous photometry are described in section 2 . in section 3 , we discuss the reddening law in the direction of ic 1848 and present fundamental parameters estimated from the photometric diagrams . the imf of ic 1848 and the mass accretion rates of pms stars with uv excesses are presented in section 4 and 5 , respectively . finally , we summarize the results from this study in section 6 . @xmath22 , j2000 ) . squares outline the observed regions.,scaledwidth=45.0% ] the observations of two regions associated with ic 1848 were made on 2009 january 19 using the azt-22 1.5 m telescope ( f/7.74 ) at maidanak astronomical observatory in uzbekistan . all imaging data were acquired using the fairchild @xmath23 ccd ( snucam @xcite ) with standard bessell @xmath0 @xcite and h@xmath1 filters . @xcite have described the characteristics of the ccd in detail . the mean seeing was better than 1@xmath24 , and sky conditions were good . the observations comprised a total of 20 frames that were taken in two sets of exposure times for each band 5s and 60s in @xmath25 , 5s and 180s in @xmath26 , 7s and 300s in @xmath27 , 15s and 600s in @xmath28 , and 30s and 600s in h@xmath1 . we present the finder chart for the stars brighter than @xmath29 mag in figure [ fig1 ] using the guide star catalogue version 2.3 @xcite . the photometry for 12 stars that were saturated in our data were taken from previous studies . the photometric data and the adopted spectral types of the stars are presented in table [ tab1 ] . [ cols="<,^,^,^,<,<,<",options="header " , ] identified in our photometry . as seen in the ( @xmath30 ) cmd of figure [ fig3 ] or 8 , our membership selection constrains the pms locus well . however , current membership are certainly incomplete , because the membership criteria of pms stars mentioned above are biased toward stars with an active accretion disk . however , it is worth comparing the detection efficiency of low - mass pms stars between h@xmath1 photometry and mir data for stars brighter than @xmath31 ( our photometric completeness is about or better than 88% ) . a total of 94 stars were classified as low - mass pms members either from mir excess emission or h@xmath1 photometry . among them , 32 stars were detected from both mir and h@xmath1 photometry . interestingly , none of 5 bright transitional disk candidates ( @xmath32 ) shows h@xmath1 emission . however , the h@xmath1 detection fraction of faint transitional disk candidates ( @xmath33% to @xmath3419 mag ) is very similar to that of class ii objects ( @xmath35% ) . this fact implies that the faint transitional disks have similar characteristics to class ii objects , i.e. they are probably pre - transition disk objects @xcite . the reason for the absence of h@xmath1 emission among bright pms stars with a transitional disk may be related either to the absence of a hot inner disk or to weak h@xmath1 emission with h@xmath1 emission equivalent widths smaller than 10 @xcite , or both . the detection efficiency of transitional disk candidates ( @xmath36 mag ) from h@xmath1 photometry may be less than 23 % with respect to mir excess emission . in this work , mir excess appears to be a more efficient way to find pms members than h@xmath1 photometry . however , the detection efficiency from mir and h@xmath1 photometry is not as high as x - ray observation and strongly depends on the age of a cluster because the excess emission is prominent in young pms stars with a circumstellar disk and accretion activity . in addition , we know that weak - line t - tauri stars ( wtts ) are unlikely to be selected as pms members from mir or h@xmath1 photometry because most of the material in the disk of such stars has dissipated . indeed , in ngc 2264 , many wtts without h@xmath1 emission or a mir excess were identified as pms by x - ray observations @xcite , and their photometric properties were found to be similar to those of ms stars @xcite . almost complete lists of pms members in ngc 2264 and ngc 6231 have now been identified down to their limiting magnitude using x - ray sources from the _ chandra _ data archive @xcite or from xmm - newton observation @xcite . up to now , no extensive x - ray observations for the ic 1848 region exist , so we note that a large number of wtts or low - mass pmss with weak accretion activity are likely missed in our membership selection . 0.60 ( dashed line ) , 0.65 ( dotted line ) , and 0.70 mag ( solid line ) . large open circles indicate the o stars . , scaledwidth=40.0% ] the interstellar reddening toward young open clusters is , in general , derived by comparing the observed color indices of the early - type members with the intrinsic ( @xmath37 ) diagram . using the intrinsic color - color relation from table 1 in paper 0 we determined the individual reddening @xmath38 of early - type members of ic 1848 . the reddening slope was simply assumed to be @xmath39 because the dependence on @xmath38 is negligible for less reddened stars [ @xmath40 mag ] . the mean reddening value was estimated to be @xmath41 ( s.d . ) mag and displayed in the left panel of figure [ fig2 ] as a dashed line . this value is in good agreement with that of previous studies , e.g. , @xmath38 = 0.55 0.70 mag @xcite , 0.66 mag @xcite , 0.72 mag @xcite , and 0.60 mag @xcite . from the spread of the early - type members in the @xmath42 diagram , one can see that there is a non - negligible amount of differential reddening across the observed region . in order to investigate the spatial variation of the reddening we constructed the reddening map of the observed regions using the position and reddening of individual early - type members . the observed field was divided into 10,000 rectangular areas , each of which has the size of @xmath43 in right ascension and declination , respectively . the weighted - mean reddening of each point is calculated from the reddening of 105 early - type members , where the weight is exponentially decreasing with the distance from individual early - type members . we present the reddening map of ic 1848 in figure [ fig4 ] . the massive molecular cloud w5 nw in w5 west lies in the northern part of the cluster ( see figure 1 of @xcite and @xcite ) . o type stars , hd 17505 , hd 17520 , and hd 237007 , in w5 west are responsible for the ionization of the molecular clouds @xcite , and bright rims lie along the ionization front @xcite . these could be evidence that the stellar wind and uv radiation from three o type stars are compressing the molecular cloud . on the other hand , the reddening map of the southern part is rather transparent compared to the northern part , and one could expect that the material in the southern part might have been rapidly swept away . thus , the large scale variation of reddening shows the systematic difference between the northern and southern part of the cluster . another o star bd + 60 586 ( @xmath44 ) is located in the valley of reddening between two molecular clouds w5 ne and w5 nw . the structure in the reddening map may reflect the interaction between the o stars and the surrounding materials . a similar feature was found in ngc 6231 @xcite . from this , we deduce that the strong stellar wind from o stars can sweep away the surrounding medium . in contrast , @xcite found the highest reddening value to be in the center of the starburst cluster westerlund 1 and interpreted this as resulting from dust formed from the ejected material from the many wolf - rayet stars . these observations indicate that within ic 1848 , dust destruction may be more dominant than dust production . however , unlike the other o type stars , hd 237019 ( @xmath45 ) lies in a highly reddened region . according to a formation scenario of w5 west suggested by @xcite the star could be ejected from the cluster by gravitational interaction between massive stars in the cluster center . if this scenario is true , the star may not have had enough time to interact with the surrounding material . to check the reliability of the reddening corrections for pms stars in our analysis , we compared the reddening @xmath46 from figure [ fig4 ] with that from the spectral type . recently @xcite published the spectral types of 389 stars in the w5 region . among them , 63 stars with @xmath47 , @xmath48 , and @xmath49 color indices in our data , were used in this comparison . the spectroscopic reddening was obtained by comparing the observed @xmath47 color with the intrinsic color inferred from a given spectral type ( paper 0 ) . one can expect that the mean difference would be close to zero with a random scatter , and that spectroscopic reddening could be higher for few stars due to their internal extinction . we present the comparison in the left panel of figure [ fig5 ] , where negative values mean that spectroscopic reddening is larger than that from our reddening map . for b early - f and k5 late - k type stars , the reddening shows reasonably good agreement . however , the spectroscopic reddening is 0.2 mag higher than ours for late - f k4 type stars , and systematically lower for m type stars . in order to find the main causes of the differences we checked our intrinsic relations in paper 0 , which is based on the ( @xmath48 , @xmath47 ) relation by @xcite with the spectral type - color relation of @xcite . the difference is less than 0.02 mag , and therefore it is unlikely to be caused by the spectral type ( @xmath47 ) relation . we also compared the reddening map from optical photometry with that from the 2mass data . with @xmath50 and @xmath51 of the early - type members ( see next section ) the nir reddening map was constructed through the same procedure as the optical reddening map . the spatial variation of the @xmath52 reddening is compatible with that of the optical photometry . in addition , we also checked the spatial distribution of the g m type stars which showed a systematic difference in reddening derived from the two different methods . if our reddening map exclusively provides small / large value of reddening toward any specific direction , this method of reddening correction may be inappropriate for the pms members . however , the late - type pms stars were distributed uniformly , thus the application of the reddening map may have nothing to do with the discrepancy . the other possibility is that the spectral types of the stars were classified as earlier types due to veiling filling in some of the lines . @xcite described their classification schemes without any discussion on the veiling effect . since the software sptclass @xcite provides the spectral type based on the equivalent width of the spectral lines of o8m6 type ms stars ( see their webpagehernandj / sptclass / sptclass.html ] ) , veiling effect could make the spectral classification of pms stars tricky . the classification schemes in sptclass are optimized for herbig ae / be , late f mid k , and mid k m type stars , respectively . the classification for late f mid k type stars includes 11 spectral indices in the relatively narrow spectral region ( from 4200 to 6200 ) , and thus it can be affected by veiling . furthermore there are fewer number of spectral indices than for herbig ae / be and later type stars . hence , the spectral classification of pms stars could be misled by the presence of non - photospheric contributions . according to the spectral indices for f , g , and k type stars , e.g. ca i @xmath53 , fe i @xmath544458 , and fe i @xmath545329 , the equivalent width of the lines increases with spectral type . if optical spectra of pms stars with veiling exhibit decreased line depths , the equivalent widths of the spectral indices will appear to be that of earlier types . therefore veiling would shift the spectral classification to earlier types . if the speculation is correct , the stars should exhibit a uv excess . we plot uv excess , @xmath55 , derived in section 5 , with respect to the discrepancy in the reddening @xmath46 in the right panel of figure [ fig5 ] . the stars possibly classified as earlier spectral type are expected to be located in the lower left side of the panel . indeed , they ( f0 k4 type pms stars - circle , open square , and triangle ) locate in the expected region , while stars without a uv excess are clustered at @xmath56 and @xmath57 . however , k5 m type stars show a systematic and opposite trend . as the strength of tio and cah bands are used for the classification of mid - k to m type stars , the spectral classification should be correct . in addition , these red bands would be little affected by non - photospheric contributions . as shown in the right panel of figure [ fig5 ] , k5 k9 type stars ( filled square ) are close to zero in @xmath58 , although most of them exhibit a uv excess . however , 3 ( i d 10938 , 13480 , and 14932 ) out of 4 m type stars with @xmath47 and @xmath59 color indices show a small spectroscopic reddening @xmath60 of @xmath61 mag , which is an unrealistic value for ic 1848 . we also checked the position of the stars in figure [ fig8 ] and found them to lie in the pms locus . these facts imply that they have too - late a spectral type . according to @xcite those stars are identified as class ii objects with strong h@xmath1 emission , and our h@xmath1 photometry also tags them as h@xmath1 emission stars . they must be very young stars of ic 1848 , and therefore we attribute this discrepancy to a gravity effect on the spectra of young pms stars . many studies @xcite remark that the spectral features of young pms stars between 6000 and 9000 resemble those of giant stars and we note that the inclusion in the spectral classification methodology of a na i @xmath62 index , that is very gravity sensitive , would help discriminate between giant- and dwarf - like m spectra . given the weaker tio bands of ms stars compared to giant stars , it is possible that the stronger bands of a pms star can lead to them being assigned a later dwarf spectral type . the same trend as in figure [ fig5 ] can also be found in the published data ( @xmath63 and spectral type ) of @xcite . we investigated their @xmath63 values with respect to the spectral types . the @xmath63 values of g k type stars appear to be higher than those of other spectral types . the mean @xmath63 of b f0 type stars is about 1.5 mag , but the @xmath63 values of g5 type stars have a range of 2.0 - 2.6 mag . the differences between two groups with different spectral types in @xmath46 is about 0.21 - 0.45 mag . these values are in good agreement with those found in figure [ fig5 ] . in addition , the @xmath63 values of k m type stars decrease with the subclasses . the feature is the same as that found in figure [ fig5 ] . these facts imply that the discrepancy found in figure [ fig5 ] originates from the published data of @xcite . we therefore conclude that the application of our reddening map is more reliable than the reddening correction inferred from previously published spectral types in the current state and does not cause serious problems , except for a few pms stars with nearly edge - on disks . dust evolution in star forming regions is one of the more interesting issues in astronomy , since the grain size is directly related to the reddening law toward an object of interest . many studies @xcite have confirmed that the extinction law in the solar neighbourhood , the galactic center , or in moderately young open clusters and associations is nearly the normal law ( a total - to - selective extinction ratio of @xmath64 ) . however the value is significantly larger for several extremely young star forming regions ( see table 3 in @xcite ) . for ic 1848 several previous photometric studies @xcite have assumed a normal reddening law . recently , @xcite found different @xmath65 in four different regions within w5 east . since the ratio of total - to - selective extinction is a useful tool to study the state of dust evolution in star forming regions , and since the value is critical in the canonical determination of distance , we tested the reddening law in ic 1848 using color excess ratios . using the nir 2mass data and mir _ data of @xcite it is possible to determine the reddening law toward ic 1848 in a consistent way from the optical to the mir photometry . to calculate the color excess ratios @xmath66 , observed color indices were compared with the intrinsic colors for the 2mass bands , from table 2 in paper 0 , and the _ spitzer _ irac bands ( sung et al . 2013 , in preparation ) . in this analysis , we used 21 of the most probable normal ms stars among 105 early - type members with @xmath49 color index smaller than -0.2 mag , to exclude the young active stars which could potentially affect the color excess ratios . in addition , three foreground stars were also used to investigate the foreground reddening law . the color excess ratios obtained from such stars are plotted in figure [ fig6 ] . we confirmed that the color excess ratios of foreground stars ( open circle ) and early - type members ( dot ) of ic 1848 are well - consistent with the solid line corresponding to the normal reddening law ( @xmath3 ) . unlike the optical band and the nir band , there is a slightly large scatter in the mir bands . there are two possible sources of scatter . we carried out psf photometry for the point sources in four sets of the long - exposed _ spitzer _ images ( program i d : 20300 , pi : allen , lori ) for the w5 region . the photometric data were compared with those of @xcite and the glimpse 360 catalogue where 3.6 and 4.5 bands are available . the photometric data are consistent with each other within 0.05 mag in the zero points , except for the 4.5 band ( @xmath67 = 0.07 - 0.08 $ ] mag ) . we also found a scatter of 0.03 - 0.07 mag among mir data , and therefore some stars can show a deviation of up to 0.1 mag . in addition , the inherent uncertainty of mir photometric data arising from the nonlinearity of the mir detector and the difficulty in the flux calibration is , in general , up to 10% . the photometric error could be also increased for some stars lying in crowded fields or pah emission nebulae . on the other hand , it is a well - known fact that ir excess emission arises from a circumstellar disk or surrounding material of young stars . thus , the scatter in the mir bands may be partly due to uncertainties in the mir photometry and partly due to circumstellar dust emission of the stars . allowing for the excess emission , the ratios of total - to - selective extinction ( @xmath65 ) , which we obtained here , are consistent for all wavelengths , and therefore the reddening law toward ic 1848 is normal . it implies that the size distribution of dust in ic 1848 is very similar to that of the general diffuse interstellar medium . @xcite found a normal reddening law toward the cluster around hd 18326 in the w5 east h ii region , while the @xmath65 of bright - rimmed clouds shows various values from 2.8 to 3.5 . the authors argued that the dust grains in the regions have inhomogeneously evolved . given their results , the strength of uv radiation from o type stars in the clusters affects the evolution of dust grains in the natal cloud . zams fitting is the most important method to estimate the distance to open clusters . in order to obtain a reliable distance from all cmds , reddening should be corrected for properly . as mentioned above , the total extinction in the @xmath26 band for individual early - type members was determined from @xmath68 . the mean value was @xmath69 ( s.d . ) mag . after correcting for the reddening and extinction , zams fitting to the de - reddened cmds was performed . since multiplicity and evolution effects make stars brighter , zams fitting to the lower ridge of the ms band is recommended , allowing for the photometric error . several relatively faint early - type members ( mid late b type ) appears to be bluer than the zams color indices . @xcite have reported the correlation between uv excess and x - ray luminosity of ms stars in the intermediate - age open cluster ngc 2516 ( age @xmath70160 myr ) . according to them , the higher the x - ray luminosity , the bluer the @xmath49 and @xmath48 color indices . due to the absence of x - ray observations for ic 1848 , we could not obtain the x - ray luminosity of the blue early - type members . in the case of the young open cluster ngc 1893 , the late b type ms stars with bluer color indices were found to be x - ray emitters ( lim et al . 2013 , in preparation ) . if this property is universal , high x - ray luminosity is responsible for the bluer color indices . if the zams relations were fitted to such blue ms stars , the distance would be overestimated . thus , the most probable normal ms members ( @xmath71 ) , as done in the previous section , were used in the zams fitting procedure . from the zams fitting to all reddening - corrected cmds , the distance modulus of ic 1848 was estimated to be @xmath72 mag , which is equivalent to 2.2 kpc . in figure [ fig7 ] , we also obtained the same result from four reddening - independent indices , as shown in paper 0 ( equation 9 12 ) . the uncertainty in the distance modulus may be 0.2 mag from figure [ fig7 ] . the distance derivations of previous investigators ranged from 1.7 kpc to 2.4 kpc . @xcite obtained a distance of 1.7 kpc from their spectral types and photometric data for o stars . however , the author did not discuss the treatment of extinction correction nor considered binarity . the distance to the cluster from other photometric studies @xcite clusters around @xmath73 kpc , assuming the normal reddening law . there are two independent measurements . @xcite constructed new galactic rotation curves derived from the distances of exciting stars and h@xmath1 radial velocities for the northern and southern hemispheres , respectively . the authors obtained a distance of 2.3 kpc for ic 1848 from a @xmath74 of @xmath75 . this value is consistent with that of the photometric studies . however , the adopted galactocentric distance ( 10 kpc ) of the sun is rather larger than the currently accepted value . on the other hand , @xcite measured a trigonometric parallax for 12-ghz methanol masers in the w3oh region and obtained a distance of @xmath76 kpc . given the little annual variation of methanol masers , the result seems to be reliable . although there is an implicit assumption that the distances to w3/w4/w5 are the same , no careful discussion of the assumption has been made . @xcite and @xcite have measured the trigonometric parallax for the young open cluster ngc 281 and two h@xmath77o maser sources , iras 00259 + 5625 and iras 00420 + 5530 , using vlbi exploration of radio astrometry . according to their studies , the three h@xmath77o maser sources ( @xmath78 kpc , @xmath79 kpc , and @xmath80 kpc ) are along the line of sight , sequentially ( see figure 5 in @xcite ) . from these measurements they could estimate the size of the expanding ring of molecular clouds , which was found by @xcite , where the diameter along the line of sight is larger than 650 pc . given these studies , the difference between the distances estimated from the photometric methods and the trigonometric parallax for the w3oh region ( @xmath70 250 pc ) can be interpreted as the extent along the line of sight , and therefore the w3oh region is likely in front of the w3/w4/w5 complex . since our result supports the estimates from previous photometric studies for ic 1848 , we use the distance modulus ( 11.7 mag ) obtained here . in figure [ fig3 ] we present the cmds with a zams obtained by adopting the mean reddening and distance . vs. @xmath81 color - magnitude diagram . the reddening of early type stars is estimated from the @xmath49 vs @xmath48 color - color diagram . the reddening map was used to correct the reddening of low - mass pms members . the pre - main sequence locus is defined by the two solid lines which enclose the distribution of most pms members in the diagram . the other symbols are the same as figure 2.,scaledwidth=41.0% ] we directly corrected for the reddening of individual early - type members , while the reddening map ( figure [ fig4 ] ) was used for the reddening correction of the other members . the dereddened @xmath82 cmd is presented in figure [ fig8 ] . in order to constrain the upper limit of the imf ( see next section ) , the locus of pms stars was defined as shown in the figure ( solid lines ) . we describe the calibrations required for the construction of the hertzsprung - russell diagram ( hrd ) . for stars earlier than o9 , the spectral type effective temperature relation from table 5 in paper 0 was used to estimate the effective temperature . we applied the relation between @xmath81 and effective temperature of @xcite to the pms members with @xmath83 and another relation @xcite to the remaining pms members to avoid the effects of the uv excess , arising from accretion activities , on the @xmath49 and/or @xmath48 colors . for the other stars we averaged the temperature derived from the spectral type effective temperature relation and the color effective temperature relations with an appropriate weight . finally the bolometric corrections were obtained from the relations in paper 0 . we present the hrd of ic 1848 in figure [ fig9 ] . several evolutionary tracks ( dashed lines ) with an initial mass and isochrones ( solid lines ) were over plotted in the figure . the stellar evolution models of @xcite , which take into account the effect of stellar rotation on the evolution of stars for a given mass , were used for post - ms and ms stars , while we used the models of @xcite for pms stars . the isochrones ( 0.5 , 2.0 , and 5.0 myr ) have been constructed from the two models . the stars with masses larger than @xmath84 seem to be moving away from the ms stage , while intermediate - mass stars @xmath85 seem to be approaching the zams . the isochrone with an age of 5 myr well predicts the position of the most massive stars in the hrd . however , according to @xcite , hd 17505 , hd 17520 , and bd+60 586 are binaries or multiple systems , and therefore it is probable that the turn - off age may be overestimated . we have conducted a deconvolution of the brightness of hd 17505 by assuming that the multiple system consists of an o6.5 iii , two o7.5v stars , and an o8.5 v @xcite . as a result , the hd 17505 system ( the o6.5iii component deconvolved from the others ) should be dimmed by @xmath86 mag from its current position , and is then well consistent with the 5 myr isochrone . although we adopted a luminosity class iii for the brightest component of hd 17505 , it is worth noting that the genuine luminosity may be very close to ib or iab supergiant because the deconvolved absolute magnitude is -6.29 mag ( see table 4 . in paper 0 ) . we have also estimated the age of ic 1848 from low - mass pms stars using the pms evolution models of @xcite . figure [ fig10 ] shows the age distribution of pms stars . the majority of pms stars cover a span of 6 myr with a median age of about 3 myr . given the age spread of 3 6 myr found in other young open clusters , such as ngc 2244 @xcite , ngc 2264 @xcite , ngc 6530 @xcite , and trumpler 14 and 16 @xcite , the age spread estimated in this work is reasonable , although the age determined for pms stars strongly depends on the pms evolution models used . in figure [ fig10 ] , the long tail toward older age may arise from overestimating the age of stars in the kelvin - helmholtz contraction phase in the pms models @xcite , and from stars with an edge - on disk lying below the pms locus @xcite . it is worth noting that some of stars in the long tail may be real , old pms stars , since the star formation history of the w5 region could be quite complex . in order to understand the star formation history in w5 west , including ic 1848 , the age distribution based on more complete , membership selection is needed . @xcite estimated a lower limit of 2.4 myr for the age of ic 1848 , assuming a sound speed for the ionized gas , and an upper limit of 5 myr , from the lifetime of o stars . @xcite inferred that the age of the w5 region is younger than 5 myr using the evolution model of @xcite . our estimates are in good agreement with those of the previous studies . on the other hand , @xcite claimed that the age of ic 1848 could not exceed 1 myr given the existence of the binary or multiple system , hd 17505 , hd 17520 , and bd+60 586 . however , no supporting discussion of the luminosity class or the lifetime of such stars was provided in the study , and we believe that an age of @xmath87 1 myr is unlikely . as the imf is a very important tool to understand star formation processes , a number of studies have attempted to derive the imf in different star forming environments . @xcite reviewed imf studies and the relationship with environmental conditions and mass range . since ic 1848 is one of the large - scale star forming regions in the perseus spiral arm , this object must be an ideal laboratory to study the imf in a low metallicity and low spatial density environment . given the fact that the cluster contains a few massive stars it is possible to derive the imf over a wide mass range . for all post - ms or ms members , the masses were estimated by comparing their position in the hrd to post - ms evolutionary tracks . the mass of pms stars , or stars in the pms locus without any membership criterion , were estimated by comparing their position in the hrd to pms evolution model tracks . in order to derive the imf we counted stars within a given mass bin ( @xmath88 ) . for the most massive bin , the bin size ( @xmath89 ) was 0.64 because of the small number of massive stars . the star counts were then normalized by the logarithmic mass bins and the observed area . the imf of ic 1848 is shown in figure [ fig11 ] . to avoid the binning effect we shifted the mass bin by 0.1 and re - derived the imf ( open circles ) , using the same procedure as above . in the figure , the circle and dot represent the imf of members with h@xmath1 emission or ysos from @xcite , while the dotted line corresponds to the imf including all stars within the pms locus . the shape of the imf of members in the low - mass regime is very similar to that of previous studies for other young open clusters , such as ngc 2244 @xcite , ngc 2264 @xcite , ngc 6530 @xcite . with complete membership criteria , @xcite showed a smooth increase in the imf of ngc 2264 down to 0.25 @xmath90 compared to that of @xcite . @xcite and @xcite have discussed the efficiencies of detecting pms stars using h@xmath1 photometry ( 47 % from a single observation ) and mir excess ( 38 % ) , respectively . hence we attribute the dip in the imf of ic 1848 to the incomplete membership selection for the stars in the kelvin - helmholtz contraction phase , while the incompleteness of our photometry may influence the imf in the lower mass regime ( @xmath91 ) . the imf of the members ( @xmath92 ) should be a lower limit , whereas that of all stars within the pms locus may be an upper limit in the same mass range . we first derived the imf of ic 1848 complete down to @xmath6 through this study . the slope ( @xmath93 ) of the imf is about @xmath94 for @xmath95 . it appears to be steeper than the single imf @xcite and the kroupa imf @xcite , but close to that of @xcite for massive stars . the slope of the imf for ic 1848 is also steeper than that of other young open clusters ( @xmath96 for ngc 6530 @xcite , @xmath97 for ngc 2244 @xcite , @xmath98 for orion nebula cluster ( onc ) @xcite , @xmath99 for w5 east @xcite , @xmath96 for trumpler 14 and 16 @xcite , @xmath100 for ngc 6231 @xcite ) . on the other hand , the slope ( @xmath101 ) and shape of the imf for ngc 2264 @xcite are similar to those of ic 1848 in the same mass range . in order to arrive at a firm conclusion on the shape of the imf down to subsolar mass regime , further deep optical and/or nir observation with extensive x - ray observations are needed . the complete imf of the cluster will give an insight into the star formation processes in the large - scale star forming region . vs. @xmath81 diagram . several pre - main sequence stars exhibit a remarkable uv excess driven by their accretion activity . a total of 51 uv excess stars ( plus symbols ) were found . the other symbols are the same as in figure 2.,scaledwidth=45.0% ] the mass accretion rate gives important information on the disk evolution around pms stars . since the emergence of the magnetospheric accretion model for pms stars , the physical quantities associated with the accretion process have been studied by observation and modeling uv excess emission or / and spectral lines such as h@xmath1 , pa@xmath102 , br@xmath103 , [ o i ] 6300 , etc , within such a paradigm . in this section we estimate the mass accretion rate of those pms stars with a uv excess found from our photometric data using the relation between the accretion luminosity and @xmath28-band brightness @xcite and compare the results with that of other studies which used independent ways to estimate accretion luminosities for different star forming regions . the first step is to find uv excess stars among pms members which exhibit either h@xmath1 emission or a mir excess . several pms stars exhibit a uv excess in the @xmath104 $ ] diagram as shown in figure [ fig12 ] . the majority of stars with a uv excess are late - type members ( @xmath105 mag ) . these late - type members appear to be bluer in @xmath59 than normal ms stars without excess emission at a given @xmath81 . the deviation from the expected photospheric color of ms stars is larger than the value expected from differential reddening ( @xmath106 mag ) . hence the large uv excess may be related to the intrinsic properties of late - type pms stars rather than to inappropriate reddening corrections ( we have further discussions on the reddening corrections for pms members in section 3.2 ) . similar features can be found in @xcite for the low - mass members in the onc . the authors discussed several sources affecting the calculated uv excess , such as chromospheric activity , accretion activity , and the difference between the intrinsic colors of ms and giant stars . @xcite have examined the influence of chromospheric activity on the size of the uv excess using field dme stars and young stars in the taurus - auriga star forming region . they suggested that the limit of uv excess from chromospheric activity is about -0.5 mag . this value is adopted as a criterion for the pms stars with a uv excess in the @xmath104 $ ] diagram . a total of 51 members were identified as pms stars with a uv excess as shown in figure [ fig12 ] ( plus symbols ) . we computed the @xmath107 magnitude expected for the photospheric color of ms stars , the extinction - corrected @xmath108 magnitude of stars with uv excess emission , then transformed them to luminosity ( @xmath109 and @xmath110 ) using a bandwidth ( 700 ) and zero magnitude flux of @xmath111 @xmath112 for the bessell @xmath28 band @xcite . the accretion luminosity @xmath113 was estimated from the relation of @xcite : @xmath114 where @xmath115 . in order to obtain mass accretion rates we estimated the mass of individual stars from the evolution models of @xcite . the effective temperature and bolometric magnitude of pms stars obtained above allow us to estimate the radii of the stars . with mass ( @xmath116 ) , radius ( @xmath117 ) , and accretion luminosity ( @xmath118 ) the mass accretion rate of pms stars is obtained by using the equation below @xcite : @xmath119 where @xmath120 and @xmath121 represent the gravitational constant and mass accretion rate , respectively . we present the mass accretion rate of pms members with uv excess in figure [ fig13 ] . the mass range does not fully cover the low - mass stars ( @xmath122 ) as shown in @xcite and @xcite for the onc , because ic 1848 is more distant than the onc . in the same mass range our results are comparable to those studies by @xcite and @xcite for the onc . we also plotted results from other studies for different star forming regions ( @xcite , and data therein ) . the authors used independent ways to estimate the accretion luminosity , such as modelled balmer excess emission and the correlation between accretion luminosity and the strength of hydrogen recombination lines , pa@xmath102 , and br@xmath103 . the accretion luminosity and mass accretion rate in this study seem to be compatible with the estimates of other studies for different star forming regions . the mean mass accretion rate of stars in the mass range of @xmath9 to @xmath10 is about @xmath8 . because of the limited mass range of our sample , we can not confirm whether or not @xmath121 is proportional to @xmath123 ( @xmath124 , @xcite ) . however , with the previous estimates for masses below @xmath10 our results are well consistent with the previously known correlation between the two quantities . on the other hand , as shown in @xcite , three intermediate - mass pms stars exhibit higher mass accretion rates than expected from the correlation found for low - mass stars . in that study , the authors found that the age of herbig ae / be stars in their sample is systematically younger than that of low - mass stars , and therefore the steep slope in the mass accretion rate with respect to stellar mass was interpreted as the active accretion of young intermediate - mass pms stars . in this study , we also found that the pms star ( i d 5923 ) with the highest mass accretion rate is younger than the other intermediate - mass pms stars ( 2.5 - 5 @xmath90 ) . however , @xcite obtained lower mass accretion rates for the same herbig ae stars . the estimates of @xcite ( except the upper limits ) , also reveal 3 times lower values for herbig ae / be stars . the systematic difference in the mass accretion rate for the herbig ae / be stars among authors may have arisen from many uncertain factors such as different uses of evolution models , calibration schemes , reddening correction , distance , disk geometry with mass , etc . however the details are not within our scope of study based on limited data . unlike low - mass star forming regions , our target is a young open cluster hosting several high - mass ms stars . thus , the fundamental parameters , such as reddening and distance , can be constrained very well to estimate the mass accretion rate of pms stars ( @xmath125 ) . furthermore , in forthcoming papers as part of the sos project , homogeneous photometric data for many young open clusters within 3 kpc from the sun will provide homogeneous estimates of the mass accretion rates and contribute to understanding the evolution of the circumstellar disks of pms stars and formation of planetary systems . ic 1848 is one of the most interesting targets in the large - scale star forming environment of the cas ob6 association . we carried out @xmath0 and h@xmath1 photometry for the young cluster ic 1848 in the w5 west region as part of the sos project . this study provided not only homogeneous photometric data but also detailed and comprehensive results for ic 1848 as shown below . using the photometric properties of early - type stars , a total of 105 stars were selected as members . we identified 397 young stars with mir excess emission from the _ spitzer _ irac data @xcite and using our h@xmath1 photometry and spectroscopy by @xcite , 257 h@xmath1 emission stars were also identified . a total of 462 pms stars were selected as pms members . from the @xmath42 diagram we obtained the reddening of individual early - type stars yielding a mean reddening of @xmath126 mag . in common with most young star clusters or associations , differential reddening was not negligible in the observed regions . we constructed the reddening map to correct the reddening of pms members . the photometric data from the optical to mir for the early - type stars ( @xmath71 ) in the observed regions allowed us to examine the reddening law of ic 1848 . we obtained a consistent ratio of total - to - selective extinction of @xmath3 over a wide wavelength coverage from the optical to the mir , and concluded that the reddening law toward ic 1848 is normal . in order to determine the distance to the cluster , zams fitting to the lower boundary of the ms band was carried out both in the reddening - corrected and reddening - independent cmds , and we obtained a distance modulus of @xmath127 mag , which is in good agreement with that of previous photometric studies . the isochrone fitting to the evolved stars gives an age of 5 myr , while the age distribution of pms stars shows a median age of 3 myr with an age spread of 6 myr . the imf complete down to @xmath6 was derived from cluster members for the first time . due to the incomplete membership selection of low - mass pms stars , the slope of the imf ( @xmath7 ) was computed for the limited mass range ( @xmath95 ) . that was similar to the scalo imf rather than the salpeter / kroupa imf . we also found that the imf of ic 1848 closely resembles that of ngc 2264 @xcite among many young open clusters in the same mass range . we found 51 pms stars with a strong uv excess from our photometry . the @xmath28 luminosity was transformed to an accretion luminosity by adopting an empirical relation from the literature . finally the mass accretion rate of the pms stars was estimated by using the accretion luminosity , the computed stellar radius , and the stellar mass inferred from the pms evolution models . the stars in the mass range of 0.5 @xmath90 to 2 @xmath90 exhibit a mean mass accretion rate of @xmath8 . our result is well consistent with the estimates of other studies for different star forming regions . we also found that high - mass stars show a higher mass accretion rate than the accretion rate expected from the correlation . the authors express deep thanks to the anonymous referee for many useful comments and suggestions and dr . colette salyk for useful discussions . this work was supported by a national research foundation of korean ( nrf ) grant funded by the korea government ( mest ) ( grant no.20120005318 ) . bastian , n. , covey , k. r. , & meyer , m. r. 2010 , araa , 48 , 339 becker , w. , & fenkart , r. 1971 , a&as , 4 , 241 bessell , m. s. 1990 , pasp , 102 , 1181 bessell , m. s. 1995 , in proc . eso workshop , the bottom of the main sequence and beyond , ed . c. g. tinney ( berlin : springer ) , 123 bessell , m. s. , & brett , j. m. 1988 , pasp , 100 , 1134 bessell , m. s. , castelli , f. , & plez , n. 1998 , a&a , 333 , 231 bigay j. h. 1963 , journal des observateurs , 46 , 319 briceo , c. , hartmann , l. , stauffer , j. , & martn , e. 1998 , aj , 115 , 2074 carpenter , j. m. , heyer , m. h. , & snell , r. l. 2000 , apjs , 130 , 381 chauhan , n. , pandey , a. k. , ogura , k. , et al . 2011 , mnras , 415 , 1202 conti , p. s. , & leep , e. m. 1974 , apj , 193 , 113 cousins , a. w. j. 1978 , mnssa , 37 , 62 cox , a. n. ( ed . ) 2000 , in allen s astrophysical quantities ( 4th ed ; new york : springer ) , p387 donehew , b. , & brittain , s. 2011 , aj , 141 , 46 ekstrm , s. , georgy , c. , eggenberger , p. , et al . 2012 , a&a , 537 , 146 elmegreen , b. g. , & lada , c. j. 1977 , apj , 214 , 725 fehrenbach , c. 1966 , publ . haute - provence , 8 , 25 flaccomio , e. , micela , g. , sciortino , s. , et al . 1999 , a&a , 345 , 521 garcia lopez , r. , natta , a. , testi , l. , & habart , e. 2006 , a&a , 459 , 837 garmany , c. d. , & stencel , r. e. 1992 , a&as , 94 , 211 georgelin , y. m. , & georgelin , y. p. 1976 , a&a , 49 , 57 greve , a. 2010 , a&a , 518 , 62 guetter , h. h. , & vrba , f. j. 1989 , aj , 98 , 611 gullbring , e. , hartmann , l. , briceo , c. , & calvet , n. 1998 , apj , 492 , 323 johnson h.l . , & hiltner , w.a , 1956 , apj , 123 , 267 johnson , h. l. , hoag , a. a. , iriarte , b. , mitchell , r. i. , & hallam , k. l. 1961 , lowell obs . bull . , 5 , 133 hartmann , l. , calvet , n. , gullbring , e. , & dalessio , p. 1998 , apj , 495 , 385 haug u. 1970 , a&as , 1 , 35 hernndez , j. , calvet , n. , briceo , c. , hartmann , l. , & berlind , p. 2004 , aj , 127 , 1682 hillwig , t. c. , gies , d. r. , bagnuolo , jr . , et al . 2006 , apj , 639 , 1069 hoag a. a. , johnson h. l. , iriarte b. , et al . 1961 , publ . u.s . naval obs . , 17 , 347 hoag , a. a. , & applequist , n. l. 1965 , apjs , 12 , 215 h@xmath128 g , e. , fabricius c. , makarov v. v. , et al . 2000 , a&a , 355 , l27 hur , h. , sung , h. , & bessell , m. s. 2012 , aj , 143 , 41 i m , m. , ko , j. , cho , y. , et al . 2010 , jkas , 43 , 75 karr , j. l. , & martin , p. g. 2003 , apj , 595 , 900 koenig , x. p. , allen , l. e. , kenyon , s. j. , su , k. y. l. , & balog , z. 2008 , apj , 687 , l37 koenig , x. p. , allen , l. e. , gutermuth , r. , et al . 2008 , apj , 688 , 1142 koenig , x. p. , & allen , l. e. 2011 , apj , 726 , 18 kook , s .- h . , sung , h. , & bessell , m. s. 2010 , jkas , 43 , 141 kroupa , p. 2002 , science , 295 , 82 lada , c. j. , elmegreen b. g. , cong , h .- , & thaddeus , p. 1978 , apj , 226 , 39 lasker , b. m. , lattanzi , m. g. , mclean , b. j. , et al . 2008 , aj , 136 , 735 lefloch , b. , lazareff , b. , & castets , a. 1997 , a&a , 324 , 249 lim , b. , sung , h. , karimov , r. , & ibrahimov , m. 2008 , pkas , 23 , 1 lim , b. , sung , h. , bessell , m. s. , karimov , r. , & ibrahimov , m. 2009 , jkas , 42 , 161 lim , b. , sung , h. , karimov , r. , & ibrahimov , m. 2011 , jkas , 44 , 39 lim , b. , sung , h. , chun , m .- y . , et al . 2013 , aj , 145 , 46 loktin , a. v. , gerasimenko , t. p. , & malysheva , l. k. 2001 , astron . astrophys . trans . , 20 , 607 loren , r. b. , & wootten , h. a. 1978 , apj , 225 , 81 luhman , k. l. , briceo , c. , rieke , g. h. , & hartmann , l. 1998 , apj , 493 , 909 luhman , k. l. , liebert , j. , & rieke , g. h. 1997 , apj , 489 , 165 manara , c. f. , robberto , m. , rio , n. d. , et al . 2012 , apj , 755 , 154 massey , p. , johnson , k. e. , & degioia - eastwood , k. 1995 , apj , 454 , 151 megeath , s. t. , biller , b. , dame , t. m. , et al . 2002 , in asp conf . ser . 267 , hot star workshop iii : the earliest stages of massive star birth , ed . p. a. crowther ( san francisco : asp ) , 257 megeath , s. t. , biller , b. , dame , t. m. , et al . 2003 , rev . , 15 , 151 mendiguta , i. , calvet , n. , motesinos , b. , et al . 2011 , a&a , 535 , 99 menzies , j. w. , marang , f. , laing , j. d. , coulson , i. m. , & engelbrecht , c. a. 1991 , mnras , 248 , 642 moffat , a. f. j. 1972 , a&as , 7 , 355 morgan , w. w. , code , a. d. , & whitford a. e. 1955 , apjs , 2 , 41 muench , a. a. , lada , e. a. , lada , c. j. , & alves , j. 2002 , apj , 573 , 366 muzerolle , j. , hillenbrand , l. , & calvet , n. 2003 , apj , 592 , 266 muzerolle , j. , luhman , k. l. , briceo , c. , hartmann , l. , & calvet , n. 2005 , apj , 625 , 906 natta , a. , testi , l. , & randich , s. 2006 , a&a , 452 , 245 niwa , t. , tachihara , k. , itoh , y. , et al . 2009 , a&a , 500 , 1119 normandeau , m. , taylor , a. r. , & dewdney , p. e. 1997 , apjs , 108 , 279 park , b .- , & sung , h. 2002 , aj , 123 , 892 park , b .- g . , sung , h. , bessell , m. s. , & kang , y. h. 2000 , aj , 120 , 894 rebull , l. m. , hillenbrand , l. a. , strom , s. e. , et al . 2000 , aj , 119 , 3026 reed , b. c. 2003 , aj , 125 , 2531 rieke , g. h. , & lebofsky , m. j. 1985 , apj , 288 , 618 sakai , n. , sato , m. , motogi , k. , et al . 2013 , arxiv:1308.3004 salpeter , e. e. 1955 , apj , 121 , 161 sana , h. , rauw , g. , naz , y. , gosset , e. , & vreux , j .- 2006 , mnras , 372 , 661 sato , m. , hirota , t. , honma , m. , et al . 2008 , pasj , 60 , 975 scalo , j. m. 1986 , fundam . , 11 , 1 sharma , s. , pandey , a. k. , ojha , d. k. , et al . 2007 , mnras , 380 , 114 sharpless , s. 1955 , aj , 1955 , 60 , 178 siess , l. , dufour , e. , & forestini , m. 2000 , a&a , 358 , 593 skrutskie , m. f. , cutri , r. m. , stiening , r. , et al . 2006 , aj , 131 , 1163 sung , h. , & bessell , m. s. 1999 , mnras , 306 , 361 sung , h. , & bessell , m. s. 2004 , aj , 127 , 1014 sung , h. , & bessell , m. s. 2010 , aj , 140 , 2070 sung , h. , bessell , m. s. , lee , b .- w . , & lee , s .- g . 2002 , aj , 123 , 290 sung , h. , bessell , m. s. , & chun , m .- y . 2004 , aj , 128 , 1684 sung , h. , bessell , m. s. , chun , m .- y . , karimov , r. , & ibrahimov , m. 2008 , aj , 135 , 441 sung , h. , bessell , m. s. , & lee , s .- w . 1997 , aj , 114 , 2644 sung , h. , bessell , m. s. , & lee , s .- w . 1998 , aj , 115 , 734 sung , h. , chun , m .- y . , & bessell , m. s. 2000 , aj , 120 , 333 sung , h. , lim , b. , bessell , m. s. , et al . 2013 , jkas , 46 , 103 ( paper 0 ) sung , h. , sana , h. , & bessell , m. s. 2013 , aj , 145 , 37 sung , h. , stauffer , j. r. , & bessell , m. s. 2009 , aj , 138 , 1116 thronson , h. a. , thompson , r. i. , harvey , p. m. , rickard , l. j. , & tokunaga , a. t. 1980 , apj , 242 , 609 thompson , m. a. , white , g. j. , morgan . l. k. , et al . 2004 , a&a , 414 , 1017 valle j. p. , hughes , v. a. , & viner , m. r. 1979 , a&a , 80 , 186 westerhout , g. 1958 , bull . netherlands , 14 , 215 white , r. j. , ghez , a. m. , reid , i. n. , & schultz , g. 1999 , apj , 520 , 811 wilking , b. a. , harvey , p. m. , lada , c. j. , joy , m. , & doering , c. r. 1984 , apj , 279 , 291 xu , y. , reid , m. j. , zheng , x. w. , & menten , k. m. 2006 , science , 311 , 54

this paper is a continuation of a recent one in which , apparently for the first time , the existence of polaritons in ionic crystals was proven in a microscopic electrodynamic theory . this was obtained through an explicit computation of the dispersion curves . here the main further contribution consists in studying electric susceptibility , from which the spectrum can be inferred . we show how susceptibility is obtained by the green kubo methods of hamiltonian statistical mechanics , and give for it a concrete expression in terms of time correlation functions . as in the previous paper , here too we work in a completely classical framework , in which the electrodynamic forces acting on the charges are all taken into account , both the retarded forces and the radiation reaction ones . so , in order to apply the methods of statistical mechanics , the system has to be previously reduced to a hamiltonian one . this is made possible in virtue of two global properties of classical electrodynamics , namely , the wheeler feynman identity and the ewald resummation properties , the proofs of which were already given for ordered system . the second contribution consists in formulating the theory in a completely general way , so that in principle it applies also to disordered systems such as glasses , or liquids or gases , provided the two general properties mentioned above continue to hold . a first step in this direction is made here by providing a completely general proof of the wheeler feynman identity , which is shown to be the counterpart of a general causality property of classical electrodynamics . finally it is shown how a line spectrum can appear at all in classical systems , as a counterpart of suitable stability properties of the motions , with a broadening due to a coexistence of chaoticity . the relevance of some recent results of the theory of dynamical systems in this connection is also pointed out . 1.em * pacs * 78.20.-bh 41.20.q 05.45.a * keywords * susceptibility , wheeler feynman identity , ewald resummation , dynamical systems , order and chaos .

1407.6994

this paper complements and generalizes the results obtained in @xcite on the microscopic foundations of the optics of materials . the main new result of that paper was a proof of the existence of polaritons in ionic crystals , that was obtained by calculating the normal modes and exhibiting the explicit form of the dispersion curves . apparently , the existence of polaritons , whose qualitative importance is evident since it explains why crystals are transparent to visible light , was previously understood only at a phenomenological level , in terms of a macroscopic polarization field ( see for example @xcite , page 239 ) . an interesting point is that the microscopic proof was obtained in @xcite in a completely classical framework , on the basis of the system of newton s equations for each charge , in which the full electrodynamic forces are taken into account , both the mutual retarded ones and the individual radiation reaction forces . for example it is just retardation that makes it possible that the new polaritonic branches occur , and born and huang @xcite could nt get this result just because they did nt fully take the role of retardation into account . the result was obtained in @xcite by previously reducing the original electrodynamic model to a hamiltonian conservative one . this in turn was made possible by exploiting two global properties of the original microscopic electrodynamic system , namely , the wheeler - feynman identity @xcite and the ewald oseen resummation of the far fields ( see @xcite@xcite and @xcite , page 101 ) , which , jointly used , provide both a cancellation of the radiation reaction force acting on each charge , and an elimination of the problems related to delay . both properties were proven in @xcite ( following @xcite and @xcite ) , for the case of ionic crystals . it is then natural to ask whether such a result concerning the dispersion curves may be complemented by providing a microscopic expression for the electric susceptibility of the system , which would allow one to determine the expected absorption and emission spectra . moreover one might also look for an extension of the methods , formulating the theory in such a general frame that it can apply to disordered dielectric systems such as glasses , or even liquids or gases . in the present paper we show how a microscopic expression for susceptibility is obtained for ordered systems , and how the result can be extended , at least partly , to cover the case of disordered systems . indeed we will show how , if the two mentioned global properties hold ( so that the system can be reduced to a conservative hamiltonian one ) , then the statistical mechanical methods of green kubo type @xcite@xcite can be used to provide a microscopic expression for macroscopic polarization , and so for susceptibility . in particular , it will be explicitly exhibited that the phenomena of absorption and emission are not related , at least in a any direct way , to the radiation reaction force , and can in fact be understood as symmetrical features of a time reversible dynamics . in order to obtain such results , we have to overcome a difficulty which arises if one tries to imitate in a strict way the green kubo type methods generally used in the quantum case . indeed , the available procedure makes use , in an apparently essential way , of the gibbs measure in phase space , whereas gibbs measure does not even exist in the classical case , due to the divergence induced by the attractive coulomb potentials . we however show how susceptibility can actually be proven to exist , obtaining for it an expression in terms of time correlations . then we study its properties , and in particular deduce the @xmath0sum rule , the essentially classical character of which was already pointed out by van vleck and huber @xcite . the existence of susceptibility and its general expression are completely independent of the qualitative nature of the motions of the system . it is instead the form of the spectrum that depends on the stability properties of the motions . we show how a pure line spectrum occurs for stable ( almost periodic ) motions of the system , while a broadening of the lines or even a continuous spectrum occur when chaoticity sets in . we also discuss the relevance that in this connection have some quite recent results on the theory of dynamical systems , in particular the results that made possible to extend to systems of interest for statistical mechanics the methods of perturbation theory @xcite@xcite@xcite which allow one to estimate when a transition from ordered to chaotic motions should occur ( see the numerical works @xcite@xcite ) . for what concerns the extension to disordered systems , all depends on proving the two mentioned global electrodynamic properties . for the wheeler - feynman identity , we do here more than required , because we give a proof which applies to completely general systems , and not just to dielectrics . in fact , the identity is shown to be equivalent to a general form of causality of electrodynamics , which is reminiscent of a general property assumed in quantum electrodynamics . the properties related to the ewald resummation methods are instead assumed to hold for dielectrics , just by analogy with the case of ordered systems . in section [ 2 ] it is recalled how a first step in passing from microscopic to macroscopic electromagnetism consists in performing a local space average . our treatment is standard , apart from a minor point . in section [ 3 ] the second step is performed , which involves a phase space ( or ensemble ) average , and leads to a green kubo type formula for macroscopic susceptibility , in a completely symmetrical way for absorption and emission . the proof is obtained without using the gibbs measure . preliminarily , it is recalled how the reduction to a conservative hamiltonian system is obtained through the ewald resummation methods , making use of the wheeler feynman identity . in section [ 4 ] the analyticity properties of susceptibility are recalled , and the @xmath0sum rule is proven . in section [ 5 ] it is shown how under quite general conditions susceptibility is expressed in terms of equilibrium time correlation functions between positions and velocities of the charges . in section [ 6 ] it is discussed how the spectrum depends on the qualitative stability properties of the motions of the system , and in particular how a pure spectrum arises in the presence of suitable stability properties ( almost periodicity ) of the orbits . instead , a broadening of the lines , or even a continuous spectrum are expected to occur as chaoticity sets in . in section [ 7 ] this is illustrated by studying the particular case of ionic crystals . some final comments are added in section [ 8 ] . an appendix is devoted to a proof of the wheeler feynman identity ( and of the consequent cancellation of the radiation reaction forces ) , which applies in a completely general situation , irrespective of the ordered or disordered structure of the system . as we know , @xcite@xcite@xcite@xcite@xcite@xcite macroscopic electromagnetism is characterized by four fields : the electric field @xmath1 , the magnetic induction field @xmath2 , the electric induction field @xmath3 and the magnetic field @xmath4 . since the times of lorentz , the first two are thought of as local space averages of corresponding microscopic fields @xmath5 , @xmath6 , while the latter ones are defined as @xmath7 and @xmath8 , where the polarization vector @xmath9 and the magnetization vector @xmath10 are the response of a material body to the presence of an external electric or magnetic field . in the macroscopic treatments one assumes that there hold the constitutive relations @xmath11 and @xmath12 , or rather that analogous relations hold frequency by frequency , i.e. , that one has @xmath13 where @xmath14 , @xmath15 , @xmath16 and @xmath17 , are the time fourier transforms of the corresponding fields . in this section we recall how , in order to obtain a macroscopic expression for polarization , a first step is accomplished through a local space averaging procedure . this is a completely standard passage , and only a minor modification to the familiar procedure will be introduced . consider a dielectric body , thought of as microscopically constituted of a certain number @xmath18 of neutral molecules or atoms , each containing a stable aggregate of point charges . in such a case the microscopic maxwell equations read @xmath19 where @xmath20 is the position of the @xmath21the particle ( of charge @xmath22 ) in the @xmath23th molecule or atom . now , following lorentz , the macroscopic fields @xmath1 and @xmath2 are defined as local space averages of the values the microscopic fields take in what is sometime called a `` physically infinitesimal domain''@xcite , or a `` physically small volume element '' @xcite , of volume @xmath24 located about the considered point @xmath25 . think for example of a cubic volume element with side @xmath26 mstrong , which , in a solid or in a liquid , in ordinary conditions contains about one million molecules . due to the linearity of the maxwell equations , the space averaged fields are expected to be solutions of those same equations , having as sources the averaged charge and current densities . this becomes a rather simple theorem if the space averaging procedure at @xmath25 is mathematically implemented through a convolution with a suitable smooth ( @xmath27 class ) function @xmath28 centered at @xmath25 , which essentially vanishes outside the chosen volume element , while having inside it essentially a constant normalizing value , namely , @xmath29 . the macroscopic fields are thus defined as @xmath30 as the microscopic fields are distributions ( because @xmath31 functions occur in the sources ) , it turns out that the differential operators commute with the convolution , i.e. , one has @xmath32 exactly as it would occur if the fields were smooth . thus , multiplying the maxwell equations by @xmath33 and integrating , due to the linearity of the equations the macroscopic fields are found , as expected , to satisfy the maxwell equations with charge density @xmath34 and current density @xmath35 which now are smooth fields rather than distributions , and are obtained by averaging with the same procedure . so the macroscopic fields satisfy the equations @xmath36 which involve the space averaged sources @xmath37 @xmath38 we now show how the space averaged charge density @xmath34 can be written as the divergence of a field , which should be interpreted as a still microscopic form of the polarization field . this is obtained by expanding the positions of the charges entering the function @xmath39 , about the centers of mass of their molecules or atoms . this makes the single microscopic dipoles come in . denote by @xmath40 the position of the center of mass of the @xmath23th molecule or atom , as a fixed position of a cell , for example a given corner . ] and by @xmath41 the corresponding displacements ( which are assumed to be bounded ) of the charges . we have now to find which expression does the space averaged charge density @xmath34 take , as a function of the displacements @xmath42 . here the familiar procedure consists in introducing a multipole expansion and a truncation , through which @xmath34 is shown to be the divergence of a vector field . we obtain this result , perhaps in a simpler and more rigorous way , by making use of the finite increment lagrange formula , according to which for a smooth function @xmath0 one has @xmath43 indeed one then has @xmath44 thus , substituting this formula in the expression ( [ eq : carica ] ) for the space averaged charge density @xmath34 , and recalling that the molecules are neutral so that @xmath45 one finds @xmath46 where the field @xmath47 is given by @xmath48 without much error this can be written in the simplified form @xmath49 i.e. , as the sum of the dipole moments of the single molecules or atoms with respect to their centers of mass , as one might have expected . on the other hand we know that , in a dielectric , the macroscopic charge density is expressed as the divergence of polarization . so one might be tempted to altogether identify @xmath47 with the macroscopic polarization @xmath9 itself . this however is not correct . the reason is that the field @xmath50 still is a dynamical variable , by which we mean a function defined on the global `` mechanical phase space '' of the charges , a point of which , call it @xmath51 , is identified through the positions and the momenta of all charges . now , @xmath50 evidently depends on the phase point , and thus may be called the _ microscopic polarization field _ . the microscopic magnetization field could be given along similar lines . however we do nt need it for our aims , because with good approximation in dielectrics one can put @xmath52 , unless one is just interested in magneto optical phenomena . as usual in statistical mechanics , a macroscopic quantity is defined as the average over phase space of a microscopic quantity ( a function of @xmath51 ) , with respect to a given measure . denoting such an averaging in the mechanical phase space by @xmath53 , the macroscopic polarization field will then be defined by @xmath54 i.e. , by @xmath55 now , the microscopic polarization , being itself a space mean over many molecules , should already satisfy some central limit theorem and so should not fluctuate very much as the phase space point @xmath51 is varied . in such a case the ensemble average just provides a `` typical value '' , so that the use of a further ensemble average may appear to be redundant . this is not so , because it is just by performing ensemble averages that analytical manipulations can be performed which lead to significant results . one such result , as we will see , is the existence itself of electric susceptibility , namely , the fact that polarization responds linearly to an external perturbation even if the unperturbed system presents highly nonlinear motions . this is obtained by green - kubo methods in phase space , just because of the linearity of the equation of motion for the probability density . a further result is the proof of the @xmath0sum rule . however , it is not at all obvious how phase space methods can be used in a microscopic model which involves both retarded forces and dissipative ones . how to do this , and how to use hamiltonian techniques in phase space will be shown in the next section . the reduction of the original electrodynamic problem to a purely mechanical one in the mechanical phase space is quite hard a task . first of all , the original problem is different from those usually studied in statistical mechanics because , due to the finite propagation speed of the electromagnetic interactions among the charges , the equations of motion for the displacements @xmath42 of the charges turn out to be differential equations with delay . notice that the delay can not be neglected , as it produces qualitatively essential features . for example , in the case of ionic crystals it is just retardation that produces the two new branches of the dispersion relation which correspond to polaritons ( see formula ( 15 ) of ( @xcite ) , thus explaining why visible light can propagate inside them . thus , in the original electrodynamic problem , having to deal with equations with delay we know nothing about the properties of the corresponding dynamical system , not even how to correctly frame a cauchy problem . neither do we know which is the phase space suited to the system , nor can we know which measure should be used to define the averages . finally , the system is not a conservative one , at least not in any obvious way , inasmuch as the charges should radiate energy away during their necessarily accelerated motions . from a heuristic point of view such problems can be overcome in the following way . due to the long range character of the field produced by any single charge ( a range much longer than the purely coulomb one ) , in order to determine the force acting on any charge and produced by all the other ones , one necessarily has to perform a `` resummation '' of the forces . this can be done in an exact way in the case of crystals ( through the so called ewald method , as implemented for example in @xcite ) by suitably splitting the field into two contributions . the first one essentially comes from the near ( in a microscopic sense ) charges , and can thus be considered to all effects as being instantaneous , while the second one is essentially due to the far charges . in turn , the contribution of the far charges too can be divided into two parts . one of them exactly cancels the radiation reaction force ( which necessarily is nonvanishing , because of the accelerated motions of the charges ) . this indeed is the so called wheeler feynman identity , which was postulated by those authors in their paper of the year 1945 and was proven , in the case of ionic crystals , in @xcite , following @xcite and @xcite . the second part of the contribution of the far charges enters in the same way as an external electromagnetic field , which propagates inside matter with a suitable refractive index ( see the first term in the force entering formula ( 15 ) of ( @xcite ) , notwithstanding the fact that the microscopic far fields entering the original problem do propagate with the speed of light in vacuum ( this is the so called ewald oseen cancellation property ) . so we have to deal both with the wheeler feynman property ( or identity ) and with the ewald oseen resummation properties . in the case of ionic crystals both properties were proved to hold , so that the original electrodynamic equations of motion for the charges could be consistently dealt with as a system of non dissipative differential equations ( possibly depending on time ) , of the form @xmath56 where @xmath57 is a microscopic ( namely , much smaller than @xmath58 ) neighborhood of @xmath40 , while the field @xmath59 is what ewald calls the `` exciting '' electric field ( _ `` erregende feld '' _ in his words , see @xcite , page 7 ) . this is the field produced by the far charges that actually enters the equations of motion as if it were an external field , propagating with a macroscopic refractive index . analogous proofs should be provided here for the disordered case of interest for dielectrics . for what concerns the wheeler feynman identity , we here do more than required , because we give in an appendix a proof which holds in any situation , and actually shows the deep significance of the identity , as corresponding to some general form of causality . instead , the ewald oseen property is not proven here for the case of disordered systems , and its validity is assumed to hold by analogy with the case of crystals . we are confident that a proof may be provided on another occasion . so our phase space can be taken to be the usual one of statistical mechanics , namely , the space having as coordinates the positions @xmath42 and the momenta @xmath60 of all the charges of the system , and our aim is now to obtain an expression for the electric susceptibility following the standard methods of green kubo type of quantum statistical mechanics . here however a difficulty arises . indeed the analogous methods transported to the classical case amount to studying the liouville equation for the probability density in phase space , looking for its time evolution under the action of a perturbation . however , in the quantum case it is first of all assumed that an unperturbed ( or equilibrium ) solution exists , given exactly by the gibbs ensemble . now , if one looks at the procedures used in the proofs , one might have the impression that the role of the gibbs density is essential , and that the proof could nt be obtained without using it . on the other hand we have to deal with coulomb attractive interactions , which have the effect that the gibbs measure does not even exist , in the classical case . we show here how any reference to the equilibrium gibbs measure can be avoided , and even in a rather simple way . indeed in this section the existence of susceptibility is proven , and a quite general expression for it is provided , essentially without introducing any requirement at all on the equilibrium measure . then in section [ 5 ] it will be shown how susceptibility is expressed in terms of time correlation functions , if an assumption of a quite general character for the measure is introduced ( validity of the large deviation principle for momenta ) . so we only assume that an equilibrium probability density exists , which will be denoted by @xmath61 ( no confusion with the space averaged charge density should occur ) , and its form will not need be specified . in other terms , @xmath61 is only assumed to be invariant under the flow determined by the equations of motion , i , e . , to be a stationary solution of the continuity equation @xmath62 where @xmath63 is the vector field defined by the equations of motion in phase space for the isolated system . has vanishing divergence . nothing should change in the general case . ] consider now the case in which there is an external electromagnetic field @xmath64 ( for example a monochromatic wave of frequency @xmath65 ) which incides on the body , with an intensity that starts increasing slowly and then reaches a stationary value ( the so called case of an adiabatically switched on perturbation ) . then a change , say @xmath66 , will be induced on the ewald exciting field , which is the one actually entering the equations of motion for the charges . the change is due both to the presence itself of the incoming external field , and to the fact that the far charges which are responsible for that field are now moving in a modified way . for the sake of consistency , the relation between @xmath67 and the incoming external field @xmath64 should be determined , and to this end the validity of the lorentz lorenz relation should be established . this is in any case a necessary step , if macroscopic optics should be deduced at all . this problem will not be dealt with in the present paper . under the perturbation induced by the external field , the density @xmath34 will evolve according to the equation @xmath68 inasmuch as the equation of motion for @xmath42 contains the further force term @xmath69 . as @xmath67 is assumed to be a small perturbation , one can look for the solution as a series expansion @xmath70 and the first order term @xmath71 is immediately seen to satisfy the equation @xmath72 clearly the suited `` initial '' condition is the asymptotic one @xmath73 and the corresponding well known solution is then @xmath74 where @xmath75 is the flow relative to the _ unperturbed _ equations of motion , the macroscopic polarization @xmath76 can now be computed to first order , as the average of the microscopic polarization @xmath77 with respect to the density @xmath78 . assuming that @xmath9 vanishes at equilibrium ( absence of ferroelectricity ) , one remains with the contribution of @xmath71 only , which gives @xmath79 one has now to insert the expression ( [ eq : polarizzazione ] ) for the microscopic polarization @xmath77 . then , first of all one performs two elementary transformations ( namely , interchange of the integration orders of @xmath80 and @xmath51 , an change of variable @xmath81 taking into account that the modulus of the jacobian determinant of @xmath75 is unitary ) . moreover , one uses the fact that @xmath82 , being a macroscopic field , takes on essentially the same value @xmath83 at all points of the volume element @xmath58 . this eventually produces the result that macroscopic polarization depends linearly on the exciting field . so the macroscopic polarization can be written in the familiar form of linear response theory , namely as @xmath84 in terms of a dielectric response function @xmath85 , which is given by @xmath86 actually , in this expression for the response function we have introduced one more simplification . this consists in the fact that , when the expression ( [ eq : polarizzazione ] ) for the microscopic polarization @xmath47 is introduced into formula ( [ eq : polarizza ] ) , one has two sums over @xmath23 and @xmath87 , corresponding to two volume elements , whereas now the first sum was restricted to just the molecules that belong to the volume element entering the second sum . this amounts to presuming that the microscopic dynamics in two different macroscopic volume elements be totally uncorrelated . this point will be discusses later . we add now some comments . the first one concerns the fact that in the deduction of the formula for the dielectric response function no reference at all was made to nonconservative forces . indeed , it was explicitly assumed that in the equation of motion for each charge the radiation reaction force be canceled by a part of the retarded forces due to the `` far '' charges of the dielectric body . the first scientist who realized the occurring of this cancellation ( already in the year 1916 ) is the swedish physicist oseen @xcite . however , his result was ignored , having even been qualified as wrong ( _ `` irrig '' _ ( see @xcite , page 266 ) , as also was essentially ignored the work of wheeler and feynman , in which the same property was conjectured to hold quite in general . so we are dealing with a time reversible dynamical system . an asymmetry in the proof was however introduced above through the choice of the incoming external field @xmath88 ( which was adiabatically switched on ) , and through the corresponding choice ( [ incoming ] ) for the initial ( or rather , asymptotic in the past ) condition needed to solve the continuity equation for the probability density ( vanishing of @xmath71 as @xmath89 ) . these are the choices which are responsible for the fact that the formula just found corresponds to an absorption process . how an emission process can be analogously described in the present time reversible frame , will be shown in the next subsection . the second remark is that the proof shows how the existence of a linear response to the external field is quite independent of the nature of the unperturbed motions , which may have either an ordered or a disordered character . the linearity of the response is inherited from that of the liouville equation , under the only assumption that the higher order corrections ( beyond the first one ) to the equilibrium solution be negligible.this fact is characteristic of linear response theory , and so also occurs in its present classical formulation in phase space . the situation was quite different with the older approaches . in the oldest one , typically described in drude s book @xcite but still somehow surviving in the born wolf book @xcite , to each observed spectral line was associated the motion of a material oscillator , which was supposed to perform linear oscillations , forced by the inciding field . for example , in the words of kronig @xcite , in that approach one is dealing with _ an electric charge , elastically bound to an equilibrium position , having _ as he even adds _ a damping proportional to its velocity_. a different attitude was taken by van vleck @xcite who , working in the spirit of bohr s approach , thought it appropriate to formulated a theory of susceptibility by assuming that the unperturbed system performs quasi periodic motions . here , instead , essentially no property is required for the unperturbed motions . the proof of the existence of a linear response was given above in a way suited to describe an absorption process . however , the proof was formulated in the general frame of a time - reversible dynamics , in such a way that different types of nonequilibrium processes can be looked upon as determined by an asymmetry of the asymptotic conditions . so an emission process should be described by the same equations previously considered , just choosing a suitable asymptotic condition , and external field ( see @xcite ) . the suitable asymptotic condition can be inferred in the following way . recall how the absorption process was described . for @xmath90 we have a stationary state described by an equilibrium probability density @xmath61 , in the presence of a well defined exciting field @xmath59 . a perturbation is then introduced through a `` free '' field @xmath88 , incoming from infinity . during the process , one has a density @xmath91 and a corresponding exciting field @xmath92 , and one presumes that eventually , for @xmath93 , one will have a new equilibrium ( at a higher energy ) , with a density @xmath94 , together with a new exciting field @xmath95 and a new free field @xmath96 . moreover , one should have @xmath97 , as the whole incoming field is supposed to have been absorbed . let us now consider the inverse process , namely , the one which is obtained with the interchanges @xmath98 and @xmath99 ( the hamiltonian being assumed to be even in the momenta ) . so one starts up with a density @xmath100 at @xmath101 , and asymptotically when @xmath102 one gets a density @xmath61 , whereas the electric field is now the sum of the exciting field @xmath59 and of the free field @xmath64 . this means that the field @xmath64 was emitted from the body , in passing from the state @xmath100 to the state @xmath61 . mathematically , the process is still described through the perturbed continuity equation ( [ rho1 ] ) , provided the asymptotic condition @xmath103 be assumed . if , as in the case of the absorption process , we look for the solution in the form of a series , the first correction @xmath71 has to satisfy the same equation ( [ xxx ] ) as before , but now with the `` final '' condition @xmath104 so the solution now has the form @xmath105 and thus , in the same hypotheses as before , the final polarization can be written as @xmath106 with @xmath107 given exactly by the expression ( [ eq : suscettivita ] ) that occurs in the absorption process . susceptibilities are defined as responses to forcings of given frequencies , and thus are obtained from the formulas ( [ eq : rispostalineare ] ) and ( [ eq : polarizza2 ] ) if the latter are expressed in the form of convolutions , namely , with integrals over the whole real axis @xmath108 . thus we introduce the functions @xmath109 @xmath110 so that through the change of variables @xmath111 formulas ( [ eq : rispostalineare ] ) and ( [ eq : polarizza2 ] ) for the polarizations in an absorption or an emission process take the form @xmath112 @xmath113 namely , of convolutions between the change of exciting field and the function @xmath114 or @xmath115 respectively . now , as the fourier transform of a convolution is the product of the fourier transforms ( which we denote by a hat ) , the relations between polarization and exciting field can be written in the familiar form @xmath116 where @xmath117 as @xmath107 is odd ( see below ) , by the change of variable @xmath118 in the second integral one gets that @xmath119 is the complex conjugate of @xmath120 . so the emission and the absorption spectra coincide . to show that @xmath121 is an odd function , we notice that , from the definition , one has @xmath122 so that , performing into the integral the substitution @xmath99 , one finds @xmath123 ( indeed , as @xmath61 is even , its derivatives are odd , whereas , by changing sign to the momenta , @xmath124 goes into @xmath125 ) . it is well known that , as the function @xmath114 vanishes for @xmath126 , then its fourier transform enjoys two relevant properties : * it is analytic in the half plane @xmath127 ; * the kramers kronig relations hold @xmath128 from a conceptual point of view the kramers kronig relations are often interpreted as expressing the causality principle , the latter being meant in the sense that the affect ( here , polarization ) can not precede the cause ( the exciting field ) . on the other hand , analogous relations obviously hold also for the function @xmath129 , which clearly is not causal in that sense , as @xmath115 vanishes after the field is applied . a second remark concerns the poles of the two susceptibilities . since the original work of kramers , the emission was attributed to the presence of the radiation reaction force ( proportional to the time derivative of acceleration ) in the equations of motion . in such a way , however , in the expression for the susceptibility , calculated by considering a single damped and forced oscillator , there appeared a pole in the wrong half plane , and kramers himself had to patch the expression in some suitable way . instead , with the full electrodynamic treatment considered here , in virtue of the wheeler feynman cancellation the radiation reaction forces entering the original equations of motion eventually disappears , and the expressions of the susceptibilities have poles in the correct half plane . we finally come to the @xmath0sum rule . the reason of the name will be recalled in the next section . for the sake of concreteness we here concentrate on the case of the absorption susceptibility , because the formulas for the case of emission are simply obtained by passing to the conjugate complex . in order to have simpler notations , we also omit the superscript @xmath130 , i.e. , we let @xmath131 . the @xmath0sum rule states that @xmath132 so that it essentially relates the total absorption to the electron charge density . indeed one should take into account that for nuclei the ratio @xmath133 is negligible with respect to that of the electrons , so that the sum at the right hand side can be restricted to the electrons present in the considered volume . thus , denoting by @xmath134 and @xmath135 the charge and the mass of the electron , the r.h.s . just reduces to @xmath136 times the electron density ( number of electrons per unit volume ) . the next part of this section is devoted to a proof of the @xmath0sum rule ( [ eq : fsumrule ] ) . we start noting that for a smooth functions @xmath137 one has @xmath138 indeed , on the one hand the fourier transform of @xmath139 is given by @xmath140 , as one immediately checks by an integration by parts . on the other hand the inversion theorem for the fourier transform gives @xmath141 so the thesis should follow by simply taking @xmath142 . however , in our case @xmath143 presents a discontinuity of first type at @xmath142 , as it vanishes for @xmath144 , while being equal to @xmath145 for @xmath126 . now , the inversion theorem tells us that at a discontinuity points the integral equals the semi sum of the right and the left limits , so that eventually one has @xmath146 however , as is easily checked , ] so that , changing @xmath65 into @xmath147 , the value of the integral does not change . ] @xmath148 is an even function of @xmath65 , so that one has @xmath149 now it turns out that @xmath150 can be evaluated exactly and , as will be seen in a moment , one has @xmath151 which indeed proves the @xmath0sum rule ( [ eq : fsumrule ] ) . in order to show the latter relation , we differentiate the expression ( [ eq : suscettivita ] ) for @xmath121 . exchanging derivative and integral one gets @xmath152 where in the second line use was made of @xmath153 . now there just remains to integrate by parts . the boundary term vanishes ( due to the vanishing of the probability for a particle to have an infinite momentum ) , so that @xmath154 inasmuch as @xmath155 , whereas the density @xmath61 is assumed to be normalized to @xmath156 . after the detour on the analyticity properties of the dielectric response functions and susceptibilities , which were based on the general expression ( [ eq : suscettivita ] ) , we show here how more transparent expressions are obtained if a further property of a quite general character is introduced for the equilibrium density @xmath61 . the point is that formula ( [ eq : suscettivita ] ) involves sums of integrals of the type @xmath157 the computation of which requires to have available a definite expression for the derivative of @xmath61 with respect to @xmath158 . now , if we were allowed to take for @xmath159 the gibbs form , the above quantity would be proportional to @xmath160 . on the other hand , essentially the same result is guaranteed under much milder conditions , essentially under conditions which allow for a large deviation principle to hold with respect to the momenta only , irrespective of the positions ( which , through the attractive coulomb potential , introduce divergences in the classical form of gibbs measure ) . indeed this allows one to get @xmath161 where the constant @xmath162 is nothing but the mean square deviation of momentum , which would just reduce to temperature if the density were the gibbs one . for the large deviation argument one can see the classical book of khinchin @xcite . so we have @xmath163 namely , the integrals ( [ integrale ] ) are just equilibrium time correlations between position and momentum of each charge . this fact , by the way , makes reasonable a property that was assumed in the last part of section [ 3 ] , when passing from ( [ eq : polarizza ] ) to ( [ eq : polarizza2 ] ) . namely , the property that the integrals ( [ integrale ] ) should present a fast decay with respect to spatial separation of the charges , i.e. , that one should have @xmath164 if the molecules @xmath40 e @xmath165 belong to different volume elements . in conclusion , the expression ( [ eq : suscettivita ] ) for the dielectric response function can be rewritten in the form @xmath166 which involves equilibrium time correlations of momenta and positions of the charges . now there remains the problem that we have to compute phase averages with respect to the equilibrium probability density @xmath61 , the form of which is still essentially undetermined . a great step forward is accomplished by making use of a general principle of statistical mechanics according to which , under extremely mild conditions , the phase space equilibrium averages can be computed as corresponding time averages ( see for example @xcite , page 63 ) . so we estimate the required phase space integral as time averages , i.e. as @xmath167 here we show how it can at all happen that a conservative hamiltonian system ( to which our original electrodynamic system has been reduced ) presents a line spectrum . this depends of the qualitative properties of the dynamical orbits ( or motions ) of the system , because it turns out that a discrete spectrum occurs if the motion of the representative point in phase space is , informally speaking , `` non chaotic '' . indeed in dynamical systems theory the property of presenting a continuous spectrum is sometimes even assumed to be the characteristic property for an orbit to be chaotic . more precisely , one certainly has a pure line spectrum if the motion is assumed to be `` almost periodic '' in the sense introduced in the year 1924 by harald bohr , the brother of niels bohr . almost periodicity can be defined in several equivalent ways . however , the following characteristic property ( which thus can be taken as a definition ) , is more significant for our purposes : if an orbit , say the motion @xmath168 of a particle , is almost periodic , then it can be represented by a generalized fourier expansion @xmath169\ ] ] where the sequence @xmath170 of _ positive _ frequencies is determined in the following way . having defined the functions @xmath171 and @xmath172 by @xmath173 @xmath174 then these functions turn out to vanish for all frequencies but for a discrete set of frequencies @xmath170 . this determines the frequencies . then , the coefficients of the expansion simply are the values of the expansion simply are the values of the the above functions at @xmath175 , i.e. , one has @xmath176 corresponding to the expansion ( [ eq : sviluppoq ] ) for the position as a function of time , one also has an analogous expansion for the momenta , namely , @xmath177 which is obviously obtained by differentiating with respect to time the expansion for @xmath168 . one thus obtains @xmath178 \ . \end{split}\ ] ] this relation is obtained by evaluating the integrals through the familiar prosthaphaeresis formulas , recalling that the time average of any non constant trigonometric function vanishes . the result is the following one . defining @xmath179 one finds that all the @xmath180 s vanish for @xmath181 , whereas for @xmath182 one has @xmath183 now , substitute into formula ( [ eq : suscettivita2 ] ) the expression ( [ eq : corrpq ] ) just found for the correlations . remarking that , due to the antisymmetry with respect to the interchange @xmath184 of the terms occurring in the sum . one has @xmath185 one obtains @xmath186 in order to find the susceptibility there just remains to compute the fourier transform of @xmath121 . a not difficult computation shows that one has @xmath187 thus , defining @xmath188\ , \ ] ] for the real and the imaginary parts of susceptibility one finds the expressions @xmath189 due to the delta functions appearing in the imaginary part of susceptibility , formula ( [ orchestra ] ) shows that the spectrum of a macroscopic dielectric body performing almost periodic motions presents infinitely tight absorption lines , in correspondence of the frequencies @xmath175 . this is the way in which , in the spectrum of a macroscopic dielectric body , `` lines '' show up without necessarily having to make reference to energy levels of the single molecule or atom . this result is exactly the property of a spectrum which , before the advent of quantum mechanics , ( starting from lorentz @xcite and drude up to kronig @xcite and even born and wolf @xcite ) , was interpreted in microscopic terms by thinking that each line should be attributed to the motion of a material harmonic `` resonator '' , of exactly that frequency . analogously the molecules were thought of as constituted of charges with mutual elastic bonds . so there would exist corresponding normal modes , which could be equivalently described as harmonic oscillators with characteristic frequencies @xmath175 ( which were introduced from outside , in correspondence with the observed ones ) . however , as the lines are infinite in number , one was meeting with the absurd situation that any atom or molecule had to be composed of an infinite number of oscillating charges for this reason such oscillators were denoted as `` virtual '' i.e. , as somehow non physical ( see @xcite ) , and each of them was weighted with a suitable weight ( usually called `` force '' ) @xmath190 . in the year 1925 the `` @xmath0sum rule '' was empirically discovered , according to which the `` forces '' of the virtual oscillators were not arbitrary , but had to satisfy the rule @xmath191 namely , the sum of the `` forces '' of the oscillators just equals the number of electrons per atom or per molecule , times the factor @xmath192 ( indeed , as already explained , the contribution of the nuclei is negligible ) . one of the big triumphs of quantum mechanics was to `` explain '' the @xmath0sum rule in terms of the quantum commutation rules . on the other hand , such a rule holds in the classical case too . indeed an explicit computation gives @xmath193 which , using the general formula ( [ eq : fsumrule ] ) , actually gives the @xmath0sum rule ( [ eq : fsumrulequan ] ) . so , a pure line spectrum occurs for stable ( almost periodic ) motions , whereas a broadening of the lines and even a continuous spectrum are expected to occur when chaoticity of the motions sets in . this connection between optical properties of the system and qualitative properties ( order or chaos , or their coexistence ) of the corresponding orbits can be illustrated in a particularly clear way in the case of ionic crystals . if one is interested in the infrared spectrum , in the expression ( [ eq : suscettivita2 ] ) for the dielectric response function it is sufficient to limit oneself to the motions of the ions only . in such a case it is convenient to choose as a reference point @xmath40 ( with respect to which the displacements @xmath42 are computed ) , an arbitrary fixed point inside each cell of the lattice . in such a way the index @xmath23 is now labeling also the cells . following @xcite one can pass to the normal modes of the lattice , which we here denote by @xmath194 and are defined by @xmath195 here , the integration is performed over the brillouin zone @xmath196 , the vectors @xmath197 are the eigenvectors of the dynamical matrix of the crystal , while the vector @xmath198 specifies the equilibrium position of the @xmath21th atom inside the cell @xmath23 . the index @xmath199 is now a label for the different branches of the dispersion relation . ( @xmath18 being the number of ions in the fundamental cell ) , instead , when the interaction with the field of the far ions is taken into account , the number of branches can vary , and polaritonic branches can appear . ] so , one gets the relation @xmath200 because , in summing over a volume element , one has @xmath201 thus , in the case of a ionic crystal the dielectric response function for the ions can be written as @xmath202 so that the relevant quantities now are the time correlations of the modes @xmath203 . if the harmonic approximation , each mode performs a periodic motion with frequency @xmath204 , so that one has @xmath205 being @xmath206 the kronecker s delta , and one ends up with a formula of the type ( [ orchestra ] ) , now however with only a finite number of terms , each corresponding to a branch of the dispersion relation ( omitting the `` acoustic '' branches , for which it is @xmath207 ) . the response function @xmath208 versus time . solid line refers to the system at low temperature , while broken line refers to the system at high temperature.,scaledwidth=80.0% ] on the other hand , if the nonlinear terms are taken into account the motion is no more integrable , and the previous analysis has to be changed . in the case of a `` small '' nonlinearity , the behavior of the correlations over some ( large ) time scale does not change with respect to the unperturbed ( i.e. , linear ) case , whereas over a larger time scale the correlations are expected to decay to zero , so that one should have @xmath209 in conclusion , passing to the fourier transform , one can presume that in the case of a small nonlinearity one should get @xmath210 i.e. , the classical expression of lorentz and drude @xcite@xcite , that such authors interpreted in terms of motions of material damped `` resonators '' . thus the line broadening corresponds to a decay of the time correlations which is induced by the nonlinearity and the presumably associated chaoticity ( or rather partial chaoticity ) of the motions . here no damping is active , neither the linear one which was heuristically introduced by lorentz and drude , nor that of the radiation reaction , which was always taken into consideration by van vleck , planck and many others . indeed the radiation reaction , although being actually present in the original full electrodynamic model , turns out to be eliminated by the electrodynamic action of the far charges , through the wheeler feynman mechanism . so much for the case of a small nonlinearity , i.e. , for the case of what may called the `` perturbation regime '' ( with respect to the linear one ) . instead , in the case of a large nonlinearity the motion is expected to be completely chaotic , displaying time correlations completely different from those of the linear case . in particular the spectrum should be now a continuous one , with no peaks anymore . plot of @xmath211 versus @xmath65 for two different temperatures . circles refer to the system at a high temperature ( no peak ) , while the triangles , which exhibit a peak for @xmath212 , refer to the system at a low temperature . in the inset , which concerns the system at low temperature , the plot of @xmath211 is reported for @xmath65 near @xmath213 , together with the best fit lorentzian curve ( solid line ) . here , @xmath213 is the frequency of the optical branch.,scaledwidth=75.0% ] on the other hand , when in statistical mechanics one makes reference to the qualitative properties of the motions with respect to order ( stability ) or chaoticity type , it is usually presumed that in the thermodynamic limit the motions should always be chaotic . this has the consequence that , in our case , which is concerned with macroscopic dielectric systems dealt with in a classical frame , one would always meet with a continuous spectrum . now , in the domain of the theory of classical dynamical systems , particularly in connection with the so called fermi pasta ulam problem , a long debate is going on about this point , and the results of numerical computations appeared to be not yet conclusive . however , rather recently it was analytically proven @xcite that in the perturbation regimes significant stability properties do persist in the thermodynamic limit , and indeed in a form suited for applications to statistical mechanics . in particular , in the works @xcite and @xcite it was proven that in the fpu and in related models the normal mode energies remain correlated for long times also in the thermodynamic limit ( see also the numerical work @xcite , or the work @xcite concerning plasmas ) . thus one can conclude that the conjecture that macroscopic systems should perform chaotic motions is , at least , not always appropriate , and should be checked in any particular case . just in order to give an example which should exhibit in a qualitative way the features described above , we report here the results of a numerical computation performed on the classical one dimensional alternating masses model ( with 1024 paticles ) , introduced already in the year 1912 by born and von krman . through a numerical simulation of the dynamics we computed the response function @xmath214 , defined by ( [ eq : suscettivita2 ] ) with the sum extended over all particles of the crystal , and then the corresponding spectrum . we considered two cases relative to a low temperature and to a larger one . the response functions for the two temperatures are reported in fig . [ fig : due ] , whereas the corresponding spectra ( computed as the discrete fourier transforms ) are reported in fig . [ fig : tre ] . in the case of low temperature the response function presents a well distinct profile , apparently not very dissimilar from a periodic one . however a decay occurs at much longer times , as witnessed by the broadened form of the spectrum ( shown in the inset of fig . [ fig : tre ] ) . further results not reported here show that with increasing temperature the broadening , and a shift too , become larger and larger . finally , at some high temperature , the results reported in the figures show that the response function presents a decay at a short time , and the corresponding spectrum is essentially a continuum . for an analogous phenomenon occurring in a model of interest for plasma confinement , in which a transition from an ordered to a chaotic motion is witnessed by the form of the spectrum , see @xcite . we leave for a future work the numerical study of the spectrum for a realistic three dimensional model of a ionic crystal involving the microscopic electrodynamic forces , already considered in @xcite in connection with the dispersion curves . so we have complemented the result obtained in @xcite , by showing how electric susceptibility can be consistently proven to exist for a dielectric macroscopic body , in a classical microscopic theory in which the full electrodynamic interactions among the charges are taken into account . preliminarily we had to make essential use of two global properties of the electrodynamic interactions , i.e. , the wheeler feynman identity and the ewald oseen resummation properties . the former was proved here for a general system in the thermodynamic limit , whereas the latter were proven in @xcite for crystals , their proof for a general system being still lacking . thus our result is at the moment proven only for crystals , although we are firmly convinced that it can be extended to cover the case of a generic dielectric body . on the basis of such global electrodynamic properties , the dynamical system can be dealt with as if it were a hamiltonian one , and in particular the radiation reaction forces are completely eliminated , so that absorption and emission appear as completely symmetrical phenomena of a time reversal invariant system . susceptibility turns out to be expressed in terms of the time correlation functions of positions and velocities of the charges , calculated for motions of the system at equilibrium , i.e. , in the absence of an external field . notice however that the system still contains a trace of the electrodynamic interactions , because the equations of motion of the charges , that have to be solved in order that the time correlation functions may be computed , still contain the force of the `` exciting field '' , namely , the field originated by the far charges , that propagates in the body as an external field , having however the correct refractive index . having reduced the original electrodynamic system to a hamiltonian one , susceptibility was proven to exist through methods of green kubo type . however , this required to overcome the difficulties of working in the absence of a gibbs measure , which does not exist for systems with attractive coulomb interactions , for what concerns the spectrum , which is the same for absorption and for emission , we have illustrated how it reflects the stability properties of the unperturbed equilibrium motions of the system . for stable ( almost periodic ) motions , as occurs with a crystal in the linear approximation , one has a purely line spectrum . so , the susceptibility presents the standard form that , since the first work of lorentz of the year 1872 , was explained by thinking of the system as if it were composed of single linear material oscillators with proper frequencies equal to the observed ones ( see the booklet @xcite of pauli ) . when chaoticity sets in , as occurs in a crystal in the presence of nonlinearities , one might conjecture that the motions be completely chaotic , so that the lines completely disappear , and a continuous spectrum occurs . we have however pointed out that the most recent analytical result appear to support the conjecture that , at least in the case of crystals , partially ordered motions persist in the thermodynamic limit ( i.e. , for a macroscopic system ) . thus the time correlations in general should decay only after a sufficiently long time , with the consequence that the lines are in general broadened . in such a case the spectrum has the form that would occur if the system were composed of single linear material oscillators with the observed frequencies , having in addition suitable linear dissipative forces . however , no physical dissipative force is actually present in our system , because , in virtue of the wheeler feynman identity , the radiation reaction forces are canceled by the electrodynamic forces due to the far charges . so , the decay of correlations occurs in the familiar dynamical way which characterizes autonomous hamiltonian systems that are ( at least partially ) chaotic , and has nothing to do with the radiation reaction force , to which for example planck , van vleck and many others were thinking . correspondingly , the poles of susceptibility in the complex frequency plane quite naturally do lie in the correct half plane . in any case , while in the theory of dynamical systems the presence of a continuous or partly continuous spectrum is sometimes used as a tool to qualify the ordered or chaotic character of motions , here the situation is reversed , and it is the spectrum itself , in its original physical optical connotation , that is a pure line spectrum in the case of ordered motions , while presenting broadened lines or a fully continuous aspect in the case of partly or fully chaotic motions . the wheeler feynman identity deals with the classical problem of the solutions of the inhomogeneous wave equation @xmath215 for the four potential @xmath216 , with a given four current @xmath217 , and states that , possibly under suitable conditions , the advanced potential coincides with the retarded one , or more precisely , in terms of their difference which is a regular function , that one has @xmath218 clearly this in not true for an arbitrary current , and the authors , on the basis of four arguments , advanced the conjecture that the identity should hold if the problem is considered as a global one involving , as they said , all charges `` of the universe '' . a much more innocuous setting in which the problem can be framed , is the standard one of statistical mechanics , where reference is made to the `` thermodynamic limit '' . so we consider the microscopic current inside a domain of volume @xmath219 ( i.e. , the `` truncated '' function @xmath220 which coincides with @xmath21 inside the domain and vanishes outside ) , and take the limit in which both the volume and the number of elementary charges constituting the current tend to infinity , in such a way that the charge density ( number of charges per unit volume ) remains constant . such a framing of the problem is immediately reflected in a deep mathematical property of the current , because for the current density one clearly has to give up any property of decrease at infinity , and one should assume for example only the property @xmath221 , i.e. , that the density @xmath222 be only locally integrable . as a possible substitute for the global integrability condition there is one that quite naturally comes to one s mind for its physical significance . moreover , it is somehow analogous to what is sometimes called the locality condition of quantum field theory , although it rather appears to express a kind of causality condition . precisely , we start up defining the autocorrelation of the current density @xmath223 by @xmath224 where the symbol @xmath219 denotes both the space region of integration and its ( lebesgue ) measure . it is implicitly assumed that the average of @xmath222 over the whole space time vanishes . in other terms we are assuming that there exists no correlation between space separated points of space time . this requirement is natural from the physical point of view , because one should assume that the interactions can not propagate faster than light , so that it seems natural to assume that space separated events be independent . [ teo : main ] consider the wave equation @xmath230 having as source a current @xmath225 satisfying the causality condition [ hyp:1 ] . let @xmath231 and @xmath232 be the retarded and the advanced solutions respectively . then for all @xmath228 one has @xmath233 to prove the theorem , let us start defining by @xmath234 , the `` truncated '' current , i.e. the function coinciding with @xmath225 inside @xmath219 , and vanishing outside it . the wave equation ( [ eq : onde ] ) can be written in fourier space ( with respect to the spatial coordinates ) as @xmath235 where @xmath236 , whereas @xmath237 is the space fourier transform of the truncated current . the retarded and advanced solutions are then given by @xmath238 so that one gets @xmath239 where @xmath240 is the fourier transform , with respect to time , of @xmath237 . now one uses the plancherel theorem , which states @xmath241 to get ( use @xmath242 ) @xmath243 where @xmath244 is the surface element on the unit sphere in the @xmath245 space . we now use the causal property of the current . in fact one has the following theorem , which will be proven below : [ teo : due ] if @xmath225 is a causal current in the sense of definition [ hyp:1 ] , then one has @xmath246 on each circular cone @xmath247 , where @xmath244 is the surface element on the unit sphere in the @xmath245 space , while @xmath248 runs along the cone generatrix . as a comment , one may add that from ( [ yyy ] ) it is rather easily seen that the validity almost everywhere of the wheeler feynman identity implies the vanishing of the `` spectrum of the current '' , i.e. of the limit of @xmath250 , on almost the whole light cone @xmath251 . so , the problem of proving the wheeler feynman identity is reduced to proving formula ( [ eq : ccm ] ) of theorem [ teo : due ] . to this end , we start defining the function @xmath252 which , apart from the factor @xmath253 , is nothing but the correlation of the truncated current , integrated over @xmath80 , as one would naturally do in defining correlations for functions having domain in space time . one then immediately sees that : indeed the first property is just a translation of the fact that @xmath234 is causal , i.e. , that ( [ eq:2 ] ) holds , whereas the second one is nothing but the `` faltung '' theorem on the fourier transform of a convolution . now , considering the spherical mean of the spectrum @xmath259 , one gets @xmath260 where @xmath261 is the spherical mean of @xmath256 . now , if one makes use of the of parity property of the correlation @xmath262 , which easily follows from the very definition , one finds that the spherical mean @xmath261 is an even function of time , so that the imaginary part of the integral in the last line of vanishes , and one gets @xmath263 \ . \end{split}\ ] ] consider now `` a ray '' in the momentum @xmath264 space , i.e. all vectors of the form @xmath265 , @xmath266 , and integrate relation along this ray : one gets @xmath267 \ . \end{split}\ ] ] now using the relation @xmath268 one gets @xmath269 where the domain @xmath270 ( depicted in figure [ fig : uno ] ) is the domain in the half plane @xmath271 , bounded by the two half lines @xmath272 . now , dividing by @xmath219 and taking the limit , the integral is seen to vanish if @xmath273 . in fact , by the causality property ( [ eq : cck ] ) , in that limit @xmath274 vanishes for all points inside the region bounded by the lines @xmath275 , i.e. , in particular , for all points of @xmath270 . so ( [ eq : ccm ] ) holds and theorem [ teo : main ] is proven . in their paper @xcite , wheeler and feynman showed how the condition @xmath276 implies the vanishing of the radiation reaction force ( or self force ) acting on each charge . one starts from the relativistic equation of motion for the charge @xmath277 where @xmath135 and @xmath134 are the charge and the mass of the particle , dots represent derivatives with respect to proper time , repeated index means summation ( einstein convention ) , @xmath278 is a four force of mechanical ( non electromagnetic ) type , while @xmath279 is the retarded electromagnetic field due to all other charges , evaluated at the four position @xmath280 of the considered charge , and finally the term @xmath281 is the relativistic expression for the radiation reaction force . the electromagnetic field @xmath279 , or rather the field @xmath282 , is defined as @xmath283 where @xmath284 is the retarded field produced by the @xmath23th charge , and the summation is extended over all charges but the considered one . the field @xmath285 can be rewritten in a more useful form as @xmath286 because , as we will show below , the wheeler feynman identity implies that @xmath287 so that the equations of motion , at the end , can be written as @xmath288 with @xmath289 the new form of the equations of motion clearly shows that they are indeed reversible and the radiation reaction has disappeared . so , such a force force can not be held responsible for the emission . to show how relation ( [ eq : canc ] ) follows from the wheeler feynman identity , one first has to notice that such an identity states that one has @xmath290 where the sum is extended to all charges . thus , at all points @xmath291 ( i.e. , at all points different from the four position of the considered charge ) one has @xmath292 because the vanishing of the potentials implies the vanishing of their derivatives . now , it was shown by dirac ( see @xcite ) that for the field @xmath293 created by the particle @xmath280 itself one has @xmath294 while on he other hand the remaining fields are regular at @xmath280 . so taking the limit of the previous relation ( [ eq : zerotot ] ) for @xmath295 , one gets ( [ eq : canc ] ) . a. lerose , a. sanzeni , a. carati , l. galgani , _ eur . d * 68 * ( 2014 ) 35 . g. grosso , g. pastori parravicini , _ solid state physics _ , ( academic press , san diego and london , 2000 ) . m. born , k. huang , _ dynamical theory of crystal lattices _ , ( oxford u.p . , oxford , 1951 ) . wheeler , r.p . feynman , rev . * 17 * , ( 1945 ) 157 . ewald , ann . * 49 * ( 1916 ) 1 . w. oseen , ann . * 48 * ( 1915 ) 1 . m. born , e. wolf , _ principles of optics _ , ( pergamon press , oxford , 1959 ) . a. carati , l. galgani , nuovo cim . b * 118 * , ( 2003 ) 839 . m. marino , a. carati , l. galgani , ann . phys . * 322 * , ( 2007 ) 799 . green , j. chem . * 22 * ( 1954 ) 398 . * 12 * ( 1957 ) 570 . van vleck , d.l . huber , rev . * 49 * ( 1977 ) 939 . a. carati , j. stat . phys . * 128 * ( 2007 ) 1057 . a. carati , a. maiocchi , comm . * 314 * ( 2012 ) 129 . a. maiocchi , d. bambusi , a. carati , j. stat . phys . * 155 * ( 2014 ) 300 . a. carati , l. galgani , a , giogilli , s. paleari . phys . e * 76 * ( 2007 ) 022104 . a. carati , f. benfenati , a. maiocchi , m. zuin , l. galgani , chaos * 24 * ( 2014 ) 013118 . p. drude , _ the theory of optics _ ( dover , new york , 1950 ) ; _ lehrbuch der optik _ ( leipzig , 1900 ) . lorentz , _ the theory of electrons _ , ( dover , new york , 1952 ) ; first edition 1906 . m. born , _ optik _ , ( springer , berlin , 1933 ) . van vleck _ the theory of electric and magnetic susceptibilities _ , ( oxford u.p . , oxford , 1932 ) . de groot , _ the maxwell equations _ , ( north holland , amsterdam , 1969 ) . de groot , l.g , suttorp , _ foundations of electrodynamics _ , ( north holland , amsterdam , 1972 ) . kirkwood , j. chem . * 4 * ( 1936 ) 592 . oseen , physik . z. * 17 * ( 1916 ) 341 . g.jaff , _ dispersion und absorption_. in handbuch der experimentalphysik ( akademische verlaggesellshaft , leipzig , 1928 ) . kronig , j. opt . * 12 * ( 1926 ) 57 . van vleck , phys . * 24 * ( 1924 ) 330 . callen , t.a . welton , phys . rev . * 83 * ( 1953 ) 34 . dirac , proc . royal soc . ( london ) a * 167 * , ( 1938 ) 148 . khinchin , _ mathematical foundations of statistical mechanics _ , ( dover , new york , 1949 ) . nemytskii , v.v . stepanov , _ qualitative theory of differential equations _ , ( dover , new york , 1989 ) . besicovitch , _ almost periodic functions _ , ( dover , new york , 1954 ) . n. bohr , h.a . kramers , j.c . slater , phil . mag . * 47 * ( 1924 ) 785 , reprinted in b.l . van der waerden _ sources of quantum mechanics _ ( dover , new york , 1967 ) . e. nelson , _ quantum fluctuations _ ( princeton u.p . , princeton , 1985 ) . w. pauli , optics and the theory of electrons , ( dover , new york , 1973 ) .

interfaces play an essential role in phonon - mediated heat conduction in solids , impacting applications ranging from thermoelectric waste heat recovery to heat dissipation in electronics . from the microscopic perspective , interfacial phonon transport is described by transmission coefficients that link vibrational modes in the materials composing the interface . however , direct experimental determination of these coefficients is challenging because most experiments provide a mode - averaged interface conductance that obscures the microscopic detail . here , we report a metrology to extract thermal phonon transmission coefficients at solid interfaces using ab - initio phonon transport modeling and a thermal characterization technique , time - domain thermoreflectance . in combination with transmission electron microscopy characterization of the interface , our approach allows us to link the atomic structure of an interface to the spectral content of the heat crossing it . our work provides a useful perspective on the microscopic processes governing interfacial heat conduction .

1612.08401

interfaces play an essential role in phonon - mediated heat conduction in solids.@xcite material discontinuities lead to thermal phonon reflections that are manifested on a macroscopic scale as a thermal boundary resistance ( tbr ) , also called kapitza resistance , @xmath0 , that relates the temperature drop at the interface to the heat flux flowing across it . tbr exists at the interface between any dissimilar materials due to differences in phonon states on each side of the interface.@xcite typical interfaces often possess defects or roughness which can lead to additional phonon reflections and hence higher tbr . tbr plays an increasingly important role in applications , particularly as device sizes decrease below the intrinsic mean free paths ( mfps ) of thermal phonons.@xcite at sufficiently small length scales , tbr can dominate the total thermal resistance . for instance , the effective thermal conductivity of a superlattice can be orders of magnitude smaller than that of the constituent materials due to tbr.@xcite this physical effect has been used to realize thermoelectrics with high efficiency@xcite and dense solids with exceptionally low thermal conductivity@xcite . on the other hand , tbr can lead to significant thermal management problems@xcite in applications such as leds@xcite and high power electronics@xcite . numerous works over several decades have investigated the microscopic origin of tbr at solid - solid interfaces , starting with studies performed at low temperatures ( @xmath1 k ) , in which heat is carried predominantly by phonons with frequencies less than 1 thz.@xcite at these low temperatures and for pristine , ordered interfaces , transmission coefficients can be obtained from continuum elastic theory in an analogy with snell s law for light ; this model is known as the acoustic mismatch model ( amm).@xcite the amm was shown to explain the experimentally measured values of tbr at various solid - solid interfaces.@xcite at higher temperatures ( above 1 k ) , heat transport across the interfaces was found to be sensitive to surface roughness . for the limit of completely diffuse scattering in which transmitted and reflected phonons can not be distinguished , swartz constructed the diffuse mismatch model ( dmm).@xcite despite the success of these models at explaining tbr at low temperatures , they generally fail at temperatures larger than 40 k and are unable to account for the atomistic structure of the interface . recent works have focused on remedying these deficiencies . optical methods enable the routine measurement of tbr over a wide range of temperatures for various metal - dielectric interfaces @xcite as well as at interfaces with variable bonding strength . @xcite other works have examined the temperature dependence of the thermal conductivity@xcite in nanocrystalline samples . computational atomistic methods such as molecular dynamics@xcite and atomistic green s functions@xcite have been extensively applied to obtain the transmission coefficients at interfaces with realistic atomic structure . these calculations generally predict the coefficients to decrease with increasing phonon frequency due to reflections of short wavelength phonons by atomistic roughness , a trend that is supported by basic wave physics and indirectly by experiment.@xcite however , a direct determination of the spectral transmission coefficients at an actual interface has not yet been reported . here , we report a metrology to extract the thermal phonon transmission coefficients at a solid interface . our approach , based on combining experimental observations with ab - initio phonon transport modeling , exploits quasiballistic transport near the interface to significantly narrow the possible transmission coefficient profiles at a solid interface compared to the bounds obtained from traditional approaches . applying our approach in conjunction with transmission electron microscopy ( tem ) , we are able to directly link atomic structure to the spectral content of heat crossing the interface . our approach is a useful tool to elucidate the microscopic transport properties of thermal phonons at solid interfaces . our approach is based on interpreting data from the tdtr experiment with ab - initio phonon transport model . briefly , tdtr is a widely used optical pump - probe technique that is used to characterize thermal properties . in this experiment , a sample consists of a metal transducer film on a substrate . a pulsed laser beam from an ultrafast oscillator is split into a pump and a probe beam . the pump pulse train is modulated at a frequency from 1 to 15 mhz to enable lock - in detection , and is then used to impulsively heat the metal film coated on the sample . the transient temperature decay @xmath2 at the surface is detected as a change in optical reflectance by the probe beam.@xcite in the traditional tdtr approach , this transient signal is related to the desired thermal properties by a macroscopic transfer function based on a multilayer heat diffusion model.@xcite this function maps thermal properties such as substrate thermal conductivity and metal - substrate interface conductance to the tdtr signal , and thus these properties are obtained by varying these parameters until the simulated results match the measured data sets . put another way , one must solve an inverse problem that links the data sets to the unknown parameters ; this calculation is often performed using a nonlinear least squares algorithm . this approach is widely used and has provided important insights into a wide range of metal - semiconductor interfaces * ? ? ? a drawback , however , is that the microscopic information about the interface is averaged into a single number , the interface conductance , obscuring the microscopic detail . hopkins et al used tdtr measurements on a variety of metal films with varying phonon cutoff frequencies to extract spectral information about phonon transmission@xcite . however , determining transmission coefficients is still challenging due to variations in phonon density of states in each metal . in this work , we aim to directly extract the transmission coefficients from tdtr data by replacing the macroscopic transfer function based on fourier s law with a microscopic transfer function based on ab - initio phonon transport modeling . just as in the traditional approach , we seek to identify the parameters that best fit the tdtr data sets . although this fitting process is much more complex than the traditional approach , in principle it is the same widely - used procedure . there are several conditions that must be satisfied to successfully extract transmission coefficients using this approach . first , it is essential that part of the non - equilibrium phonon distribution emerging from the interface propagate into the substrate ballistically . as illustrated in fig . [ fig : schematics ] ( a ) , when mfps are much shorter than the characteristic length scale of the thermal gradient , information about the phonon distribution at the interface is lost due to scattering . in this case , the tdtr signal is largely insensitive to the transmission coefficient profile so long as the overall interface conductance remains fixed . on the other hand , if some phonons have sufficiently long mfps , the non - equilibrium phonon distribution penetrates into the substrate . in this case , as we will show in subsequent sections , the tdtr signal depends not only on the magnitude of the interface conductance but on the spectral profile of the transmission coefficients . it is this sensitivity that we will exploit to retrieve the transmission coefficients from the tdtr data sets . this discussion implies that not every substrate will be suitable for our approach as sufficiently long phonons mfps compared to the induced thermal gradient are required . fortunately , many experimental reports have demonstrated clear evidence of this quasiballistic heat transport regime in different material systems @xcite . in particular , mfps in si are generally accepted to exceed one micron at room temperature . @xcite considering that thermal penetration depths in si are on the same order in tdtr * ? ? * , si is a suitable substrate for this work . second , we must determine the microscopic transfer function that maps the transmission coefficients directly to the tdtr signal without any artificial fitting parameters . this step is challenging due to the difficulty of efficiently solving the bte for the tdtr experiment . a number of simplified models@xcite have been proposed , but these models make various approximations that limit their predictive capability . in this work , we overcome this challenge using two recent advances we reported for rigorously solving the spectral bte under the relaxation time approximation ( rta ) that yield a factor of @xmath3 speedup compared to existing methods and allows the first ab - initio phonon transport modeling of tdtr free of artificial parameters or simplifications of the phonon dispersion . this solution is derived in the next section . before deriving the necessary functions , we briefly summarize the necessary steps in our approach . first , we obtain tdtr data for a given metal / substrate sample . we then calculate the intrinsic phonon dispersion and lifetimes for each material that are then inserted into the ab - initio transfer function derived in the next section . finally , the transmission coefficients are varied until the optimized transmission profiles are identified that best match the experimental data . thus , the procedure is identical to that used in traditional tdtr measurements excepting the complications of deriving and inverting the microscopic transfer function . we now describe the derivation of the microscopic transfer function @xmath4 that maps transmission coefficients to the tdtr amplitude and phase data as a function of time . the result of the derivation is a function for which the only inputs are the phonon dispersions and lifetimes for each material composing the interface , and the only unknown parameters are the spectral transmission coefficients @xmath5 . the output of the function is the tdtr amplitude and phase signal versus delay time . for this work , the phonon dispersion and lifetimes for si were calculated from first - principles with no adjustable parameters by lucas lindsay . we have also used first - principles inputs from mingo _ et al_. with no appreciable difference in results or conclusions . let us first briefly review the signal formation in tdtr . since the thermal response given by the bte is a linear time - invariant system , the output transient signal @xmath2 of tdtr can be represented in terms of frequency response solution through the following equation@xcite @xmath6 where @xmath7 is the reference angular frequency of the periodic heating , @xmath8 is the angular sampling frequency set by the repetition rate of the laser pulses , and @xmath9 is surface temperature response subject to a periodic heating at frequency @xmath10 given the transmission coefficients @xmath5 . therefore , identifying @xmath4 requires computing @xmath9 , or the surface temperature frequency response to a periodic heating in a double - layer structure of a metal film on a substrate as shown in fig . [ fig : schematics ] ( b ) . thermal transport in an isotropic crystal , assuming only cross - plane heat conduction , is described by the one - dimensional ( 1d ) spectral boltzmann transport equation ( bte ) under the relaxation time approximation ( rta)@xcite , @xmath11 where @xmath12 $ ] is the deviational distribution function , @xmath13 is the equilibrium distribution function , @xmath14 is the directional cosine , @xmath15 is the phonon group velocity , @xmath16 is the phonon relaxation time , and @xmath17 is the spectral volumetric heat generation . assuming a small temperature rise , @xmath18 , relative to a reference temperature , @xmath19 , the equilibrium distribution is proportional to @xmath20 , as shown in eq . ( [ eq : bedist_linearized ] ) . here , @xmath21 is the reduced planck constant , @xmath22 is the phonon frequency , @xmath23 is the phonon density of states , @xmath24 is the bose - einstein distribution , and @xmath25 is the mode specific heat . the volumetric heat capacity is then given by @xmath26 and the fourier thermal conductivity @xmath27 , where @xmath28 and @xmath29 is the phonon mfp . to close the problem , energy conservation is used to relate @xmath30 to @xmath20 , given by @xmath31d\omega d\omega = 0 , \label{eq : energyconservation}\ ] ] where @xmath32 is the solid angle in spherical coordinates and @xmath33 is the cut - off frequency . note that summation over phonon branches is implied without an explicit summation sign whenever an integration over phonon frequency or mfp is performed . we now divide our discussion into three parts : transducer film , substrate , and interface . the bte in the transducer film can be reformulated as a fredholm integral equation of the second kind@xcite ; the solution in the substrate can be treated as the green s function to the bte.@xcite the solutions in the two layers depend on each other through the interface conditions that enforce conservation of heat flux . the metal thin film serves as an optical transducer that absorbs the incident optical energy while also enabling the observation of temperature decay through the thermoreflectance coefficient . in our work , we neglect electrons and consider that heat is only carried by phonons in al . justification for this approximation is given in appendix [ sec : electrons ] . since the system is modulated at a given frequency @xmath10 , we can assume that both @xmath34 and @xmath35 are of the form @xmath36 to define @xmath37 and @xmath38 . the volumetric heat generation in thin film is given by @xmath39 , where the amplitude of heating source @xmath40 . we also assume that phonons are specularly reflected at @xmath41 , _ i.e. _ @xmath42 . substituting the definition of @xmath43 and @xmath44 and the specular boundary condition at @xmath41 into eq . ( [ eq : bte ] ) leads to a first - order ode for @xmath45 . its solution is given by @xmath46 ) \\ \label{eq : odesolution_gminus}\nonumber \widetilde{g}^-_{1\omega}(x_1,\mu ) & = & b_{\omega}e^{\frac{\gamma_{1\omega}}{\mu}(d - x_1 ) } \\ & - & \int_x^d\frac{c_{1\omega}\delta \widetilde{t}(x'_1 ) + q^0_{\omega}e^{-x/\delta}\tau_{1\omega}}{4\pi\lambda_{1\omega}\mu}e^{\frac{\gamma_{1\omega}}{\mu}(x'_1-x_1)}dx'_1\ ( \mu \in [ -1 , 0]),\end{aligned}\ ] ] where @xmath47 , @xmath48 is the film thickness , and @xmath49 are the unknown coefficients determined by the interface condition at @xmath50 . here , @xmath51 indicates the forward - going phonons and @xmath52 the backward - going phonons . to close the problem , we plug eqs . ( [ eq : odesolution_gplus ] ) & ( [ eq : odesolution_gminus ] ) into eq . ( [ eq : energyconservation ] ) and obtain an integral equation for temperature as : @xmath53 where @xmath54 . the kernel function @xmath55 is given by @xmath56+e_1[\widehat{\gamma}_{1\omega}|\widehat{x}'_1-\widehat{x}_1|]\}d\omega\ ] ] and the two inhomogeneous functions are given by @xmath57+e_2[\widehat{\gamma}_{1\omega}(1-\widehat{x}_1)]\}\\ \label{eq : inhomof2}\nonumber f^2(\widehat{x}_1 ) & = & \frac{2\pi}{\int_0^{\omega_m}\frac{c_{1\omega}}{\tau_{1\omega}}d\omega}\int_0 ^ 1\int_0^{\omega_m}\frac{q^0_{\omega}e^{-\rho\widehat{x}'_1}}{\text{kn}_{1\omega}}\{e_1[\widehat{\gamma}_{1\omega}(\widehat{x}'_1+\widehat{x}_1)]+e_1[\widehat{\gamma}_{1\omega}|\widehat{x}'_1-\widehat{x}_1|]\}d\omega d\widehat{x}'\\\end{aligned}\ ] ] where kn@xmath58 is the knudsen number , @xmath59 , and @xmath60 is the exponential integral given by@xcite : @xmath61 recently , we have developed a spectral method to efficiently solve eq . ( [ eq : filmtemperature ] ) in ref . briefly , the functions in eq . ( [ eq : filmtemperature ] ) can be expanded as a finite cosine series , such as @xmath62 and @xmath63 where @xmath64 is the truncated basis number , and @xmath65 s and @xmath66 s are the fourier coefficient . similarly , @xmath67 and @xmath68 are also expanded in term of cosines . following the steps in the above reference , we can express the temperature as @xmath69^{t}\underline{\phi ( x)}\ ] ] where the matrix @xmath70 contains elements @xmath71 , @xmath72 , @xmath73 , @xmath74 and @xmath75 ( @xmath76 ) and @xmath77 is a @xmath78 column vector of the unknown coefficients @xmath49 , where @xmath78 is the number of discretization in phonon frequency . @xmath79 is a @xmath80 matrix , consisting of the fourier coefficients of @xmath81 evaluated at each phonon frequency @xmath82 and @xmath83 is a n column vector , consisting of the fourier coefficients of @xmath68 . then , @xmath84 and @xmath85 can be expressed in terms of the unknown coefficients @xmath49 by plugging eq . ( [ eq : filmtemp ] ) into eqs . ( [ eq : odesolution_gplus ] ) and ( [ eq : odesolution_gminus ] ) . the substrate can be treated as a semi - infinite region subject to a surface heat flux . therefore , the bte for the substrate becomes @xmath86 where the unknown coefficients @xmath87 s are determined through the interface conditions . we then apply the green s function method given in ref . 61 . the unknown distribution function in spatial frequency domain is then written as @xmath88 and the temperature profile @xmath89d\omega}{\int_0^{\omega_m}\frac{c_{2\omega}}{2\pi\tau_{2\omega}}\left[1-\frac{1}{\lambda_{2\omega}\xi_2}\text{tan}^{-1}\left(\frac{\lambda_{\omega}\xi_2}{1+i\eta\tau_{2\omega}}\right)\right]d\omega},\ ] ] where @xmath90 is the fourier variable of @xmath91 . again , to express @xmath92 only in terms of unknown coefficients @xmath87 , we simply plug eq . ( [ eq : substratetemp ] ) into eq . ( [ eq : substratedistribution ] ) . the unknown coefficients in the solutions of transducer film and substrate are obtained by applying appropriate interface conditions . here , we use the elastic transmission interface condition with mode conversion , closely following the work by minnich _ et al_.@xcite briefly , for a given mode @xmath93 , the heat fluxes outgoing from the interface , @xmath94 and @xmath95 , must be equal to the reflected and transmitted heat fluxes incident to the interface , @xmath96 and @xmath97 . by assuming elastic and diffuse scattering , the transmission and reflection process for each phonon frequency is treated independently and the heat flux equality condition must be satisfied for each frequency and polarization . the interface conditions are @xmath98 where @xmath99 is the transmission coefficient of mode @xmath100 at frequency @xmath22 from side 1 to side 2 as mode @xmath93 , @xmath101 is the reflection coefficient of mode @xmath100 at frequency @xmath22 from side 2 back into side 2 as mode @xmath93 , and so on . the next question is how @xmath102 is related to the other reflection and transmission coefficients . the reflection coefficients are related to the transmission coefficients by energy conservation given by @xmath103 and @xmath104 @xmath105 is related to @xmath102 through the principle of detailed balance , which requires that no net heat flux can transmit across the interface when both materials are at an equilibrium temperature @xmath106 . applying this condition to every phonon mode on each side of the interface for each polarization and frequency gives : @xmath107 therefore , we need to specify @xmath102 , @xmath108 , and @xmath109 . let us first consider a special case where no mode conversion is allowed ( @xmath102 , @xmath110 , @xmath108 , @xmath111 for @xmath112 ) . then , the interface conditions become @xmath113 and the detail balance becomes @xmath114 therefore , once @xmath115 is specified , all the other transmission and reflection coefficients are determined . for now , we only consider this special case and neglect the mode conversion in our bte simulations . later , we show that the mode specific transmission coefficients can not be resolved by the tdtr measurements and the measurable quantity is @xmath116 instead of individual transmission coefficients . for simplicity , we will use @xmath117 rather than the summation . in this section , we justify the approximations made in the derivation of the transfer function . first , we have neglected the role of electrons , either in carrying heat in the metal or in carrying heat across the interface by direct coupling to substrate phonons . for the first point , in appendix [ sec : electrons ] , we solve the bte explicitly including electron - phonon coupling and show that it has negligible effect on the tdtr signal and hence the transmission coefficients . for the second point , our approach can not rule out this mechanism . however , the available evidence in the literature @xcite shows that the metal electron - substrate phonon coupling effect is too small to be observed for several material systems . therefore , little evidence exists to support a large contribution from this mechanism . as a result , we assume that heat is carried across the interface solely by phonons . the second assumption is the neglect of mode conversion , or the change of phonon polarization as phonons transmit or reflect at the interface . in appendix [ sec : modeconversion ] , we explicitly calculate the transfer function including this mechanism , showing that it has very little impact on the results . therefore , fitting the data with or without mode conversion yields identical results . we now describe how these pieces fit together to provide the transfer function @xmath118 . at any given frequency @xmath10 , the analytical expression of the unknown distributions from both sides , @xmath119 and @xmath120 can be obtained according to sec . [ subsec : film ] and sec . [ subsec : substrate ] , which then are evaluated at @xmath50 and at @xmath121 , respectively . given the values of @xmath5 , the values of @xmath122 , @xmath123 and @xmath124 can be inferred , and the unknown coefficients @xmath87 and @xmath49 s are obtained by plugging eqs . ( [ eq : odesolution_gplus ] ) , ( [ eq : odesolution_gminus ] ) , and ( [ eq : substratedistribution ] ) into eqs . ( [ eq : ic_1_special ] ) and ( [ eq : ic_2_special ] ) and solving the linear system . once @xmath49 is known , the temperature profile in the transducer film is solved using eq . ( [ eq : filmtemp ] ) as well as the surface temperature @xmath125 . at a given modulation frequency @xmath126 , @xmath10 is chosen to be @xmath127 , where integer @xmath128 is typically ranging from -50 to 50 . using eq . ( [ eq : tdtrsignal ] ) yields @xmath4 , which is directly compared to experimental data @xmath129 . with the transfer function obtained , it is useful to re - examine the conditions that are required to successfully extract transmission coefficients from tdtr data . first , the number of unknowns should be on the order of the number of data points . as an estimate , consider that we take tdtr data at five modulation frequencies . as in eq . ( [ eq : tdtrsignal ] ) , in the frequency domain each tdtr signal consists of surface temperature responses at different frequencies , the lowest of which is the modulation frequency . typically , among those surface temperature responses , the responses at the lowest four to five frequencies contain most of the thermal information.@xcite with around five tdtr data sets containing amplitude and phases at five different modulation frequencies , we obtain 40 - 50 unique data points . now consider the number of unknowns . if we take the materials to be isotropic , the transmission coefficients depend only on phonon frequency . in our typical discretization , we find that there around one hundred transmission coefficients that must be fit . however , these coefficients are not all independent due to a smoothness constraint - physically , the transmission profile can not fluctuate arbitrarily , an intuition supported by atomistic simulations @xcite . qualitatively , this requirement implies that each coefficient depends on the adjacent coefficients , decreasing the effective number of unknowns by a value on the order of two to three . therefore , the number of unknowns is comparable to the number of data points . ample data points are a necessary but not sufficient condition to enable extraction of transmission coefficients . the last requirement is that the tdtr signal should be sensitive to the shape of the transmission coefficient profile . in the case of heat diffusion , this requirement is not satisfied : so long as the interface conductance is unchanged , the tdtr signal will not change because scattering in the substrate obscures the interfacial phonon distribution . to demonstrate this point , we simulated tdtr signals using two different transmission coefficient profiles as shown in fig . [ fig : tdtr_sensitivity ] ( a ) with the scattering rates of si modified such that no mfp exceeds 50 nm . the chosen transmission profiles possess the same interface conductance . the amplitude and phase of the simulated tdtr signals for these two transmission coefficient profiles are shown in fig . [ fig : tdtr_sensitivity ] ( b ) . the figure shows that they are nearly identical even though the two transmission coefficient profiles are completely different , demonstrating that in the diffusion regime the tdtr signal only depends on the magnitude of interface conductance . on the other hand , in the quasiballistic regime the tdtr signal is sensitive to the shape of the transmission coefficient profile , enabling the coefficients to be obtained with tight constraints . we demonstrate this sensitivity in fig . [ fig : tdtr_sensitivity ] ( c ) . for this calculation , the same transmission coefficient profiles are used but the bulk scattering rates of si from lindsay et al are used without modification . the longest mfp in silicon is on the order of micrometers , @xcite and the characteristic length scale of a tdtr experiment with a modulation frequency around 1 mhz is also on the order of a micron . therefore , the transport in the tdtr experiment for al / si is quasiballistic . in this case , the tdtr signals using two different transmission coefficient profiles are no longer identical as shown in fig . [ fig : tdtr_sensitivity ] ( c ) , demonstrating that in the quasiballistic regime the tdtr signal is sensitive to both the magnitude of interface conductance and the spectral profile . [ fig : tdtr_sensitivity ] ( c ) plots a measured tdtr signal . from the plot , one can immediately tell the transmission coefficient profiles used in the calculations are not correct as the experimental and calculation signals do not match . the correct transmission coefficient profile will reproduce the measured tdtr signal . our procedure to identify this profile is described in the next section . the final step is to solve the inverse problem that identifies the transmission coefficients that best explain the observed data . from the bte model , we obtain a surface temperature decay curve as a function of time just like the one measured in the experiments . for a given sample , the actual transmission coefficient profile as a function of phonon frequency will minimize the difference between the simulation curves and experimental tdtr traces at all modulation frequencies . we solve the inverse problem using a particle swarm optimization ( pso ) method to search for the optimal profile . the goal of the pso method is to minimize the objective function defined as @xmath130 the first part of the equation evaluates the norm of the difference between the experimentally measured and bte - simulated tdtr signals given a transmission profile profile @xmath5 . the second part of the equation evaluates the second derivative of the transmission coefficient profile , serving as the smoothness penalty function . note that the smoothness of the profiles is the only constraint we impose in the objective function . the smoothing parameter @xmath131 determines the relative importance of the second part to the first part . if @xmath132 , then no smoothness constraint is imposed . here , we use @xmath133 where @xmath134 is the initial profile . the formula is chosen such that the first and second parts of the equation have the same order of magnitude . to search for the optimal profile that minimizes the objective function , the pso algorithm randomly initializes a collection of transmission coefficient profiles and evolves them in steps throughout the phase space which contains all possible transmission coefficient profiles . at each step and for each profile , the algorithm evaluates the objective function defined as above . after this evaluation , the algorithm decides how each profile should evolve according to the current best profile . the profile evolves , then the algorithm reevaluates . the algorithm stops when the objective function reaches the desired value . the transmission coefficient profile that achieves the minimum value of the objective function is the optimal profile that explains the data however , since the inverse problem is ill - posed , a unique solution does not exist . we generate a probability density plot for the transmission coefficients using gibbs sampling to explore adjacent regions of the optimal transmission coefficient profile . we first randomly generate around 1000 profiles by perturbing the optimal profile with a smooth function defined using the following formula @xmath135 , \label{eq : perturbation}\ ] ] where the amplitude of the perturbation @xmath136 is 0.1 , and @xmath137 , @xmath138 , @xmath139 , @xmath140 , @xmath141 and @xmath142 are random numbers between 0 to 1 . we evaluate the objective function at all the perturbed profiles and recorded the values . then , we start the gibbs sampling process . at each iteration , we randomly draw a profile , @xmath143 , from the stored population and compare the value of its corresponding objective function , @xmath144 to the one from the previous step , @xmath145 evaluated at profile @xmath146 . if @xmath144 is less than @xmath145 , we accept @xmath143 and kept @xmath144 . if not , a random number @xmath147 is drawn and compared to @xmath148 , where @xmath149 if @xmath147 is smaller than @xmath150 , then we accept @xmath143 and kept @xmath144 . if not , we reject @xmath143 and update @xmath144 to be @xmath145 . the system temperature , @xmath19 , is chosen such that the stationary distribution is gradually changing . here , @xmath19 is set to be the mean value of the objective functions of all the perturbed samples . we keep track of how many times each profile was chosen at each iteration and generated a histogram of the occurrence frequency of each profile . we stop the sampling process when the histogram becomes stationary . this occurrence frequency is also called the likelihood of the transmission coefficient profiles . the higher the value of a profile s likelihood is , the better the fit with the experimentally measured tdtr signals at different modulation frequencies . thus by combining the pso method with the gibbs sampling algorithm , we are able to determine the most likely transmission coefficients at the interface between si and al . we demonstrate our transmission coefficient measurements on an al film on si substrate with the native oxide removed by hydrofluoric acid prior to al deposition , yielding a clean interface . the tem image in fig . [ fig : tdtrsignal_clean_300k ] ( a ) shows the interface thickness is less than 0.5 nm . the amplitude and phase of signals from the lock - in amplifier at different modulation frequencies are given in fig . [ fig : tdtrsignal_clean_300k ] . for reference , solving the usual inverse problem with the macroscopic transfer function on this data set yields @xmath151 mw / m@xmath152-k and @xmath153 w / m - k , in good agreement with prior works and literature values for the thermal conductivity of si.@xcite although the good agreement is often taken as evidence that the macroscopic transfer function is valid for si , this conclusion is incompatible with several independent ab - initio calculations that clearly show that heat is carried by phonons with mfps exceeding the thermal penetration depth of tdtr.@xcite this prediction has recently been experimentally confirmed by cuffe _ et al _ using thermal measurements on variable thickness silicon membranes.@xcite this fact implies that quasiballistic transport should be readily observable in a typical tdtr experiment on si , despite the seemingly correct thermal properties measured . this apparent contradiction is resolved by observing that the signal measured in tdtr strongly depends on the spectral profile of the transmission coefficients in the quasiballistic regime as shown in fig . [ fig : tdtr_sensitivity ] . we represent the transmission coefficient as a probability density plot , with the color intensity indicating the likelihood that a single transmission coefficient curve passing through a particular point at a given phonon frequency is able to simultaneously explain all of the data in fig . [ fig : tdtrsignal_clean_300k ] , without any other adjustable parameters . the result is shown in fig . [ fig : tdtrsignal_clean_300k](c ) . the figure shows that the transmission coefficient from si to al for longitudinal phonons , t@xmath154 , starts at unity , its maximum possible value , and decreases steadily to near zero for high phonon frequencies ( @xmath155 thz ) . the transmission coefficient profiles for the other polarizations have similar shapes , and so throughout the paper we plot only the longitudinal transmission coefficients for simplicity . the transmission coefficients from al to si , t@xmath156 are calculated by satisfying the principle of detailed balance ; the relationship between t@xmath154 and t@xmath156 reflects the differences in density of states and group velocity between the two materials . the transmission coefficients for each side of the interface and for the other polarizations are given in appendix [ sec : transmissioncoeff ] . our measured transmission coefficient profile thus indicates that longitudinal phonons with frequencies less than 6 thz are transmitted to the maximum extent allowed by the principle of detailed balance , while longitudinal phonons with frequencies larger than 8 thz are nearly completely reflected at the interface . we now examine this result in context with the common models for transmission coefficients . the amm is often cited as an appropriate model for transmission coefficients at sufficiently low phonon frequencies . calculating the amm transmission coefficients for normal incidence , we obtain a value of 0.95 , which is quite consistent with our observation . [ fig : lowfreqjustification ] shows how the tdtr signal varies as the low phonon frequency transmission coefficients are reduced . below 90 % transmission , the computed signal is outside of the error bounds . therefore , the low frequency transmission must exceed this value to match the data . for short wavelength phonons , the dmm would be expected to apply . at the highest phonon frequencies ( shortest wavelengths ) , the dmm correctly predicts the trend of the measured transmission coefficients tending to zero . however , for most of the phonon spectrum , the dmm is inconsistent with our measurement . we provide additional evidence for the inadequacy of conventional models to explain our measurements by considering two models : the gray model in which the transmission coefficient is a constant , independent of phonon frequency , and the diffuse mismatch model ( dmm ) . the dmm is only determined by the phonon properties of the materials , such as density of states and phonon group velocity . the constant transmissivity value is chosen to yield an interface conductance @xmath157 mw / m@xmath152-k using the formula of ref . the measured value for the clean interface is @xmath158 mw / m@xmath152-k . here , we demonstrate that neither of the models can explain the experimental tdtr data . as shown in fig . [ fig : comparison9mhz ] , the use of a constant transmission coefficient in the bte model overpredicts the phase values . similarly , the dmm underpredicts both the amplitude and phase at the early time of the signals . in figs . [ fig : comparison9mhz ] ( c ) & ( d ) , we show the deviation in amplitude and phase between the averaged experimental data at a given modulation frequency and the bte simulations using a constant profile and dmm , demonstrating that the deviation is far beyond the uncertainty in experimental data . the uncertainty is computed by calculating the standard deviation of both amplitude and phase data for multiple runs and multiple locations on a sample . a better comparison for our measurements is with atomistic calculations that are not subject to the highly restrictive assumptions of the amm and dmm . performing this comparison , we observe that our measurements agree with numerous molecular dynamics and atomistic green s function calculations , essentially all of which predict the general trend of decreasing transmission coefficients with increasing phonon frequency.@xcite in particular , our measurement of high transmission for longitudinal phonons with frequencies less than approximately 4 thz is consistent with atomistic calculations on acoustically - matched materials.@xcite our result also agrees with the experimental studies of polycrystalline silicon by wang _ et al_,@xcite which suggested that the transmission coefficients should decrease with increasing frequency . using this transmission coefficient profile , we plot the spectral interfacial heat flux versus phonon frequency and accumulative heat flux versus phonon wavelength in fig . [ fig : interfaceheatflux ] . our results show that most of interfacial heat flux is carried by phonons with frequencies less than 4 thz , with the contribution from higher frequencies strongly reduced due to their small transmission coefficients . in fact , we find that the contribution of phonons with frequencies less than 4 thz is essential to explain our observations : we are unable to explain the measured data without the contribution of phonons with frequencies less than 4 thz . similarly , we find that we can only explain the measurements using the exact phonon dispersion for al computed from dft ; simple dispersion relations such as debye model can not explain the data because they underestimate the contribution of low frequency phonons to thermal transport . we conducted several additional experiments to confirm the robustness of the measured transmission coefficients . first , since the energy transmission at the interfaces is considered elastic , the transmission coefficients in theory should be independent of temperature . we performed tdtr measurements on the same al / si sample at several temperatures higher than 300 k and compared the experimental results with the calculations using the same transmission coefficient profile measured at 300 k. as shown in figs . [ fig : tdtr_clean_400k ] , the calculation is in excellent agreement with experimental data at 400 k using exactly the same transmission coefficient profile obtained at 300 k. note that this comparison does not require any adjustable parameters . additional measurements at various temperatures are given in the supplementary information , and all give excellent agreement . we were unable to conduct measurements at lower temperatures due to the onset of radial heat conduction that is not accounted for in our model . second , we measured the transmission coefficients for al on sige . while this material has an additional point defect scattering mechanism compared to pure si , we expect the transmission coefficients to be nearly the same given that the host lattice is unchanged . the details about point defect scattering in sige are given in appendix [ appsec : sige ] . [ fig : tdtr_clean_400k ] ( c ) & ( d ) plots the amplitude and phase of the surface temperature decays at different modulation frequencies , demonstrating that the same transmission coefficient profile shown in fig . [ fig : tdtrsignal_clean_300k](c ) yields a signal that agrees well with this independent data set , again without any adjustable parameters . this result confirms that the measured transmission coefficients for si and sige substrates are indeed the same . interpreting the tdtr data on the sige with the conventional model results in a number of physical inconsistencies that are eliminated with our approach . first , the thermal conductivity of sige obtained with the conventional fourier model is around 35 w / mk and varies with modulation frequency as reported previously @xcite . however , the actual value is around 50 w / mk ( see appendix [ appsec : sige ] ) . thus tdtr does not necessarily provide the actual thermal conductivity of a material . second , interpreting these data with the traditional fourier model results in interface conductances for si and sige differing by more than 30 % even though the host lattice is the same @xcite . this inconsistency is removed when interpreting the data with our microscopic model as the same transmission coefficients explain both data sets . finally , we examine how the atomic structure of the interface affects the spectral content of the phonons carrying heat across the interface . we conducted additional measurements for al on si with a native oxide layer ( thickness @xmath159 1 nm as shown in a tem image in fig . [ fig : tem_oxide_nativeoxide ] ( a ) ) and si with thermally grown oxide layer ( thickness @xmath159 3.5 nm as shown in a tem image in fig . [ fig : tem_oxide_nativeoxide ] ( b ) ) . since the oxide layers are sufficiently thin to neglect their thermal capacitance , we can treat them as part of the interface@xcite that modifies the net transmission coefficient profile that describes transmission between al and si . by solving the inverse problem with the measurement as in figs . [ fig : tem_oxide_nativeoxide ] ( c ) & ( d ) as input , we are able to find the transmission coefficient profiles for these two cases as shown in figs . [ fig : tem_oxide_nativeoxide ] ( e ) & ( f ) . compared to a clean interface , the transmission coefficients for al on si with a native oxide are reduced for most of the phonon modes , except those with long wavelength longer than 1 nm . when the roughness of the interface increases with a thicker oxide layer , the transmission coefficient keeps decreasing and more phonons , especially those with wavelengths between 1 and 3 nm , are reflected at the interface . therefore , our measurements show that phonons with wavelength shorter than the interface roughness are more likely to be reflected by the interface than phonons with wavelength longer than the interface roughness , and as the interface gets rougher , a larger fraction of the phonon spectrum is affected by the interface . in contrast to prior approaches that measure only interface conductance , here we are able to provide microscopic insight into which phonons are more likely to be reflected due to atomic - scale changes in the interface structure . our work has implications for thermal metrology and technological applications . first , we have shown that tdtr is capable of providing considerable microscopic detail about thermal phonons if the measurements can be properly interpreted using a microscopic transfer function with ab - initio input . unlike with the macroscopic transfer function , our approach provides quantitative details on the spectral content of the heat carried by phonons in the sample . as a corollary , using the macroscopic transfer function to interpret tdtr data on certain samples can lead to erroneous results . for example , the apparently correct measurement of silicon thermal conductivity is a coincidence that occurs due to the transmission coefficient profile of a typical al / si interface . as recently showed by wilson et al , if the interface is modified by the introduction of an oxide layer , the measured thermal conductivity no longer coincides with the literature value . @xcite similarly , the traditional fitting approach yields a thermal conductivity for sige that does not agree with its actual value measured with a bulk method . these inconsistencies are eliminated if the data are interpreted with our approach . therefore , the conventional tdtr interpretation does not necessarily provide the actual physical properties of materials . second , our measurements show that the spectral profile of transmission coefficients is essential to understanding thermal transport across interfaces . due to a lack of knowledge about interfaces , the phonon transmission coefficients are often predicted with a variety of simple models . however , this work shows that none of these models are capable of explaining the experimental measurements . therefore , including an accurate spectral transmission coefficient profile is essential to properly describing thermal phonon transport across interfaces . third , our work provides evidence that elastic transmission of phonons across an interface is the dominant energy transmission mechanism for materials with similar phonon frequencies . our microscopic transfer function does not incorporate electrons or inelastic scattering yet is able to explain all of the measurements we performed . this observation shows that the consideration of inelastic transmission and coupling between electrons in metals and phonons in semiconductors is not necessary to explain the tdtr data sets . fourth , our results provide quantitative information into which phonons transmit across interfaces of a given atomic structure . in particular , the strong frequency dependence of the transmission coefficients can be exploited to create thermal phonon filters to selectively remove parts of phonon spectrum , analogous to optical long - pass filters . phonons with wavelength much longer than the characteristic roughness of an interface are more likely transmitted through the interface while short - wavelength phonons are mostly reflected . our approach provides a means to determine which of these phonon modes are transmitted or reflected and thus identify which phonon modes are filtered by the interface . finally , our work demonstrates the insights into heat conduction at the atomic scale that can be obtained through the interwoven application of experimental measurements in the quasiballistic heat conduction regime , ab - initio phonon transport modeling , and electron microscopy . through our approach , we are able to provide useful microscopic detail on the spectral content of heat cross an interface and its atomic structure . such a capability will permit the rational understanding and control of interfacial heat transport at the atomic level , a capability that could impact numerous applications . the authors thank l. lindsay , j. carrete and n. mingo for providing the first - principles calculations for silicon , prof . nathan lewis group for the access to the ellipsometer , and the kavli nanoscience institute ( kni ) at caltech for the availability of critical cleanroom facilities . x. c. thanks melissa a. melendes , matthew h. sullivan and carol m. garland from the kni for fabrication assistance , and victoria w. dix from the lewis group at caltech for the help with the ellipsometer measurements . this work was sponsored in part by the national science foundation under grant no . cbet 1254213 , and by boeing under the boeing - caltech strategic research & development relationship agreement . our simulations do not include electron - phonon coupling across the al - si interface or electron heat conduction in the metal film . the first approximation was justified in the text in sec . [ sec : approxjustification ] . we justify the second approximation by performing the simulations based on an electron - fourier / phonon - bte model that accounts for electron conduction in the metal . briefly , this model uses spectral phonon bte described in sec . [ sec : modeling ] coupled with a heat diffusion equation for electrons in the al thin films after absorption of an optical laser impulse . the coupled equations are given as following : @xmath160 where @xmath161 and @xmath162 are the temperatures of the electrons and phonons , respectively , and @xmath163 and @xmath164 are the volumetric heat capacity and the thermal conductivity of the electrons in al , respectively . the phonon temperature is linearly coupled to the electron temperature through the electron - phonon coupling coefficient g. the values of all the constants in eq . [ eq : electrons ] are tabulated in table [ tab : constants ] . this system of the equations is solved by a standard finite difference method in a two layered geometry . we compare the surface temperature responses to a heat impulse with and without the effects of electrons . as shown in fig . [ fig : electronjustification ] , due to strong electron - phonon coupling , electrons only affect the heat conduction shortly after the absorption of a heat pulse . after the first 100 ps , the heat conduction is dominated by the phonons . since a typical signal in a tdtr experiment is usually measured after 500 ps , whether heat is attributed to phonons or electrons in the metal has a negligible effect on the signal on the timescales that is interested in the experiments . therefore , our neglect of electrons has no effect on our transmission coefficient measurement . when phonons cross an interface they can change their frequency , in an inelastic process , or polarization , known as mode conversion , which can influence thermal interface conductance.@xcite in our work , we do not consider inelastic scattering . we justify the neglect of inelastic scattering through the work of prior numerical studies , which have provided evidence that the phonon transmission between two slightly dissimilar crystalline solids is elastic.@xcite considering the phonon frequencies between al and si are very similar , there is no evidence that inelastic processes should play a role in the interfacial transport for al / si interfaces . we have carefully examined the effect of mode conversion on our conclusions by rigorously including this process in our bte model . to examine whether conversion between polarizations would affect the phonon transport across the interface , we conducted our bte simulation assuming modes maintain their polarization after crossing the interface , or allowing them to change to any polarization while keeping the total transmission coefficient the same . keeping the total transmission coefficient for a given polarization . keeping @xmath116 invariant , we randomly partitioned @xmath116 , @xmath165 and @xmath166 into two different combinations of @xmath102 , @xmath108 and @xmath109;in other words , randomly between all the polarizations on the opposite side of the interface . in fig . [ fig : modeconversionjustification ] , we show that the surface temperature decay with and without conversion between polarization are essentially identical . moreover , the spectral interfacial heat flux is also identical with and without conversion . therefore , we conclude that mode conversion does not have an observable effect on the signal . the reason that our measurement is not sensitive to mode conversion is that the polarizations in si do not have extremely dissimilar mean free paths . as our measurement approach relies on the lack of scattering of some modes near the interface , the only way the mode conversion could affect our measurements would be if one polarization consistently changed to another polarization after transmitting through the interface with a drastically different mean free path than the original polarization . our calculations clearly show that the difference in mean free paths between the polarizations is not sufficient to affect our calculations and hence have any effect on our conclusions . in the main text , we only show the transmission coefficient from si to al for longitudinal phonons for the three samples . here , in figs . [ fig : trans_sitoal ] and [ fig : trans_altosi ] , we plot the transmission coefficient profiles as a function of phonon frequency and wavelength from both sides of the materials for each polarization with a clean interface , with a native oxide layer and with a thermally grown oxide layer . the color intensity indicates the likelihood that a single transmission coefficient curve passing through a particular point at a given phonon frequency is able to simultaneously explain all of the experimental data . we emphasize that the only fitting parameters are the transmission coefficients from si to al for the three polarizations . all other transmission and reflection coefficients are determined from detailed balance and energy conservation.@xcite for the clean interface , the only constraint used in the fitting process is the smoothness of the profile . in particular , note that we do not enforce any type of monotonicity or shape requirement on the coefficients other than smoothness . for the native oxide interface , we additionally require that the transmission coefficients of the native oxide interface do not exceed the values for the clean interface . similarly , the transmission coefficients of the thicker oxide interface should always be smaller than those of the native oxide interface . in figs . [ fig : decaycurves_varioustemperatures ] & [ fig : decaycurves_variousinterfaces ] , we plot all the original raw data from the tdtr experiments used in the manuscript along with the bte fitting results . in all the cases , we show excellent agreement between simulation and experiments . commercial high - purity natural si ( 100 ) wafer and si - ge ( 1.5 - 2 at % ge ) wafer ( 100 ) from mti corp . were used in the experiments . before coating al on the samples , three different surface conditions of the samples were prepared . first , the native oxide was removed with buffered hf acid to obtain a clean surface of si and sige . after etching , the samples were immediately put into a vacuum chamber for al deposition . second , the native sio@xmath167 layer was left in place . no further treatment was taken for this condition before al deposition . finally , a thermally grown sio@xmath167 layer as fabricated by putting the si samples into a tube furnace for three hours . the thickness of the native sio@xmath167 layer and thermally grown sio@xmath167 layer was measured by ellipsometry and tem to be @xmath168 nm and @xmath169 nm , respectively . a thin film of al was deposited on all samples using electron beam evaporator . the thickness of the al transducer layer was 70 nm , measured by atomic force microscopy . the measurements are taken on two - tint tdtr . the details are available in ref . the probe diameter is 10 @xmath170 m and the pump diameter is 60 @xmath170 m . both beam sizes are measured using a home - built two - axis knife - edge beam profiler . with 60 @xmath170 m pump heating size , the heat transfer problem can be treated as one - dimensional . all the measurements at @xmath171 k are performed under ambient conditions , and the additional measurements at @xmath172 and @xmath173 k are performed in an optical cryostat ( janis st-500 ) under high vacuum of @xmath174 torr . the tem samples were prepared by standard fib lift - out technique in the dual beam fe - sem / fib ( fei nova 600 ) . to protect the top surface , a pt layer with thickness @xmath159 300 nm was deposited with electron beam evaporation followed by another pt layer with thickness @xmath159 3 - 4 @xmath170 m by ga ion beam . the lamella was cut parallel to the chip edge which was aligned to the wafer flat edge during initial cutting in tdtr sample preparation . as a result , the cutting surface normal was along ( 110 ) direction and all the tem images were taken parallel to the si ( 110 ) crystallographic zone axis . high resolution transmission electron microscopy ( hrtem ) analyses were carried out in a fei tecnai tf-20 tem / stem at 200 kv . to avoid damage from the high energy electron beam , the beam exposure on region of interest was minimized especially at high magnification during operation . for sige , the mass difference scattering rate is calculated using the tamura formula,@xcite given by @xmath175 where @xmath22 is phonon frequency , @xmath23 is the phonon density of states per unit volume , and @xmath176 is the volume per atom . @xmath177 is a measure of the mass disorder , @xmath178 and @xmath179 are the concentration and the atomic mass of species @xmath93 , respectively , and @xmath180 is the average mass for the given composition . the tamura formula has been proven to effectively calculate the impurity scattering in sige with different ge concentration.@xcite the values of all the constants in eq . [ eq : tamura ] are tabulated in table [ tab : constants ] we have sent the sige wafer to the third party , thermotest , for bulk thermal conductivity measurements . the measured value , using transient plane source method on a bulk sample , is @xmath181 w / m - k . using the measured value , we are able to obtain the ge concentration to be about @xmath182 at % based on calculations with the tamura formula while the measured ge concentration using energy dispersive x - ray spectrometry is @xmath183 at % , which gives sige thermal conductivity around @xmath184 w / m - k . these differences in atomic concentration have only a minimal effect on the transport calculations and have been incorporated in the uncertainty of bte simulations in figs . [ fig : tdtr_clean_400k ] ( c ) & ( d ) of the main text . we assume a constant mfp for all modes in al ; the value @xmath185 nm is chosen to yield a lattice thermal conductivity @xmath186 w / m - k so that no size effects in the thin film occur . although the literature value of al thermal conductivity is about @xmath187 w / m - k , we verified that the resulting surface temperature decay curves by using these two al thermal conductivities in the tdtr diffusion model could not be distinguished as shown in fig . [ fig : alk_justification ] . since the transmission coefficients are extracted by fitting our model to the data , if a parameter in the model has little effect on the results of the model , then it can not affect the measured transmission coefficients . here , we demonstrate that the calculations are completely insensitive to al thermal conductivity , provided that it is larger than @xmath159 30 w / m - k . therefore , our choice of al thermal conductivity has no impact on our results . the relaxation time for each mode is then obtained through @xmath188 . we also verified that the particular value of the al mfp does not affect the results . note that although the al mpf is a constant , the dispersion of al is directly from the first - principle calculations , and the transmission coefficients depend heavily on the density of states and phonon group velocity in both metal and substrate . therefore , al is still modeled with a spectral phonon bte . al heat capacity ( j / m@xmath189-k ) : & @xmath190 + al lattice thermal conductivity ( w / m - k ) : & 123 + al total thermal conductivity ( w / m - k ) : & 230 + si heat capacity ( j / m@xmath189-k ) : & @xmath191 + si thermal conductivity ( w / m - k ) : & 155 + sige heat capacity ( j / m@xmath189-k ) : & @xmath191 + sige thermal conductivity ( w / m - k ) : & 51 + + heat capacity ( j / m@xmath189-k ) : & @xmath192 + thermal conductivity ( w / m - k ) : & @xmath193 + electron - phonon coupling coefficient @xmath194 ( w / m@xmath189-k ) : & @xmath195 + + volume per si atom @xmath176 ( nm@xmath189 ) : & 0.02 + measure of the mass disorder @xmath196 : & 0.0568 + + al / si with a clean interface ( nm ) : & 69 + al / sige with a clean interface ( nm ) : & 72 + al / si with a native oxidized interface ( nm ) : & 70 + al / si with a thermally - grown oxidized interface ( nm ) : & 70 + + optical penetration depth @xmath197 ( nm ) : & 10 + laser repetition frequency ( mhz ) : & 76 + david g. cahill , paul v. braun , gang chen , david r. clarke , shanhui fan , kenneth e. goodson , pawel keblinski , william p. king , gerald d. mahan , arun majumdar , humphrey j. maris , simon r. phillpot , eric pop , and li shi . nanoscale thermal transport . ii . 20032012 . , 10 ( 1):0 011305 , 2014 . . jayakanth ravichandran , ajay k. yadav , ramez cheaito , pim b. rossen , arsen soukiassian , s. j. suresha , john c. duda , brian m. foley , che - hui lee , ye zhu , arthur w. lichtenberger , joel e. moore , david a. muller , darrell g. schlom , partick e. hopkins , arun majumdar , ramamoorthy ramesh , and mark a. zurbuchen . crossover from incoherent to coherent phonon scattering in epitaxial oxide superlattices . , 13:0 168172 , 2014 . peixuan chen , n. a. katcho , j. p. feser , wu li , m. glaser , o. g. schmidt , david g. cahill , n. mingo , and a. rastelli . role of surface - segregation - driven intermixing on the thermal transport through planar @xmath198 superlattices . , 111:0 115901 , sep 2013 . . bed poudel , qing hao , yi ma , yucheng lan , austin minnich , bo yu , xiao yan , dezhi wang , andrew muto , daryoosh vashaee , xiaoyuan chen , junming liu , mildred s. dresselhaus , gang chen , and zhifeng ren . high - thermoelectric performance of nanostructured bismuth antimony telluride bulk alloys . , 3200 ( 5876):0 634638 , 2008 . k. biswas , j. he , i. v. blum , c. i. wu , t. p. hogan , d. n. seidman , v. p. dravid , and m. g kanatzidis . high - performance bulk thermoelectrics with all - scale hierarchical architectures . , 489:0 414418 , 2012 . zonghui su , li huang , fang liu , justin p. freedman , lisa m. porter , robert f. davis , and jonathan a. malen . layer - by - layer thermal conductivities of the group iii nitride films in blue / green light emitting diodes . , 1000 ( 20):0 201106 , 2012 . . nam han , tran viet cuong , min han , beo deul ryu , s. chandramohan , jong bae park , ji hye kang , young - jae park , kang bok ko , hee yun kim , hyum kyu kim , jae hyoung ryu , y. s. katharria , chei - jong choi , and chang - hee hong . improved heat dissiapation in gallium nitride light - emitting diodes with embedded graphene oxide pattern . , 4:0 1452 , 2013 . ramez cheaito , john t. gaskins , matthew e. caplan , brian f. donovan , brian m. foley , ashutosh giri , john c. duda , chester j. szwejkowski , costel constantin , harlan j. brown - shaklee , jon f. ihlefeld , and patrick e. hopkins . thermal boundary conductance accumulation and interfacial phonon transmission : measurements and theory . , 91:0 035432 , jan 2015 . . peter j. obrien , sergei shenogin , jianxiun liu , philippe k. chow , danielle laurencin , p. hubert mutin , masashi yamaguchi , pawel keblinski , and ganpati ramanath . bonding - induced thermal conductance enhancement at inorganic heterointerfaces using nanomolecular monolayers . zhaojie wang , joseph e. alaniz , wanyoung jang , javier e. garay , and chris dames . thermal conductivity of nanocrystalline silicon : importance of grain size and frequency - dependent mean free paths . , 110 ( 6):0 22062213 , 2011 . . pmid : 21553856 . robert j. stevens , leonid v. zhigilei , and pamela m. norris . effects of temperature and disorder on thermal boundary conductance at solid solid interfaces : nonequilibrium molecular dynamics simulations . , 500 ( 1920):0 3977 3989 , 2007 . . . woon ih choi , kwiseon kim , and sreekant narumanchi . thermal conductance at atomically clean and disordered silicon / aluminum interfaces : a molecular dynamics simulation study . , 1120 ( 5):0 054305 , 2012 . r. e. jones , j. c. duda , x. w. zhou , c. j. kimmer , and p. e. hopkins . investigation of size and electronic effects on kapitza conductance with non - equilibrium molecular dynamics . , 1020 ( 18):0 183119 , 2013 . . n. yang , t. luo , k. esfarjani , a. henry , z. tian , j. shiomi , y. chalopin , b. li , and g. chen . thermal interface conductance between aluminum and silicon by molecular dynamics simulations . , in press , 2013 . patrick e. hopkins , pamela m. norris , mikiyas s. tsegaye , and avik w. ghosh . extracting phonon thermal conductance across atomic junctions : nonequilibrium green s function approach compared to semiclassical methods . , 1060 ( 6):0 063503 , 2009 . . mark . e. siemens , qing li , ronggui yang , keith a. nelson , erik h. anderson , murnane margaret m. , and henry c. kapteyn . quasi - ballistic thermal transport from nanoscale interfaces observed using ultrafast coherent soft x - ray beams . , 9:0 2930 , 2010 . a. j. minnich , j. a. johnson , a. j. schmidt , k. esfarjani , m. s. dresselhaus , k. a. nelson , and g. chen . thermal conductivity spectroscopy technique to measure phonon mean free paths . , 107:0 095901 , aug 2011 . . k. regner , d. sellan , z. su , c. amon , a. mcgaughey , and j. malen . broadband phonon mean free path contributions to thermal conductivity measured using frequency - domain thermoreflectance . , 40 ( 1640 ) , 2012 . jeremy a. johnson , a. a. maznev , john cuffe , jeffrey k. eliason , austin j. minnich , timothy kehoe , clivia m. sotomayor torres , gang chen , and keith a. nelson . direct measurement of room - temperature nondiffusive thermal transport over micron distances in a silicon membrane . , 110:0 025901 , jan 2013 . . bjorn vermeersch , amr m. s. mohammed , gilles pernot , yee rui koh , and ali shakouri . superdiffusive heat conduction in semiconductor alloys . truncated lvy formalism for experimental analysis . , 91:0 085203 , feb 2015 . . john cuffe , jeffery k. eliason , alexei a. maznev , kimberlee c. collins , jeremy a. johnson , andrey shchepetov , mika prunnila , jouni ahopelto , clivia m. s. torres , gang chen , and keith a. nelson . reconstructing phonon mean free path contributions to thermal conductivity using nanoscale membranes . , 2014 . jaeho lee , elah bozorg - grayeli , sangbum kim , mehdi asheghi , h .- s . philip wong , and kenneth e. goodson . phonon and electron transport through ge2sb2te5 films and interfaces bounded by metals . , 1020 ( 19):0 191911 , may 2013 . . ashutosh giri , john t. gaskins , brian f. donovan , chester szwejkowski , ronald j. warzoha , mark a. rodriguez , jon ihlefeld , and patrick e. hopkins . mechanisms of nonequilibrium electron - phonon coupling and thermal conductance at interfaces . , 1170 ( 10):0 105105 , march 2015 . . . john c. duda , patrick e. hopkins , justin l. smoyer , matthew l. bauer , timothy s. english , christopher b. saltonstall , and pamela m. norris . on the assumption of detailed balance in prediction of diffusive transmission probability during interfacial transport . , 140 ( 1):0 2133 , march 2010 . . . kwangu kang , yee kan koh , catalin chiritescu , xuan zheng , and david g. cahill . two - tint pump - probe measurements using a femtosecond laser oscillator and sharp - edged optical filters . , 790 ( 11):0 , 2008 . . jivtesh garg , nicola bonini , boris kozinsky , and nicola marzari . role of disorder and anharmonicity in the thermal conductivity of silicon - germanium alloys : a first - principles study . , 1060 ( 4 ) , january 2011 . .

given the recently deduced relationship between x - ray temperatures and stellar velocity dispersions ( the `` relation '' ) in an optically complete sample of elliptical galaxies ( ) , we demonstrate that ellipticals contain substantial amounts of dark matter _ in general_. we present constraints on the dark matter scale length and on the dark - to - luminous mass ratio within the optical half - light radius and within the entire galaxy . for example , we find that minimum values of dark matter core radii scale as kpc and that the minimum dark matter mass fraction is% within one optical effective radius and is% within , depending on the stellar density profile and observed value of . we also confirm the prediction of davis & white ( 1996 ) that the dark matter is characterized by velocity dispersions that are greater than those of the luminous stars : . the relation implies a nearly constant mass - to - light ratio within six half - light radii :ml . this conflicts with the simplest extension of cdm theories of large scale structure formation to galactic scales ; we consider a couple of modifications which can better account for the observed relation .

astro-ph9901242

there is a strong consensus that spiral galaxies are embedded in massive , non - luminous halos , the structure and scaling properties of which have been intensively investigated ( e.g. , @xcite ) . studies of galaxy kinematics and of the density and temperature distributions of hot intergalactic gas in galaxy groups ( @xcite ) and clusters ( e.g. , @xcite ) have demonstrated the dominance of dark matter on larger scales . in the context of existing theories of the formation of galaxies and large scale structure , it is therefore natural to expect non - luminous material within elliptical galaxies as well . however , until very recently , the evidence for dark matter in ellipticals has been less forthcoming and more controversial . dark matter in ellipticals has been elusive largely due to the lack of internal dynamical indicators at optical and radio wavelengths that are as ubiquitous , spatially extensive , and unambiguous as hi rotation curves in spirals . although mergers have endowed some early - type galaxies ( many classified as s0 ) with neutral ( e.g. , @xcite ) or ionized ( @xcite ) gas disks , these are rare and may not be representative of the population of ellipticals as a whole . studies of stellar kinematics have , until recently , been inadequate in quality and extent to significantly constrain the large scale mass distribution in ellipticals in light of the complicating effects of projection and anisotropic velocity distributions . most , if not all , giant elliptical galaxies contain extended distributions of hot x - ray emitting gas that provide a means for probing the large scale dark matter distribution a means potentially as powerful as rotation curves in spiral galaxies . since these hot gaseous halos should be close to hydrostatic equilibrium ( @xcite ) , measurements of the density and temperature distributions from x - ray imaging and spectroscopy yield constraints on the mass distribution on the scale and with the accuracy and spatial resolution of x - ray temperature profiles . observations of elliptical galaxies by the _ einstein observatory _ ipc were presented as strong evidence for the presence of extensive dark halos in early - type galaxies ( @xcite ; @xcite ; @xcite ) ; however , the temperature profiles were of sufficiently poor quality to foster a continuing skepticism ( @xcite , @xcite ) . progress on a number of fronts has been rapid during recent years . in the optical , improvements in detector capabilities , data analysis , and modeling have now provided robust stellar dynamical constraints on the mass distribution out beyond the optical half - light radius @xmath5 for several ellipticals ( @xcite , @xcite , @xcite , @xcite ) , while gravitational lensing observations are providing the first evidence for dark matter in intermediate redshift galaxies ( @xcite ; @xcite ; @xcite ) . x - ray observations using the _ rosat _ and _ asca _ satellites have improved the accuracy , spatial resolution , and extent of derived hot gas density and temperature profiles . of particular interest is the recent study of _ asca _ observations of @xmath15 early - type galaxies by matsushita ( 1997 ) that shows that gas temperature profiles are relatively flat out as far as can be reliably measured ( 5@xmath16 , depending on the intrinsic luminosity and galaxy distance ) . the existence of isothermal 0.51 kev gaseous halos extending , in some cases , to well beyond 10@xmath5 is strong evidence for the presence of extended dark matter in elliptical galaxies . the case for dark matter halos in a number of elliptical galaxies is now overwhelming , and the following pattern seems to be emerging from the latest results on individual galaxies : dark matter comprises a significant but not dominant fraction of the total mass within @xmath5 ( @xcite ; @xcite ; @xcite ; @xcite ; but , see @xcite for an exceptional case : ngc 4636 ) ; dark matter becomes increasingly dominant at large radii : @xmath17m@xmath18 kpc@xmath19 to within a factor of two for @xmath20 kpc ( @xcite , @xcite , @xcite ) , corresponding to mass - to - light ratios @xmath21m@xmath13l@xmath22 measured at ( or extrapolated to ) @xmath23 kpc ( @xcite ; @xcite ) . while there is now little doubt that some elliptical galaxies do contain dark matter , a number of broader issues remain to be addressed . among the most important with respect to our understanding of galaxy formation is whether dark matter in ellipticals is ubiquitous , and how the relative scaling of dark and luminous matter varies both within galaxies as a function of radius , and between galaxies with different optical properties . in this paper we use the recently derived collection of x - ray temperatures in a complete , optically selected sample from davis & white ( 1996 ; hereafter dw ) to constrain the _ average _ properties of dark halos in the _ population _ of @xmath2 elliptical galaxies . dw analyzed x - ray spectra from 42 of the 43 optically brightest elliptical galaxies , using _ pspc data if available and _ einstein _ ipc data otherwise ( _ i.e. _ for four galaxies ) . see dw and white & davis ( 1998 ) for details about the sample and data analysis . for the purposes of this work , we use only the mean relationship between x - ray temperatures , @xmath24 , and central projected optical velocity dispersions , @xmath25 ( hereafter , the `` @xmath0@xmath1 relation '' ) @xmath26 or equivalent x - ray / optical correlations . more precisely , we consider the variation of the observable @xmath9 with optical luminosity @xmath27 ( figure 1 ) , where @xmath28 and @xmath29 is the mean mass per particle . if dark matter dominates the gravitational potential on large scales , then @xmath24 is essentially a measure of the dark matter content within the extraction radii used by dw ( @xmath8 , where @xmath5 is the optical effective radius ) . since the observed scaling relations ( _ i.e. _ the fundamental plane ) for elliptical galaxies provide a link between @xmath25 and the global luminosity , @xmath9 is an excellent diagnostic of the dark - to - luminous matter ratio within the optical radius . temperature profiles can provide additional dark matter constraints , but are measurable for only the x - ray brightest galaxies in the sample . likewise , extended stellar velocity dispersion profiles are useful , but data of sufficient detail to be unambiguously interpreted are not generally available . on the other hand , central velocity dispersions have been accurately measured for the entire sample ; central velocity dispersions are known to correlate with other optical properties and are less likely to suffer from the uncertainties caused by anisotropic stellar orbits . our goal in this paper is to make general inferences about the presence of dark matter halos and their mean systematic variation with optical luminosity . toward that end , we utilize a set of physically and observationally well - motivated mass models and scaling relations in order to reproduce the @xmath0@xmath1 relation . although we address the scatter in the @xmath0@xmath1 relation , a detailed examination of individual galaxies is beyond the scope of this paper and is deferred to future investigation . our results can be considered in three stages . first , we demonstrate how the observed range of @xmath9 _ necessarily _ requires the existence of dark halos in elliptical galaxies . second , we derive conservative lower limits on the dark matter scale length and dark - to - luminous mass fraction , and discuss how the dark matter content must scale with optical luminosity so as to reproduce the observed @xmath0@xmath1 relation . finally , we discuss possible implications of our results for cosmology and galaxy formation , by embedding our models within particular scenarios for large scale structure formation . as stated above , we use the observable @xmath9 , defined in equation ( 1 ) , to constrain the dark matter distribution in elliptical galaxies . this parameter is the ratio of suitable averages ( to be defined below ) of the square of the stellar velocity dispersion and the gas temperature . therefore , we must derive the expected distributions of stellar velocity dispersions @xmath1 and gas temperatures @xmath0 , given specified total gravitational mass distributions . recently published _ hubble space telescope _ ( _ hst _ ) observations of the centers of elliptical galaxies show that their surface brightness profiles can generally be characterized by the function @xmath30^{(c - b)/a},\ ] ] where @xmath31 is the projected radius and @xmath32 is a break radius such that @xmath33 for @xmath34 and @xmath35 for @xmath36 , while @xmath37 characterizes the sharpness of the break ( @xcite ) . in order to accurately calculate the central stellar velocity dispersion , this observed departure from distributions with `` analytic cores '' must be accounted for . in the appendix we demonstrate that , for @xmath38 and @xmath39 , equation ( 2 ) provides a good approximation to the projection of a space density profile @xmath40 ( @xmath41 ) for @xmath42 and @xmath43 for @xmath44 . to derive the luminosity density distribution @xmath45 , we numerically deproject the surface brightness profile : @xmath46 where @xmath47 , and the change of variables @xmath48 has been made . the dw sample is magnitude - limited , and we are primarily interested in the most luminous ellipticals , since they have the most accurate x - ray temperatures ( see 3.3 ) . we adopt the typical values for @xmath49 galaxies ( adopting @xmath50 km s@xmath19 mpc@xmath19 ) of @xmath38 , @xmath51 , and @xmath52 ( @xcite ) . for the nine galaxies in common with dw ( excluding the highly flattened ngc 4697 ) , the mean surface brightness parameters and variances from faber et al . ( 1997 ) are as follows : @xmath53 , @xmath54 , @xmath55 . the resulting stellar density profile diverges slightly faster than @xmath56 for @xmath42 . the asymptotic surface brightness slope @xmath57 is flatter than in conventional surface brightness parameterizations and the integrated luminosity @xmath58 does not converge . therefore , we truncate the optically luminous distribution at radius @xmath59 , determined so that @xmath60 consequently @xmath61 . as an alternative stellar model , we consider the density profile from hernquist ( 1990 ) , @xmath62 where @xmath63 is the total luminosity and @xmath64 ( as adopted here ) provides the best match to a de vaucouleurs ( dev ) model with effective radius @xmath5 ( @xcite ) . an advantage of this model is that the density and mass profiles are analytic . we shall subsequently refer to the stellar distribution of equation ( 3 ) as the `` hst '' model , and that of equation ( 5 ) as the `` hernquist '' model . a comparison of the luminosity density ( in units of the average density , @xmath65 ) as a function of normalized radius ( @xmath66 ) in these models with the corresponding dev model is shown in figure 2 , where @xmath67 . all three models produce similar slopes at very small @xmath68 ; however , dev and hernquist models fail to reproduce the flattening at @xmath69 observed by _ hst_. on the other hand , while @xmath70% of the total luminosity is contained within @xmath71 in all three models , the hst model concentrates the entire remaining @xmath72% of the mass between @xmath73 and @xmath8 as opposed to spreading it out from @xmath73 to @xmath74 . in order to accurately characterize the observed central stellar properties as required by the definition of @xmath9 without introducing additional parameters ( _ i.e. _ an additional break radius and a distinct asymptotic form for the density profile ) , such a redistribution of the stellar light is necessary . if the light profile does indeed steepen outside of the _ hst _ field - of - view , then the only effect on our results for the hst model is that , by overestimating the stellar mass at large radii , we would underestimate the dark halo mass required to reproduce any observed value of @xmath9 . the stellar velocity dispersion is derived by solving the jeans equation in the form @xmath75 where @xmath76 is the stellar density distribution obtained from multiplying the luminosity density ( see equations 3 and 5 , above ) by a constant stellar mass - to - light ratio ; @xmath77 is the total mass within radius @xmath68 ( the dark matter component of which is described in the next section ) ; @xmath78 is the velocity dispersion anisotropy parameter ( @xmath79 ) ; and @xmath80 ( @xmath81 ) is the radial ( tangential ) component of the velocity dispersion . we make the reasonable assumption that the velocity dispersion becomes isotropic at small @xmath68 but radial at large @xmath68 ( @xcite ) , and adopt the functional form @xmath82 where the transitional scale radius @xmath83 is a free parameter of our models . the dark matter mass distribution is assumed to follow the `` universal '' density profile ( @xcite ) , @xmath84 that , when integrated , yields @xmath85.\ ] ] it is not clear how appropriate equations ( 8) and ( 9 ) are for elliptical galaxy dark halos . on one hand , there is evidence for cores in spiral galaxy dark matter distributions ( e.g. , @xcite ) ; on the other , dissipational collapse of the baryonic component in ellipticals can increase the dark matter concentration . moreover , models with cores can not generally be consistently coupled with stellar mass distributions as cuspy as those often observed in elliptical galaxies ( @xcite ) . we have verified the existence of self - consistent two - component mass models with dark matter following equation ( 8) and hernquist stellar models ( equation 7 ) , in the parameter range of interest , according to the criterion of ciotti and pelligrini ( 1992 ) . from the virial theorem it follows that the integrated mass can be robustly estimated from the mean temperature . constraints on the dark matter concentration can also be placed from consideration of the @xmath0@xmath1 relation ; however , any inferences about dark matter at small or large radii are contingent on the corresponding assumed asymptotic dark matter density slopes . our choice of equation ( 8) for the dark matter distribution does allow us to connect our results with predictions of dark halo structure from numerical structure formation simulations . however , the true shape of the dark matter spatial distribution can only be constrained from temperature and stellar velocity dispersion _ profiles _ of sufficient extent and resolution . the hot , x - ray emitting gas is assumed to be in hydrostatic equilibrium so that @xmath86 the mass in hot gas is not significant compared to the stellar mass , so its contribution to the gravitational potential is neglected . since we are interested in the average properties of the dw sample , and because spatial analysis is not available for most of the sample , we assume that @xmath87 ; this has been found to be a good approximation for x - ray bright galaxies , and is expected if heating is balanced by cooling locally ( @xcite ) . both assumptions , hydrostatic equilibrium and @xmath87 , are more likely to break down for galaxies with low x - ray - to - optical luminosity ratios , where a partial or transient outflow might obtain ; however , the @xmath0@xmath1 relation for these relatively x - ray faint systems does not differ , within the statistics of the sample , from that of the sample as a whole . a significant unifying simplification for the hst stellar models follows from the observations that the ratios of core mass to total stellar mass ( @xmath88 , where @xmath89 is @xmath90 [ equation 4 ] multiplied by the stellar mass - to - light ratio ) , and break radius @xmath32 to @xmath5 ( the `` true '' @xmath5 as observed , not the half - light radius in our models ) are approximately constant : @xmath91 and @xmath92 ( @xcite ) . our adoption of these scaling relations , along with the empirical relation @xmath93 , insures that our model galaxies are consistent with the fundamental plane ( fp ) as long as the core is baryon - dominated , since it follows that @xmath94 where @xmath95 is the central velocity dispersion .. ] in order to provide us with a one - parameter family of models ( the quality and quantity of the x - ray data is insufficient to attempt to derive a `` @xmath0@xmath1@xmath5 '' or equivalent fp - like relation ) , we take a `` faber - jackson cut '' through the fp such that @xmath96 , normalized so that @xmath97 kpc ( and @xmath98 km s@xmath19 ) at @xmath99 for @xmath50 km s@xmath19 mpc@xmath19 . the root - mean - square deviation of galaxies in the dw sample from this projection of the fp is only 34 km s@xmath19 ( 13.5% of the mean ) . the hernquist models ( equation 5 ) likewise adhere to the fp , since we universally adopt @xmath64 . equations ( 6 ) and ( 10 ) are solved in dimensionless form . there are two fundamental ( luminosity - dependent ) dimensional quantities that we denote as @xmath100 ( chosen to be @xmath32 for the hst model or @xmath101 for the hernquist model ) and @xmath102 ( @xmath89 for the hst model or total stellar mass @xmath103 for the hernquist model ) . thus , adopting the above scaling relations , the length- , mass- , density- , velocity- , and temperature - scaling relations are as follows : @xmath104 @xmath105 @xmath106 @xmath107 and @xmath108 here @xmath109 is the hubble constant in units of 80 km s@xmath19 mpc@xmath19 , @xmath110 with @xmath111l@xmath14 , and @xmath112 is defined such that the stellar mass - to - light ratio @xmath113m@xmath13l@xmath14 for @xmath114 and @xmath115 . @xmath116 corresponds to @xmath99 , _ i.e. _ @xmath117 . the dark matter mass distribution is scaled accordingly , _ i.e. _ @xmath118 where @xmath119 is the ratio of dark to stellar mass at the hst stellar model cutoff radius , i.e. @xmath120 and @xmath121 is the dimensionless dark matter scale length . in the above , @xmath122 , @xmath123 , and the function @xmath124 is defined in equation ( 9 ) . to determine the average values of @xmath0 and @xmath1 , the hydrostatic and jeans equations are integrated inward from a zero - pressure boundary at @xmath125 ( see @xcite ) . although the stellar mass has a cutoff ( @xmath59 ) for the hst model , the gas and dark matter halos are likely to extend beyond this and are not truncated in the models . this results in a pressure contribution to the gas virial temperature within @xmath59 that leads to lower values of @xmath9 for a given @xmath119 , and is therefore conservative in the sense of minimizing the required amount of dark matter . the average quantities that constitute the observable @xmath9 are calculated as follows : we define @xmath25 as the average projected stellar velocity dispersion within a circular aperture of size @xmath32 , and @xmath24 as the emission - weighted projected gas temperature within @xmath59 ( the same aperture over which the temperature is measured in dw , corresponding to the radius enclosing @xmath126% of the optical light ) . the value of @xmath9 is robust to the precise choice of these radii since @xmath127 . the aperture - averaged , projected velocity dispersion for all central radii @xmath128 ( @xmath129 ) is within 10% of its value at @xmath32 for the hst ( hernquist ) model , assuming isotropic orbits and that the central potential is dominated by the stellar component . by averaging over constant multiples of the fundamental stellar scale length ( @xmath100 ) , we assure that our computed values of @xmath9 depend only on the _ relative _ ( to stellar ) dimensionless dark matter and velocity dispersion anisotropy parameters . these parameters are the following : the ratio of dark to luminous matter scale length , @xmath121 ; the mass ratio @xmath130 within @xmath59 , @xmath119 ; and the dimensionless anisotropy scale length , @xmath131 . the only additional model parameters are the stellar distribution slopes . for a given triplet ( @xmath131 , @xmath132_{r_{\rm max}}$ ] , @xmath133 ) , a general solution for the dimensionless gas temperature and stellar velocity dispersion radial distributions follows that can be scaled to physical units for any @xmath109 and @xmath27 using equations ( 1216 ) ; @xmath112 is determined by normalizing to @xmath134 km s@xmath19 for @xmath99 and @xmath115 ; note that @xmath135 . in this section we consider the dependences of @xmath9 on @xmath131 , @xmath133 , and @xmath132_{r_{\rm max}}$ ] for both hst and hernquist models . while the latter provides a more accurate global representation of elliptical galaxy stellar distributions , the former is more accurate around and inside the break radius @xmath32 where @xmath25 is calculated . for models that are dark matter dominated beyond several @xmath5 , the redistribution of the stellar mass at large radii in the hst model is irrelevant . in the absence of dark matter , the only free parameter in our models is the anisotropy length scale @xmath83 ; recall that this is defined such that orbits become isotropic for @xmath136 and radial for @xmath137 . therefore , if stars dominate the gravitational potential within @xmath59 , and if the shape of the velocity ellipsoid does not vary systematically with @xmath27 , then @xmath9 should be independent of @xmath27 and @xmath25 ( _ i.e. _ , @xmath138 ) , contrary to the observed @xmath0@xmath1 relation ( dw ) . moreover , as we now show , the typical observed value of @xmath139 can not be reproduced in the absence of dark matter . the solid and dotted curves in figure 3 show the variation of @xmath9 with the dimensionless anisotropy scale length @xmath140 for , respectively , the hst and hernquist stellar models without dark matter . evidently , @xmath9 falls with increasing @xmath83 , but is essentially constant for @xmath141@xmath142 ( @xmath143@xmath144 ) , since @xmath25 is an average over a much smaller radius . the asymptotic values of @xmath9 are @xmath145 and @xmath146 for the the hst and hernquist stellar models , respectively . the lower value for the hernquist model is the result of a temperature maximum at several @xmath32 that substantially contributes to the emission - averaged temperature because of the high stellar ( and , by assumption , gas ) density there . as dark matter is introduced , the temperature profile becomes more nearly isothermal , and differences in @xmath9 between hst and hernquist models lessen ( see next paragraph , and figure 3 ) . varying the slope parameters of the hst models ( equation 2 ) does not result in any values of @xmath9 less than 0.75 . the values of @xmath9 derived without dark matter lie well above what is observed in a typical elliptical galaxy ( in fact , no galaxy in the dw sample has a best - fit value of @xmath9 greater than 1 ; see figure 1 ) . one of our major conclusions is already apparent . while a typical bright elliptical galaxy has @xmath139 , _ even under the most extreme assumptions , @xmath9 is never less than @xmath147 without the presence of an extended dark halo_. while 10 of the 30 galaxies in the dw sample have upper limits to @xmath9 greater than 0.75 , only 3 have upper limits greater than 1.2 ( see figure 1 ) . of the 16 galaxies where @xmath9 is determined to better than 20% , only 1 has an upper limit greater than 0.75 . the dash - dotted and dashed curves show the corresponding models containing dark matter halos with dark matter scale length @xmath148 set equal to @xmath5 ( _ i.e. _ @xmath149 ) , and with @xmath150_{r_{\rm max}}$ ] chosen so that @xmath139 for @xmath151 . the non - luminous fraction of the total mass within @xmath5 is @xmath152 and @xmath153 for the hst and hernquist models , respectively . from here on we assume isotropic stellar orbits . as shown in figure 3 , for the practical purposes of calculating @xmath9 , this is equivalent to assuming that the orbits are mostly isotropic inside @xmath154 . since @xmath9 increases with decreasing @xmath140 , more dark matter is required to obtain a given value of @xmath9 if this assumption is discarded once again we adopt the most conservative assumption as regards the required amount of dark matter . figure 4a shows @xmath9 as a function of the ratio of dark matter scale length to stellar break radius , @xmath155 , for the hst ( solid curve ) and hernquist ( dotted curve ) models with @xmath150_{r_{\rm max}}$ ] . ( for the hernquist model , the scaling relation @xmath156 has been utilized . ) as @xmath148 decreases below @xmath157 , dark matter plays an increasingly important role in determining @xmath25 , so @xmath9 rises . for fixed @xmath132_{r_{\rm max}}$ ] , @xmath9 is nearly independent of @xmath155 provided that @xmath158 . it is clear that measurements of @xmath9 can only place _ lower _ limits on @xmath148 . there is a fair amount of freedom in how the dark matter can be distributed and still produce @xmath139 . this is illustrated in figure 4b , which displays the variation in baryon fraction at @xmath5 ( solid curve for the hst stellar model , dotted curve for the hernquist model ) and at @xmath32 ( dashed curve for the hst stellar model , dot - dashed curve for the hernquist model ) . @xmath9 changes very slowly with @xmath155 for @xmath159 , while the baryon fraction inside @xmath32 varies from @xmath160 and @xmath161 for the hst and hernquist stellar models , respectively , to @xmath162 for both models as @xmath148 becomes very large . meanwhile , the baryon fraction inside @xmath163 increases from @xmath164 ( @xmath165 ) if @xmath166 , to @xmath147 ( @xmath167 ) for the hst ( hernquist ) model as @xmath148 becomes large compared to the stellar cutoff radius @xmath59 . figure 5 shows @xmath9 as a function of @xmath132_{r_{\rm max}}$ ] for hst and hernquist stellar models with @xmath148 set equal to 0.5 , 1.0 , and @xmath168 . as discussed above , @xmath169 ( 0.75 ) as @xmath170 for the hst ( hernquist ) model . the curves pass through @xmath139 at @xmath171_{r_{\rm max}}$ ] if @xmath172 or @xmath173 , while @xmath174_{r_{\rm max}}$ ] is required if @xmath175 for hst ( hernquist ) models . as @xmath132_{r_{\rm max}}$ ] becomes sufficiently large for fixed @xmath148 , dark matter comes to dominate the mass even at small radii ( _ i.e _ within @xmath32 where @xmath25 is calculated ) and @xmath9 becomes independent of @xmath119 . that is , in order to maintain consistency with the fp and the faber - jackson relation , the physical mass must remain constant and the increase in dark - to - luminous mass ratio @xmath119 is compensated for by decreasing the stellar mass - to - light ratio ( _ i.e. _ , @xmath112 in equations 1316 ) accordingly . we have shown ( section 4.1 ) that the lower limits on @xmath9 for the hst and hernquist stellar models without dark matter are 0.75 and 1.2 , respectively . to explain observed values of @xmath9 that are less than this requires the addition of dark matter halos with sufficient mass to raise the value of @xmath24 , but with a large enough scale length to leave @xmath25 ( which is averaged over a much smaller scale ) relatively unaltered . by calculating solutions to equations ( 6 ) and ( 10 ) for the entire ( @xmath132_{r_{\rm max } } , r_{\rm dm}/r_*$ ] ) parameter space , we have derived limits on the amount and scale length of any dark halo able to reproduce a given observed value of @xmath9 . as discussed in section 4.3 and shown in figure 5 , for a fixed value of @xmath155 there is a minimum value of @xmath9 that is obtained when @xmath119 becomes sufficiently large that both @xmath25 and @xmath24 are determined by the non - luminous mass component . since this minimum @xmath9 increases as @xmath155 decreases , there is a minimum value of @xmath155 required to explain any observed @xmath9 . figure 6a shows these minimum values of @xmath155 for the hst ( solid curve ) and hernquist ( dotted curve ) models . the minimum value of @xmath148 for a @xmath176 galaxy with @xmath139 is @xmath177 kpc ( @xmath143 ) ; the dimensional value scales as @xmath178 . these extreme models are dark matter dominated , even inside the break radius @xmath32 , and have very low stellar mass - to - light ratios . requiring models to have @xmath179m@xmath13l@xmath14 for @xmath180 and @xmath115 ( or , equivalently , that @xmath460% of the mass within @xmath32 be stellar ) roughly doubles the minimum values of @xmath155 for @xmath181 . this reasonable requirement then implies that @xmath3 kpc . the maximum baryon fractions within @xmath32 , @xmath5 , and @xmath59 for any given @xmath9 . are shown in figure 6b . the lower limits on the dark matter fractions in models required to explain @xmath182 are less than 5% within @xmath32 , and less than 20% within @xmath5 non - luminous halos could be quite inconspicuous unless data well beyond @xmath5 are considered . the maximum baryon fractions at @xmath59 for the hst ( hernquist ) model are 0.43 ( 0.15 ) , 0.53 ( 0.27 ) , and 0.61 ( 0.45 ) to produce @xmath183 , 0.5 , and 0.6 , respectively . these models have very large dark matter scale lengths ( @xmath184 ) and are analogous to `` maximum disk '' models of spiral galaxies in that they maximize the contribution of stars to the mass - to - light ratio . in order to investigate the scaling characteristics of dark halos with optical luminosity , we consider the equivalent of the @xmath0@xmath1 relation in the @xmath9@xmath27 plane . correlations of @xmath9 and log@xmath185 with log(@xmath27 ) are derived using the dw data set , blue magnitudes and @xmath186 colors from the rc3 catalog ( @xcite ) , and velocity distances from faber et al . ( the @xmath9log(@xmath27 ) data and best - fit correlation are re - plotted in figure 7b . ) galaxies in the centers of cluster cooling flows and those dominated by nuclear point source emission are excluded . the log@xmath185log(@xmath27 ) correlation , equivalent to @xmath187 where @xmath188 is the visual luminosity in units of @xmath111l@xmath14 , is essentially identical to the convolution of the @xmath0@xmath1 correlation and faber - jackson relation ( see section 3.4 ) . nevertheless , requiring that models conform to the observations in this form makes use of the tilt of the fp for scaling of the the stellar component , thus circumventing concerns about the thickness of the projection of the fp onto the @xmath27@xmath1 plane . note that the galaxies considered span @xmath189 , although most of the sample lies within the range @xmath190 . given our homologous elliptical galaxy models , which are based on the scaling properties of the fp , the observed variation in @xmath9 implies that there is a systematic variation of relative dark matter properties which breaks the self - similarity for galaxies of different luminosity . the sense of the observed @xmath0@xmath1 relation could also be explained by a fine - tuned systematic variation of the velocity dispersion anisotropy ( having more radial orbits for more luminous galaxies ) , or if more luminous galaxies have cuspier stellar distributions ( larger values of @xmath191 in equation 2 ) . however , the latter is contrary to observations ; and as we have shown , the observed values of @xmath9 are too low to be explained without the introduction of dark matter . it is possible to explain the @xmath9@xmath27 trend , assuming the dark - to - luminous mass ratio within @xmath59 is constant , if the ratio of dark matter to stellar break ( or , equivalently , effective ) radius , @xmath155 decreases with increasing galaxy luminosity @xmath27 . however , since @xmath9 is flat for @xmath192 and steeply rises for @xmath193 ( figure 4a ) this requires fine - tuning . moreover , as we discuss shortly , @xmath193 is not expected . more naturally , the increase in @xmath9 with @xmath27 could be explained if @xmath132_{r_{\rm max}}$ ] _ decreases _ with _ increasing _ @xmath27 ( that is , if more luminous galaxies are less dark matter dominated ) and @xmath155 is relatively large ( see figure 5 ) . we now discuss the plausibility of such a correlation within the context of the hierarchical clustering formalism of structure formation , simulations of which produced the dark matter distribution expressed by equations ( 8) and ( 9 ) . this formalism provides specific predictions for @xmath148 as a function of virial mass ( defined here as the mass within a radius where the mean dark matter density is 200 times the average density in the universe ) for a given cosmology and power spectrum of initial fluctuations . in order to extend our models to the scale of the virial radius and mass , we transform the dark matter parameters @xmath194 and @xmath132_{r_{\rm max}}$ ] that are directly constrained by observations into the following pair of global parameters : the dark matter concentration parameter @xmath195 and the global dark - to - luminous mass ratio @xmath196_{r_{200}},\ ] ] where @xmath197 refers to the radius within which the overdensity of the dark matter ( only ) is 200 . ( @xmath198 and @xmath197 refer to the original mass of the dark halo , but this may have been subsequently reduced in rich environments through tidal stripping . ) note that for @xmath199 and @xmath200 , @xmath201 kpc , @xmath202 , and @xmath132_{r_{\rm max}}=1.6 $ ] for @xmath203 ( @xmath204 ) . in bottom - up hierarchical clustering theories of structure formation , lower mass halos are more concentrated a reflection of the higher density in the universe at the earlier epoch of their formation ( @xcite ) . we parameterize this dependence with the function @xmath205 for the concentration , where @xmath206 and @xmath207m@xmath18 is on the order of the present - day characteristic mass . the second term in the denominator of equation ( 21 ) is introduced to prevent the concentration from exceeding a maximum value @xmath208 , that we set equal to 100 , and driving the stellar mass - to - light ratio to very small values . note that @xmath209^{-(1+\gamma)/\gamma}$ ] becomes insignificant for @xmath210 ( e.g , in standard cdm models ) . utilizing the `` characteristic density '' program kindly provided by j. navarro , a standard cold dark matter ( cdm ) fluctuation spectrum results in @xmath211 and @xmath212 . for these parameters , @xmath213 ( @xmath214 for @xmath215 and @xmath216 ) and @xmath148 is indeed much greater than @xmath217 in the luminosity range of interest . the dot - dashed curve in figure 7a shows @xmath132_{r_{\rm max}}$ ] as a function of @xmath188 for @xmath218 ( _ i.e. _ baryon fraction 0.06 ) cdm models ( and for two alternative models to be discussed in the next section ) . clearly @xmath132_{r_{\rm max}}$ ] _ increases _ with optical luminosity , implying a _ decrease _ of @xmath9 with increasing @xmath27 as shown by the dot - dashed curve in figure 7b . a comparison with the observed @xmath9@xmath188 correlation ( solid curve in figure 7b ) , demonstrates that _ a constant baryon fraction cdm model for elliptical galaxy dark matter halos clearly fails to match the observed correlation _ , even over the limited range of @xmath188 where the correlation is well - established . a successful alternative model must produce dark halos where @xmath132_{r_{\rm max}}$ ] _ decreases _ with increasing optical luminosity in this range . we will discuss the form such possible alternatives must take in the next section . a natural explanation for why @xmath9 , the ratio of the central stellar velocity dispersion to the globally - averaged gas temperature , _ increases _ with @xmath25 and therefore @xmath27 is that the dark - to - luminous mass ratio within the luminous parts of elliptical galaxies ( @xmath219 ) _ decreases _ with @xmath27 ( see figures 5 , 7a and 7b ) . in fact , the observed trend requires that the total mass - to - light ratio within @xmath59 be nearly constant . this can be understood from a simple argument based on the virial theorem and fp , as follows . the virial temperature within @xmath59 , @xmath220 since @xmath221 ( all masses are evaluated at @xmath59 ) . since , from the fp @xmath222 , it follows that @xmath223 , or @xmath224 if we associate @xmath225 with the integrated x - ray temperature , which should be a fair approximation . the implication from the observed @xmath9@xmath27 trend that the total mass - to - light ratio within @xmath59 be nearly constant ( see below ) is contrary to expectations from cdm models of large scale structure if all elliptical galaxies have the same global baryon fraction . the dark - to - luminous mass ratio is predicted to _ increase _ with @xmath27 , in such models , since the ratio of the dark matter scale length @xmath148 to the optical half - light radius is large and weakly varying , thus producing a predicted correlation between @xmath9 and @xmath27 in the opposite sense to what is observed ( figures 7a and 7b ) . we consider two simple variations on the above model to try and recover the observed @xmath9@xmath27 trend . first we assume that the dark matter concentration depends only weakly on halo mass , as cdm predicts , but relax the constant baryon fraction assumption . then we assume that the concentration varies much more steeply with virial mass than the cdm prediction . if the weak dependence of concentration on halo mass predicted in cdm simulations ( equation 21 with @xmath226 and @xmath227 ) is maintained , then the assumption of a constant baryon fraction within @xmath197 must be relaxed : the required increase in dark matter fraction within @xmath59 for less luminous galaxies must be a reflection of a larger _ total _ ( _ i.e. _ within @xmath197 ) dark - to - luminous mass ratio . if all galaxies start off with identical baryon fractions , this then implies that smaller galaxies lose an increasingly large fraction of their original baryonic mass . ( similar conclusions for spiral galaxies have been reached by @xcite . ) we parameterize this variation in the dark / stellar mass ratio as @xmath228,\ ] ] where , as previously defined , @xmath229_{r_{200}}$ ] , @xmath188 is the visual luminosity in units of @xmath111l@xmath14 , and @xmath230 is a fitting parameter . equation ( 22 ) has the properties @xmath231 as @xmath232 ( complete baryonic mass loss ) , and @xmath233 as @xmath234 ( presumably , @xmath235 is the average baryon fraction in the universe ) . the observed @xmath9@xmath27 correlation is satisfactorally reproduced in models with @xmath236 , @xmath237 , and @xmath2382.5 . the dotted curves in figures 7a and 7b represent hst stellar models with @xmath239 ; the observed trend is very well matched ( figures 7b ) . ( hernquist stellar models require even larger values of @xmath240 ; however , this is an artifact of the underestimate of @xmath9 for low @xmath241 ; see section 4.1 ) . alternatively , if the dark matter halo concentration increases much more steeply with virial mass than predicted for a cdm fluctuation spectrum , the observed @xmath9@xmath27 correlation can be reproduced assuming a constant total baryon fraction ( _ i.e. _ , @xmath242 ) . in this case , the necessary decrease in @xmath132_{r_{\rm max}}$ ] for more luminous galaxies results not from a smaller relative global amount of dark matter , but from a larger fraction of the dark matter lying outside of the luminous portion of the galaxy ( _ i.e. _ @xmath59 ) . we have found that equation ( 21 ) with @xmath243 , @xmath244 , and @xmath245 provides models that adequately reproduce the data ( smaller values of @xmath246 require larger values of @xmath242 ) . @xmath132_{r_{\rm max}}$ ] and @xmath9 are shown as functions of @xmath27 , for @xmath218 , by the dashed curves in figure 7a and 7b . such a steep dependence of the concentration on total mass would seem to indicate a flatter - than - cdm primordial fluctuation spectrum ( @xcite ) . these alternative scenarios were designed to reproduce the @xmath9@xmath27 correlation without consideration of the scatter or that of galaxies about the mean fp . we address these concerns by computing @xmath247 for individual pairs @xmath248 , @xmath249 corresponding to each of the 25 non - cluster - cooling - flow , non - point - source - dominated galaxies in the dw sample , for the three scenarios of figures 7a b . a comparison with the observed values of @xmath9 , as shown in figures 8a c , confirms that the constant baryon fraction cdm scenario provides the poorest match to the observations , on average , and further illustrates the magnitude of variations about the mean correlation . further investigation into these variations will be pursued in a future work that examines the individual galaxies in detail . both of these alternative scenarios have drawbacks . mass loss induced by supernovae - driven galactic winds during the star - forming epoch of elliptical galaxies has been inferred from both the color - magnitude diagram and the enrichment of the intracluster medium ( e.g. , @xcite , @xcite ) . however the magnitude of this effect must be extreme ( _ i.e _ @xmath240 , in equation 22 , on the order of 1 or greater ) if it is to explain the increasing dominance of dark matter in less luminous galaxies required to explain the observed @xmath0@xmath1 relation . the mass loss implied by equation ( 22 ) with the required set of parameters , when integrated over the elliptical galaxy luminosity function ( @xcite ; @xcite ) , corresponds to @xmath250% of the original baryonic mass of @xmath2 galaxies . moreover , in such a scenario , all elliptical galaxies with @xmath251l@xmath14 would span a range in total mass of less than a factor of three ; the large range in mass loss could not , therefore , be due to less luminous galaxies lying within shallower potential wells . instead , one must presume that variations in the relative timing of star formation and the merging of pre - galactic fragments is responsible for the large range in mass loss , with galaxies with smaller present - day luminosities having lost the largest fraction of their original baryonic mass . light would be a very poor tracer of mass for elliptical galaxies , and the luminosity function would be a reflection of the rarity of elliptical galaxies sufficiently relaxed at the star formation epoch to prevent copious mass loss from galactic winds . it might therefore seem more appealing to resort to a concentration - mass relationship for dark halos ( equation 21 ) that is steeper than predicted in cdm simulations in such a way that optically less luminous galaxies are more dark matter dominated within @xmath59 , despite having the same total ( within @xmath197 ) baryon fraction . the basic requirement is that the dark matter scale length @xmath148 increase from @xmath252 at the low - luminosity end of the observed range , to @xmath253 at @xmath254l@xmath14 , to @xmath255 ( @xmath256 ) at the high luminosity range . these criteria are met for only a rather narrow volume in ( @xmath257 ) parameter space , and thus this scenario seems to suffer from a fine - tuning problem ; nonetheless , the range that successfully reproduces the observed @xmath0@xmath1 relation is not out of line for cosmogenic models with flat primordial fluctuation spectra ( @xcite ) . figures 9a , 9b , and 9c show the luminosity dependences of mass , mass - to - light ratio , and baryon fraction at various radii for the two models discussed above that reproduce the observed @xmath0@xmath1 ( @xmath9@xmath27 ) relation : dotted lines show the model with cdm - predicted halo concentration as a function of mass and increasing mass loss with decreasing luminosity ( @xmath226 and @xmath227 in equation 21 , @xmath239 , @xmath236 , and @xmath258 in equation 22 ) ; dashed curves show the model with steeper - than - cdm concentration as a function of mass and constant baryon fraction ( @xmath244 and @xmath243 in equation 21 , @xmath259 and @xmath260 in equation 22 ) . we refer to these as `` mass loss '' and `` non - cdm '' scenarios , respectively . the stellar model is assumed to follow equation ( 3 ) , and the luminosity range in the plots roughly corresponds to that of the dw sample ( @xmath152@xmath261 ) . the radii of interest are @xmath262 kpc , @xmath263 , and @xmath264 ; @xmath265 increases from 270 to 400 kpc for the mass loss scenario and @xmath266 kpc for the @xmath218 , non - cdm scenario , where @xmath110 with @xmath111l@xmath267 . there are a number of robust properties that hold for any model ( including hernquist models ) to successfully reproduce the @xmath0@xmath1 relation . since the models are constrained by the observed properties of the stars and by the average temperature of hot gas in hydrostatic equilibrium within @xmath264 , the inferred quantities at @xmath59 are nearly model - independent : @xmath268m@xmath18/l@xmath14 , @xmath269 . note also that @xmath270m@xmath18 for any model where the central potential is dominated by the stellar component . these values agree strikingly well with the constraints from statistical studies of gravitational lensing ( @xcite ; @xcite ; @xcite ) . the solid lines in figure 9a shows the mass as a function of luminosity , evaluated at @xmath5 and @xmath59 for the softened isothermal sphere ( sis ) model that provides a best - fit to the observed weak gravitational shear from @xcite . it is remarkable that the mass - to - light scaling derived from the shear that is dominated by @xmath271 galaxies at intermediate redshifts is in such excellent agreement with our results using a completely independent method on local galaxies primarily with @xmath2 . for the constant baryon fraction cdm model , the total mass within @xmath59 is a significantly steeper function of luminosity than is demonstrated by the curves in figure 9a . on scales both larger and smaller than @xmath59 , the two scenarios depicted in figures 9a - c diverge . in the mass loss scenario the baryon fraction within @xmath5 is generally greater than 0.7 and @xmath272m@xmath13l@xmath14 , while in the non - cdm scenario the mass - to - light ratio is nearly constant , @xmath273m@xmath13l@xmath14 , so that the baryon fraction within @xmath5 becomes less than 0.5 below @xmath274 . extrapolated out to the dark matter virial radius , @xmath197 , the total mass scalings differ ( by assumption ) . for the mass loss scenario , @xmath275 h_{80}{\rm m}_{\odot}/{\rm l}_{v_\odot},\ ] ] where acceptable models have @xmath2765 and @xmath2778 ( @xmath278 , @xmath279 for the model shown by dotted curves in figures 7a - b and 9a - c ) . for the non - cdm ( or any other constant baryon fraction ) scenario , @xmath280 the ( constant ) total baryon fraction @xmath281 for the dashed curves shown in figures 7a - b and 9a - c . we found that @xmath282 in non - cdm models is required to reproduce the observed @xmath9@xmath27 correlation . non - cdm models with relatively small global @xmath283 have larger baryon fractions within @xmath5 for @xmath284 ( because of larger dark matter scale lengths ) and agree somewhat better with the lensing constraints at this radius . integrating over the early - type galaxy luminosity function ( @xcite ; @xcite ) yields an average total ( _ i.e. _ within the virial radius ) mass - to - light ratio @xmath285m@xmath13l@xmath14 for the non - cdm ( constant baryon fraction ) scenario and @xmath286 , which can be compared to an average total mass - to - light ratio of @xmath287m@xmath13l@xmath14 for spirals ( @xcite ) , and @xmath2880.06 . in the mass loss scenario , the average total mass - to - light ratio is sensitive to the lower luminosity cutoff , @xmath289 , for the luminosity function and can be as high as @xmath290m@xmath13l@xmath14 ( @xmath291 ) for @xmath292l@xmath14 since even the lowest luminosity elliptical galaxy has a total mass of @xmath293m@xmath18 . the average density in the universe of virialized mass on galactic scales is @xmath294 of the critical density . we have calculated the velocity dispersion distributions for both the dark matter and stellar density distributions , assuming isotropic orbits . these are compared in figure 10 for an @xmath180 galaxy for both the mass loss and non - cdm scenarios . both velocity dispersion profiles have maxima since the total gravitational potential is not isothermal . the ratio ( dark - matter - to - stars ) of the squares of these maxima is greater than 1.4 over the luminosity range in figures 9a - c for the two models under consideration , and is @xmath295 over the range @xmath296 . in fact , the minimum value of this ratio for any hst or hernquist model that produces @xmath297 is greater than one . in this sense , the dark matter is hotter than the stars , as predicted by davis & white ( 1996 ) . in this work , we have addressed two essential features of the x - ray temperatures derived by davis & white ( 1996 ) for an optically complete sample of elliptical galaxies : ( 1 ) the x - ray emitting gas is always hotter than the stars and , typically , twice as hot ( @xmath298 ) ; ( 2 ) the gas / stellar temperature ratio tends to be higher for galaxies with lower velocity dispersions ( the `` @xmath0@xmath1 relation '' ) . we have constructed physically plausible models of the mass distribution in bright elliptical galaxies in an effort to constrain their average dark matter properties by matching these observations . the stellar models ( described in 3 ) are fully consistent with the fundamental plane scaling relations , and are designed to either conform to the latest published _ hst _ results on the structure of the centers of elliptical galaxies or to follow the hernquist ( 1990 ) approximation to a de vaucouleurs profile ( see figure 2 ) . we have made other plausible , but conservative , assumptions , such as ( 1 ) maximizing the non - gravitational ( _ i.e _ pressure ) contribution to the gas temperature by allowing the gas and dark matter distributions to extend to infinity and ( 2 ) assuming that the stellar orbits are isotropic for @xmath299 . a basic and general result of our calculations is that , in the absence of dark matter , @xmath300 ( see figure 3 ) . since @xmath139 is observed , a convincing case is made that dark matter is an extremely common , if not ubiquitous , constituent of elliptical galaxies . although x - ray ( and other ) observations have been used to infer the presence of dark matter in individual cases , we have shown that dark halos are generic to luminous , nearby elliptical galaxies . furthermore , the dark matter / stellar temperature ratios ( derived at the peak of their velocity dispersion distributions , assuming isotropic orbits ) are greater than one for models with @xmath297 . thus , the observation that the temperature of the extended hot gas exceeds the central stellar temperature is a reflection of the fact that the dark matter is dynamically hotter " than the stars , as suggested by davis & white ( 1996 ) . in 4 we described how @xmath9 varies as functions of the two relative parameters of the `` universal '' dark matter distribution ( equation 8) the ratio of dark matter to stellar scale lengths , @xmath301 , and the ratio of dark - to - luminous matter within @xmath264 , @xmath302_{r_{\rm max}}$ ] ( see figures 4a - b and 5 ) . because we attempt to match only the single global observable @xmath9 , there is an allowed range in the details of the dark matter spatial distribution ; however , we have derived absolute limits on the dark matter parameters required to obtain any particular value of @xmath9 ( see figures 6a - b ) . the observations do not require that dark matter dominate the inner luminous regions of elliptical galaxies : more than half of the mass within @xmath5 is baryonic for models with @xmath303 if @xmath158 . the most natural explanation of the tendency for galaxies with lower stellar temperatures to have larger gas - to - stellar temperature ratios is that @xmath132_{r_{\rm max}}$ ] decreases with @xmath27 in such a way that , on average , the total mass - to - light ratio inside @xmath59 is nearly independent of optical luminosity . this ratio , @xmath304m@xmath13l@xmath14 , is exactly what is predicted for mass models of elliptical galaxies designed to explain the gravitational shear of background field galaxies . if one specifies a scaling relation for the dark halo concentration , one can extend the dark matter distribution out to the virial radius and calculate the total baryon fraction and mass - to - light ratio . when we attempt to embed our models within the cdm theory of hierarchical halo formation , the implied dark matter scaling badly fails to reproduce the observed @xmath0@xmath1 relation unless smaller galaxies lose an increasingly larger fraction of their initial baryonic content ( see figure 7b ) , such that the average @xmath2 galaxy has lost most of its initial baryonic mass . alternatively , the global dark - to - luminous mass ratio could be constant if the dark halo concentration declines much more steeply with virial mass than cdm models predict , so that the decrease in @xmath132_{r_{\rm max}}$ ] for larger systems is a result of a more diffuse dark matter halo rather than a less massive ( relative to the stars ) halo . in this latter scenario , dark matter may become increasingly important inside @xmath5 as @xmath27 decreases , becoming dominant for @xmath305 . this deviation from cdm predictions of dark halo scaling could conceivably be due to a relatively flat primordial fluctuation spectrum on mass scales @xmath306m@xmath18 or to the effects on the dark matter density profile of the evolution of the baryonic component ; to date , large scale structure numerical simulations that can resolve halos on galactic scales _ and _ include a dissipational component have not been attempted . in this paper we have shown that the observed relationship between optical velocity dispersions and x - ray temperatures in giant elliptical galaxies implies that they have dark matter halos with @xmath12m@xmath13l@xmath14 within @xmath8 . in the future , we plan to fully utilize the available x - ray and optical imaging and spectroscopic data for our sample on a case - by - case basis in order derive the mass distributions in individual galaxies in the highest possible detail and to investigate the scatter in the @xmath0@xmath1 relationship . for galaxies with x - ray temperature profiles , we will be able to constrain the detailed form of the dark matter distribution as well as its integrated properties . we thank lars hernquist for reminding us of the consistency considerations of ciotti & pelligrini ( 1992 ) , and richard mushotzky for feedback on the original manuscript . comments from an anonymous referee led to significant improvements in the quality of this paper . r.e.w . acknowledges partial support from nasa grants nag 5 - 1718 and nag 5 - 1973 . suppose the stellar density profile is given by a broken power - law of the form @xmath307 the projected surface mass , @xmath308 can be recast , utilizing the change of variable @xmath309 , into the form @xmath310,\ ] ] where @xmath311 @xmath312 , and @xmath313 . note that the bracketed term in equation ( a3 ) has the asymptotic slopes at small and large @xmath31 that one would expect from dimensional analysis ( _ i.e. _ @xmath314 ) . the definite integral of equation ( a4 ) can be evaluated as follows ( @xcite ) : @xmath315 where @xmath316 is one - half the asymptotic density slope as @xmath317 , @xmath318 is the beta function , and @xmath319 is the hypergeometric function ( @xcite ) . we have empirically found that we can introduce a parameter @xmath320 , recast equation ( a3 ) as @xmath321 where @xmath322 and determine a value of @xmath320 ( for any @xmath323 ) that results in a very nearly constant @xmath324 as a function of @xmath31 ( with no significant structure near @xmath325 ) . equation ( a6 ) has the expected slope as @xmath326 ; and , we find @xmath327 as @xmath328 ( the `` @xmath329 model '' ) , and @xmath330 as @xmath331 . we conclude that the projection of a density profile of the form expressed by equation ( a1 ) is well - described by equation ( 2 ) ( at least for @xmath38 ) , thus perhaps lending some physical justification to its use . bahcall , n. a. , lubin . l. m. , dorman , v. 1995 , , 447 , l81 bertola , f. , pizzella , a. , persic , m. , & salucci , p. 1993 , , 416 , l45 brainerd , t. g. , blandford , r. d. , & smail , i. 1996 , , 466 , 623 buote , d. a. , & canizares , c. r. 1997 , , 474 , 650 carollo , c. m. , de zeeuw , t. , van der marel , r. p. , & danziger , i. j. 1995 , , 441 , l25 ciotti , l. , & pelligrini , s. 1992 , , 255 , 561 davis , d. s. , & white , r. e. iii 1996 , , 470 , l35 ( dw ) de paolis , f. , ingrosso , g. , & strafella , f. 1995 , , 438 , 83 de vaucouleurs , g. , et al . 1991 , third reference catalogue of bright galaxies ( new york : springer ) de zeeuw , t. , & franx , m. 1991 , , 29 , 239 fabbiano , g. 1989 , , 27 , 87 faber , s. m. , et al . 1997 , , 114 , 177 faber , s. m. , wegner , g. , burstein , d. , davies , r. l. , dressler , a. , lynden - bell , d. , & terlevich , r. j. 1989 , , 69 , 763 fabian , a. c. , thomas , p. a. , white , r. e. iii , & fall , s. m. 1986 , , 221 , 1049 forman , w. , jones , c. , & tucker , w. 1985 , , 293 , 102 franx , m. , van gorkom , j. , & de zeeuw , t. 1994 , , 436 , 642 gradshteyn , i. s. , and ryzhik , i. m. 1980 , table of integrals , series , and products , new york : academic press , 287 and 413 gerhard , o. , jeske , g. , saglia , r. p. , & bender , r. 1998 , , 295 , 197 gibson , b. k. 1997 , , 290 , 471 griffiths , r. e. , casertano , s. , i m , m. , & ratnatunga , k. u. 1996 , , 282 , 1159 hernquist , l. 1990 , , 356 , 359 irwin , j. a. , & sarazin , c. l. 1996 , , 471 , 683 kochanek , c. 1995 , , 445 , 559 lin , h. , kirshner , r. p. , shectman , s. a. , landy , s. d. , oemler , a. , tucker , d. l. , & schechter , p. l. 1996 , , 464 , 60 loewenstein , m. 1994 , , 431 , 91 loewenstein , m. , & mathews , w. g. 1987 , , 319 , 614 loewenstein , m. , & mushotzky , r. f. 1996 , , 466 , 695 marzke , r. o. , huchra , j. p. & geller , m. j. 1994 , , 428 , 43 matsushita , k. 1997 , ph.d . thesis , university of tokyo mulchaey j. s. , davis , d. s. , mushotzky , r. f. , & burstein , d. 1996 , , 456 , 80 mushotzky , r. f. 1998 , proc . usa , 95 , 72 mushotzky , r. f. , loewenstein , m. , awaki , h. , makishima , k. , matsushita , k. , & matsumoto , h. 1994 , , 436 , l79 navarro , j. f. , frenk , c. s. , & white , s. d. 1997 , , 490 493 persic , m. , salucci , p. , & stel , f. 1996 , , 281 , 27 rix , h .- w . , de zeeuw , t. , c. m. , cretton , n. , van der marel , r. p. & carollo , c. m. 1997 , , 488 , 702 saglia , r. p. et al . 1993 , , 403 , 567 white , r. e. iii , & davis , d. s. 1998 , in preparation zepf , s. e. , & silk , j. 1996 , , 466 , 114

we study the effect of a magnetic field on the behaviour of a conducting elastic rod subject to a novel set of boundary conditions that , in the case of a transversely isotropic rod , give rise to exact helical post - buckling solutions . the equations used are the geometrically exact kirchhoff equations and both static ( buckling ) and dynamic ( whirling ) instability are considered . critical loads are obtained explicitly and are given by a surprisingly simple formula . by solving the linearised equations about the ( quasi-)stationary solutions we also find secondary instabilities described by bifurcations , the usual signature of incipient ` breathing ' modes . the boundary conditions can also be used to generate and study helical solutions through traditional non - magnetic buckling due to compression , twist or whirl .

1201.6351

a straight current - carrying wire held in tension between pole faces of a magnet is well known to buckle into a ( roughly ) helical configuration at a critical current ( see fig . [ fig : hel : exp - setup ] ) . a photograph of this phenomenon is shown in section 10.4.3 of @xcite , where a linear stability analysis is carried out for a simple string model . ( a string is here meant to be a perfectly flexible elastic wire . ) the problem was studied by wolfe @xcite by means of a rigorous bifurcation analysis for a ( nonlinearly elastic ) string suspended between fixed supports and placed in a uniform magnetic field directed parallel to the undeformed wire . he found that an infinite number of solution branches bifurcate from the trivial straight solution , much like in the euler elastica under compressive load . in this case the non - trivial solutions are exact helices . that this should be so , is easily explained by the fact that the ( lorentz ) body force is everywhere normal to the deformed configuration and hence the wire necessarily in a uniform state of tension . some ( statics ) stability results ( i.e. , minimisation of the potential energy ) were also obtained , indicating that the first branch of solutions is stable while the others are unstable . in a subsequent paper wolfe @xcite extends the analysis to a uniformly rotating ( whirling ) string and shows again the existence of bifurcating branches of whirling non - trivial solutions . due to centrifugal effects no closed - form solutions could be obtained in this case . this result was further extended by healey @xcite using equivariant bifurcation theory in order to deal with the symmetries of the problem . wolfe also considered a conducting _ rod _ in a uniform magnetic field @xcite . in addition to extension a rod can undergo flexure , torsion and shear , and for the case of welded boundary conditions it was found that in certain cases bifurcation occurs , with the usual infinity of non - trivial equilibrium states . all the works cited above were content with showing the existence of bifurcating solutions and did not study their post - buckling behaviour . in this paper we consider the post - buckling behaviour of a conducting rod . wolfe considered welded boundary conditions in @xcite and encountered degeneracies ( even - dimensional eigenspaces ) because of rotational symmetry of the problem . in previous work @xcite we showed that further complications occur and that magnetic buckling of a welded transversely isotropic rod ( i.e. , a rod with unequal bending stiffnesses about the two principal axes of its cross - section ) is described by a remarkably degenerate pitchfork bifurcation . wolfe also reported numerical evidence of helical post - buckling solutions . however , exact helical solutions can not be supported by ( coaxial ) welded boundary conditions . here we formulate a novel set of what we call ` coat hanger ' boundary conditions that do support ( i.e. , are compatible with ) exact helical solutions , and show that subject to these boundary conditions an isotropic rod does indeed buckle ( exclusively ) into a helix , or more precisely , that there is an infinite series of helical modes bifurcating at increasing load , each successive mode having one more ( half ) helical turn . all helical solutions can be obtained explicitly and it is found that the pitchfork bifurcations for these coat hanger boundary conditions are non - degenerate and that the critical loads are given by a remarkably simple formula . unlike in string buckling a rod does not require a tensile force in the trivial state , but we allow for such an applied force as well . the pertinent dimensionless parameter that governs buckling measures the product of current and magnetic field against the bending force . we also study steady whirling solutions for which we introduce a rotating coordinate system . this extends wolfe s analysis of whirling strings to whirling rods . an interesting feature of helical solutions is that since all points on a helix have equal distance to the whirling axis , and are therefore equally affected by centrifugal forces , solutions remain helical when spun . we perform a stability analysis by computing eigenvalues of the linearised boudary - value problem about a ( quasi - stationary ) whirling solution . for this we use a continuation ( or homotopy ) approach that takes advantage of the fact that exact expressions for the ( imaginary ) eigenvalues can be obtained in an appropriate limit ( no spin , no magnetic field ) . the eigenvalues in this limit are then traced as system parameters are varied . whirl tends to destabilise the helical solutions , but stable solutions can be obtained by adding the effect of internal viscoelastic damping . we find hopf bifurcations on the first bifurcating branch where a stable whirling solution becomes unstable under an increase of the angular velocity . we also briefly consider anisotropic rods . critical loads can still be obtained analytically , but these rods buckle into coiled but non - helical solutions . secondary instabilities are found due to hamiltonian - hopf bifurcations , a common signature of ` breathing ' or ` flutter ' instabilities in mechanical systems . helical solutions are widely studied in a whole range of applications . often these solutions are thought to arise through buckling of a straight rod under the action of end loads . however , as commented above , exact helical solutions are not supported by the usual set of boundary conditions . consequently , boundary conditions are often not mentioned , or the rod is implicitly assumed to be infinitely long in order to prevent end effects @xcite . one of the contributions of this paper is to present and highlight boundary conditions for an elastic rod that do support helical solutions . one could apply these experimentally if one was interested in generating or studying helical solutions in a finite - length rod . the paper is organised as follows . in section 2 we present the rod mechanics formulation , in which the magnetic field enters the force balance equation through the lorentz body force . the coat hanger boundary conditions are introduced together with a sketch of a device that can be constructed to realise these boundary conditions in a testing rig . for the study of whirling solutions the equilibrium equations are transformed to a coordinate system rotating at constant angular velocity . we use numerical bifurcation and continuation methods to find the buckling loads and to compute post - buckling solution paths , both for the statics and dynamics case . after the nondimensionalisation in section 3 , the linearisation is presented in section 4 . section 5 first presents analytical stability results for the statics case and then introduces our continuation approach to numerical stability analysis of the full system . in section 6 results are presented in the form of bifurcation diagrams and curves of hopf bifurcations in appropriate parameter planes . conclusions are drawn in section 7 and the study closes with two appendices : one giving details about the linearised system of equations and one deriving exact buckling results for helical solutions . the latter is complementary to the bifurcation analysis ( not assuming any shape ) in section 5 ; together these analyses give a complete picture of helical magnetic buckling . we describe the elastic behaviour of a conducting cable by the kirchhoff equations for the dynamics of thin rods . the rod is assumed to be uniform , inextensible , unshearable and intrinsically straight and prismatic . the assumptions of inextensibility and unshearability are appropriate for thin rods with relatively low external ( here electrodynamic ) forces . for the background of the kirchhoff equations the reader is referred to @xcite . these equations were also used in @xcite and @xcite to analyse the dynamics of a spinning tether . let @xmath0 denote the position of the rod s centreline and let @xmath1 be a right - handed orthonormal frame of directors ( the cosserat triad ) defined at each point along the centreline . since the centreline is assumed to be inextensible we can take @xmath2 in the direction of the local tangent : @xmath3 where the prime denotes differentiation with respect to arclength @xmath4 measured along the centreline , and @xmath5 is time . the directors @xmath6 and @xmath7 will be taken to point along the principal bending axes of the cross - section ( see fig . [ fig : cosserat ] ) . the unstressed rod is taken to lie along the basis vector @xmath8 of a fixed inertial frame @xmath9 . looking at fig . [ fig : cosserat ] we note that the position vector of an arbitrary point of the rod can be expressed as @xmath10 where @xmath11 are the components of @xmath12 in the cross - section relative to @xmath13 . the rod is thus viewed as a set of infinitesimal slices centred at all @xmath4 . a one - dimensional description will be obtained by averaging of forces and moments over each cross - section . the internal traction , which is the projection of the stress tensor onto the cross - sectional plane , is given by a force which we denote by @xmath14 ( see fig . [ fig : cosserat ] ) . the resultant elastic force exerted in a section @xmath15 is given by @xmath16 where @xmath17 is an infinitesimal area element . this force can be expressed in the director basis as @xmath18 . the resultant moment in the section @xmath15 is given by @xmath19 and will be expressed as @xmath20 . the rod is assumed to carry an electric current for which we can write @xmath21 here we have assumed the current to have the same direction as the rod , which is consistent with a one - dimensional rod theory . it amounts to the assumption that the cross - section of the conducting wire is small enough to make currents within the cross - section ( eddy currents ) induced by the motion negligible . the current @xmath22 interacts with the magnetic field @xmath23 to generate a ( lorentz ) body force given by @xmath24 following @xcite we assume the magnetic field to be uniform and directed along the unstressed rod , i.e. , @xmath25 the balancing of forces and moments across an infinitesimal rod element then yields the following set of partial differential equations @xcite : @xmath26 @xmath27 where @xmath28 is the ( volumetric ) mass density , @xmath29 the cross - sectional area , @xmath30 and @xmath31 the second moment of area of the cross - section about @xmath6 and @xmath7 respectively , and @xmath32 denotes differentiation with respect to time . for a closed system of equations these balance equations need to be supplemented by constitutive relations that characterise the material behaviour of the rod . we assume the rod to be made of homogeneous isotropic linear viscoelastic material so that stress - strain relations , based on a model by valverde et al . @xcite , are @xmath33 where @xmath34 and @xmath35 are the curvatures about @xmath6 and @xmath7 , respectively , while @xmath36 is the twist about @xmath2 . the constant @xmath37 is the viscoelastic coefficient of the material , @xmath38 is young s modulus , @xmath39 is the shear modulus and @xmath40 is the second moment of area of the section about @xmath2 . we shall assume that the section is symmetric with respect to the principal axes , in which case @xmath41 . the @xmath42 are the components of the curvature vector @xmath43 which governs the evolution in space of the frame of directors as one moves along the centreline : @xmath44 the constitutive relations ( [ eq : constitutive1 ] ) can be used to replace the @xmath42 in ( [ eq : twist ] ) by moments , after which the equations ( [ eqtangent1 ] ) , ( [ eq : lmomentum2 ] ) , ( [ eq : amomentum2 ] ) and ( [ eq : twist ] ) form a system of 18 differential equations for the 18 unknowns @xmath45 . + * remark : * we ignore in this study secondary electrodynamic effects ( such as an induced emf and hence additional current in the conductor ) as a result of the motion of the wire in the magnetic field @xcite . since we are considering a steadily rotating wire these effects would be null on the configuration of the wire . however , the same would not be true for the stability analysis , which considers arbitrary time - dependent perturbations . we assume that these induction effects are negligible . we shall also be interested in steadily rotating solutions and therefore we transform the equilibrium equations ( [ eq : lmomentum2 ] ) and ( [ eq : amomentum2 ] ) to a coordinate frame @xmath46 that rotates with constant angular velocity @xmath47 about the @xmath8 axis ( and the axis of the rod in its trivial unstressed state ) . noting that the derivative with respect to time of an arbitrary vector @xmath48 is given by @xmath49 where @xmath50 indicates the derivative with respect to time in the inertial frame and @xmath51 stands for the derivative with respect to time in the moving frame , the equations ( [ eq : lmomentum2 ] ) and ( [ eq : amomentum2 ] ) expressed relative to @xmath46 become @xmath52 @xmath53 the second term on the right - hand side of equation ( [ eq : lmomentum4 ] ) is the coriolis force , while the third term is the centrifugal force , as a result of the rotating coordinate system . steadily rotating ( whirling ) solutions satisfy the equations ( [ eq : lmomentum4 ] ) and ( [ eq : amomentum4 ] ) with the dotted variables set to zero : @xmath54 @xmath55 the other equations ( [ eqtangent1 ] ) and ( [ eq : twist ] ) do not change their form , but all vectors are now to be considered as expressed relative to the rotating frame @xmath46 . statical solutions are simply obtained by setting @xmath56 equal to zero . for a well - posed problem the final 18 odes require 18 boundary conditions to be specified , which we do next . helical solutions in rods are usually studied in infinitely long rods , which avoids the need for imposing boundary conditions . indeed , it is not immediately clear how an exact helix can be supported : the boundary conditions can not be simply welded as no two points on a helix have coaxial tangents , nor can they be simply pinned because a helix has curvature and therefore carries a bending moment . here we formulate a set of boundary conditions that support exact helical solutions . we call them coat hanger boundary conditions , for obvious reasons . consider fig . [ fig : hel : bc ] where a rod is suspended between two axes @xmath57 ( at @xmath58 ) and @xmath59 ( at @xmath60 ) lying in two parallel planes normal to @xmath61 . axis @xmath59 is taken to be fixed in space , while @xmath57 is free to move along @xmath61 . we assume the axes to have a fixed relative rotation @xmath62 , i.e. , @xmath63 . the rod is free to hinge about and slide along both @xmath57 and @xmath59 . for definiteness we assume that the rod is mounted in such a way that both axes @xmath57 and @xmath59 are directed along the vector @xmath7 in the rod s cross - section . this situation is described by the following boundary conditions : @xmath64 at @xmath58 , and @xmath65 at @xmath60 , where the position vector has been decomposed as @xmath66 and @xmath67 is an applied end force ( positive for tension ) . conditions ( [ eq : bc41 ] ) and ( [ eq : bc46 ] ) restrict the movement of the ends of the rod to the planes spanned by @xmath68 and @xmath69 , respectively . to these 12 conditions we have to add conditions that ensure the orthonormality of the director basis , for which we can take @xmath70 for a total of 18 boundary conditions , as required . we shall take @xmath71 so that the initial rod , lying straight along @xmath61 , is untwisted . this choice implies that any helical solutions will have an integer number of half helical periods . also , the directors , and hence the cross - section of the rod , will make a half - integer number of turns between @xmath58 and @xmath72 . without loss of generality we may choose @xmath73 , so that at the ends of the initial rod the directors @xmath1 are aligned with @xmath46 . since @xmath57 and @xmath59 are chosen aligned , there is a rigid - body degree of freedom of translation of any solution along these axes . to eliminate this degeneracy we replace condition ( [ eq : bc45 ] ) above by @xmath74 where the slide @xmath75 along @xmath59 ( see fig . [ fig : hel : bc ] ) is chosen as follows . first note that a helix is a curve of constant axial radius @xmath76 and ( total ) curvature @xmath77 . the two are related by @xmath78 , where @xmath79 is the helical angle defined by @xmath80 if @xmath61 is along the axis of the helix ( the angle @xmath81 is usually called the pitch angle ) . since the rod is hinged about @xmath7 , we have @xmath82 . so , in order to ensure that any bifurcating helix is centred at @xmath61 we take @xmath83 giving a nonlinear boundary condition . note that @xmath34 can here be taken with its sign , so that ( [ eq : bc45a ] ) also specifies which way the rod moves along @xmath59 . when the rod buckles , the axis @xmath57 lifts up and the rod is free to find its own radius @xmath84 . we stress that condition ( [ eq : bc45a ] ) has no effect on the bifurcation behaviour . in particular , it does not suppress any non - helical solutions . it merely ensures that if a helical solution bifurcates it will be centred at the axis of rotation . this is important when we start rotating the axes @xmath57 and @xmath59 about @xmath61 . a centred helix will experience a uniform centrifugal force and is therefore expected to remain helical . of course the above coat hanger boundary conditions merely _ allow _ for helical solutions . they need not exist . however , if the equilibrium equations do have helical solutions and one - parameter curves of such solutions intersect the trivial path of straight solutions , then one might expect to detect them as ( pitchfork ) bifurcations at critical buckling loads . the results presented in section [ results ] show that this is indeed the case . we make the system of equations dimensionless by scaling the variables in the following way @xmath85 , \quad \bar \omega=\frac{\omega}{\omega_c } , \quad \bar{{\mbox{\boldmath $ x$}}}=\frac{{\mbox{\boldmath $ x$}}}{l } , \\ & \bar { { \mbox{\boldmath $ f$}}}={\mbox{\boldmath $ f$ } } \frac{l^2}{ei_1},\quad \bar{t}=t \frac{l^2}{e i_1 } , \quad \bar{{\mbox{\boldmath $ m$}}}={\mbox{\boldmath $ m$}}\frac{l}{e i_1 } , \quad \bar{{\mbox{\boldmath $ \kappa$}}}= { \mbox{\boldmath $ \kappa$ } } l. \end{split } \label{eq : nondim}\ ] ] here @xmath86 is a reference characteristic bending frequency of the rod . with this nondimensionalisation the equations become ( dropping the overbars for simplicity and letting a prime denote @xmath87 and an overdot @xmath88 ) : @xmath89 , \label{eq : amomentum}\\ & { \mbox{\boldmath $ x$}}'={\mbox{\boldmath $ d$}}_3 , \label{eq : tangent}\\ & { \mbox{\boldmath $ d$}}_i'={\mbox{\boldmath $ \kappa$ } } \times { \mbox{\boldmath $ d$}}_i , \label{eq : twist2}\end{aligned}\ ] ] and the constitutive relations can be written as @xmath90 where the dimensionless parameters are @xmath91 and ( @xmath92 ) is equal to poisson s ratio . for the boundary conditions we can still use ( [ eq : bc36 ] ) to ( [ eq : bc47 ] ) if we assume that they now refer to dimensionless variables and that the right - hand conditions are imposed at @xmath93 . we consider whirling solutions ( relative equilibria ) that are stationary in the moving frame @xmath46 . such solutions are found by solving the set of equations ( [ eq : lmomentum])([eq : constitutive ] ) with the dotted variables set to zero ( thus obtaining an ode ) . to study their stability we linearise the full pde ( [ eq : lmomentum])([eq : constitutive ] ) about these whirling solutions . we follow the approach in @xcite , which is similar to approaches in @xcite . the stability of static ( non - whirling ) solutions can be investigated by simply setting the angular velocity @xmath56 to zero . we start our perturbation analysis by writing @xmath94 where @xmath95 are the basis vectors of a quasi - stationary solution , @xmath96 are the basis vectors of a time - dependent perturbation and @xmath97 is a small bookkeeping parameter introduced to separate scales . note that , in order to preserve orthonormality to @xmath98 ( @xmath99 ) , we must have @xmath100 where the matrix @xmath101 is skew - symmetric and can be written as @xmath102 thus , the nine components of the director basis perturbation are described by only three independent parameters , and if we introduce @xmath103 ( with respect to the unperturbed director basis ) then the perturbed director basis can be expressed as @xmath104 using ( [ per : dirfinal ] ) , the perturbation of an arbitrary vector @xmath105 can be written on the basis @xmath106 as @xmath107 { \mbox{\boldmath $ d$}}_i^0 + o(\delta^2 ) , \label{per : varfinal}\ ] ] where @xmath108 denotes the component along @xmath109 and time and space dependence of the variables have been suppressed for the sake of simplicity @xcite . applying this perturbation scheme to the pdes ( [ eq : lmomentum])([eq : constitutive ] ) and the boundary conditions , we arrive at an @xmath110 nonlinear ode for the quasi - stationary solutions and an @xmath98 linear pde governing their stability . the @xmath110 equations are time - independent . recalling that @xmath111 , we find the @xmath110 terms of the linear momentum equation ( [ eq : lmomentum ] ) , projected on the director basis @xmath106 , to give @xmath112 where subscripts are used to indicate components relative to the basis vectors @xmath106 ( but the @xmath109 components are relative to @xmath46 ) . similarly , the @xmath110 term of the angular momentum equation ( [ eq : amomentum ] ) , projected on the director basis @xmath106 gives @xmath113 the @xmath110 term of equation ( [ eq : tangent ] ) can be expressed as @xmath114 and the twist equation ( [ eq : twist2 ] ) by @xmath115 where @xmath116 . the @xmath110 term of constitutive relations ( [ eq : constitutive ] ) can be expressed as @xmath117 which can be used to express the @xmath118 in ( [ stat : twist ] ) in terms of the moments @xmath119 . proceeding in the same way , the @xmath110 part of the boundary conditions is given by @xmath120 @xmath121 where ( [ eq : stat : bc46 ] ) is the @xmath110 contribution from ( [ eq : bc45a ] ) , with the radius of the helix also affected by the perturbation scheme , i.e. , @xmath122 . the @xmath98 part of the linear momentum equation ( [ eq : lmomentum ] ) can be written as @xmath123 where the @xmath124 matrices @xmath125 are given in appendix a. here we have expressed @xmath126 relative to @xmath106 and @xmath127 relative to @xmath46 . for the @xmath98 part of the angular momentum equation ( [ eq : amomentum ] ) we can write @xmath128 where the matrices @xmath129 are again given in appendix a. @xmath130 is expressed relative to @xmath106 . the 9 twist equations ( [ eq : twist2 ] ) at @xmath98 are reduced to only 3 independent equations that relate @xmath131 and @xmath132 as @xmath133 introducing these relations into the @xmath98 part of the constitutive relations gives @xmath134 where the matrices @xmath135 are given in appendix a. finally , the @xmath98 part of equation ( [ eq : tangent ] ) yields @xmath136 applying the perturbation scheme to the boundary conditions at @xmath98 , we obtain @xmath137 after elimination of the @xmath118 by means of ( [ stat : constitutive ] ) , the set of 12 equations ( [ per : lmomentum ] ) , ( [ per : amomentum ] ) , ( [ per : constitutive ] ) and ( [ per : tangent ] ) together with the 12 boundary conditions ( [ per : bc1])([per : bc12 ] ) , with appropriate initial conditions form a well - posed initial - boundary - value problem . since we are interested in stability of solutions we look for solutions of the @xmath98 equations of the form @xmath138 when these expressions are inserted into ( [ per : lmomentum])([per : tangent ] ) a linear eigenvalue problem for a 12-dimensional ode is obtained in terms of the variables @xmath139 . the eigenvalue @xmath140 measures the growth of small perturbations and is to be found as part of the solution . eigenvalues come as complex conjugate pairs . a whirling state is unstable if at least one of the ( in general infinitely many ) @xmath140 s has positive real part . to solve a real system of equations we split the eigenvalues and variables ( eigenfunctions ) into real and imaginary parts , @xmath141 , @xmath142 , @xmath143 , @xmath144 and @xmath145 . the equations ( [ per : lmomentum])([per : tangent ] ) along with the boundary conditions ( [ per : bc1])([per : bc12 ] ) , are similarly split into real and imaginary parts . thus we end up with a doubled 24-dimensional linearised boundary - value problem . the trivial solution of the @xmath110 equations ( [ stat : forces1])([stat : constitutive ] ) , representing a straight and untwisted rod , is given by @xmath146 . \label{trivial_sol}\ ] ] it satisfies the coat hanger boundary conditions . for the statics case ( @xmath147 ) without end force ( @xmath148 ) the @xmath98 equations ( [ per : lmomentum ] ) , ( [ per : amomentum ] ) , ( [ per : constitutive ] ) , ( [ per : tangent ] ) about this trivial solution , on inserting ( [ per : sol1])([per : sol4 ] ) , can be written as @xmath149 with boundary conditions @xmath150 while @xmath151 , @xmath152 . note that the torsional ( @xmath153 ) modes decouple from the bending ( @xmath154 ) modes . to find the static magnetic buckling loads we set @xmath155 . the bending equations then reduce to @xmath156 subject to @xmath157 on setting @xmath158 we obtain the characteristic equation @xmath159 with solutions @xmath160 , where @xmath161 . application of the boundary conditions ( [ bc_x ] ) to the general solution @xmath162 leads to the remarkably simple condition : @xmath163 these critical loads correspond to pitchfork bifurcations where non - trivial solutions bifurcate from the trivial straight solution . we stress that the above calculation is only possible for the statics case . if @xmath164 then the @xmath165 and @xmath166 equations do not decouple and no simple characteristic equation is obtained . however , bifurcating branches of helical solutions , and hence critical loads , can be computed explicitly , even for non - zero @xmath56 ; see appendix b. we shall call the case where @xmath148 , @xmath167 , @xmath147 and @xmath168 the _ unperturbed problem_. for this problem explicit expressions can be obtained for the eigenvalues of the linearisation about the straight solution . the @xmath165 and @xmath166 equations in ( [ lin_eqs ] ) decouple into two fourth - order beam equations : @xmath169 subject to boundary conditions ( [ bcs ] ) . since we anticipate imaginary eigenvalues we set @xmath170 , @xmath171 , @xmath172 , and find for the @xmath165 equation @xmath173 while for the @xmath166 equation @xmath174 where @xmath175 , @xmath176 , @xmath177 , @xmath178 are non - negative real numbers . the general solutions are @xmath179 application of the boundary conditions ( [ bcs ] ) leads to @xmath180 the first equation implies @xmath181 the second equation is transcendental and needs to be solved numerically to obtain the eigenvalues @xmath182 , for instance by using a newton - raphson scheme . meanwhile , the torsional eigenvalues for the @xmath153 equation in ( [ lin_eqs ] ) are given by @xmath183 these are all the eigenvalues for the unperturbed problem . they will be used as starting values in the numerical procedure described next . the main idea is to use the known eigenvalues in the unperturbed problem as starting values in a continuation procedure in order to compute the eigenvalues and corresponding eigenfunctions for general values of the parameters @xmath67 , @xmath184 , @xmath56 and @xmath185 . for this we use the well - tested code auto @xcite ( specifically auto2000 ) . auto solves boundary - value problems by means of orthogonal collocation . it requires a starting solution and can then trace out solution curves as a parameter of the problem is varied . bifurcations are detected where branches of solutions intersect . at such points auto is able to switch branches and compute curves of bifurcating solutions . our procedure takes advantage of the fact that @xmath140 appears only quadratically in the linearisation ( [ per : lmomentum ] ) , ( [ per : amomentum ] ) , ( [ per : constitutive ] ) and ( [ per : tangent ] ) if @xmath167 and @xmath147 . to explain the method consider the typical @xmath98 equation @xmath186 where @xmath187 and @xmath188 are functions of the @xmath110 solution . writing @xmath189 , @xmath190 , we can decompose the @xmath191 equation into @xmath192 the important thing to note here is that these equations decouple into two identical equations if the eigenvalue is either imaginary ( @xmath193 ) or real ( @xmath194 ) . this suggests the following sequence of steps , involving boundary - value problem of increasing dimension , to compute eigenvalues of static or uniformly whirling solutions . 1 . consider the unperturbed problem of section [ sect : unperturbed ] and , noting that all eigenvalues are purely imaginary , solve the 30-dimensional system of 18 @xmath110 equations and one 12-dimensional system for the imaginary part of the @xmath98 equations ( cf . the @xmath166 equation in ( [ re_im_eq ] ) ) . set @xmath193 and use @xmath195 as the continuation parameter in auto in order to compute the eigenvalues ( instead of solving the transcendental equations in section [ sect : unperturbed ] ) . these eigenvalues will show up as branching points ( bp ) , or pitchfork bifurcations , as eigenvalues by definition are those values for which non - zero bvp solutions exist . by symmetry it is only necessary to consider @xmath196 . 2 . keeping the same 30-dimensional system , switch branches at a bp to compute ( ` grow ' ) the corresponding ( imaginary ) eigenfunction . since the equations are linear the value of @xmath195 will not change in this run . for later use we monitor the non - zero solution by means of some measure @xmath197 ( not necesarily a proper norm ) on the space @xmath198 of imaginary linearised variables . 3 . now consider the full system of 42 equations ( 18 @xmath110 equations and two sets of 12-dimensional @xmath98 equations ( cf . ( [ re_im_eq ] ) ) . fix the measure @xmath197 on the imaginary part and release @xmath199 instead in order to compute the real eigenfunction ( since the imaginary part of the solution is fixed there is only one branch of solutions through the starting point and there is nowhere else to go for the continuation but to ` grow ' the real eigenfunction ) . again we monitor this function by means of a suitable measure @xmath200 . in this run neither @xmath199 nor @xmath195 will change . + this approach works because the solution obtained in step 2 also solves the full 42-dimensional system when the extra 12 variables @xmath201 are set to zero . this is a consequence of the fact that the real and imaginary parts of the @xmath98 equations decouple if @xmath193 , as a result of the quadratic dependence of the eigenvalue problem on @xmath140 ( cf . ( [ re_im_eq ] ) ) . steps 2 and 3 can be performed for as many of the bps computed in step 1 as required and will give the corresponding eigenvalues and eigenfunctions . once these have been obtained both measures @xmath200 and @xmath197 can be fixed and an extra system parameter such as @xmath185 or @xmath56 released in order to trace the eigenvalues ( and hence monitor stability changes ) as system parameters are varied . ( note that fixing @xmath200 and @xmath197 makes sense as eigenfunctions are only defined up to a multiplicative factor . ) the above 3-step procedure is not limited to linearisations about the trivial straight solution . it can be applied to any _ starting solution _ that has no eigenvalue with both @xmath199 and @xmath195 non - zero , as these would not be picked up in step 1 . ( it is of course no problem if eigenvalues become fully complex ( hopf bifurcation ) in the course of further continuations . ) for instance , we find that at the first critical @xmath185 , given by ( [ char_coat ] ) , the lowest conjugate pair of eigenvalues @xmath202 goes to zero and becomes a real pair of eigenvalues , signalling a stability change of the straight rod . the ( first - mode ) solution bifurcating at this point is stable with all eigenvalues being imaginary and the above procedure can be applied to find the eigenvalues . we end this section with a few comments : 1 . there are infinitely many eigenvalues and the above procedure only finds the lowest order ones . this is of course a limitation of any numerical scheme . we find that eigenvalues vary slowly with system parameters , suggesting that stability is governed by the lowest - order eigenvalues . we typically consider 5 or 6 eigenvalues . 2 . note that in steps 1 and 2 above we could not have taken the full 42-dimensional system of equations as that would have made the branching points ( pitchfork bifurcations ) degenerate and auto would not detect a bp . this is because if @xmath193 ( or @xmath194 ) the two sets of 12-dimensional linearised equations are identical ( cf . ( [ re_im_eq ] ) ) . ) . ] fig . [ fig : hel : bif - helix ] shows the bifurcation diagram obtained by varying the magnetic field parameter @xmath185 in the isotropic case ( @xmath203 ) . throughout this entire section the dimensionless parameters taken , unless stated otherwise , are those listed in table [ table : hel : parameters ] , where realistic dimensional parameters are also given . pitchfork bifurcations on the trivial branch occur at @xmath204 , @xmath205 , @xmath206 , etc . , in agreement with ( [ char_coat ] ) ( only the positive-@xmath207 branch is shown ) . the bifurcating solutions are found to be exact helices and therefore the dimensionless helical radius @xmath208 is used as solution measure on the vertical axis . a slight complication in computing this diagram occurs because of the denominator in ( [ eq : radius ] ) , which is zero for the straight rod . however , this problem is easily resolved by replacing boundary condition ( [ eq : bc45a ] ) by @xmath209 along the trivial branch and switching back to ( [ eq : bc45a ] ) once an incipient non - trivial solution has been obtained . .parameters used for the coat hanger boundary conditions . [ cols="^,^,^,^",options="header " , ] [ table : hel : parameters ] -@xmath61 } and \{@xmath210-@xmath61 } projections of the first 4 helical modes at constant curvature @xmath211 . values of @xmath185 are : 61.68 ( b1 ) , 270.69 ( b2 ) , 868.72 ( b3 ) and 2024.70 ( b4).,title="fig : " ] -@xmath61 } and \{@xmath210-@xmath61 } projections of the first 4 helical modes at constant curvature @xmath211 . values of @xmath185 are : 61.68 ( b1 ) , 270.69 ( b2 ) , 868.72 ( b3 ) and 2024.70 ( b4).,title="fig : " ] fig . [ fig : hel : hel - modes - shape ] shows \{@xmath212-@xmath61 } and \{@xmath210-@xmath61 } projections of bifurcating solutions along the first four branches , taken at constant curvature @xmath213 . it was noted in section [ coat_hanger ] that the coat hanger boundary conditions with @xmath71 ( i.e. , parallel end supports @xmath57 and @xmath59 ) only allow helices of a half - integer number , @xmath214 , of helical turns . we find that each successive bifurcating solution in fig . [ fig : hel : bif - helix ] has one more half helical turn . for large @xmath185 the solutions approach a circular shape in the @xmath215 plane , with corresponding value @xmath216 along the vertical axis in fig . [ fig : hel : bif - helix ] . the bifurcating branches have handedness . that is , regardless of the sign of @xmath207 , the bifurcating solutions are right - handed helices if , as here , @xmath217 and would be left - handed helices if @xmath218 , i.e. , had we run @xmath185 in the other direction . [ fig : hel : hel - shape-3d ] shows three - dimensional views of two solutions along the sixth branch . they have three full turns and nicely illustrate the exact helical shape . * ( a ) ( b ) * + to @xmath219.,title="fig : " ] to @xmath219.,title="fig : " ] to @xmath220 and then switching to branch b1 . all real parts of the eigenvalues are zero . ] * ( a ) ( b ) * + fig . [ fig : hel : helix_w0-stab-0bp1bp2-imag ] show the evolution of the imaginary and real parts , respectively , of the first five pairs of eigenvalues along the trivial solution , from @xmath168 to @xmath219 . at successive pitchfork bifurcations , pairs of imaginary eigenvalues collide at zero and become real , one of the eigenvalues of the pair with positive real part , signalling that the trivial solution becomes unstable at @xmath220 , while a further loss of stability occurs at @xmath221 where a second pair of eigenvalues becomes real . [ fig : hel : helix_w0-stab-0bp1-b1-imag ] shows the evolution of the first five pairs of imaginary eigenvalues when switching at bp1 from the trivial branch to b1 , which is found to be stable . as expected , solutions along b2 are found to be unstable ( see figs [ fig : hel : helix_w0-stab-0bp2-b2-imag]a , b ) . on the helical radius of the solution along b1 in fig . [ fig : hel : bif - helix ] at @xmath223 . the dot indicates a hopf bifurcation where the real parts of two complex conjugate eigenvalues become positive ( see fig . [ fig : hel : helix_stab - b1_b35_gamma05_w - real](a ) ) and the solution loses stability . ] ) . ] * ( a ) ( b ) * + for a solution of the first branch b1 of fig . [ fig : hel : bif - helix ] for @xmath223 , bifurcation diagram in fig . [ fig : hel : bif_w_b35 ] . note that first panel only represents the positive imaginary eigenvalue , as the negative is symmetric with respect to the real axis.,title="fig : " ] for a solution of the first branch b1 of fig . [ fig : hel : bif - helix ] for @xmath223 , bifurcation diagram in fig . [ fig : hel : bif_w_b35 ] . note that first panel only represents the positive imaginary eigenvalue , as the negative is symmetric with respect to the real axis.,title="fig : " ] + * ( c ) ( d ) * + for a solution of the first branch b1 of fig . [ fig : hel : bif - helix ] for @xmath223 , bifurcation diagram in fig . [ fig : hel : bif_w_b35 ] . note that first panel only represents the positive imaginary eigenvalue , as the negative is symmetric with respect to the real axis.,title="fig : " ] for a solution of the first branch b1 of fig . [ fig : hel : bif - helix ] for @xmath223 , bifurcation diagram in fig . [ fig : hel : bif_w_b35 ] . note that first panel only represents the positive imaginary eigenvalue , as the negative is symmetric with respect to the real axis.,title="fig : " ] + * ( e ) ( f ) * + for a solution of the first branch b1 of fig . [ fig : hel : bif - helix ] for @xmath223 , bifurcation diagram in fig . [ fig : hel : bif_w_b35 ] . note that first panel only represents the positive imaginary eigenvalue , as the negative is symmetric with respect to the real axis.,title="fig : " ] for a solution of the first branch b1 of fig . [ fig : hel : bif - helix ] for @xmath223 , bifurcation diagram in fig . [ fig : hel : bif_w_b35 ] . note that first panel only represents the positive imaginary eigenvalue , as the negative is symmetric with respect to the real axis.,title="fig : " ] * ( a ) ( b ) * + of the imaginary part of the first five pairs of eigenvalues for a solution along b1 at @xmath223 and @xmath224 . the first two panels only show the positive imaginary part.,title="fig : " ] of the imaginary part of the first five pairs of eigenvalues for a solution along b1 at @xmath223 and @xmath224 . the first two panels only show the positive imaginary part.,title="fig : " ] + * ( c ) * + of the imaginary part of the first five pairs of eigenvalues for a solution along b1 at @xmath223 and @xmath224 . the first two panels only show the positive imaginary part.,title="fig : " ] * ( a ) ( b ) * + of the real part of the first five pairs of eigenvalues for a solution along b1 at @xmath223 and @xmath224.,title="fig : " ] of the real part of the first five pairs of eigenvalues for a solution along b1 at @xmath223 and @xmath224.,title="fig : " ] + * ( c ) * + of the real part of the first five pairs of eigenvalues for a solution along b1 at @xmath223 and @xmath224.,title="fig : " ] next we consider whirling motions in which the supports @xmath57 and @xmath59 are spun about @xmath8 with constant angular velocity @xmath56 . solutions remain helical , as expected . [ fig : hel : bif_w_b35 ] shows the effect of @xmath56 on a solution taken on the first bifurcating branch ( b1 ) of fig . [ fig : hel : bif - helix ] . the helical radius increases with @xmath56 . fig . [ fig : hel : bif - diag - w2 ] shows the bifurcation diagram for a fixed value of @xmath225 . pitchfork bifurcations along the trivial branch occur at @xmath226 , 247.44 , 836.80 , etc . the angular velocity tends to destabilise the helical solutions ( see fig . [ fig : hel : helix_w - stab - b1_real_imag ] , where the eigenvalues are shown as a function of @xmath56 ) . we now introduce damping by continuing in the parameter @xmath184 , and to further investigate the stability of the first mode we fix @xmath224 @xcite and perform continuation in @xmath56 . figs [ fig : hel : helix_stab - b1_b35_gamma05_w - imag ] and [ fig : hel : helix_stab - b1_b35_gamma05_w - real ] show that at @xmath147 the solution is stable and that around @xmath227 the real part of the first eigenvalue becomes positive . thus the system loses stability in a hopf bifurcation . the point where this occurs is indicated in the bifurcation diagram in fig . [ fig : hel : bif_w_b35 ] , which is still valid as @xmath184 has no effect on relative equilibria . fig . [ fig : hel : hopf - loci ] shows curves of hopf bifurcations in the @xmath56-@xmath185 parameter plane for various values of @xmath184 . at the end points of these curves ( indicated by dots ) the hopf bifurcation coalesces with the pitchfork bifurcation in which the curve b1 is created . consequently , the whirling helical solution is stable below these curves and unstable above . for small @xmath184 the curve collapses onto the line @xmath147 . note that for relatively large @xmath56 damping has a stabilising effect , but that for small positive and for negative @xmath56 damping has a destabilising effect . the latter behaviour is known from the classical linear stability theory of gyroscopic systems @xcite . : from left to right , @xmath167 , @xmath228 , @xmath229 and @xmath224 . dots indicate points where the hopf bifurcation coalesces with the primary pitchfork bifurcation . ] it is known that in the absence of damping and inertial effects ( i.e. , @xmath167 , @xmath147 ) the equations for a rod in a magnetic field have a hamiltonian structure @xcite . in hamiltonian systems a common mechanism for loss of stability is through a so - called hamiltonian - hopf bifurcation @xcite . in this bifurcation two imaginary eigenvalues move on the imaginary axis , meet at some non - zero value and then leave the axis to become a real pair , as illustrated in fig . [ fig : eigen - type ] . since eigenvalues come as conjugate pairs this event involves four eigenvalues . in structural problems hamiltonian - hopf bifurcations usually mark oscillatory instabilities such as flutter . , @xmath230 ) . bifurcating solutions are not exact helices . ] * ( a ) ( b ) * + of the first and second ( ( a ) and ( b ) ) and fifth and sixth ( ( c ) and ( d ) ) eigenvalues for an initially stable anisotropic - rod solution on the first bifurcating branch ( b1 ) . both pairs of eigenvalues undergo a hamiltonian - hopf followed by a reverse hamiltonian - hopf bifurcation . ( @xmath167 , @xmath147 , @xmath231.),title="fig : " ] of the first and second ( ( a ) and ( b ) ) and fifth and sixth ( ( c ) and ( d ) ) eigenvalues for an initially stable anisotropic - rod solution on the first bifurcating branch ( b1 ) . both pairs of eigenvalues undergo a hamiltonian - hopf followed by a reverse hamiltonian - hopf bifurcation . ( @xmath167 , @xmath147 , @xmath231.),title="fig : " ] + * ( c ) ( d ) * + of the first and second ( ( a ) and ( b ) ) and fifth and sixth ( ( c ) and ( d ) ) eigenvalues for an initially stable anisotropic - rod solution on the first bifurcating branch ( b1 ) . both pairs of eigenvalues undergo a hamiltonian - hopf followed by a reverse hamiltonian - hopf bifurcation . ( @xmath167 , @xmath147 , @xmath231.),title="fig : " ] of the first and second ( ( a ) and ( b ) ) and fifth and sixth ( ( c ) and ( d ) ) eigenvalues for an initially stable anisotropic - rod solution on the first bifurcating branch ( b1 ) . both pairs of eigenvalues undergo a hamiltonian - hopf followed by a reverse hamiltonian - hopf bifurcation . ( @xmath167 , @xmath147 , @xmath231.),title="fig : " ] + * ( a ) ( b ) * + ( @xmath167 and @xmath147 ) of the imaginary part ( real part equals zero ) of the third , fourth and seventh pairs of eigenvalues for non - circular cross - section , @xmath231.,title="fig : " ] ( @xmath167 and @xmath147 ) of the imaginary part ( real part equals zero ) of the third , fourth and seventh pairs of eigenvalues for non - circular cross - section , @xmath231.,title="fig : " ] + * ( c ) * + ( @xmath167 and @xmath147 ) of the imaginary part ( real part equals zero ) of the third , fourth and seventh pairs of eigenvalues for non - circular cross - section , @xmath231.,title="fig : " ] , @xmath231 . all curves terminate on the dotted curve of pitchfork bifurcations given by ( [ char_coat ] ) . we have instability of the solution inside the regions bounded by solid and dotted lines . @xmath232 labels the curve generated by the @xmath233th and @xmath234th eigenvalue . ] no hamiltonian - hopf bifurcation was found in our study of the isotropic rod in section [ coat_statics ] . however , after breaking the cross - sectional symmetry by taking @xmath235 hamiltonian - hopf bifurcations are found to occur under @xmath185 continuation . [ fig : hel : bif - diag - r055w0 ] shows a bifurcation diagram for @xmath230 . pitchfork bifurcations along the trivial branch are found at @xmath236 , 184.56 , 622.55 , etc . , in agreement with ( [ char_coat ] ) . bifurcating solutions are not pure helices . figs [ fig : hel : hopf1 ] and [ fig : hel : hopf2 ] show the eigenvalue behaviour for a solution on the first bifurcating branch , b1 , in fig . [ fig : hel : bif - diag - r055w0 ] . the first and second eigenvalues collide at @xmath237 , where the solution becomes unstable . the fifth and sixth eigenvalues then collide at @xmath238 but shortly after become imaginary again in a reverse hamiltonian - hopf bifurcation at @xmath239 . the first and second eigenvalues then also become imaginary again at @xmath240 , restabilising the solution ( as far as we checked no other eigenvalues cause instability , see fig . [ fig : hel : hopf2 ] , where @xmath241 , @xmath242 and @xmath243 are shown ) . fig . [ fig : hel : b_r - hopf ] shows loci of hamiltonian - hopf bifurcations in the @xmath244-@xmath185 parameter plane . at the end points of the curves the hamiltonian - hopf bifurcation coalesces with the pitchfork bifurcation . the curve of pitchfork bifurcations is given by ( [ char_coat ] ) and is included in dotted lines . we have instability inside the region bounded by solid and dotted curves . note that the stability of the b1 solution is almost entirely , but not completely , determined by the first two eigenvalues . we have shown that whirling current - carrying transversely isotropic rods bifurcate under increasing magnetic field ( or current ) into exact helical shapes provided one applies what we call coat hanger boundary conditions . the first bifurcating branch , containing solutions with half a helical turn , is stable while higher - order branches are all unstable . we stress that the stability analysis of section 5.1 together with the helical analysis in appendix b gives a complete picture of magnetically - induced helical buckling subject to these boundary conditions . in section 5.1 we show that the critical loads for a straight rod are given by equation ( [ char_coat ] ) . these bifurcation points are non - degenerate and valid for both isotropic and anisotropic rods . the analysis says nothing , however , about the type of solutions that bifurcate . they could be helices or not . in appendix b we then compute branches of helical solutions and show that , for isotropic rods ( i.e. , @xmath203 ) , they intersect the trivial branch of straight rod solutions precisely at the critical loads computed in section 5.1 . taken together these results prove that for an isotropic rod all solutions bifurcating from the straight rod are helical . for an anisotropic rod , on the other hand , the critical loads are still given by ( [ char_coat ] ) ( with @xmath245 ) but no branch of helical solutions in appendix b is found to intersect the trivial branch and therefore the bifurcating solutions are non - helical , as confirmed numerically in section 6.3 . we have also investigated the stability of post - buckling solutions and found hopf bifurcations where stable helical solutions lose stability . in the case of a non - rotating anisotropic rod we found secondary instabilities given by hamiltonian - hopf bifurcations . unfortunately , our method allows us only to study stationary or quasi - stationary ( whirling ) solutions , so it is not clear what type of stable solutions occur after these secondary bifurcations . to investigate this one would have to do simulations based on direct discretisation of the pdes ( [ eq : lmomentum4 ] ) and ( [ eq : amomentum4 ] ) . for whirling isotropic rods subject to welded boundary conditions magnetic buckling is described by a doubly - degenerate pitchfork bifurcation ( see @xcite , where we also showed that the same is true for non - rotating states of anisotropic rods ) . this is because the equations are invariant under rotation about the axis ( @xmath61 axis ) of the supports . this symmetry property complicated wolfe s analysis ( for non - rotating states of isotropic rods ) , which had to take account of the variational nature of the problem to prove existence of non - trivial bifurcating states @xcite . by contrast , the coat hanger boundary conditions here introduced break the @xmath246 symmetry down to @xmath247 symmetry ( reflection symmetry along @xmath57 and @xmath59 ) , and no problems in the application of standard results from bifurcation theory should arise . indeed , we find the ( isolated ) critical values of the magnetic field to be given by a remarkably simple explicit expression . helical solutions are often used and studied in applications of elastic rods or filaments . these solutions are incompatible with the common ( aligned ) clamped , pinned , cardan joint and other boundary conditions . boundary conditions are therefore often ignored in these studies , making it impossible to do a stability analysis . here we have introduced a set of boundary conditions that does allow for helical solutions , not only in magnetic buckling but also in traditional buckling due to compression , twist or whirl . these boundary conditions allow one to study exact helical solutions in finite - length rods . we have also proposed a mechanical device that can be used for generating helical solutions in the laboratory . j.v . wishes to thank the andalusian regional ministry for financial support through the excellence research programme , under the fqm-4239 project . doedel , a.r . champneys , t.r . fairgrieve , yu.a . kuznetsov , b. sandstede , x.j . wang , auto2000 : continuation and bifurcation software for ordinary differential equations ( available by anonymous ftp from ftp.cs.concordia.ca/pub/doedel/auto ) , 2000 . assume _ a helical shape for the magnetically buckled rod then we can derive exact solutions as well as buckling loads . for this it is convenient to introduce euler angles @xmath268 , @xmath269 , @xmath270 and relate the director frame @xmath1 to the rotating frame @xmath46 as follows : @xmath271 the rate of change of the director frame is given by @xmath272 where @xmath273 is the curvature vector . inverting ( [ strains ] ) and using orthogonality of the directors gives @xmath274 which , on inserting ( [ angles ] ) , yields @xmath275 so for the total curvature @xmath276 we find @xmath277 now we make the assumption of a helical centreline and uniformly rotating directors by taking @xmath79 to be a constant and setting @xmath278 , @xmath279 ( @xmath280 and @xmath281 constants ) . ( that @xmath282 may be taken constant is not a priori clear . however , performing the following calculation without this assumption , while still taking @xmath283 constant , one is quickly led to conclude that @xmath282 is in fact constant , so to simplify the presentation we assume constancy of @xmath282 from the start . ) by integrating @xmath284 we then find for the shape @xmath285 and with this solution the boundary conditions ( [ eq : bc36 ] ) ( [ eq : bc47 ] ) reduce to the four kinematical conditions @xmath286 where we have allowed for a non - zero angle @xmath62 between the axes @xmath57 and @xmath59 ( see fig . [ fig : hel : bc ] ) . compatibility of the director and fixed frame for the case of a straight ( but twisted ) rod ( i.e. , with @xmath287 ) requires @xmath288 ( both frames are aligned at @xmath72 and have a relative rotation @xmath62 about @xmath289 at @xmath58 ) . equations ( [ bc2 ] ) and ( [ bc3 ] ) imply @xmath290 to find the reduced equilibrium equation we integrate the force balance equation ( [ eq : lmomentum4_whirl ] ) to get ( letting @xmath291 , @xmath292 , @xmath293 ) @xmath294 and insert these expressions , together with ( [ curvs ] ) and the constitutive relations ( [ eq : constitutive1 ] ) with @xmath295 , in the moment balance equation ( [ eq : amomentum4_whirl ] ) written out in the director frame . the twisting moment is directly given by its constitutive relation to be @xmath296 i.e. , a constant . in the isotropic case ( @xmath297 ) the moment balance equation for @xmath153 is therefore identically satisfied , while the equations for @xmath298 and @xmath299 each give @xmath300 with @xmath280 and @xmath281 given by ( [ omega ] ) . the torque @xmath301 of the lower axis @xmath57 about @xmath8 is given by @xmath302 thus we have an infinite set of post - buckling helical solution branches parametrised by @xmath303 $ ] . the bifurcating curves in fig . [ fig : hel : bif - helix ] , for instance , are obtained by plotting ( nondimensionalised ) @xmath207 against @xmath304 for @xmath305 , using equations ( [ shape ] ) and ( [ equil_eq ] ) . the handedness of helical solutions is determined by the sign of @xmath280 , right - handed for @xmath306 . the two solutions for @xmath307 in ( [ omega ] ) correspond to the two branches emanating from the pitchfork bifurcations , both with the same handedness but with opposite signs for both @xmath34 and @xmath35 . the buckling condition is obtained by setting @xmath287 : @xmath308 or , for @xmath71 and in dimensionless parameters , @xmath309 this expression shows that an applied tension ( @xmath310 ) and inertia ( @xmath311 ) stiffen the rod against helical buckling , while whirl ( @xmath312 ) softens it , as expected . the critical values of @xmath185 agree with the pitchfork bifurcations in figs [ fig : hel : bif - helix ] and [ fig : hel : bif - diag - w2 ] , and with ( [ char_coat ] ) for the present case of an isotropic rod . formula ( [ buckling ] ) is of course also valid in the absence of a magnetic field ( @xmath313 ) , in which case it gives critical values for buckling due to compression ( @xmath67 ) , twist ( @xmath62 ) or whirl ( @xmath56 ) . in the anisotropic case ( @xmath314 ) the moment balance equation for @xmath153 becomes @xmath315 , whose only solution compatible with the boundary conditions ( [ omega ] ) is @xmath316 or @xmath317 . but this implies @xmath318 and hence , by ( [ omega ] ) , @xmath319 and @xmath320 . we conclude that in the case @xmath71 , i.e. , in the case of parallel axes @xmath57 and @xmath59 , an anisotropic rod can not buckle into a helical solution . this agrees with our numerical results , which show that the solutions bifurcating at the critical loads ( [ char_coat ] ) when @xmath245 are non - helical .

it has recently been suggested that high - density star clusters have stellar age distributions much narrower than that of the orion nebula cluster , indicating a possible trend of narrower age distributions for denser clusters . we show this effect to likely arise from star formation being faster in gas with a higher density . we model the star formation history of molecular clumps in equilibrium by associating a star formation efficiency per free - fall time , , to their volume density profile . we focus on the case of isothermal spheres and we obtain the evolution with time of their star formation rate . our model predicts a steady decline of the star formation rate , which we quantify with its half - life time , namely , the time needed for the star formation rate to drop to half its initial value . given the uncertainties affecting the star formation efficiency per free - fall time , we consider two distinct values : and . when , the half - life time is of the order of the clump free - fall time , . as a result , the age distributions of stars formed in high - density clumps have smaller full - widths at half - maximum than those of stars formed in low - density clumps . when the star formation efficiency per free - fall time is 0.01 , the half - life time is 10 times longer , i.e. 10 clump free - fall times . we explore what happens if the duration of star formation is shorter than 10 , that is , if the half - life time of the star formation rate can not be defined . there , we build on the invariance of the shape of the young cluster mass function to show that an anti - correlation between the clump density and the duration of star formation is expected . we therefore conclude that , regardless of whether the duration of star formation is longer than the star formation rate half - life time , denser molecular clumps yield narrower star age distributions in clusters . published densities and stellar age spreads of young clusters and star - forming regions actually suggest that the time - scale for star formation is of order- . we also discuss how the age - bin size and uncertainties in stellar ages affect our results . we conclude that there is no need to invoke the existence of multiple cluster formation mechanisms to explain the observed range of stellar age spreads in clusters .

1407.4119

a crucial constraint on models of star cluster formation is the star formation duration which reveals itself as the age spread of stars in clusters . cluster star age spreads are notoriously difficult to assess , as illustrated by the contrasting results reached by @xcite and @xcite for the orion nebula cluster ( see also * ? ? ? * for a review ) . nevertheless , cluster star age spreads seem to range from a fraction of a million years up to 10 million years @xcite . based on such a variety of results , one may conclude that different star cluster formation mechanisms exist . building on the concept of star formation efficiency per free - fall time , we will show that this is not necessarily the case . the star formation efficiency per free - fall time , first defined by @xcite , is the gas mass fraction of an object turned into stars over one free - fall time at the object mean density . @xcite recently generalized the concept by associating a _ radially - varying _ free - fall time to the observed volume density profile of proto - cluster clumps . note that their model applies to molecular clumps ( i.e. , star cluster progenitors ) and not to molecular clouds ( i.e. , cluster complex progenitors ) . in this class of models , molecular gas with a higher volume density forms stars more rapidly than gas with a low volume density ( see also * ? ? ? * ; * ? ? ? . under the assumption of a constant star formation efficiency per free - fall time , the slope predicted for the star formation relation , namely the relation between the surface densities in molecular gas and young stars , actually matches that observed by @xcite . this suggests that the star formation efficiency per free - fall time is independent of the distance from the clump center , hence from the gas density . such a conclusion agrees with the earlier result of @xcite who found no significant variations of the star formation efficiency per free - fall time over three orders of magnitude in gas density , from giant molecular clouds to hcn - traced molecular clumps . here we take the model of @xcite one step further and we show that high - density molecular clumps may yield clusters with shorter stellar age spreads than those formed by low - density clumps . our study has been stimulated by the recent finding that stellar age spreads in the central regions of the starburst clusters ngc3603 yc and westerlund 1 ( wd 1 ) may be an order of magnitude smaller than in the orion nebula cluster @xcite . in what follows , we call age spread of a cluster the full - width at half - maximum ( fwhm ) of its stellar age distribution , regardless of the shape of the age distribution . we obtain and discuss the relation between stellar age spreads in clusters and cluster - parent clump characteristics . a star - forming molecular clump has typically a power - law density profile with a slope of @xmath6 to @xmath7 @xcite . we adopt a slope of @xmath7 for the volume density profile of molecular clumps , @xmath8 , with an additional term to avoid an infinite central density : @xmath9 @xmath8 also defines the gas density profile at the onset of star formation . in this equation , @xmath10 is the distance to the clump center ( spherical symmetry is assumed ) , @xmath11 is the central density of the clump and @xmath12 its core length - scale . we take @xmath13pc , which is of the order of the radius of the circumstellar envelope of a protostar @xcite . for the models discussed here , this means that @xmath14 where @xmath15 is the clump outer radius . the central density @xmath16 is such that the mass enclosed within the clump radius @xmath15 is the clump total mass @xmath17 : @xmath18}\;. \label{eq : rhoc}\ ] ] in this model , we assume that the molecular gas first gathers into a massive and compact clump whose density profile is described by eq . [ eq : cpl ] , _ then _ starts forming stars while retaining a near - equilibrium configuration . do such compact yet starless molecular clumps exist ? the answer seems to be yes . in the apex telesope large area survey of the galaxy at 870@xmath19 m ( atlasgal , * ? ? ? * ) , @xcite identify 210 such clumps . their size of order 1pc and masses up to @xmath20 make them cluster - progenitor candidates . in addition , the absence of mid - ir tracers of ongoing star formation ( either class i young stellar objects or warm dust ) lead @xcite to consider these clumps as starless . an alternative model is that a cluster - forming clump is contracting from volumes with radius @xmath21pc over timescales of several myr during which star formation is already proceeding . the key difference between both models resides in the evolution of the star formation rate . while our model predicts a decrease with time of the star formation rate ( see below ) , a contracting clump leads to a star formation rate increase . a third possibility is that the actual star formation history of molecular clumps consists of the combination of both models , with an increasing star formation rate during the contraction of the clump followed by a decline once stellar feedback has forced the clump to a position of near - equilibrium . such a scenario inevitably leads to star age distributions broader than those derived here . in what follows , we assume that the clump radius @xmath15 remains constant in time . such a static configuration may be maintained as protostellar outflows compensate for the decay of the turbulence inherited from the clump formation , thereby allowing the clump to resist a global gravitational contraction . collimated protostellar outflows are especially efficient at transporting the outflow momentum and energy to large distances , eventually driving the turbulence on the full pc - scale of the cluster - forming clump ( @xcite ; see @xcite for an early model ) . in fact , the clump will oscillate around a position of near - equilibrium . [ see also @xcite for a model of the evolution of _ giant molecular clouds _ where a cloud can oscillate around a given radius until its eventual dispersal by stellar feedback ] . we will neglect such clump oscillations . we also stress that the star formation rate we consider is the stellar _ mass _ formed per time unit . its evolution with time may differ from the evolution of the _ number _ of stars formed per time unit , especially if high - mass stars formed first . we come back to these points in section [ sec : acc_dec ] . assuming that star formation starts at a time @xmath22 in an initially starless clump of mass @xmath17 , radius r , and density profile given by eq . [ eq : cpl ] , its star formation rate ( sfr ) at time @xmath23 is : @xmath24 here , @xmath23 is the time elapsed since the onset of star formation , @xmath25 is the radial density profile of the stars at time @xmath23 and @xmath26 describes therefore the evolution with time of the volume density of the star formation rate . equation [ eq : sfr1 ] can be written as a function of the initial gas density profile , @xmath8 , and star formation efficiency per free - fall time , @xmath27 , by combining eqs . 18 - 19 of @xcite : @xmath28 one model parameter which still needs to be specified is the star formation efficiency per free - fall time , @xmath0 . based on the relation between the observed surface densities of the gas and star formation rate in unresolved disk and starburst galaxies in the local and high redshift universe , as well as in kpc - scale regions of local group galaxies , @xcite infer @xmath29 with @xmath30 . their result is similar to that obtained by @xcite , from co - traced clouds to hcn - traced clumps ( i.e. clumps with a mean density of about @xmath31 ) . in contrast , @xcite derive an - order - of - magnitude higher estimate , i.e. @xmath32 , by matching their model onto the observed star formation relation of @xcite : @xmath33 in eq . [ eq : sfl ] , @xmath34 and @xmath35 are the local surface densities of gas and young stellar objects ( ysos ) in molecular clouds of the solar neighborhood . figure [ fig : eff ] shows eq . [ eq : sfl ] ( black dotted line ) as well as two models of @xcite : @xmath36 and @xmath37myr ( filled circles ) , and @xmath38 and @xmath37myr ( open circles ) . although the model with @xmath36 provides a better match to eq . [ eq : sfl ] than its @xmath38 counterpart , one should keep in mind that the @xmath0 estimate is directly related to the assumed time - span @xmath23 since the onset of star formation . for a given star formation efficiency per free - fall time , the longer the time - span , the higher the surface density in ysos . @xcite assume an averaged star - formation timespan of @xmath37myr in the star - forming regions studied by @xcite . this estimate is based on the presence of class ii objects in these regions and on an estimated duration of the class ii phase of @xmath39myr @xcite . should the class ii phase be longer , the estimated star formation efficiency per free - fall time would be smaller ( since a smaller @xmath0 is needed to achieve the same stellar surface densities over a longer star - formation time - span ) . the assumption of a 2myr time - scale for the class ii phase has recently been challenged by @xcite . they study star - forming regions where massive stars have already settled on the main sequence and infer a class ii phase duration of @xmath40myr . this is about twice as long as the earlier result of @xcite . it also implies a longer class i phase duration , i.e. , @xmath41myr . if the result of @xcite stands , and if star formation in the regions surveyed by @xcite covers the full class ii phase , then the time - span since the onset of star formation is closer to @xmath42myr , a factor of 3 longer than assumed by @xcite . that would accordingly decrease their @xmath0 estimate by a factor 3 ( @xmath43 ) , bringing it in agreement with the upper limit on the result of @xcite this degeneracy between the time elapsed since star formation onset and the star formation efficiency per free - fall time is illustrated in fig . [ fig : eff ] . one can see that the models with @xmath44myr and @xmath45 on the one hand , and with @xmath37myr and @xmath46 on the other hand , lead to identical star formation relations . that @xcite and @xcite derive different estimates of the star formation efficiency per free - fall time might also be related to the use of different star formation relations . while @xcite build on the @xmath35 vs. @xmath47 space of @xcite , @xcite work in the @xmath48 vs. @xmath49 space . specifically , they obtain the @xmath0 by matching their collected data onto their eq . 2 which we reproduce here for the sake of clarity : @xmath50 @xmath51 is the fraction of molecular gas which we take equal to unity . note that the slope of eq . [ eq : sfl ] is two , while the slope of the star formation relation implemented by @xcite ( eq . [ eq : kdm12 ] ) is unity . to test whether this can lead to different @xmath0 estimates , we plot in fig . [ fig : eff_kdm ] eq . [ eq : kdm12 ] for @xmath2 ( green dotted line ) and @xmath52 ( black dashed line ) . the green dotted line ( @xmath2 ) is the star formation relation shown in fig . 3 of @xcite . in the same figure , we plot the outputs of the @xcite model for the same efficiencies per free - fall time and two different times after the onset of star formation ( @xmath53myr and @xmath37myr ) . we approximate @xmath48 by @xmath54 . @xmath3 is the free - fall time relevant for star formation , namely , this is the free - fall time of the gas in a spherical shell whose radius is the distance from the clump center at which @xmath35 and @xmath47 are measured . in other words , @xmath3 is a three - dimensional , local and instantaneous measurement ( @xmath55 ) . for a given star formation efficiency per free - fall time , [ fig : eff_kdm ] does not reveal any significant difference between the star formation relations of @xcite ( dashed and dotted lines ) and the predictions of our model ( lines with symbols ) . we therefore conclude that the difference between the @xmath0 estimates of @xcite and @xcite does not stem from using two different star formation relations . given the uncertainties and caveats described above , in what follows , we apply eq . [ eq : sfr2 ] for two values of the star formation efficiency per free - fall time , i.e. @xmath29 and @xmath56 . @xmath0 is assumed to be constant in time and uniform in space . as we shall see , these two values lead to two different regimes when defining the stellar age spread of a cluster . we now proceed to obtain the evolution with time of the star formation rate . combining eq . [ eq : cpl ] and eq . [ eq : sfr2 ] , one obtains : @xmath57 \ , , \label{eq : sfrcpl}\end{aligned}\ ] ] where @xmath58 is a function of time @xmath23 : @xmath59 figures [ fig : princ ] and [ fig : princb ] present the evolution of the sfr normalized to its initial value for two molecular clumps , clump a and clump c , and two star formation efficiencies per free - fall time , @xmath1 ( fig . [ fig : princ ] ) , and @xmath2 ( fig . [ fig : princb ] ) . the characteristics of the two model clumps are listed in table [ tbl : t1 ] . table [ tbl : t1 ] also gives the mean free - fall time of the gas at star formation onset : @xmath60 where @xmath61 is the clump mean density . the mass and radius of clump c are similar to what was found by @xcite for the gas - embedded orion nebula cluster ( onc ) . clump c ( @xmath62 ) may thus be a good proxy to the onc parent clump . given our assumption that the clump remains in equilibrium during star formation , we set an upper limit of five initial free - fall times to the duration of star formation . this represents 0.13myr and 3.5myr for clump a and clump c , respectively ( see table [ tbl : t1 ] ) . during this time - span , the clump gas content is solely depleted by the conversion of gas into stars . that is , we ignore the possibility that all or a fraction of the gas may be expelled beyond the clump radius @xmath15 by stellar feedback processes . should gas removal take place before five initial free - fall times , star formation would be slowed down through a decrease of the gas mass and an expansion of the gas clump . as star formation depletes the clump gas reservoir , the gas free - fall time steadily increases and the star formation efficiency per free - fall time is applied to an ever lower gas mass . therefore , unless the clump contracts or is replenished in gas by inflows , the star formation rate decreases with time . this decrease is faster at higher @xmath0 , as well as in high - density regions ( here clump a ) since in essence they evolve on a faster time - scale . lccc & clump a & clump c + r [ pc ] & 0.5 & 2.0 + @xmath63 $ ] & @xmath64 & @xmath65 + @xmath66 $ ] & @xmath67 & @xmath68 + @xmath69 $ ] & @xmath70 & @xmath71 + @xmath72 $ ] & @xmath73 & @xmath74 + @xmath75 $ ] & @xmath76 & @xmath77 + @xmath78 $ ] & @xmath79 & @xmath80 + @xmath81 [ myr ] & @xmath82 & 0.74 + @xmath12 [ pc ] & 0.01 & 0.01 + @xmath83 $ ] & @xmath84 & @xmath85 + @xmath86 [ myr ] ( @xmath87 ) & 0.05 & 0.80 + given that the model is valid for a few free - fall times and that @xmath88 , eq . [ eq : sfrcpl ] can be simplified . introducing @xmath89 , the gas volume density at the edge of the clump , and using eq . [ eq : cpl ] , we find : @xmath90 one can then express the product @xmath91 as a function of @xmath92 , the free - fall time at the clump edge . from eq . [ eq : cround ] , we successively derive : @xmath93 given the limitations imposed to the parameter space , that is , @xmath88 , @xmath94 several @xmath92 and @xmath95 , one gets : @xmath96 and eq . [ eq : sfrcpl ] becomes : @xmath97 \\ \nonumber & \simeq & 4 \pi \sqrt{\frac{32g}{3\pi } } { \epsilon_{ff}}\rho_e^{3/2 } r^3 \\ \nonumber & \cdot & \left [ - ln \left ( \frac{r_c}{r } + \frac{{\epsilon_{ff}}}{2 } \frac{t}{\tau_{ff , e } } \right ) - \frac{3}{2 } + 2\frac{r_c}{r } + { \epsilon_{ff}}\frac{t}{\tau_{ff , e}}\right]\;. \\ \nonumber & \simeq & \frac{4\pi { \epsilon_{ff}}\rho_e r^3}{\tau_{ff , e } } \\ \nonumber & \cdot & \left [ - ln \left ( \frac{r_c}{r } + \frac{{\epsilon_{ff}}}{2 } \frac{t}{\tau_{ff , e } } \right ) - \frac{3}{2 } + { \epsilon_{ff}}\frac{t}{\tau_{ff , e}}\right]\;.\end{aligned}\ ] ] the mean density of an isothermal sphere is three times the density at its edge , i.e. @xmath98 . therefore , @xmath99 , with @xmath3 the free - fall time at the clump mean density , and : @xmath100\;. \label{eq : sfr_simpl}\ ] ] equation [ eq : sfr_simpl ] is shown as the dashed green lines in the top panels of figs . [ fig : sfh ] and [ fig : sfhb ] which we discuss in the next section . it can be integrated to provide the growth with time of the global star formation efficiency of the clump . with @xmath101 and @xmath102 , we derive : @xmath103 we emphasize that this equation is valid only when @xmath88 , @xmath95 and for star formation durations not longer than several free - fall times . to derive the stellar age distribution , we need the stellar mass formed at time @xmath23 over a constant time - interval @xmath104 . let us assume that @xmath104 is small enough so that the star formation rate does not vary significantly over an interval @xmath105 $ ] . the stellar mass formed over such a time interval is then : @xmath106 if the cluster is observed at a time @xmath107 after star formation onset , its stellar age distribution is : @xmath108 where @xmath109 is the age of the stars formed at time @xmath23 . in other words , for constant and small enough @xmath104 , the function @xmath110 quantifies the stellar age distribution once time @xmath23 is swapped for the star age @xmath109 . the top @xmath111-axis of figs [ fig : princ ] and [ fig : princb ] shows the age of the stars formed at time @xmath23 for @xmath112myr . therefore , this top @xmath111-axis provides us with the normalized star age distribution of the newly - born cluster . here , we assume that no stars are lost from the cluster between the time of their formation and the time of cluster observation . reasons for a cluster to lose stars at an early stage of its evolution include star - encounter - driven scattering @xcite and gas expulsion @xcite . here , a caveat regarding the size of the @xmath104 interval ( equivalent to the bin size of the age distribution ) . it can not necessarily be chosen as small as requested by the star formation rate variations . for instance , the age bin must be large enough so as to contain enough stars to limit the amplitude of the poisson noise . there may thus be cases where the star formation rate varies so rapidly that it can not be considered constant over the adopted time interval @xmath104 . in other words , the stellar age distribution of a cluster can not necessarily capture all the star formation rate variations of its gaseous progenitor . this happens preferentially for high - density clumps and/or high star formation efficiency per free - fall time . the next section will present such a case . if one wants to define the cluster star age spread as the fwhm of the stellar age distribution , we see two possible regimes depending on how fast the star formation rate decreases with time . we introduce them here , and present the corresponding age distributions in the next section . * if the star formation rate decreases by more than a factor of 2 over the duration of star formation ( here when @xmath52 ; see fig . [ fig : princ ] ) , the fwhm of the age distribution mirrors the time @xmath86 required for the star formation rate to drop by a factor of 2 compared to its initial value . we refer to @xmath86 as the half - life time of the clump star formation rate . half the initial star formation rate is visualized in figs [ fig : princ ] and [ fig : princb ] by the horizontal ( black ) dashed line , and the @xmath86 time of each model clump is given in table [ tbl : t1 ] . we see that @xmath113 , which suggests that , in the @xmath52 regime , the star age distribution of a young cluster can be used to probe its parent - clump volume density . note that @xmath86 also depends on the gas initial density profile , with shallower profiles yielding longer half - life times . for instance , a top - hat profile implies @xmath114 when @xmath52 . * if the decrease of the star formation rate is so slow that @xmath86 can not be defined ( here when @xmath2 , see fig . [ fig : princb ] ) , the fwhm of the age distribution is primarily determined by the duration of star formation . the top panels of figs [ fig : sfh ] and [ fig : sfhb ] show the evolution of the star formation rate without the normalization of figs [ fig : princ ] and [ fig : princb ] . the bottom panels depict the time evolution of the stellar mass fraction formed per constant time - interval @xmath104 . for the sake of an easy comparison , we choose @xmath115myr , identical to the age - bin width of the stellar age distributions presented in the next section . the mass fractions are defined with respect to the final stellar mass , here the clump stellar mass at @xmath116 ( 3.5myr for clump c and 0.13myr for clump a ) . as an example , when @xmath2 , clump c forms at any time @xmath117% of its final stellar mass per time - interval of 0.1myr ( fig . [ fig : sfhb ] ) . note also that the star formation rate is approximately constant when @xmath2 since only a few per cent of the gas are turned into stars . we now move to estimating the star age distributions generated by our model clumps . they are presented in figs [ fig : ad ] ( @xmath2 ) and [ fig : adb ] ( @xmath1 ) . the solid red lines with open circles are the intrinsic ( i.e. error - free ) age distributions . star formation started 5myr ago and lasted for 0.13myr and 3.5myr for clump a and clump c , respectively . each age distribution is normalized with respect to the total stellar mass eventually formed . for the sake of an easy comparison with their results , we adopt the same age - bin size as @xcite , that is , 0.1myr . we come back to this choice later in this section . these intrinsic distributions thus correspond to the models shown in the bottom panels of figs [ fig : sfh ] and [ fig : sfhb ] mirrored around a vertical axis , that is , the time @xmath23 since star formation onset has been swapped for the age of the stars formed at time @xmath23 . note that the @xmath118-axis scaling is logarithmic in figs [ fig : sfh ] and [ fig : sfhb ] , and linear in figs [ fig : ad ] and [ fig : adb ] . the intrinsic age spreads , @xmath119 , are quoted in figs [ fig : ad ] and [ fig : adb ] . clump c clearly illustrates the two regimes introduced in the previous section . in the first regime , when @xmath2 ( bottom panel of fig . [ fig : ad ] ) , the star formation rate does not decrease significantly over the duration of the star - formation episode . as a result , its half - life time can not be defined , and the fwhm of the intrinsic age distribution is driven by the duration of star formation , @xmath120 . this happens if the gas density is low and/or @xmath0 is small . a constraint on the duration of star formation as a function of the cluster - forming gas density is provided by the study of cluster mass functions . the mass function of young clusters has a power - law of slope @xmath121 , i.e. @xmath122 . at young ages ( e.g. , age @xmath123 50myr ) , it retains its shape while evolving towards lower masses @xcite . this means that , following the expulsion of their residual star - forming gas , all clusters experience the same mass fraction of infant weight - loss , regardless of their mass . in turn , this implies that all cluster parent - clumps achieve , irrespective of their mass , similar star formation efficiencies at the time of gas expulsion . for a given star formation efficiency per free - fall time , a common star formation efficiency suggests a star formation duration equal to a common number @xmath124 of free - fall times , namely , @xmath125 for all cluster - forming clumps ( see eq . [ eq : sfet ] ) . the invariance of the cluster mass function shape at young ages therefore suggests a star formation duration in clusters scaling with the clump free - fall time and , hence , shorter age spreads at higher clump densities . note , however , that the formation of a bound gas - free star cluster in @xmath126 with @xmath2 requires an extremely large amount of gas . with a final star formation efficiency of @xmath1275% , the star mass bound fraction is about 10% ( see fig . 15 in * ? ? ? that is , the formation of a bound gas - free cluster with a mere mass of @xmath128 requires @xmath129 of gas . however , should the star formation duration be longer , either because the present model is valid for more than 5@xmath130 , or because the actual star formation history of molecular clumps consists of more stages than described by our model , the formation of massive bound clusters when @xmath2 will require less extreme gas masses since both the star formation efficiency and the bound fraction will be higher . in the second regime , when @xmath52 ( bottom panel of fig . [ fig : adb ] ) , the fwhm of the intrinsic age distribution is driven by the half - life time of the star formation rate . note , however , that the fwhms are larger than @xmath86 ( see table [ tbl : t1 ] ) . this effect , especially prominent for the fast - evolving clump a , is a consequence of binning . the star formation rate of clump a drops by a factor of 2 within @xmath131 years . therefore , its half - life time is shorter than the age - bin width ( @xmath132myr ) , and can not be resolved in the intrinsic age distribution . here , we note that the age - bin size adopted by @xcite and in this contribution ( 0.1myr ) is smaller than the uncertainties typically affecting derived ages of pre - main - sequence stars . @xcite estimate an error of @xmath133myr in their study of wd 1 ( see below ) , and @xcite quotes a gaussian error of 0.5myr when stellar variability , photometry errors and extinction uncertainties are taken into account ( see section [ sec : acc_dec ] ) . in strict logic , we should therefore have adopted an age - bin size of that order of magnitude . however , we think it a valuable exercise to show that , even if stellar age errors were negligible ( e.g. , smaller than 0.1myr ) , a proper reading of the half - life time of the star formation rate from the binned age distribution may already be hindered when it comes to gas with high density / short free - fall times . in the hypothetical absence of stellar age uncertainties , the age - bin size will be chosen so as to limit the poisson noise in the star age distribution and , therefore , will depend on the size of the stellar sample and the age distribution extent . should the age bin be larger than 0.1myr , the fast evolution of high - density gas will be concealed even more than presented here . to model the impact of stellar age uncertainties , we assume a gaussian error of @xmath134myr ( as estimated for wd 1 by * ? ? ? * ) and obtain the ` measured ' star age distributions ( solid blue lines with asterisks in figs [ fig : ad ] and [ fig : adb ] ) . the measured age spreads , @xmath135 , are given in figs [ fig : ad ] and [ fig : adb ] . unsurprisingly they can be significantly larger than their intrinsic counterparts . when @xmath52 ( fig . [ fig : adb ] ) , the steady decrease of the star formation rate leads to a measured age distribution of clump c more prominent on its young side than on its old side . as for clump a , the asymmetry is negligible because its intrinsic age distribution is narrower than the adopted error of @xmath136myr . @xcite also find an asymmetry on the young side of the stellar age distribution of the cluster wd 1 . they interpret it as resulting from the presence of binaries ( their fig . 3 ) . figure [ fig : adb ] suggests that an asymmetric age distribution could also be the imprint of the star formation rate decrease driven by gas depletion . the top panel of fig . [ fig : sum ] summarizes the half - life time ( @xmath86 ) , intrinsic ( @xmath119 ) and measured ( @xmath135 ) fwhms found for the second regime , that is , when @xmath1 . here we included an additional model clump , with a density between that of clumps a and c ( clump b : @xmath137 and @xmath138 ) . the dependence of @xmath86 and @xmath119 on the initial mean density of the gas , @xmath139 , is clearly highlighted . one could argue here that the model is degenerate in the sense that narrow fwhms can be obtained from either high - density gas , or short star - formation duration in low - density gas . this is not the case . for low - density clumps to give rise to @xmath140myr , comparable to that of clumps a and b , star formation must be halted on a similar time - scale . at @xmath141 , the clump c stellar mass is 250@xmath142 . the star formation efficiency , @xmath143 , gives a bound fraction of @xmath144 ( fig . 15 in * ? ? ? the star cluster mass when gas expulsion effects are over is therefore @xmath145 ( i.e. @xmath146 ) . such a low - mass cluster would be swiftly disrupted by the galactic tidal field . it is therefore highly unlikely to observe clusters characterized by narrow age spreads _ and _ formed in low - density environments . the bottom panel shows the relation between @xmath86 and @xmath139 . it is obtained by setting eq . [ eq : sfrcpl ] equal to @xmath147 and solving for @xmath23 . the clump radii considered are quoted in the key , although note that the parameter of relevance here is the ratio @xmath148 . also shown are half and twice the free - fall time at the gas initial mean density ( see eq . [ eq : tff ] ) . one can see that , as suggested in section 2.2 , the half - life time of the star formation rate is of the order the clump free - fall time when @xmath36 . using eq . [ eq : sfet ] derived in section [ ssec : mod ] , we can estimate the global star formation efficiency achieved by the clump after one half - life time ( equivalent to @xmath149 when @xmath36 ) . for clumps b and c , @xmath150 , @xmath151 and @xmath152 at a time @xmath153 ( the numerical solution predicts @xmath154 ) . at first glance , one may be surprised that after one free - fall time , the global star formation efficiency is higher than @xmath1 . this effect is a direct consequence of the gas mass concentration inside the clump . at @xmath155 , the ( three - dimensional ) half - mass radius of the forming cluster is @xmath156pc . this shows that the bulk of star formation takes place in the clump inner regions where the free - fall time is shorter than the free - fall time @xmath130 averaged over the whole clump . this effect was already pointed out by @xcite and is also discussed by @xcite . for clump a , eq . [ eq : sfet ] predicts a slightly lower global star formation efficiency , i.e. , @xmath157 ( @xmath158 in the numerical solution ) . this is because the central core of the initial gas mass distribution occupies a greater fraction of the clump radius ( @xmath159 ) , which decreases the gas mass central concentration . after five free - fall times , the global star formation efficiency achieved by clumps b and c is @xmath160 ( @xmath161 in the numerical solution ) . this number reflects the decreasing star formation rate over five free - fall times of star formation activity . these efficiencies are slightly higher than those observed , which range from a few per cent to 0.33 @xcite . this may indicate that either the star formation efficiency per free - fall time is slightly lower than @xmath1 , or that the duration of star formation is shorter than five free - fall times . note , however , that star formation efficiency estimates for embedded clusters are mostly available for the solar neighborhood . in addition , we remind that not only do measured star formation efficiencies depend on time , they also depend on the spatial scale on which they are measured , with smaller scales ( i.e. more limited to the cluster inner regions ) leading to higher efficiencies ( see fig . 7 in lcccc & @xmath162 & age range & fwhm + & [ @xmath163 & [ myr ] & [ myr ] + lh95 & 0.47 & & 3.6 + w3 main ( stars only ) & 37 & 3 . & + w3 main ( stars + gas ) & 83 & 3 . & + onc ( stars only ) & 56 & & 3.0 + onc ( stars + gas ) & 140 & & 3.0 + wd 1 ( @xmath164 ) & @xmath165 & & 0.3 + wd 1 ( @xmath166 ) & @xmath167 & 1 . & + ngc3603 yc ( @xmath168 ) & @xmath169 & & 0.1 + ngc3603 yc ( @xmath170 ) & 550 & & 8.0 + chamelon ii & 18 & & 4.0 + lupus & 7.5 & & 5.0 + ophiuchus & 18 & & 2.0 + ic348 & 50 & 3.5 & + in this section , we assess whether the currently available observations support a star formation scenario where higher - density clumps lead to narrower star age distributions . to that purpose , we have surveyed the literature and gathered densities and age distribution widths for star clusters and star - forming regions . they are presented in table [ tbl : t2 ] and in the top panel of fig . [ fig : litdat ] . depending on the available data , the width of the star age distribution is either its fwhm , or an age range estimate . when comparing such data to model predictions , there are two caveats to keep in mind . first , our model refers to the total mean density @xmath139 of cluster - forming clumps ( i.e. , gas and newly - formed stars ) , while the density given in table [ tbl : t2 ] is sometimes the stellar density alone ( e.g. , ngc3603 yc , wd 1 ) . due to residual star - forming gas expulsion and the ensuing star loss and cluster spatial expansion , the current density of a cluster is a lower limit to the density of its parent clump . second , star formation is not yet over in gas - embedded clusters , which prevents the estimate of the star age spread in a fully - formed cluster . with these caveats in mind , we now describe each data point . + * lh95 * is a star - forming region in the large magellanic cloud . its mean stellar density is @xmath171 ( table 2 in * ? ? ? from the gaussian distribution of pre - main sequence star ages , @xcite derive a fwhm of the order of 2.8 - 4.4myr . we adopt a mean value of 3.6myr ( open circles in top panel of fig . [ fig : litdat ] ) . we note that lh95 consists of three subclusters , each with a mean stellar density higher than that of lh95 as a whole ( fig . 4 and table 2 in * ? the range covered by the subcluster densities is depicted as the blue horizontal arrow . we note that including the residual molecular gas would move the tip of the arrow further to the right . + * w3 main : * this galactic star - forming region hosts pre - main sequence stars , stars and hii regions in various evolutionary stages . one of its massive stars is consistent with an isochrone of age 2 - 3myr and its hii regions span the range from very young hypercompact ( few @xmath172yr old ) to old and diffuse ones ( few @xmath173yr old ) . @xcite therefore estimate the age range in the embedded cluster w3 main to be at least 3myr . using their estimates for the embedded - cluster radius ( 3pc ) , the stellar mass ( 4000@xmath142 ) and the star - plus - gas mass ( 9000@xmath142 ) , we derive stellar and total volume densities of @xmath174 and @xmath175 . the corresponding data are shown as the blue open squares in fig . [ fig : litdat ] . + * the orion nebula cluster ( onc ) : * @xcite estimate its stellar and total masses within a radius of 2pc to be 1800@xmath142 and 4500@xmath142 , respectively . the corresponding volume densities are 56 and 140@xmath176 . we obtain the fwhm of the linear distribution of pre - main sequence star ages from fig . 11 in @xcite , which is based on the data of @xcite . assuming a symmetric age distribution gives a fwhm of @xmath117myr . these results are depicted as the open red triangles in fig . [ fig : litdat ] . + * wd 1 * is a young compact massive cluster of the galactic disk . its age , mass , and half - light radius are estimated to be , respectively , 5myr @xcite , @xmath177 and 0.86pc @xcite . @xcite obtain the age distribution of the stars of the cluster central region ( stars with masses between 0.5 and 11.5@xmath142 within a field - of - view 0.2pc in radius ) . they find @xmath178myr . subtracting in quadrature their photometric error estimate ( 0.25myr ) leads to @xmath179myr . we still need the mean stellar density within 0.2pc of the cluster center . for a plummer model with the above - quoted mass and half - light radius , the stellar mass enclosed within 0.2pc from the cluster center is 1200@xmath142 @xcite . the corresponding mean volume density is @xmath180 . this point is shown as the blue filled square in fig . [ fig : litdat ] . + @xcite consider a field - of - view an order of magnitude larger than that of @xcite ( radius : 2.9pc ) . they find that the star age range is very unlikely to be @xmath181 myr . assuming again a plummer model , the mean volume density within 2.9pc from the cluster center is @xmath182 . the corresponding result is shown as the green open diamond in fig . [ fig : litdat ] . we note that the cluster has likely expanded and lost mass ( gas and stars ) as a result of residual star - forming gas expulsion @xcite . this correction would move the point to higher densities as indicated by the horizontal green arrow ( the arrow size is arbitrary ) . as for the central region observed by @xcite , its star formation efficiency was probably higher than in the cluster outskirts @xcite , making it less prone to post - gas - expulsion expansion and better able to retain the initial density . + * ngc 3603 yc * is also a starburst cluster of the galactic disk , with a dynamical mass estimate of 17,600@xmath142 and a ( uncertain ) half - mass radius of 0.5pc @xcite . the study of its age distribution has yielded very contrasting results so far . @xcite infer a fwhm of 0.1myr from their linear star age distribution . the sample includes main sequence , pre - main sequence stars and transition region stars with masses between 0.8 to 6.5@xmath142 . based on proper - motion cluster membership of fainter main - sequence stars and the apparent width of the region where pre - main - sequence stars join the main sequence , @xcite find support for a modest age spread in the cluster , or for earlier star formation episodes in that region . based on pre - main sequence stars , @xcite estimate that star formation in and around the cluster has been ongoing for 10 - 20myr . in what follows , we assume that the linear fwhm of @xcite is 8myr ( by inspecting their log - log star age distribution ; see their fig . how can two fwhm estimates @xcite of the same cluster differ by almost two orders of magnitude ? part of the difference resides in both studies scrutinizing different regions of the cluster . @xcite study the region within 0.5pc from the cluster center , where the mean density is @xmath183 ( using a plummer model , the half - mass radius and total mass quoted above ) . in contrast , @xcite consider a wider field - of - view , with a radius of about 2pc and an enclosed mean density of @xmath184 . both results are shown as solid red circles in fig . [ fig : litdat ] . besides the difference in the field - of - view size , we note that the fwhm inferred by @xcite is surprisingly small , being even lower than the photometric error they derive for wd 1 . given that they do not discuss the photometric error contribution to the age spread of ngc 3603 yc , we consider their value of 0.1myr a lower limit which we indicate by an upward arrow in the top panel of fig . [ fig : litdat ] . we also note that @xcite include main sequence stars in their sample while @xcite consider pre - main sequence stars only . we further discuss the difference between both studies later in this section . + * ic348 * is an embedded cluster of the perseus molecular cloud . it contains about 400 members within a radius of @xmath185pc @xcite . adopting a mean stellar mass of 0.5@xmath142 , the mean stellar density of ic348 is @xmath186 . star formation started 3.5myr ago @xcite and is continuing today @xcite . we therefore adopt an age range of 3.5myr when plotting the corresponding point in fig . [ fig : litdat ] ( filled green diamond ) . note that including the gas mass would move the point rightwards . + * cha ii , lupus , and ophiuchus * are star - forming molecular clouds of the solar neighborhood . we infer their mean volume density from the masses and surface areas given in table 1 of @xcite , assuming spherical symmetry . the fwhm of their linear star age distribution are estimated by inspecting figs 2 , 1 and 3 in @xcite . here , we assume that each age distribution is symmetric about its peak , which may or may not be true . the mean volume densities and fwhms are 18@xmath176 and 4myr ( cha ii ) , 7.5@xmath176 and 5myr ( lupus ) , 18 @xmath176 and 2myr ( ophiuchus ) . they are shown as the plus - sign , cross and asterisk in fig . [ fig : litdat ] . here , we repeat the caveat already made in section [ sec : intro ] , that is , our model applies to molecular clumps inside molecular clouds and not to molecular clouds as a whole . the yso age distribution for a whole cloud is necessarily widened by the age differences between its constituent molecular clumps ( i.e. , not all clumps started to form stars at the same time ) . ideally , one would use the mean volume density and yso age distribution of star - clustering region _ individually_. this would move the symbols in fig . [ fig : litdat ] towards higher densities and smaller age spreads . + the data collected in the top panel of fig . [ fig : litdat ] suggest that denser clusters have narrower star age distributions . however , this conclusion is mostly driven by the central regions of the starburst clusters wd 1 and ngc 3603 yc . it is highly desirable to add more data points to the high - density regime ( @xmath187 ) to see whether the highlighted trend is confirmed or not . whether the correlation stands within the low - density regime ( @xmath188 ) can not be established . the data points occupy a fairly limited region of the parameter space so that any trend is likely concealed by the uncertainties affecting the age range / spread and volume density estimates . clusters with intermediate densities ( i.e. @xmath189 ) would also be helpful . in that respect , ngc 1333 in the perseus molecular cloud is an interesting target . with a gas mass of @xmath128 enclosed within a radius of 0.5pc @xcite , its mean density is @xmath190 @xcite infer a median cluster age of less than 1myr . to constrain the age spread of this forming cluster would allow to populate the diagram in the intermediate - density regime . the bottom panel of fig . [ fig : litdat ] presents the age spread as a function of the crossing - time for the same sample of star - forming regions and clusters . we define the crossing - time as @xmath191 . therefore , @xmath192 . as already found by @xcite , the age spread is of order 1 - 2@xmath193 , or @xmath194-@xmath5 . + we now come back to the age spread of ngc 3603 yc , and to the very different results obtained by @xcite and @xcite . the data point for ngc 3603 yc by @xcite stands out of the density - vs - age - spread correlation . their estimated age spread is about 2 orders of magnitude larger than found by @xcite ( see red circles in fig . [ fig : litdat ] ) . as noted above , both studies differ by the size of their field - of - view ( radii of 2.9pc and 0.5pc , respectively ) . @xcite therefore include lower - density regions where star formation proceeded on a slower time - scale , naturally yielding a wider age distribution . a larger aperture also encompasses more cluster stars located along the line - of - sight and more distant from the cluster center than 2.9pc . this broadens further the observed star age distribution . this effect is akin to the central ( denser ) regions of star clusters returning faster to virial equilibrium than the cluster as a whole , as highlighted by @xcite . ( a larger aperture also increases the probability of contaminating the sample with stars not belonging to the cluster ) . we therefore stress that stellar age spreads are aperture - dependent . we call for a consistent comparison of literature results in terms of aperture size and nature of the stellar sample ( main sequence vs. pre - main sequence stars ) under scrutiny . the very large age spread found by @xcite might also be driven by pre - main sequence stars being erroneously assigned old ages . @xcite have recently reconsidered two major outliers of the stellar age distribution of the onc . their estimated isochronal age is @xmath195myr and they show strong accretion activity ( as indicated by their h@xmath196 emission excess ) . for these outliers , @xcite have obtained new age estimates in agreement with the age of the bulk of the onc population , i.e. @xmath197myr . the cause for this age shift was an incorrect estimate of either the spectral type or the visual extinction . @xcite suggest that these issues affect mostly young star - forming regions where effects such as extinction and the high accretion rate from the inner protostellar disc onto the yso surface are significant ( see also * ? ? ? our model predicts a decrease with time of the star formation rate . the higher the clump volume density and/or the star formation efficiency per free - fall time , the faster the star formation rate decreases . for low - density regions with @xmath2 , the star formation rate is almost constant in time ( e.g. , clump c , fig . [ fig : sfhb ] ) . let us assume for now that the mass of the newly formed stars is independent of time , that is , the stellar _ mass _ formed per time unit and the _ number _ of stars formed per time - unit evolve in a similar way . our results are then supported by the observations of @xcite . in a submillimetre continuum survey of two embedded clusters of the perseus molecular cloud , ic348 and ngc 1333 , they find evidence for a steady or reduced ( number ) star formation rate in the last 0.5myr , not an increasing one . note , however , that their results depend on their assumed duration of the class0/i phase . indications of a star formation activity declining with time have also been obtained for the star - forming region chamaeleon i by @xcite and @xcite . our concept of a decreasing star formation rate contrasts with the scenario of @xcite in which the ( number ) star formation rate accelerates with time . we now discuss this aspect . it remains uncertain whether the increase with time of the star formation rate observed by @xcite for a number of embedded clusters and star - forming molecular clouds of the solar neighborhood is genuine or not . @xcite emphasizes that the error affecting isochronal age estimates of pre - main sequence stars can by itself carve an apparent decrease towards old age in the linear star age distribution . we illustrate his point with our model clump c in fig . [ fig : acc ] . we assume that star formation has been ongoing for 1.5myr with a star formation efficiency per free - fall time of 0.10 and we convolve the intrinsic age distribution with a gaussian error of 0.5myr ( standard deviation ) . this error is that estimated by @xcite for a population of pre - main sequence stars with masses between 0.1 and 7@xmath142 . it includes effects such as stellar variability , photometry errors and extinction uncertainties . while the high star formation efficiency per free - fall time results in a significant decrease in the number of stars per age bin towards young ages ( red curve in fig . [ fig : acc ] ) , the age distribution convolved with the gaussian error shows an _ apparently _ increasing star formation rate on its old side ( see blue curve ) . the comparison between the right sides of the measured age distribution and of the gaussian function ( depicted as the dotted black line ) shows this apparently accelerating star formation rate to be the signature of the stellar age error , as warned by @xcite . @xcite suggest yet another reason as to why the observed ( number ) star formation rate increase may be apparent only . that is , many of the old pre - main sequence stars may have migrated out of the field - of - view , which artificially decreases the number of stars at old ages . this suggestion is supported by their study of ngc 3603 yc , in which they find a correlation between the estimated ages of pre - main sequence stars and their location in the cluster . their fig . 7 shows young pre - main sequence stars ( isochronal age @xmath12310myr ) to be more centrally concentrated than their older counterparts ( isochronal age @xmath198 10myr ) . this may be a hint that the old objects are migrating away from the cluster center and observed field - of - view , thereby causing an apparent star formation rate increase ( but see the caveat about age estimates of old pre - main sequence stars by * ? ? ? * ) , or that they are indeed unrelated to the cluster and the product of an earlier star formation episode ( see , e.g. , * ? ? ? * ) . these two examples errors affecting the age estimates of pre - main sequence stars and stellar motions removing them from the observed field - of - view demonstrate the difficulty to infer the actual evolution of the star formation rate , even in nearby star - forming regions . current observational data hardly allow to disentangle between an intrinsically increasing star formation rate and an intrinsically decreasing one . our model predicts a ( mass ) star formation rate decreasing with time , starting from an isolated star - less clump . here we re - iterate that the steady decrease of the star formation rate stems from the model assumptions : the sudden onset of star formation in a molecular clump whose outer radius remains constant in time . should star formation already start as the clump contracts , there will be an initially increasing star formation rate , although that does not imply that most stars form during that phase . a ( mass ) star formation rate genuinely increasing with time is the expected star formation history of a contracting molecular clump , since the density increase of the collapsing gas shortens the clump free - fall time . this scenario naturally leads to the outer regions of the forming star - cluster to be populated by the first stars to form , hence the oldest ones . the actual evolution of a star - forming molecular clump may combine a contraction ( star formation rate increase ) followed by a near - equilibrium evolution ( star formation rate decrease ) once stellar feedback has halted the contraction @xcite . not only is the issue of the star formation rate evolution important in its own right , it also bears consequences for the estimate of the class i phase duration . @xcite estimate it by taking the ratio of number counts in class i and class ii stages , and multiplying by the assumed lifetime of the class ii phase . by so doing , they assume that star formation has been proceeding at a constant rate over a time - interval longer than the added durations of the class i and class ii phases . in our model , this hypothesis holds in the low - density star - forming regions of the solar neighborhood , provided that the star formation efficiency per free - fall time is not higher than a few per cents . if the star formation rate decreases ( as expected in our model when @xmath1 ) , the class i lifetime estimate should be extended . in contrast , if the star formation rate increases with time , as suggested by @xcite , then the class i phase duration is to be shortened . + the comparison between the mass star formation rate predicted by a model and the number star formation rate retrieved from the observations may become even more intricate than presented above if the mass of stars depend on the time of their formation . should that be the case , the star formation rates per mass and number evolve differently . for instance , if the high - mass stars form first , the mass star formation rate declining with time predicted by our model will lead to a smoother decline of the number star formation rate or , possibly , to an accelerating number star formation rate . additionally , the detection of the stars with the lowest luminosity is hindered by the presence of the star - forming gas . therefore , the stellar sample may be incomplete for the low - mass and/or old pre - main sequence stars , making the model - observations comparison even less straightforward . to discuss these two points , however , is well beyond the scope of the present paper . we have modeled the stellar age distributions of young clusters based on the density - dependent star formation model of @xcite . in this class of models , the star formation efficiency increases more rapidly in higher volume - density gas than in low volume - density gas . our model starts with starless molecular clumps in equilibrium , and we assume that a static configuration is retained throughout the star formation episode . for this reason , we limit the duration of star formation to five initial free - fall times of the clump . we also assume that the clump neither gains nor loses mass . therefore , the star formation rate is steadily decreasing with time since , as star - formation depletes the clump gas , the star formation efficiency per free - fall time is applied to an ever lower amount of gas on an ever longer time - scale . our results are equations describing the star formation rate evolution of a clump with given mass and radius ( eq . [ eq : sfr2 ] ) . we also provide analytical equations obtained for the specific case of an isothermal sphere of gas ( eqs . [ eq : sfrcpl ] and [ eq : cround ] ) . for each of our model clumps , we have presented the star formation history and the linear distributions of the star ages if the newly - formed cluster is observed 5myr after the onset of star formation . regarding the star age distribution , we have considered the _ intrinsic _ age distribution , where star ages are binned with a bin size of 0.1myr , and the _ measured _ age distribution , which accounts for the errors affecting individual star ages . that is , the measured age distribution is the intrinsic one convolved with a gaussian function . we have successively discussed the star formation history of the clump in relation to the clump properties ( mainly mean density and star formation efficiency per free - fall time ; see figs . [ fig : princ ] and [ fig : princb ] ) and how the star formation history translates into the cluster star age distribution ( figs . [ fig : ad ] and [ fig : adb ] ) . in our study of the clump star formation history , we have considered two regimes , depending on whether the star formation rate decreases by a factor of 2 during star formation or not . if the evolution is rapid enough for it to decrease by a factor of 2 , one can define the half - life time of the clump star formation rate , namely , the time needed for the star formation rate to decrease to half its initial value . this half - life time is a key - driver of the fwhm of the cluster intrinsic age distribution . if the star formation rate has decreased by less than a factor of two at the end of the star - formation episode , the half - life time of the star formation rate can not be defined , and the fwhm of the intrinsic age distribution is driven by the duration of star formation . we have embodied these two regimes with two distinct star formation efficiencies per free - fall time , @xmath1 and @xmath2 . when @xmath1 ( rapid evolution ) , the half - life time of the star formation rate is of the order of the free - fall time at the clump mean volume density . ( note that this result is valid for an isothermal sphere of gas . shallower density profiles lead to longer half - life times . ) therefore , the denser the cluster - parent clump , the shorter the half - life time of its star formation rate and the shorter the stellar age spread of the resulting cluster . caution should be taken , however , as the fwhm of the age distribution does not depend only on the half - life time of the star formation rate . firstly , star - age binning may prevent a short half - life time from being resolved in the intrinsic age distribution . this is the case encountered for our densest model clump ( model a ) . secondly , uncertainties in the age of individual stars inflate the measured age spread . in case of a very dense fastly - evolving clump , the measured age spread embodies mostly the star age uncertainty and loses memory of the short star formation rate half - life time . therefore , these two effects large age - bin size and age uncertainties weaken the contrast between the measured age spread for high- and low - density clumps ( see fig . [ fig : sum ] , top panel ) . the predicted effect that high - density clumps give rise to narrower age spreads seems to have been observed , with the fwhm of the stellar age distribution in the central regions of the high - density clusters ngc3603 yc and wd 1 found to be an order of magnitude smaller than for the orion nebula cluster ( see * ? ? ? * ; * ? ? ? for the central regions of the young starburst cluster wd 1 , @xcite derived a fwhm of 0.4myr . they estimated a photometric error contribution of 0.25 myr . assuming that the observed fwhm is the quadratic sum of the age error and of the intrinsic fwhm , we derive @xmath199myr . this value agrees well with what we found for clump b ( see top panel of fig . [ fig : sum ] ) . we therefore suggest that the central regions of wd 1 formed out of gas with a density of the order of @xmath200 if @xmath87 . matching a cluster age spread to the density of its parent clump can be done by using the bottom panel of fig . [ fig : sum ] . it provides the relation between the half - life time and the mean volume density of molecular clumps with an @xmath201-density profile when @xmath1 . observers can build on it to predict the relation between the clump mean density and the fwhm of the stellar age distribution they would measure for their own age - bin size and uncertainties in stellar ages . it can also be rescaled to another @xmath27-value : lower efficiencies lead to higher gas densities for given fwhms . if the star formation efficiency per free - fall time is lower than 0.1 , the evolution is slower . this implies that to achieve a given age spread @xmath119 , the free - fall time of the gas must be shorter ( to compensate for the slower @xmath0 ) and , therefore , the gas density must be higher . as an example , if an intrinsic age spread @xmath119 is achieved by a clump of mean density @xmath139 with @xmath36 , the same age spread is achieved by a clump of mean density @xmath202 if @xmath203 . but what if the star formation rate decreases so slowly that its half - life time can not be defined ? if the star formation efficiency per free - fall time is as low as @xmath2 , it takes about 10 clump free - fall times for the star formation rate to decrease to half its initial value ( again assuming an isothermal density profile ) . therefore , if the star formation duration is shorter than that , the half - life time can not be defined , and the fwhm of the intrinsic star age distribution is driven by the duration of the star formation episode @xmath120 ( e.g. bottom panel of fig . [ fig : ad ] ) . to estimate @xmath120 requires a detailed understanding of how long it takes for stellar feedback processes to terminate star formation inside the clump , which is beyond the scope of this paper . yet , we have shown that building on the invariance of the shape of the _ young _ cluster mass function can help us reach some conclusions . that the shape of the cluster mass function does not change with time at young ages points to a mass - invariant star formation efficiency and , therefore , to a duration of star formation equal to a given number of free - fall times for all cluster - forming clumps . therefore , in this case too , we reach the conclusion that the @xmath119 scales with the clump free - fall time @xmath3 , i.e. @xmath204 . again , denser clumps give rise to narrower star age distributions . the normalisation of the @xmath139-@xmath119 relation depends on how many free - fall times star formation lasts . we have also searched the literature for the age spreads and densities of observed star - forming regions . we find that current data support a scenario where the characteristic time - scale of star formation is of order @xmath4-@xmath5 , which is reminiscent of the earlier study by @xcite ( see fig . [ fig : litdat ] ) . note that the @xmath118-axis of fig . [ fig : litdat ] often refers to the observed fwhm of the stellar age distribution . in our model , it reflects the half - life time @xmath86 of the star formation rate . as such , the fwhm is not representative of the duration of star formation which can be significantly longer . the key conclusion of our work is that the observation in young clusters of stellar age spreads spanning 1-to-2 orders of magnitude @xcite does not require different cluster - formation mechanisms @xcite . this may simply reflect the wide range of volume densities among cluster - forming clumps . our work also suggests that the intrinsic star age spread in clusters may constitute a promising probe into the mean volume density of their gaseous progenitors , provided that the star formation efficiency per free - fall time is well - constrained . gp acknowledges support from the olympia - morata program of heidelberg university . this work was also supported by the sonderforschungsbereich sfb 881 `` the milky way system '' ( subproject b5 ) of the german research foundation ( dfg ) . we thank nicola da rio and rob jeffries for interesting discussions , and the referee for a thorough report .

current phenomenology suggests the presence of a compact baryon - poor energy source to cosmological gamma - ray bursts reacting to high - density matter . the association of short and long gamma - ray bursts with hyper- and suspended accretion onto slowly and rapidly spinning black holes predicts weak x - ray afterglow emissions from short bursts , as in grb050509b and grb050709 . long gamma - ray bursts are probably the birth place of rapidly spinning high - mass black holes in core - collapse of massive stars , as in grb030329 with supernova sn2003dh . this predicts a long - duration burst of gravitational radiation powered by the spin - energy of the black hole . in contrast to mev - neutrino emissions , as in sn1987a , this can be tested by the advanced detectors ligo and virgo about once per year up to distances of about 100mpc . detection of the expected chirps and long - duration bursts of gravitational waves promises identification of kerr black holes as the most luminous objects in the universe .

astro-ph0510348

gammma - ray bursts ( grbs ) were discovered serendipitously as flashes in the sky " by the nuclear test - ban monitoring satellites vela ( us ) and konus ( ussr ) . these were soon recognized to be a natural phenomenon . the first data were publicly released by @xcite and @xcite . the mysterious origin of this new astrophysical transient , lasting up to tens of seconds in non - thermal gamma - ray radiation , gradually triggered a huge interest from the high - energy astrophysics community . the burst and transient source experiment ( batse ) on nasa s compton gamma - ray observatory , launched in 1991 , identified an isotropic distribution in the sky and revealed the existence of short and long gamma - ray bursts , whose durations are broadly distributed around 0.3s and 30s , respectively @xcite . moreover , the sensitivity of batse showed a deficit in faint bursts in the number - versus - intensity distribution distinct from a -3/2 powerlaw . @xcite hereby established that gamma - ray bursts are of cosmological origin . more recently , the italian - dutch satellite bepposax discovered x - ray flashes @xcite , which may be closely related to long gamma - ray bursts . their non - thermal emissions are herein best accounted for by internal shocks in ultrarelativistic outflows in the fireball model proposed by @xcite with unknown inner engine producing these baryon - poor outflows . if gravitationally powered , their compact size , based on short - time scale variabilities in the gamma - ray lightcurves , may involve the formation of black holes surrounded by a high - mass torus @xcite . at cosmological distances , the isotropic equivalent luminosities are on the order of @xmath0 erg s@xmath1 as bright as @xmath2 solar luminosities . these events , widely acclaimed as the biggest bang since the big bang " , are now known to light up the universe about once per minute . what , then , is their origin ? the most powerful observational method to encyst the enigmatic energy source is through _ calorimetry on all radiation channels_. this involves measuring the energy output in emissions in the electromagnetic spectrum , as well as in yet unseen " emissions in neutrinos and gravitational waves . perhaps thus shall we be in a position to determine their constitution , and decide whether their inner engine is _ baryonic _ , such as a neutron star , or _ non - baryonic _ : a black hole energized by spin . the exact solution of roy p. @xcite see table i for a parametrization shows an energy per unit mass which is anomalously large at high spin rates , exceeding that of a rapidly spinning neutron by an order of magnitude . by this potentially large and baryon - free energy reservoir , a kerr black hole is a leading candidate for the inner engine of grbs , at least those of long - durations in view of the following observational results . the italian - dutch bepposax satellite , launched in 1996 , dramatically changed the landscape of long grbs with the discovery by e. costa @xcite of x - ray afterglows in grb970228 ( confirmed in observations by the asca and rosat satellites ) . these lower - energy emissions permit accurate localizations , enabling follow - up by optical telescopes . thus , j. van paradijs @xcite pointed the isaac newton telescope and the william herschel telescope to further discover optical emissions associated with the same event . these lower - energy x - ray and optical emissions agree remarkably well with the expected decay of shocks in ultrarelativistic , baryon - poor outflows in the previously developed fireball model . even lower - energy , radio - afterglows have been observed in some cases , including grb970228 @xcite . the same optical observations have now led to tens of grbs with individually measured redshifts , up to @xmath3 in grb050904 detected by the recently launched swift satellite @xcite . optical follow - up identified the association of long grbs with supernovae . in particular , the unambiguous association of grb030329 with sn2003dh ( @xmath4mpc ) @xcite identified type ib / c supernovae as the parent population of long grbs ( fig . 1 ) . this observation confirms their association with core - collapse supernovae of massive stars , proposed by s. woosley @xcite . massive stars have characteristically short lifetimes , whereby grb - supernovae track the cosmological event rate . this conclusion is quantitatively confirmed by consistent estimates of the true - to - observed grb - sne event rate of 450 - 500 deduced from two independent analysis , based on geometrical beaming factors @xcite and locking to the star - formation rate @xcite . these event rates correspond to a true cosmological event rate of about one per minute . core - collapse supernovae from massive stars are believed to produce neutron stars or black holes . the former are perhaps best known from the rapidly spinning neutron star ( a pulsar ) in the crab nebula , a remnant of what was probably a type ii supernova in 1054 . the second is more difficult to ascertain with certainty , such as in the more recent type ii event sn1987 in the large megellanic cloud . at close proximity , its burst in mev - neutrinos was detected by kamiokanda and imb ( see @xcite ) which provided direct evidence for the formation of matter at nuclear densities . the absence of a pulsar or a point - source of x - ray radiation in this case suggests continued collapse probably into a stellar mass black hole . type ii and type ib / c supernovae are both believed to represent the endpoint of massive stars @xcite , and possibly so in binaries such as the type ii / ib event sn1993j @xcite . this binary association suggests a hierarchy , wherein hydrogen - rich , envelope retaining snii are associated with wide binaries , while hydrogen - poor , envelope stripped snib and snic are associated with increasingly compact binaries @xcite . by tidal coupling , the primary star in the latter will rotate at approximately the orbital period at the moment of core collapse . with an evolved core @xcite , these type ib / c events in particular are believed to produce a spinning black hole @xcite . however , the branching ratio of type ib / c into grb - supernovae is small , @xmath5 as calculated from the true grb - supernova event rate relative to the observed event rates of supernovae of type ii and type ib / c . this ratio is remarkably small , suggesting a higher - order down - selection process . the identification of long grbs with core - collapse supernovae brings into scope their potential gravitational - wave emissions . this may be due to a variety of mechanisms associated with rapidly rotating fluids . in particular , emissions will be produced by bar - mode instabilities producing fragmentation of orbiting matter prior to the formation of a compact object @xcite or in the formation of multiple compact objects @xcite , non - axisymmetries in accretion disks @xcite , as well as after the formation into a rapidly spinning black hole . the first may produce an initial splash " of gravitational radiation @xcite , the second a bi - model spectrum containing high - frequency emissions produced by non - axisymmetric perturbations of the black hole and low - frequency emissions from the disk , whereas the latter creates a long - duration burst of low - frequency gravitational radiation . calculations this time - evolving spectrum of gravitational - radiation are only beginning to be addressed by computational hydro- and magnetohydrodynamics , during initial collapse @xcite and in the formation of a non - axisymmetric torus @xcite . emissions from matter surrounding a stellar - mass black hole are in the frequency range of sensitivity of the gravitational - wave experiments ligo @xcite and virgo @xcite . understanding their wave - forms can serve strategic search - and - detection algorithms , triggered by gamma - ray observations @xcite or by their associated supernova . the latter is more common by the aforementioned true - to - observed event ratio of grb - supernovae . in this review , we shall discuss a theory of grbs from rotating black holes . current phenomenology on grb - supernovae poses a challenge to model * the durations of long grbs of tens of seconds * the formation of an associated aspherical supernova * the launch of ultrarelativistic jets with beamed gamma - ray emissions * a small branching ratio of type ib / c supernovae into grbs of less than @xmath6 this phenomenology will be directly linked to the spin energy of the black hole . quite generally , we are left with the task of modeling all possible radiation channels produced by the putative kerr black hole , including gravitational radiation , mev - neutrino emissions , and magnetic winds . just as lower - energy x - ray and optical afterglow emissions linked long grbs to supernovae , we expect that the detection of a contemporaneous burst in gravitational - radiation will lift the veil on their enigmatic inner engine . a precise determination of this link will enable the identification of a kerr black hole . the same might apply to type ii and type ib / c supernovae . understanding the energy source of long grbs might tell us also about the constitution of short grbs . the first _ faint _ x - ray afterglows have been detected of the short event grb050509b by swift @xcite and grb050709 by hete - ii @xcite , further providing the low redshifts @xmath7 and @xmath8 , respectively . the nature of host galaxies , preferably ( older ) elliptical galaxies for short bursts versus ( young ) star - forming galaxies for long bursts , supports the origin of short events in binary coalescence of black holes and neutron stars @xcite . after completion of this review , @xcite outlined a very interesting prospect for long - lasting supernova - like signatures to short bursts from the debris of a neutron star binary coalescence with a similar partner or black hole . earlier , we explained the dichotomy of short and long events in terms of hyper- and suspended accreting onto slowly and rapidly rotating black holes , respectively @xcite . thus , we predicted similar x - ray afterglows with the property that those to short events are relatively faint : _ the short burst is identical to the final moment of a long burst of gamma - rays_. hyperaccreting , slowly rotating black holes can be produced through various channels . black - hole neutron star binaries are remnants of core - collapse events in binaries ( @xcite ; unless formed by capture in dense stellar clusters ) . any rapidly spinning black hole will be spun - down in the core - collapse process , possibly representing a prior grb - supernova event ( this review ) . the merger of a neutron star onto a black hole @xcite could produce a state of hyperaccretion and produce a short gamma - ray burst . this scenario is consistent with the predominance of long over short events . less clear is the potential for gamma - ray emissions from the merger of two neutron stars ( e.g. @xcite ) which , however , should also produce a slowly rotating black hole . finally , slowly rotating black holes can also be produced in core - collapse of massive star in isolation or a wide binary . here as well , may detection of their gravitational - wave emissions provide a probe to differentiate amongst these different types of short bursts . in 2 , we give a practical outlook on kerr black holes and gravitational radiation . in 3 , we review the formation and evolution ( kick - velocities , growth and spin down ) of kerr black holes in core collapse supernovae . this includes the process of converting spin energy into radiation and estimates on the lifetime of rapid spin . in 4 , we calculate multipole mass - moments in a torus due to a hydrodynamical instability . by asymptotic analysis , the energies in the various radiation channels are given in 5 . an mechanism for launching ultrarelativistic jets from rotating black holes by a small fraction of black hole spin is given in 6 . specific predictions for these flashes " as sources for the observatories ligo and virgo are given in 7 . kerr black holes and gravitational radiation are two of the most dramatic predictions of general relativity ( other than cosmology ) . while evidence for kerr black holes , in particular in their role as powerful inner engines , remains elusive , the quadrupole formula for gravitational radiation has been confirmed observationally with great accuracy . the kerr metric describes an exact solution of a black hole spacetime with nonzero angular momentum . it demonstrates the remarkable property of _ frame - dragging _ : the angular velocity @xmath9 of otherwise zero - angular momentum observers in spacetime outside the black hole . frame - dragging assumes a maximal and constant angular velocity @xmath10 on the event horizon of the black hole , and decays with the cube of the distance to zero at large distances . this introduces an angular velocity @xmath10 of the black hole , as well as _ frame - dragging by a non - zero gradient ( whereby it is not gauge effect ) . the aforementioned energy reservoir in angular momentum satisfies , @xmath11 where @xmath12 denotes the specific angular momentum of the black hole per unit mass . here , we use geometric units with newton s constant @xmath13 and velocity of light @xmath14 . these properties give kerr black holes the potential to react energetically to their environment . they hereby have to potential of serving as universal sources of energy , distinct from any known baryonic object and with direct relevance to the phenomenology of grbs . rotating objects have a general tendency to radiate away their energy and angular momentum in an effort to reach a lower energy state . in the dynamics of rotating fluids , this is described by the well - known rayleigh stability criterion . in many ways , black hole radiation processes are governed by the same principle . the second law of thermodynamics @xmath15 for the entropy @xmath16 shows that the specific angular momentum of a kerr black hole @xmath17 ) increases with radiation : @xmath18 for a black hole of mass @xmath19 and angular momentum @xmath20 which emits a particle with specific angular momentum @xmath21 . generally , black holes in isolation are stable by exponential suppression of spontaneous emissions by canonical angular momentum barriers @xcite . fortunately , magnetic fields can modulate and suppress these angular momentum barriers . the earliest studies of active black holes focus on energy - extraction processes in the ergosphere , i.e. : scattering of positive energy waves onto rotating black holes superradiant scattering of @xcite , @xcite , @xcite and @xcite as a continuous - wave analogue to the process of @xcite . a steady - state variation can be found in the process of spin down through horizon maxwell stresses , proposed @xcite and extended to force - free magnetospheres by @xcite . quite generally , therefore , magnetic fields are required as a mediating agent to stimulate luminosities which are of astrophysical interest . generally , stellar mass black holes produced in core - collapse of a massive star are parametrized by their mass , angular momentum and kick velocity @xmath22 . the equilibrium electric charge of the black hole which arises in its lowest magnetic energy state , is relevant in sustaining adequate horizon magnetic flux especially at high spin rates , but is insignificant relative to the spin - energy of the black hole . gravitational - wave emissions are predominantly due to low multipole mass - moments . the quadrupole radiation formula according to the linearized theory of general relativity has been given by @xcite @xmath23 where @xmath24 is a function of the ellipticity @xmath25 of a binary with orbital frequency @xmath9 and chirp mass @xmath26 . this linearized result has been observationally confirmed to within 0.1% by decade - long observations of the hulse - taylor binary neutron star system psr1913 + 16 ( with ellipticity @xmath27 , @xcite ) . its gravitational - wave luminosity of about about @xmath28 of a solar luminosity @xmath29 erg s@xmath1 gives a binary lifetime of about 7.4 gyr . encouraged by this observational confirmation , we apply ( [ eqn_pm ] ) to a non - axisymmetric matter surrounding a black hole . with orbital period @xmath30 , @xmath31 and ellipticity @xmath32 , this predicts @xmath33 the quadrupole moment is due to a mass - inhomogeneity @xmath34 in the torus . a quadrupole moment appears spontaneously due to non - axisymmetric waves , such as a hydrodynamical or magnetohydrodynamical instability . a detailed description of radiation by all multipole mass - moments in a torus is given in @xcite . these existing approaches to energy extraction from rotating black holes do not elucidate the loading problem : " the diversity in radiation channels and their various energy outputs . this defines a novel challenge for kerr black holes as inner engines to gamma - ray bursts . these events , perhaps including core - collapse supernovae , pose the potential for a powerful link between kerr black holes and gravitational radiation , as outlined below . while all snib / c might be producing black holes , only some are associated with grb - supernovae in view of the observed small branching ratio ( [ eqn_branching ] ) . black holes produced in _ aspherical _ core - collapse receive typical kick velocities of a few hundred km s@xmath1 @xcite , measured relative to the center of mass of the progenitor star . such objects inevitably escape from the central high - density region of the progenitor star , even before core - collapse is completed . a small sample of black holes receive low kick - velocities at birth . remaining centered , these grow into rapidly spinning high - mass black holes by infall of a substantial fraction of the progenitor he - core mass ( fig . this forms a starting point for the parametrization of rotating black holes as inner engines to grb - supernovae . the state of matter surrounding the newly formed black hole in core - collapse supernovae is unique , in attaining temperatures of in excess of 1mev @xcite and masses of up to a few percent of the mass of the black hole @xcite . we further expect these accretion disks to be magnetized , representing a remnant magnetic field of the progenitor star and modified by a dynamo action in the disk . the angular velocity @xmath10 of the black hole easily exceeds that of surrounding matter , even at relatively low spin - energies . when a surrounding disk is largely unmagnetized , the black hole continues to accrete , enlarging it and spinning it up towards an extreme kerr black hole along a _ bardeen trajectory _ @xcite , shown in fig . 2 . alternatively , the surrounding matter can be magnetized and , fairly generally , may contain an appreciable @xmath35 component of poloidal flux on average or , when time - variable , with non - zero standard deviation . topologically , this component represents a uniform magnetization , corresponding to two counter - oriented current rings @xcite . under these conditions , a torus can form in a state of suspended accretion with angular velocity @xmath36 . this introduces a channel for _ catalytic _ conversion of spin energy into various radiation channels @xcite , forcing the black hole to spin down until the angular velocities become similar @xmath37 . a magnetized torus surrounding a black hole is expected to form in both black hole - neutron star coalescence and core collapse of a massive star with very similar topological properties ( fig . a magnetized star can be represented to leading order by a single current loop or equivalently a density of magnetic dipole moments . stretching is around the black hole , as in tidal break - up , or by excision of the center of the star , as in core - collapse into a new black hole , leaves a magnetized annulus consisting of two counter - oriented current rings . thus , core - collapse supernovae and binary black hole - neutron - star coalescence both give the same outcome : a black hole surrounded by a magnetized torus . in practice , the resulting magnetic field will be modified by turbulence and , possibly , a dynamo . we can look at the structure of the magnetic field of a torus by inspecting its shape in a poloidal cross - section , and by comparison with pulsar magnetospheres . in case of a pulsar , we see magnetic flux - surfaces of closed magnetic field - lines that reach an outer light cylinder , while open magnetic field - lines extend outwards to infinity . the same structure is found with regards to the torus outer face , when viewed in poloidal cross - section . since the event horizon of the black hole is a null - surface , it has the same radiative boundary condition as asymptotic infinity @xcite , except for its finite surface area ( and surface gravity ) . it follows that the structure of magnetic field - lines found in a magnetized neutron star also applies to the torus inner face . for the same reason that a spinning neutron star transfers angular momentum to infinity , the inner face of the torus gains angular momentum from the black hole , whenever the latter spins more rapidly this produces a _ connection by topological equivalence to pulsars_. angular momentum transport is by alfvn waves emitted from the surface of the torus , carrying positive angular momentum waves off the outer face to infinity and negative angular momentum off the inner face into the black hole . in this process , the outer face becomes sub - keplerian and the inner face becomes super - keplerian . on balance of the competing torques between these two faces @xcite , this produces a state of suspended accretion @xcite . the angular velocity of a kerr black hole can reach @xmath38 , which far surpasses that of an accretion disk or torus . the black - hole spin energy ( [ eqn_rot ] ) can be transferred to a torus with angular velocity @xmath39 with efficiency @xmath40 the task is now to quantify the various radiation channels provided by the torus . we shall discuss these in the next sections . the process of formation and spin down of a kerr black hole has been proposed to quantitatively model the supernovae by various groups @xcite . this approach represents a special case of aspherical , magneto - rotationally driven core - collapse supernovae , notably discussed for type ii supernovae by @xcite . a given torus can support a finite poloidal magnetic field - energy . this is due to an instability , produced by magnetic moment - magnetic moment self - interactions in the fluid . instability criteria can be derived for both tilt and buckling @xcite . for a tilt instability between two counter - oriented current rings , representing a uniformly magnetized torus , these interactions are described by a potential @xmath41 where @xmath42 denotes the magnetic moment dipole moment of the inner ring , @xmath43 denotes the magnetic field produced by the outer ring , and @xmath44 denotes the angle between @xmath42 and @xmath43 . note that @xmath45 has period @xmath46 , is maximal ( minimal ) when @xmath42 and @xmath43 are parallel ( antiparallel ) . for the buckling instability , the same interaction energies arise in the azimuthal partition of the torus into small current rings which , combined , are equivalent to the two counter - oriented current rings . the central potential well of the black hole provides a stabilizing contribution . as the magnetic moment - magnetic moment interactions act primarily to introduce vertical displacements between two current rings ( about their equilibrium configuration in the equatorial plane ) , we can focus on vertical displacements of fluid elements along surfaces of constant cylindrical radius @xmath47 . this is distinct from motions of a rigid ring , whose fluid elements move on surfaces of constant spherical radius . in particular , the tilt of a current ring hereby changes the distance to the central black hole according to @xmath48 . in the approximation of equal mass in the inner and outer face of the torus , simultaneous tilt of one ring upwards and the other ring downwards is associated with the potential energy ( @xcite , corrected ) @xmath49 with @xmath50 upon averaging over all segments of a ring . note that @xmath51 has period @xmath52 and is minimal at @xmath53 . similar expressions hold for an azimuthal distribution of current rings . stability is accomplished provided that the total potential energy @xmath54 satisfies @xmath55 . we find the following poloidal magnetic field - to - kinetic energy ratios @xmath56 in the approximation @xmath57 , representing the poloidal magnetic field - energy of the inner torus magnetosphere in a characteristic volume @xmath58 and @xmath59 . we next discuss the physical parameters at this point of critical stability , which we interpret as a practical limit on the magnetic field energy that the torus can support . for a pair of rings of radii @xmath60 , @xmath61 , we have @xmath62 , so that the point of critical stability , @xmath63 , gives for the critical magnetic field - strength @xmath64 , or @xmath65 with critical poloidal magnetic field - energy ( [ eqn_eb1 ] ) . rotating black holes with @xmath66 dissipate most of their spin energy unseen " in the event horizon . the lifetime of rapid spin is thus @xmath67 where @xmath68 denotes the horizon flux , taking into account that most of the black - hole luminosity is incident onto the inner face of the torus . we then have @xmath69 where the subscripts denote normalization constants , i.e. : @xmath70 , @xmath71 , @xmath72 and @xmath73 corresponding to @xmath74 in ( [ eqn_rot ] ) . this estimate agrees well with the observed durations of tens of seconds of long grbs shown in fig . ( [ fig_y1 ] ) and statistics of the batse catalogue @xcite . a torus tends to be unstable to a variety of symmetry - breaking instabilities . this provides a spontaneous mechanism for the creation of multipole mass - moments . provided that the torus is not completely disrupted , such multipole mass - moments ensure that the torus is luminous in gravitational radiation , when its mass reaches a few percent of that of the black hole . these instabilities can take the form of hydrodynamic and magnetohydrodynamic instabilities . @xcite describe a hydrodynamic buckling instability in infinitesimally slender tori , due to a coupling of surface waves on a super - keplerian inner face and a sub - keplerian outer face . this theory can be extended to tori of arbitrary slenderness to be of practical interest ( below ) . in a recent numerical study , the magnetic moment - magnetic moment self - interactions of a uniformly magnetized torus are also found to produce instabilities buckling instabilities in the poloidal plane @xcite . this fully nonlinear numerical study , though one - dimensional , recovers our heuristic analytical bound on the maximal poloidal magnetic field energy - to - kinetic energy in the torus ( [ eqn_eb1 ] ) . it confirms that the dominant emission channel in gravitational radiation is through the quadrupole mass - moment . these are but two example calculations on the more general challenge of computing gravitational - wave spectra of magnetized tori in suspended accretion . below , we discuss the hydrodynamic buckling instability for wide tori . the dynamical stability of a torus with sub - keplerian outer face and super - keplerian inner face can be studied in the limit of an inviscid incompressible fluid with newtonian angular velocity @xcite @xmath75 here , @xmath76 denotes the rotation index which is bounded between the keplerian value @xmath77 and rayleigh s stability limit @xmath78 . irrotational perturbations to the underlying flow ( vortical if @xmath79 ) remain irrotational by kelvin s theorem . in studying their stability we consider the harmonic velocity potential in cylindrical coordinates @xmath80 @xmath81 the equations of motion can be expressed in a local cartesian frame @xmath82 with newtonian angular velocity @xmath83 , equal to the angular velocity of the torus at its major radius @xmath84 about a central mass @xmath19 . these cartesian coordinates are related to cylindrical coordinates @xmath85 according to @xmath86 , @xmath87 and @xmath88 . together with zero - enthalpy boundary conditions on the free inner and outer surface of the torus , we obtain a complete problem for linearized stability analysis . this problem can be solved semi - analytically , by searching numerical for points of change in stability @xcite of harmonic perturbations @xmath89 of infinitesimal amplitude at frequency @xmath90 as seen in a corotating frame at @xmath84 . 4 shows the numerical stability diagram . in general , we encounter for each @xmath91 a critical rotation index @xmath92 which depends on the minor - to - major radius @xmath93 of the torus . these curves can be found using numerical continuation methods @xcite . the stability diagram for the rotation index is shown in fig . in particular , we mention the critical values @xmath94 for the various @xmath91-modes at the rayleigh stability line @xmath78 . rotating black holes in suspended accretion dissipate most of their rotational energy ( [ eqn_rot ] ) unseen " in their event horizon , while a major fraction ( [ eqn_eta ] ) is incident into the inner face of the surrounding torus . this lasts for the lifetime of rapid spin ( [ eqn_ts ] ) , whose tens of seconds represents a secular timescale relative to the millisecond period of the orbital motion of the torus . this reduces the problem of calculating the radiation output from the torus to algebraic equations of balance in energy and angular momentum flux , taking into account the channels of gravitational radiation , mev - neutrino emissions and magnetic winds . the equations of suspended accretion are @xmath95 where ( @xmath96 denote the torques on and angular velocities of the inner and outer face of the torus due to surface maxwell stresses as those on a pulsar , @xmath97 denotes the torque on the torus due to the emission of gravitational radiation with luminosity @xmath98 and @xmath99 the power in mev - neutrino emissions due to dissipation . ( the results show temperatures of a few mev . ) these equations are closed by a constitutive relation for the dissipation process . we set out by attributing dissipative heating to magnetohydrodynamical stresses . this introduces an overall scaling with the magnetic field - energy @xmath100 . the resulting total energy emissions , @xmath101 , @xmath102 and @xmath103 , produced over the lifetime of rapid spin of the black hole ( [ eqn_ts ] ) , thus become _ independent _ of @xmath100 and reduce to specific fractions of @xmath104 . when the torus is sufficiently slender , @xmath105 modes ( two lumps swirling around the black hole ) develop . these radiate at essentially twice the orbital frequency of the torus , when the minor - to - major radius is less than 0.3260 by ( [ eqn_bam ] ) . by an asymptotic analysis of ( [ eqn_sas ] ) , the gravitational - wave emissions satisfy @xcite @xmath106 thus , an unseen " energy output ( [ eqn_egw ] ) surpasses the true energy @xmath107erg in gamma - rays @xcite by several orders of magnitude . the associated output in magnetic winds contains an additional factor @xmath108 , i.e. , @xmath109 this baryon - rich wind provides a powerful agent towards collimation of any outflows from the black hole @xcite , as well as a source of neutrinos for pick - up by the same @xcite . it is otherwise largely incident onto the remnant stellar envelope _ from within _ , which serves as energetic input to accompanying supernova ejecta . we estimate these kinetic energies to be @xmath110 with @xmath111 denoting the velocity @xmath112 of supernova ejecta relative to the velocity of light @xmath113 . in the expected aspherical geometry , @xmath114 refers to the mass - average taken over all angles . the canonical value @xmath115 refers to the observed value in grb011211 @xcite . eventually , the expanding remnant envelope becomes optically thin , which permits the appearance of x - ray line - emissions excited by the underlying continuum emission @xmath116 by dissipating @xmath102 . the estimate ( [ eqn_sn ] ) is in remarkable agreement with the kinetic energy @xmath117 erg of the aspherical supernova sn1998bw associated with grb980425 @xcite . the energy output in mev - neutrinos is intermediate , in being smaller than @xmath101 by an additional factor given by the slenderness parameter @xmath118 , where @xmath119 denotes the ratio of minor - to - major radius of the torus and @xmath76 its rotation index ( [ eqn_ome ] ) . thus , we have @xmath120 where @xmath121 . at the associated dissipation rate , the torus develops a temperature of a few mev which stimulates the production of baryon - rich winds @xcite . frame - dragging induced by angular momentum extends through the environment of the black hole and includes the spin axis . the kerr metric provides an exact description of this gravitational induction process by the riemann tensor @xcite . in turn , the riemann tensor couples to spin @xcite . specific angular momentum ( angular momentum per unit mass ) represents a rate of change of surface area per unit of time , while the riemann tensor is of dimension @xmath122 ( in geometrical units ) . therefore , curvature spin coupling produces a force ( dimensionless in geometrical units ) , whereby test particles with spin follow non - geodesic trajectories @xcite . in practical terms , the latter holds promise as a mechanism for _ linear acceleration_. by dimensional analysis once more , the gravitational potential for spin aligned interactions satisfy @xmath123 where @xmath9 denotes the local frame - dragging angular velocity produced by the black hole ( or any other spinning object ) and @xmath124 is the angular momentum of the particle at hand . spinning bodies therefore couple to spinning bodies @xcite . thus , ( [ eqn_eo ] ) defines a mechanism for accelerating baryon - poor ejecta to ultrarelativistic velocities , provided @xmath124 is large . the angular momentum @xmath124 of charged particles in strong magnetic fields , confined to individual magnetic flux - surfaces in radiative landau states , is macroscopic in the form of orbital angular momentum @xmath125 here @xmath25 denotes the elementary charge and @xmath68 denotes the enclosed magnetic flux , where @xmath126 denotes the electromagnetic vector potential of an open magnetic flux - tube along the spin axis of the black hole . geometrically , the specific angular momentum represents the rate of change of surface area traced out by the orbital motion of the charged particle , by helical motion " as seen in four - dimensional spacetime . combined , ( [ eqn_eo ] ) and ( [ eqn_eo1 ] ) describe a powerful mechanism for linear acceleration of baryon - poor matter . it radically differs from the common view , that black - hole energetic processes are limited exclusively to the action of frame - dragging in the ergosphere . it is instructive to derive ( [ eqn_eo ] ) by specializing to the kerr metric on the spin axis of the black hole . to this end , we may consider one - half the difference in potential energy of particles with angular momenta @xmath127 suspended in a gravitational field about a rotating object . in the first case , we note that the riemann tensor @xmath128 expressed relative to tetrad elements associated with boyer - lindquist coordinates @xcite gives rise to a linear force @xmath129 which can be integrated along the spin axis to give @xmath130 in the second case , we merely assume a metric @xmath131 with time - like and azimuthal killing vectors and consider two particles with velocity four - vectors @xmath132 according to angular momenta @xmath133 , @xmath134 which shows that @xmath135 is the same for both particles . the total energy of the particles @xmath136 gives rise to @xmath137 to apply ( [ eqn_eo]-[eqn_eo1 ] ) to gamma - ray bursts from rotating black holes , consider a perfectly conducting blob of charged particles in a magnetic flux - tube subtended at a finite half - opening angle @xmath138 on the event horizon of the black hole and along its spin axis . in the approximation of electrostatic equilibrium , it assumes a rigid rotation @xcite described by an angular velocity @xmath139 . ( by faraday induction , differential rotation introduce potential differences along magnetic field - lines . some differential rotation is inevitable over long length scales and , possibly , in the formation of gaps ) . in the frame of zero - angular momentum observers , the equilibrium charge - density assumes the value of @xcite , as viewed by zero - angular momentum observers . thus , given a number @xmath140 of charged particles per unit scale height @xmath141 in the flux - tube , @xmath142 a pair of blobs of scale height @xmath143 in both directions along the spin axis of the black hole each hereby receive a potential energy @xmath144 where @xmath145 g and @xmath146 is a quantity of order unity , expressed in terms of the normalised angular velocities @xmath147 and @xmath148 . electrons and positrons in superstrong magnetic fields are essentially massless . the ejection of a pair of blobs with energy ( [ eqn_eb ] ) thus takes place in a light - crossing time of about 0.3ms of , e.g. , seven solar mass black hole of linear dimension @xmath149 cm . this corresponds to in an instantaneous luminosity of about @xmath150erg / s . this produces a total kinetic energy output of up to @xmath151 erg in tens of seconds in long bursts . a fraction hereof which will be dissipated in gamma - rays and lower - energy afterglow emissions in the internal - external shock model for grbs @xcite . similar results obtain by considering the luminosity in a steady - state limit , by considering a horizon half - opening angle @xmath152 , set by the poloidal curvature @xmath153 of the magnetic field - lines . we note the paradoxical _ small _ energy output of grb - afterglow emissions , when viewed relative to the total black - hole spin energy ( [ eqn_rot ] ) . the luminosity in the jet scales with the square of the enclosed magnetic flux , while the latter scales with the enclosed surface area , and hence the square of the half - opening angle @xmath138 . thus , we encounter a geometrical scale - factor , which creates a jet luminosity @xmath154 even when @xmath138 is not small , e.g. , about @xmath155 degrees , this scaling creates a small parameter . integrated over time , the energy output in gamma - rays satisfies @xcite @xmath156 consistent with the observed true energy output @xmath150 erg in gamma - rays of long bursts @xcite , where @xmath157 and @xmath158 denotes the efficiency of converting kinetic energy to gamma - rays . in our unification scheme , the durations of short and long bursts are attributed to different spin - down times of , respectively , a slowly and rapidly spinning black hole according to ( [ eqn_ts ] ) . the total energy output in gamma - rays of short grbs is hereby significantly smaller than that of long grbs , due to both shorter durations and smaller luminosities . in this light , the recent x - ray afterglow detections to short bursts , very similar but fainter than their counterparts to long bursts , are encouraging . the output in gravitational radiation ( [ eqn_egw ] ) is in the range of sensitivity of the broad band detectors ligo and virgo shown in fig . 5 , as well as geo @xcite and tama @xcite . note that the match is better for higher - mass black holes , in view of their emissions at lower frequencies ( towards the minimum in the detector noise curve ) and their larger energy output . matched filtering gives a theoretical upper bound on the signal - to - noise ratio in the detection of long bursts in gravitational radiation from grb - supernovae . in practice , the frequency will be unsteady at least on an intermittent timescale associated with the evolution of the torus , e.g. , due to mass - loss in winds and possibly mass - gain by accretion from additional matter falling in . for this reason , a time - frequency trajectory method which correlates the coefficients of a fourier transformation over subsequent windows of durations of seconds might apply . the ultimate signal - to - noise ratio will therefore be intermediate between that obtained through correlation a second - order procedure and matched filtering a first - order procedure . the results shown in fig . 5 show the maximal attainable signal - to - noise ratio ( by matched filtering ) for sources at a fiducial distance of 100mpc . this distance corresponds to an event rate of one per year . grb - supernovae are an astrophysical source population locked to the star - formation rate . we can calculate their contribution to the stochastic background in gravitational radiation accordingly given their band - limited signals , assuming @xmath159 of around 10% , where @xmath160 denotes the average gravitational - wave frequency in the comoving frame . in what follows , we the following scaling relations are applied , @xmath161 where @xmath162 , @xmath163 and @xmath164hz@xmath165 , assuming maximal spin rates ( @xmath166 ) . for non - extremal black holes , a commensurate reduction factor in energy output can be inserted . this factor carries through proportionally to the final results , whence it is not taken into account explicitly . summation over a uniform distribution of black hole masses , e.g. , @xmath167 , and assuming that the black hole mass and @xmath108 in ( [ eqn_eta ] ) , are uncorrelated , the expected spectral energy - density satisfies @xcite @xmath168 where @xmath169max@xmath170 is a normalized frequency distribution . the associated dimensionless amplitude @xmath171 , where @xmath172 and @xmath173 denotes newton s constant satisfies @xmath174 where @xmath175max@xmath176 , @xmath177 , likewise , we have for the spectral closure density @xmath178 relative to the closure density @xmath179 @xmath180 where @xmath181max@xmath182 , @xmath183 and @xmath184 denotes the hubble constant . these cosmological results show a simple scaling relation for the extremal value of the spectral closure density in its dependency on the model parameter @xmath108 . the location of the maximum scales with @xmath185 in view of @xmath186 . the spectral closure density thus becomes completely determined by the shape of the function representing the star - formation rate , the fractional grb - supernova rate thereof , @xmath108 , and the black - hole mass distribution . 5 shows the various distributions . the extremal value of @xmath187 is in the neighborhood of the location of maximal sensitivity of ligo and virgo . it would be of interest to search for this contribution to the stochastic background in gravitational waves by correlation in the spectral domain , following fourier transformation over series of sub - windows on intermediate timescales of seconds . the sixties saw two independent discoveries : the first grb670702 by the vela satellite and the exact solution of rotating black holes by roy p. kerr . through observational campaigns with batse , bepposax , the interplanetary network ( ipn ) , hete - ii and now swift , we have come to understand the phenomenology of grbs . long bursts are association with supernovae , representing a rare and extraordinary powerful cosmological transient , taking place about once a minute and reaching the earliest epochs in the universe . independently , through the works of @xcite and others , we have come to understand the potential significance of kerr black holes as compact , baryon - free energy sources with certain universal properties . the applications to high - energy astrophysics should be enormous ( e.g. @xcite ) , of supermassive black holes in active galactic nuclei including our own galaxy @xcite , and of stellar mass black holes in microquasars @xcite and , possibly , gamma - ray bursts . while the formation - process of supermassive black holes remains inconclusive , the birthplace of stellar mass black holes is most probably core - collapse supernovae of massive stars . evidence for kerr black holes as the energy source to high - energy astrophysical processes remains elusive , however . recent measurements on frame - dragging by x - ray spectroscopy , typically during inactive states of the black hole , are encouraging in this respect ( a.c . fabian , these proceedings ) . specically , we propose that grb - supernovae are powered by rapidly rotating black holes , wherein ( 1 ) the durations of long grbs of tens of seconds are identified with the lifetime of rapid spin of the black hole in a state of suspended accretion , ( 2 ) an accompanying supernova is radiation - driven by magnetic winds from a torus in suspended accretion , ( 3 ) ultrarelativistic outflows are launched by gravitational spin - orbit coupling with charged particles along open magnetic field - lines , and ( 4 ) a small branching ratio of type ib / c supernovae into grbs is attributed to the small probability of producing black holes with small kick velocities . modeling short grbs from slowly rotating black holes , we predicted x - ray afterglows very similar but weaker ( in total energies ) than those of long bursts . the recent discovery of faint x - ray afterglows to grb050509b and grb050709 fit well within this scheme . gamma - ray bursts present a potentially powerful link between rotating black holes and gravitational radiation . strategic searches for their chirps in binary coalescence of neutron stars and black holes , or long - duration bursts in gravitational radiation during radiative spin - down of a high - mass black hole , can be pursued by the advanced detectors ligo and virgo . strategic searches are preferrably pursued in combination with upcoming optical - radio supernova surveys , e.g. , pan - starrs in hawaii @xcite in combination with the low frequency array @xcite , see also @xcite . in light of the proposed supernova - like signatures from the debris of a neutron star @xcite , these strategies might apply to both short and long bursts . * acknowledgment . * the author thanks a. levinson , r.p . kerr , r. preece , and david wiltshire for constructive comments . this research is supported by the ligo observatories , constructed by caltech and mit with funding from nsf under cooperative agreement phy 9210038 . the ligo laboratory operates under cooperative agreement phy-0107417 . this paper has been assigned ligo document number ligo - p040013 - 00-r . abramovici , a. , althouse , w.e . , drever , r.w.p . , et al . , 1992 , science , 292 , 325 acernese , f. , et al . , 2002 , class . grav . , 19 , 1421 akiyama , s. , wheeler , j.c . , meier , d.l . , lichtenstadt , i. , 2003 , apj , 584 , 954 ando , m. , and the tama collaboration , 2002 , class . grav . , 19 , 1409 bardeen , j.m . , 1970 , nature , 226 , 64 bardeen , j.m . , press , w.h . , & teukolsky , s.a . , 1972 , apj , 178 , 347 barish , b. , & weiss , r. , 1999 , phys . today , 52 , 44 bekenstein , j.d . , 1973 , apj , 183 , 657 bethe , h.a . , brown , g.e . , & lee , c .- h . , 2003 , selected papers : formation and evolution of black holes in the galaxy ( world scientific ) , p262 bisnovatyi - kogan , g.s . , 1970 , zh . , 47 , 813 bisnovatyi - kogan , g.s . , popov , yu . p. , & samochin , a.a . , 1976 , ap & s.s . , 41 , 287 blandford , r. d. , and znajek , r. l. 1977 , mnras , 179 , 433 bonnell , i.a . & pringle , j.e . , 1995 , mnras , 273 , l12 bradaschia , c. , del fabbro , r. , di virgilio , a. , et al . , 1992 , lett . a , 163 , 15 bromberg , o. , levinson , a. , & van putten , m.h.p.m . , 2005 , astro - ph/0507078 brown , g.e . , lee , c .- h . , wijers r.a.m.j . , lee , h.k . , israelian g. & bethe h.a . , 2000 , newa , 5 , 191 burrows , a. , & lattimer , j.m . , 1987 , apj , 318 , l63 chandrasekhar , s. , 1983 , the mathematical theory of black holes , oxford university press , oxford chincarini , g. , et al . , 2005 , apj , submitted costa , e. , et al . , 1997 , nature , 387 , 878 davies , m.b . , king , a. , rosswog , s. , & wynn g. , 2002 , apj , 579 , l63 danzmann , k. , 1995 , in first edoardo amaldi conf . wave experiments , e. coccia , g. pizella , f. ronga ( eds . ) , world scientific , singapore , p100 duez , m.d . , shapiro , s.l . , & yo , h .- j . , 2004 , phys . rev . d. , 69 , 104016 faber , j.a . , grandclment , p. , & rasio , f. , 2004 , phys . d. , 124036 filipenko , a.v . , 1997 , ara&a , 35 , 309 finn , l.s . , krishnan , b. , & sutton , p.j . , 2004 , apj , 607 , 384 fox , d.b . , et al . , 2005 , nature , 437 , 845 frail , d.a . , et al . , 1997 , nature , 389 , 261 frail , d.a . , kulkarni , s.r . , sari , r. , et al . , 2001 , apj , 562 , l55 fryer , c.l . , woosley , s.e . , herant , m. , & davies , m.b . , 1999 , apj , 520 , 650 fryer , c.l . , holz , d.e . , & hughes , s.a . , 2002 , apj , 565 , 430 mcfadyen , a.i . & woosley , s.e . , 1999 , 524 , 262 gal - yam , a. , et al . , 2005 , apj , submitted ; astro - ph/0508629 gehrels , n. , et al . , 2005 , nature , 437 , 851 goldreich , p. , & julian , w.h . , 1969 , apj , 157 , 869 goldreich , p. , goodman , j. , & narayan , r. , 1986 , mnras , 221 , 339 haislip , j.b . , et al . , 2005 , nature , submitted hawking s.w . , 1975 , , 43:199 heise , j. , in t zand , j. , kippen , r.m . , & woods , p.m. , in gamma - ray bursts in the afterglow era , cnr , 2000 , ed . e. costa , f. frontera & j. hjorth , berlin : springer 2001 , p16 hjrth , j. , et al . , 2005 , nature , 437 , 859 hflich , p.j . , wheeler , j.c . , wang , l. , 1999 , apj , 521 , 179 hulse , r.a . , & taylor , j.h . , 1975 , apj , 195 , l51 katz , j.i . , 1994 , 432 , l27 keller , h.b . , 1987 , numerical methods in bifurcation problems ( berlin : springer / tata inst . kerr , r.p . , 1963 , phys . , 11 , 237 klebesadel , r. , strong i. and olson r. , 1973 , apj , 182 , l85 kobayashi , s. , mszros , p. , 2003 , apj , 582 , l89 kobayashi , s. , mszros , p. , 2003 , apj , 582 , l89 kouveliotou , c. , meegan , c.a . , fishman , g.j . , et al . , 1993 , apj , 413 , l101 kudritzki , r. , 2003 , private commun . ( see http://www.ifa.hawaii.edu/pan-starrs ) kulkarni , s.r . , 2005 , astro - ph/0510256 kundt , w. , 1976 , nature , 261 , 673 landau , l.d . , & lifschitz , e.m . , 1984 , classical theory of fields , pergamon press , oxford lofar ( see http://www.lofar.org ) leblanc , j.m . , & wilson , j.r . , 1970 , apj , 161 , 541 lee , c .- h . , brown , g.e . , & wijers , r.a.m.j . , 2002 , apj , 575 , 996 levinson a. , van putten , m.h.p.m . , 1997 , apj , 488 , 69 levinson , a. , & eichler , d. , 2000 , phys . lett . , 85 , 236 levinson , a. , & eichler , d. , 2003 , apj , 594 , l19 levinson , a. , 2004 , apj , 608 , 411 maund , j.r . , smartt , s.j . , kudritzki , r.p . , podsiadiowski , p. , & gilmore , , g.f . , 2004 , nature , 427 , 129 mazets , e.p . , golenetskii , s.v . and ilinskii , v.n . , 1974 , jetp , 19 , l77 meegan , c.a . , fishman , g.j . , wilson , r.b . , et al . , 1992 , nature , 355 , 143 mszros , p. , & rees , m.j . , 1993 , apj , 405 , 278 mineshige , s. , takashi , h. , mami , m. , & matsumoto , r. , 2002 , pasj , 54 , 655 mirabel , i.f . & rodrguez , l.f . 1994 , nature , 371 , 46 misner , c.w . , thorne , k.s . , wheeler , a. , 1974 , gravitation ( san francisco ) nakamura , t. , & fukugita , m. , 1989 , apj , 337 , 466 nomoto , k. , iwamoto , k. , & suzuki , t , 1995 , phys . 256 , 173 oconnel , r.f . , 1972 , d. , 10 , 3035 okamoto , i. , 1992 , mnras , 253 , 192 paciesas , w.s . 1999 , apj suppl , 122 , 465 paczyski , b.p . , 1991 , acta . astron . , 41 , 257 paczyski , b.p . , 1998 , apj , 494 , l45 papadopoulos , p. , & font , j.a . , 2001 , d. , 63 , 044016 papaloizou , j.c.b . , & pringle , j.e . , 1984 , mnras , 208 , 721 papapetrou , a. , 1951 , proc . soc . , 209 , 248 papapetrou , a. , 1951 , proc . 209 , 259 penrose , r. , 1969 , rev . del nuovo cimento , 1 , 252 peters , p.c . , and mathews , j. , 1963 , phys . rev . , 131 , 435 pirani , f.a.e . , 1956 , act . phys . pol . , xv , 389 piran , t. , 2004 , rev . modern phys . , to appear ; astro - ph/0405503 piro , l. , 2005 , nature , 437 , 822 porquet , d. , et al . , 2004 , a&a , 407 , l17 press w.h . and teukolsky s.a . , 1972 , , 178 , 347 rampp , m. , mller , e. , & ruffert , m. , 1998 , a&a , 332 , 969 rees , m.j . , ruffini , r. , & wheeler , j.a . , 1974 , _ black holes , gravitational waves and cosmology : an introduction to current research _ ( gordon & breach , new york ) , section 7 rees , m.j . , & mszros , p. , 1992 , mnras , 258 , 41p rees , m.j . , & mszros , p. , 1994 , apj , 430 , l93 reeves , j.n . , watson , d. , osborne , j.p . , 2002 , nature , 416 , 512 ruffini , r. , & wilson , j.r . , 1975 , d. , 12 , 2959 spallici , a.d.a.m . , aoudia , s. , de freitas pacheco , j.a . , et al . 2004 , gr - qc/0406076 stanek , k.z . , matheson , t. , garnavich , p.m. , et al . , 2003 , apj , 591 , l17 starobinsky , a.a . , 1972 , zh . etf , 64 , 48 [ sov jetp,37:28 ( 1973 ) ] taylor , j.h . , 1994 , rev . phys . , 66 , 711 teukolsky s.a . , 1973 , apj . , 185 , 635 teukolsky , s.a . , & , press , w.h . , 1974 , , 193 , 443 thorne , k.s . , price , r.h . , macdonald , d.a . , 1986 , black holes : the membrane paradigm , yale university press , new haven , ct turatto , m. , 2003 , in supernovae and gamma - ray bursters , ed . weiler ( springer - verlag , heidelberg ) , p21 unruh w.g . , 1974 , phys . d. , 10 , 3194 villasenor , j.s . , et al . , 2005 , nature , 437 , 855 van paradijs , j. , et al . , 1997 , nature , 386 , 686 van putten , m.h.p.m . , & ostriker , e.c . , 2001 , apj , 552 , l31 van putten , m.h.p.m . , 2002 , apj , 575 , l71 van putten , m.h.p.m . , & levinson , a. , 2003 , apj , 584 , 953 van putten , m.h.m.p . , & regimbau , t. , 2003 , apj , 593 , l15 van putten , m.h.p.m . , 2004 , apj , 611 , l81 van putten , m.h.p.m . , levinson , a. , lee , h .- k . , regimbau , t. , & harry , g. , 2004 , phys . d. , 69 , 044007 wheeler , j.c . , yi , i. , hflich , p. , wang , l. , 2000 , apj , 537 , 810 willke , b. , et al . , 2002 , class . 19 , 1377 woosley , s.e . , 1993 , apj , 405 , 273 zeldovich ya . b. , 1971 , zh . eks . teor . , 14 , 270 [ transl . jetp lett . 14,180(1971 ) ] zhuge , x. , centrella , j.m . , mcmillam , s.l.w . d. , 50 , 6247 * table i. * trigonometric parametrization of a kerr black hole . here , @xmath19 denotes the mass of the black hole , @xmath188 denotes the specific angular momentum , @xmath104 the rotational energy and @xmath189 denotes the irreducible mass . 1 * @xmath190 a histogram of redshift - corrected distributions of 27 long bursts with individually determined redshifts from their afterglow emissions . it shows durations @xmath191 of tens of seconds at a mean redshift distance @xmath192 , indicative of their cosmological distances . @xmath193 the optical spectrum of the type ic sn2003dh associated with grb030329 is remarkably similar to that of the type ic sn1998bw of grb980425 one week before maximum . grb030329 displayed a gamma - ray luminosity of about @xmath194 times typical at a distance of @xmath195 ( @xmath196mpc ) , whereas grb980425 was observed at anomalously low gamma - ray luminosity @xmath197 times typical ) in the local universe ( @xmath198mpc ) . at the same time , their supernovae were very luminous with inferred @xmath199ni ejecta of about 0.5@xmath200 . ( reprinted from @xcitethe american astronomical society . ) * fig . 2 * ( @xmath201 ) black holes with small kick velocities remain centered in core - collapse of a uniformly rotating massive star . shown is the accumulated specific angular momentum of the central object ( arbitrary units ) versus dimensionless orbital period @xmath202 . arrows indicate the evolution as a function of time . kerr black holes exist _ inside _ the outer curve ( diamonds ) . a black hole forms in a first - order transition following the formation and collapse of a torus . this produces a short burst in gravitational radiation . when centered , the black hole surges to a high - mass object by direct infall of matter with relatively low specific angular momentum , up to the inner continuous curve ( isco ) . at this point , the black hole either spins up by continuing accretion or spins down radiatively against gravitational radiation emitted by a surrounding non - axisymmetric torus . in this state , the black hole creates a baryon poor jet as input to grb - afterglow emissions . this continues until the angular velocity of the black hole equals that of the torus ( dot - dashed line ) . this scenario fails for black holes with typical kick velocities with inevitable escape from the high - density core , prohibiting the formation of a high mass black hole surrounded by a high - mass torus . the probability of small kick velocities defines the branching ratio of type ib / c supernovae into long grbs . ( @xmath203 ) the black hole mass @xmath19 and rotational energy @xmath104 of formed after surge in case of small kick velocities , expressed relative to the mass @xmath204 of the progenitor he - star . the results are shown in cylindrical geometry ( continuous ) and spherical geometric ( dashed ) . note the broad distribution of high - mass black holes with large rotational energies of @xmath205 ( spherical to cylindrical ) of @xmath206 . ( reprinted from @xcitethe american astronomical society . ) * fig . 3 * a uniformly magnetized torus represented by two counter - oriented current rings around a black hole ( c ) forms out of both core - collapse ( a1-b1 ) in massive stars and tidal break - up ( a2-b2 ) in black hole - neutron star coalescence , followed by a single reconnection event ( b2-c ) . ( reprinted from @xcitethe american astronomical society . ) * fig . 4 * the stability diagram showing the neutral stability curves for non - axisymmetric buckling modes in a torus of an inviscid incompressible fluid of arbitrary minor - to - major radius @xmath93 . curves of critical rotation index @xmath207 are labeled with azimuthal quantum number @xmath208 , where instability sets in above and stability sets in below . ( reprinted from @xcitethe american astrophysical society . ) * fig . 5 * ( @xmath201 ) grb - supernovae from rotating black holes predict a contemporaneous long - duration burst in gravitational radiation within the range of sensitivity of upcoming gravitational - wave detectors ligo and virgo within a distance of about 100mpc . the corresponding event rate is about one per year . blue bar " denotes the distribution of dimensionless strain amplitudes for a distribution of sources corresponding to a range of black hole masses and efficiency factors , assuming matched filtering . in practice , the sensitivity will depend less on using time frequency trajectory methods with correlations in the spectral domain . @xmath193 the cosmological distribution of grb - supernovae is locked to the star - formation rate @xmath209 . this enables the calculation of the expected contribution to the stochastic background in gravitational radiation , here shown in terms of the spectral energy - density @xmath210 , the strain amplitude @xmath211 and the spectral closure density @xmath187 . the results are calculated for uniform mass - distributions of the black hole , @xmath212 ( top curves ) and @xmath213 ( lower curves ) with @xmath214 ( solid curves ) and @xmath215 ( dashed curves ) . the extremal value of @xmath187 is in the neighborhood of maximal sensitivity of ligo and virgo . ( reprinted from @xcite2004 american physical society . )

we investigate , analytically and numerically , the effects of disorder on the density of states and on the localization properties of the relativistic two dimensional fermions in the lowest landau level . employing a supersymmetric technique , we calculate the exact density of states for the cauchy ( lorentzian ) distribution for various types of disorders . we use a numerical technique to establish the localization - delocalization ( ld ) transition in the lowest landau level . for some types of disorder the ld transition is shown to belong to a different universality class , as compared to the corresponding nonrelativistic problem . the results are relevant to the integer quantum hall plateau transitions observed in graphene .

0706.3737

recent experiments @xcite have unravelled a fascinating set of phenomena in atomically thin layer of hexagonally arranged carbon atoms known as graphene . @xcite the quasiparticles of graphene are @xmath0-dimensional massless weyl fermions . @xcite in the context of condensed matter physics their properties are strikingly different from non - relativistic fermions . and phenomena that are hard to realize for the relativistic case , such as the klein paradox or the _ zitterbewegung _ are accessible in graphene . @xcite it is perhaps not an exaggeration to remark that many subtleties and a rich set of phenomenology are waiting to be discovered . a highlight has been the observation of an unconventional quantum hall effect @xcite and the corresponding theoretical development.@xcite in graphene the filling fractions are @xmath1 for magnetic field @xmath2 , where @xmath3 is an integer.@xcite the factor of @xmath4 comes from the two fold spin degeneracy and the two fold nodal degeneracy of the landau levels . the zeeman splitting is negligible compared to the cyclotron frequency and the disorder broadening of the landau levels . the factor of half is due to a zero mode in the landau level spectrum of dirac fermions.@xcite for stronger magnetic fields , @xmath5 , plateaus appear at @xmath6 , where @xmath7 is an integer.@xcite the plateaus at @xmath8 can be explained by the lifting of both the spin and the nodal degeneracies in the lowest landau level ( lll ) , but those at @xmath9 reflect only the removal of spin degeneracy in higher landau levels . the removal of nodal degeneracy requires electron - electron interaction . mechanisms suggested include @xmath10 ferromagnetism@xcite , sublattice symmetry breaking due to short range interactions@xcite and the generation of a mass gap by magnetic catalysis . @xcite @xmath10 quantum hall ferromagnetism predicts plateaus at all odd integer filling fractions . however , apart from @xmath11 , the plateaus at @xmath12 have not yet been observed . the special quantization rules in graphene are explained by the relativistic landau levels , modified perhaps by interactions , but for the existence of hall plateaus the laughlin argument is necessary.@xcite according to this argument the extended states at the center of a landau band are separated by the localized states elsewhere . if the fermi energy falls in the mobility gap , the plateaus are explained by a gauge invariance argument that is remarkably robust . the underlying phenomenon , therefore , is a localization - delocalization ( ld ) transition at the band center.@xcite the conventional integer quantum hall ( iqh ) plateau transition has been widely studied , and it is known that the localization length exponent @xmath13.@xcite can we prove that the same argument applies to graphene , and , if so , does the ld transition belong to the same universality class ? in the absence of a magnetic field , dirac fermions in the presence of disorder have been widely studied in systems as varied as gapless semiconductors , @xcite gapless superconductors , @xcite and iqh plateau transitions . @xcite as compared to nonrelativistic fermions , the localization problem of dirac fermions is richer because of a number of discrete symmetries . more specifically , if the disorder is particle - hole symmetric , for example a random gauge field , the ld transition takes place at zero energy and is reflected in the single particle density of states ( dos ) , in contrast to the conventional metal - insulator transition where the dos is smooth through the ld transition . surprisingly , there is a line of fixed points with continuously varying exponents depending on the disorder coupling constant . @xcite some of the unusual behavior of disordered dirac fermions may be expected to realize in graphene . one such effect that has received considerable attention is the weak ( anti)-localization phenomenon.@xcite however , relativistic landau levels in the presence of disorder have not yet received much attention . @xcite here we provide a reasonably complete study of the possible effects . there is another important reason why ld transitions in the relativistic landau level should be carefully analyzed . in the conventional iqh effect , the spin - degenerate plateau transition corresponds to @xmath14 when it is assumed that the ld transition takes place at a single energy at the band center.@xcite this has led to intense theoretical investigation of the ld transition in the spin - degenerate landau band . @xcite when spin - orbit scattering is included , the ld transition is found to occur at two distinct energies , away from the band center . scaling analysis about these distinct energies provide , once again , that @xmath15 , as in the spin - polarized system . the scaling about a single energy at the band center leads to the effective exponent @xmath16 . one should anticipate a similar discrepancy between the spin and the nodal polarized iqh effect and the fourfold degenerate iqh effect in graphene . for simplicity we shall concentrate on the spin polarized lowest landau level ( lll ) of graphene and analyze the ld transition both in the presence and in the absence of nodal degeneracy . an interesting example of a controlled analytic calculation in the disordered landau level problem is the dos in the lll . this was first computed exactly by wegner @xcite by examining the euler trails of the impurity diagrams for the white noise disorder and was subsequently extended by brezin _ _ @xcite by using a supersymmetric ( susy ) technique . here we also obtain some exact results for the dos in the disordered relativistic lll using susy techniques . the most general model of disorder consists of a random potential , a random mass , a random gauge field and a random internode scattering ; however , the random gauge field leaves the lll unperturbed . after projection to the spin - polarized lll , we study the following hamiltonian : @xmath17 where @xmath18 , @xmath19 represent potential and mass disorders respectively and @xmath20 and @xmath21 describe internode scattering effects . a mass @xmath22 of the fermions have been included to study the effect of the removal of the nodal degeneracy . for simplicity we have omitted the constant zeeman energy . the @xmath23 matrix @xmath24 is the identity matrix and @xmath25 , @xmath26 and @xmath27 are the three pauli matrices . because of a large number of cases involved , it is useful to summarize the results for the ld transition . let @xmath28 , @xmath29 , @xmath30 and @xmath31 denote the widths of the gaussian random distributions corresponding to the random potential , random mass , and random internode scatterings , respectively . the list of possible cases are : 1 . @xmath33 and @xmath34 . 2 . @xmath35 , @xmath36 . 3 . @xmath37 , @xmath38 . 4 . @xmath39 , @xmath40 . @xmath33 , @xmath35 and @xmath41 . 6 . @xmath33 , @xmath37 and @xmath42 . 7 . @xmath33 , @xmath39 and @xmath43 . 8 . @xmath35 , @xmath37 , @xmath44 . 9 . @xmath35 , @xmath39 , @xmath45 . @xmath37 , @xmath39 and @xmath46 . 11 . @xmath33 , @xmath35 , @xmath47 and @xmath48 . @xmath33 , @xmath35 , @xmath49 and @xmath50 . 13 . @xmath33 , @xmath37 , @xmath49 and @xmath51 . 14 . @xmath35 , @xmath37 , @xmath49 and @xmath52 . 15 . @xmath33 , @xmath35 , @xmath47 and @xmath49 . in the cases ( 1 ) and ( 2 ) , when disorder does not mix the two nodes , the ld transitions belong to the conventional iqh universality class with @xmath53 . it is interesting to note that mass disorder produces ld transition in the lll , whereas for zero magnetic field random mass is known to be an irrelevant perturbation for the @xmath0-dimensional dirac fermions .@xcite the hamiltonians for ( 2 ) , ( 3 ) and ( 4 ) involve only a single pauli matrix at a time , related to each other by unitary transformations . thus , ( 2 ) , ( 3 ) and ( 4 ) are equivalent to each other and have @xmath53 . because unitary transformations leave the identity matrix invariant , the same argument implies that ( 5 ) , ( 6 ) and ( 7 ) are equivalent to each other and once again @xmath53 . the cases ( 8) , ( 9 ) and ( 10 ) involve a pair of pauli matrices and are equivalent to each other . in ( 8) the hamiltonian has a discrete symmetry @xmath54 , often called a particle - hole symmetry . the cases ( 9 ) and ( 10 ) have the same discrete symmetry with respect to @xmath26 and @xmath27 . the case ( 10 ) has been analyzed by hikami _ et al._@xcite for a spin degenerate nonrelativistic lll . when @xmath55 , dos diverges at the band center and has two symmetrically located peaks away from it . the ld transition takes place at these three distinct energies . away from the band center the ld transition has the exponent @xmath56 and the transition at the band center corresponds to a different exponent . if @xmath57 , the divergence of the dos at the band center disappears , but the two symmetrically placed peaks away from the band center still exist . we find that the ld transition at these two energies hve continuously varying exponents depending on the ratio @xmath58 . the cases ( 11 ) , ( 12 ) and ( 13 ) are equivalent . the hamiltonians in these cases are respectively the hamiltonians for the cases ( 8) , ( 9 ) and ( 10 ) , augmented by the identity matrix corresponding to the potential disorder . potential disorder breaks the discrete symmetry mentioned above , and there is no divergence of the dos at the band center . the dos is still peaked at two symmetrically placed energies away from the band center . the ld transitions occur at energies away from the band center . if @xmath28 is much smaller than the two remaining coupling constants , @xmath59 follows trends similar to ( 8) , ( 9 ) , and ( 10 ) . if @xmath28 is comparable or larger , we find @xmath60 . in ( 14 ) all three pauli matrices are present . the discrete symmetry of ( 8) , ( 9 ) and ( 10 ) are absent , and the ld transitions take place at two symmetrically placed energies away from the band center . when all the coupling constants are equal , the exponent @xmath61 . depending on the relative strengths of the coupling constants the exponents vary continuously . if any particular coupling constant is significantly larger than the rest , @xmath60 . by adding @xmath28 we obtain ( 15 ) . if @xmath28 is smaller than the rest , the situation is similar to ( 14 ) . if @xmath28 is larger than the rest , @xmath60 . when @xmath63 , the ld transitions occur at two symmetrically placed energies about the band center , and these energies are greater than or equal to @xmath22 . in the absence of internode scattering , the transitions occur at @xmath64 and the exponent @xmath60 . if the strength of the intranode scattering is larger than @xmath22 , the bands at @xmath65 overlap and effectively correspond to the nodally degenerate case . if @xmath46 and only one of the internode couplings is present , the dos diverges at @xmath64 with an exponent of 0.5 and is identically zero for @xmath66 . the ld transitions occur at @xmath64 and have a continuously varying exponent . when the disorder is strong compared to @xmath22 , @xmath60 , and , in the opposite limit , @xmath59 approaches unity . if we include small intranode scattering the situation is similar . if the intranode scattering strength is greater than the internode scattering , @xmath60 . when @xmath67 and both internode couplings are present , the dos diverges at @xmath64 with an exponent @xmath68 . however , the ld transitions occur at energies larger than @xmath69 . we have analyzed a case where @xmath55 . the exponent varies continuously . if the internode scattering strength is larger than @xmath22 , @xmath70 , and in the opposite limit @xmath59 approaches unity . this behavior is stable against intranode scattering if its strength is smaller than both @xmath22 and the internode scattering . if intranode scattering strength is larger than the internode scattering , @xmath60 . our paper is organized as follows : in sec . ii we describe the dirac fermion model . in sec . iii we describe various possible disorders and their forms when projected to the lll . in sec iv we calculate the averaged density of states using supersymmetry . in the sections v , vi , and vii we describe the numerical studies of the ld transition projected to the lowest landau level . section viii is a brief concluding section . in the appendix [ appendixa ] we provide some mathematical details of the density of states calculation . in appendix [ appendixb ] we describe the recursive green function technique used for numerical calculations and finally in appendix [ appendixc ] we outline the procedure of data collapse involved in the finite size scaling of the localization length . the low energy quasiparticles in graphene are well described by the lorentz invariant form as the sum over two inequivalent nodes ( the fermi velocity @xmath71 ) @xmath72 where @xmath73 and the summation over spin @xmath74 is understood . the four component dirac spinor @xmath75 , where the component @xmath76 is constructed by superposing bloch functions close to one of the two inequivalent nodes @xmath77 of the brillouin zone , corresponding to one of the two sublattices @xmath78 of the hexagonal graphene lattice . the notation @xmath79 stands for the covariant derivative , @xmath80 being the vector potential . the @xmath81-matrices are defined by @xmath82 ; the pauli matrix @xmath83 operates on the two components corresponding to the sublattice indices , and the pauli matrix @xmath84 operates on the components corresponding to the nodal indices . to be explicit : @xmath85 to include a zeeman term , we add @xmath86 where @xmath87 is the zeeman energy and @xmath88 is a pauli matrix operating on the spin indices . the zeeman term breaks the @xmath89 symmetry of the spin space down to @xmath90 . the energy eigenvalues of the hamiltonian operator @xmath91 are well known : @xmath92 where @xmath93 refer to the particle and the hole branches . in the presence of disorder landau levels get broadened into a band , and the amount of broadening depends on the strength of the disorder . when the disorder is very strong , the half - width of the broadened band can be larger than @xmath94 , and experimentally this corresponds to the spin degeneracy of the landau bands . in the spin degenerate situation , the observed filling factors is given by @xmath95.@xcite the lll wave function in the absence of the disorder in the symmetric gauge @xmath96 can be written as @xmath97 where @xmath98 . the functions @xmath99 and @xmath100 are holomorphic functions of the complex coordinates @xmath101 ; @xmath102 , and @xmath103 is the magnetic length . hence , in the zero mode , the first and the second node have nonzero amplitudes coming only from the sublattices a and b respectively . two distinct onsite energies on the two sublattices correspond to a charge density modulation at the lattice scale . as a result , the particle and the hole branches acquire an energy gap . when linearized about the inequivalent nodes , this energy gap appears as a parity preserving mass of the dirac fermions . to be explicit , the linearized hamiltonian will have two new terms : the chemical potential term @xmath104\bar{\psi}\gamma^{0}\psi$ ] and the mass term @xmath105\bar{\psi}\psi$ ] , where @xmath106 and @xmath107 are the site energies at the sublattices @xmath108 and @xmath109 . although the non - interacting quasiparticles are massless in the absence of site modulation , they can acquire a parity conserving mass due to interaction effects . this spontaneous symmetry breaking is facilitated by the presence of the magnetic field , a phenomenon known as `` magnetic catalysis '' of chiral symmetry breaking.@xcite the effect has been argued to be the reason behind the quantum hall plateaus at @xmath110 observed in strong magnetic fields.@xcite though it is beyond the scope of the present paper to consider electronic interactions , we will pay some attention to the noninteracting problem with a finite mass . our philosophy is to analyze the consequences of having a mass ( possible in an interacting theory ) on the ld transition . so , we shall include the term @xmath111 in the effective hamiltonian to examine the effect of mass . in the presence of such a mass term , the nodal degeneracy of @xmath112 is removed and it splits into four levels @xmath113 and @xmath114 . each of these levels has the degeneracy @xmath115 . if the applied chemical potential is smaller than @xmath116 , there will be a plateau at @xmath117 . if @xmath118 , @xmath119 plateaus will appear depending on the sign of @xmath120.@xcite next possible values of quantized plateaus are @xmath121 . the introduction of the mass term does not , however , lift the nodal degeneracy of the higher landau levels , and the energy levels @xmath122 has the degeneracy @xmath123 . therefore , when a mass is included , quantized plateaus appear at @xmath124 , where @xmath7 is an integer . there are many sources of disorder in graphene : vacancies , interstitials , substrate disorder and lattice distortions due to dislocations . in principle there could also be random spin - orbit coupling . however , due to the small atomic mass of carbon , spin - orbit coupling is very weak compared to other energy scales . for simplicity , we shall primarily be interested in the spin polarized limit and ignore the random spin - orbit coupling . point defects and substrate disorder can be described by introducing random site energies in the tight binding model . in the presence of substrate disorder there can also be a random modulation of the charge densities between the two sublattices . these effects can be described by a random chemical potential @xmath125 and a random mass @xmath126 in the continuum limit . because true long range crystalline order is not possible in two dimensions at any finite temperature , topological defects , dislocations and disclinations will be present . effects of these topological defects will result in random hopping amplitudes @xmath127 and hence intranode as well as internode scattering . however , these scattering processes will take place between states on different sublattices . the follwing two bilinears , @xmath128 and @xmath129 , describe the internode scattering terms arising from random hopping . the two mutually anticommuting matrices , @xmath130 also anticommute with @xmath131 ; @xmath132 is the identity matrix . in the continuum limit , the most general impurity hamiltonian is a @xmath133 matrix : @xmath134 where @xmath135 are @xmath136 matrices . here @xmath137 and @xmath138 represent intranode scattering at nodes 1 and 2 respectively ; @xmath139 represent internode scattering . after projecting to the lll , the disorder matrix reduces to a @xmath136 matrix and can be represented by the pauli matrices @xmath84 . the most general disorder matrix projected to lll then takes the form @xmath140 where we have denoted the @xmath132 matrix by @xmath24 . using a four component bosonic spinor @xmath141 and a four component grassmann spinor @xmath142 the average retarded green function for a noninteracting problem can be written as @xmath143\mathcal{d}[\phi]\mathcal{d}[\chi^{*}]\mathcal{d}[\chi]\nonumber \\ \mathcal{d}[v_{j}]p[v_{j}]\phi^{*}(r)\phi(r^ { ' } ) e^{s^{r } } , \end{aligned}\ ] ] where @xmath144 $ ] is the probability distribution of @xmath145 and @xmath146.\end{aligned}\ ] ] the average density of states is given by @xmath147 after performing the disorder averages we can write @xmath148\mathcal{d}[\phi]\mathcal{d}[\chi^{*}]\mathcal{d}[\chi]\nonumber \\ \phi^{*}(r)\phi(r^ { ' } ) e^{\mathcal{a}^{r } } , \end{aligned}\ ] ] where the action @xmath149 involves interactions among the fields generated by the disorder averaging procedure . after projection to the lll , the action @xmath149 can be expressed in terms of a two - component holomorphic bosonic spinor @xmath150 and a two component holomorphic grassmann spinor @xmath151 in terms of these fields the action is given by @xmath152,\end{aligned}\ ] ] where @xmath153 and @xmath154 \mathcal{d}v_{j } \bigg),\ ] ] is the effective interaction of the fields generated by the averaging over the random variable @xmath145 . for the cauchy distribution , defined by @xmath155=\frac{g_{j}}{\pi}\frac{1}{g_{j}^{2}+v_{j}^{2}(\vec{r})},\ ] ] we have @xmath156 if the disorder distribution is gaussian white noise , defined by @xmath155={\cal n } \exp[-\frac{1}{2g_{j}}\int d^{2}r v_{j}^{2}(\vec{r})],\ ] ] we get @xmath157 the above action is invariant under the translation followed by a gauge transformation . due to this invariance , the spatial dependence of the average retarded green function is same as the spatial dependence of the pure system s green function @xmath158}{2\pi l_{b}^{2}(e+i\delta)}.\ ] ] the disorder averaged green function can be written as @xmath159,\ ] ] where @xmath160 s are coupling constants of various types of disorder and @xmath161 is a gauge invariant proportionality constant which depends on the energy and disorder strengths . this gauge invariant proportionality constant is what we need to calculate to find the average density of states . for the calculation of the average green function s dependence on the energy and disorder coupling constants we introduce two new grassmann variables @xmath162 and @xmath163 and enlarge the euclidean coordinate space into a superspace of coordinates @xmath164 . integrals over the grassmann coordinates are normalized as @xmath165 . the norm of a coordinate vector is defined as @xmath166 . this norm is invariant under the superspace rotations . in addition to the ordinary rotations in the euclidean subspace and the symplectic transformations in the grassmann subspace , the superspace rotations involve transformations which mix @xmath167 and @xmath168 in the following manner : @xmath169 in the above set of transformations @xmath170 are two arbitrary euclidean vectors and @xmath171 is a grassmann number . we also define two holomorphic superfields and their conjugates as @xmath172 in terms of these superfields the pure part of the action can be expressed as @xmath173 which is manifestly invariant under superspace rotations . after the disorder contributions to the action are expressed in terms of these new superfields , we have to demonstrate these to be invariant under superspace rotations . in order to be supersymmetric @xmath174 s have to be local in the supercoordinate space and this is only possible if they do not involve any quartic fermionic interactions . we note that @xmath175\nonumber \\ & & = h_{j}\bigg[e^{-z\bar{z}/2l_{b}^{2}}\phi^{\dagger}\eta_{j}\phi\bigg]\nonumber \\ & & + h^{'}_{j}\bigg[e^{-z\bar{z}/2l_{b}^{2}}\phi^{\dagger}\eta_{j}\phi\bigg]e^{-z\bar{z}/2l_{b}^{2}}\chi^{\dagger}\eta_{j}\chi \nonumber \\ & & + \frac{1}{2}h^{''}_{j}\bigg[e^{-z\bar{z}/2l_{b}^{2}}\phi^{\dagger}\eta_{j}\phi\bigg]e^{-z\bar{z}/l_{b}^{2}}(\chi^{\dagger}\eta_{j}\chi)^{2},\end{aligned}\ ] ] where @xmath176 and @xmath177 correspond to the first and second derivatives of @xmath178 with respect to its argument . the taylor series truncates at the quadratic order as the higher powers of @xmath179 are identically zero according to the anticommutation rules . we also note that @xmath180 for @xmath181 and @xmath182 . if @xmath177 does not vanish we get four - fermion interactions . if there were one bosonic and one grassmann fields instead of spinors , as in the problem solved by of brzin _ et al . _ , @xcite no four fermionic terms would be generated , and the action for an arbitrary disorder distribution would be local in the superspace coordinates . for the case under consideration , such a simplification is not possible in general . however , for cauchy distribution the disorder averaged action is quadratic and can be made manifestly supersymmetric . thus , the calculation of the dos reduces to a calculation of a zero dimensional field theory over two complex bosonic fields . the action is given by @xmath183.\ ] ] using the superfields @xmath184 and @xmath185 the action can be written as , @xmath186,\end{aligned}\ ] ] which is manifestly invariant under rotation and magnetic translation in superspace . because of this symmetry , the dos can be reduced to a simple expression involving integrals over two ordinary complex variables . expressed in terms of two radial and two angular variables , it is @xmath187\bigg\}. \label{eq : dencauch}\end{aligned}\ ] ] for simplicity , consider the cases where we keep only one of the internode scattering , or the random mass term , along with the potential disorder . we get : @xmath188 @xmath189,\ ] ] @xmath190 , @xmath191.\ ] ] the answer for the case @xmath192 is identical to the case @xmath193 . the dos obtained for these three cases are identical , as the hamiltonian involves only one pauli matrix at a time , and these matrices are related by unitary transformations . consider now @xmath194 and @xmath51 . we obtain , defining by @xmath132 the expression within the curly parenthesis in eq . [ eq : dencauch ] , @xmath195 , \label{eq : integral1}\ ] ] where @xmath196 . the details of the evaluation of the multiple integrals are provided in the appendix [ appendixa ] . the expression for the dos obtained from this expression is lengthy and not very illuminating , but it is important to note that because of the presence of the term @xmath197 , we obtain a @xmath198 divergence at the band center when @xmath52 . based on symmetry , similar behavior will be obtained when a combination of two pauli matrices are considered . this should be contrasted with the @xmath199 divergence obtained by hikami _ et al . _ @xcite when the fermion is massive , we will ignore the mass disorder part . the density of states is given by @xmath200\bigg\ } , \label{eq : massdencauch}\end{aligned}\ ] ] where @xmath201 . again if we take only one of the internode scattering terms ( @xmath50 ) for simplicity , the expression within the curly parenthesis in eq . [ eq : massdencauch ] , @xmath132 , becomes @xmath202 the density of states is then given by @xmath203 where @xmath204 and @xmath205 . the above expression takes particularly simple form when @xmath52 . it becomes @xmath206,\nonumber \\ \bar{\rho}(e^{2}<m^{2})&=&0.\end{aligned}\ ] ] if both internode scatterings are present and @xmath194 , the integral is given by @xmath207 \label{eq : int2}\ ] ] the expression for the dos is tedious . however for @xmath52,the feature that the dos is zero for @xmath208 is still valid . in this case for energies close to @xmath65 , @xmath209 . similar to the method described in ref . , we generate the matrix elements of the dirac hamiltonian after projecting to the lowest landau level . in our problem , the element @xmath210 itself is a @xmath23 matrix : @xmath211\psi_{k'}(x , y ) = m\eta_3\delta_{k , k'}+v(k , k'),\ ] ] where @xmath212 is the lowest landau level wave function in the landau gauge . we choose all the @xmath145 s to follow independent gaussian white noise distributions such that @xmath213 . then the elements of the @xmath136 matrix @xmath214 can be computed explicitly for example , @xmath215 , \label{exa_mat_ele}\ ] ] where @xmath216 is a complex random variable defined to be the fourier transforms of @xmath217 along the @xmath218 direction normalized by the width @xmath219 , namely : @xmath220 because each of the disorder fields has zero correlation length , and there are no correlations between them , @xmath221 it is straightforward to compute the statistical properties of the matrix elements . the averages are : @xmath222 as to correlations , the only non - vanishing pairs are : @xmath223}\delta_{k_1-k_2,k_4-k_3}\\ \overline{v(k_1,k_2)_{11}v(k_3,k_4)_{22}}&=\overline{v(k_1,k_2)_{22}v(k_3,k_4)_{11 } } = \frac{g_0 ^ 2-g_3 ^ 2}{\sqrt{2\pi}l_y}\exp{[-\frac{l_b^2}{2}((k_1-k_2)^2+(k_4-k_1)^2)]}\delta_{k_1-k_2,k_4-k_3}\\ \overline{v(k_1,k_2)_{12}v(k_3,k_4)_{12}}&=\overline{v(k_1,k_2)_{21}v(k_3,k_4)_{21 } } = \frac{g_1 ^ 2-g_2 ^ 2}{\sqrt{2\pi}l_y}\exp{[-\frac{l_b^2}{2}((k_1-k_2)^2+(k_4-k_1)^2)]}\delta_{k_1-k_2,k_4-k_3}\\ \overline{v(k_1,k_2)_{12}v(k_3,k_4)_{21}}&=\overline{v(k_1,k_2)_{21}v(k_3,k_4)_{12 } } = \frac{g_1 ^ 2+g_2 ^ 2}{\sqrt{2\pi}l_y}\exp{[-\frac{l_b^2}{2}((k_1-k_2)^2+(k_4-k_1)^2)]}\delta_{k_1-k_2,k_4-k_3}\\ \end{split}\ ] ] for numerical implementation , we discretize and use the integer @xmath224 to label the @xmath225 coordinate . we then generate a set of complex random variables @xmath226 , that are @xmath227-correlated as in ( [ cor_u ] ) . finally , we approximate the integrals by sums . explicitly , the matrix elements are @xmath228e^{-a^2 j^2}\\ v(k , k+k')_{22}&=\frac{e^{-a^2 k'^2}}{\sqrt{m a}}\sum_j \left [ g_0 u_{0}(2k+j , k')-g_3 u_{3}(2k+j , k')\right]e^{-a^2 j^2}\\ v(k , k+k')_{12}&=\frac{e^{-a^2 k'^2}}{\sqrt{m a}}\sum_j \left [ g_1 u_{1}(2k+j , k')-ig_2 u_{2}(2k+j , k')\right]e^{-a^2 j^2}\\ v(k , k+k')_{21}&=\frac{e^{-a^2 k'^2}}{\sqrt{m a}}\sum_j \left [ g_1 u_{1}(2k+j , k')+ig_2 u_{2}(2k+j , k')\right]e^{-a^2 j^2 } \end{split}\ ] ] with @xmath229 and @xmath230 . here @xmath231 is the length of the system in the @xmath218 direction , the unit being @xmath232 , that is , @xmath233 , chosen to be an integer . the integers @xmath234 and @xmath235 label the wave vectors . since the matrix elements decay exponentially , we can neglect them for @xmath236 . a cutoff is also necessary for the recursive green s function technique@xcite that we use . we compute the density of states @xmath237 by directly diagonalizing the hamiltonian . we have checked that @xmath237 is independent of @xmath231 , for sufficiently large @xmath231 ; @xmath238 seems to be sufficient ; the total number of momentum states @xmath239 is chosen to be @xmath240 , which is half the dimension of the hamiltonian matrix to be diagonalized , as there are two fermions for each @xmath234 . typically , an average over @xmath241 disorder realizations is used . the recursive green s function technique , similar to that in ref . , is used to explore the localization properties . the details are described in appendix [ appendixb ] . we first compute the localization lengths for a finite system , @xmath242 , at a set of energies , @xmath243 , in systems with transverse dimensions @xmath244 . since there are two types of fermions , in general there can be two distinct localization lengths ; however , in most cases discussed below , they are identical within our numerical accuracy , and we will not generally distinguish them . assuming finite - size scaling , @xmath245 , where @xmath246 is a universal function , the data is collapsed to obtain the localization length exponent @xmath59 , and the critical energy @xmath247 . strictly , scaling holds only for large enough systems in the vicinity of critical energy . here the energies @xmath248 are chosen close to the critical energy , @xmath247 , and the validity of the scaling law is verified by the success of data collapse . for the details of the procedure involving data collapse , see appendix [ appendixc ] . the numerical calculations about the localization properties were mostly performed for a quasi - one dimensional system with the transverse dimensions @xmath249 . the total number of momentum states is @xmath250 . because of the @xmath23 character of the hamiltonian matrix elements , the numerical calculations are more demanding than those in ref . . the data are typically averaged over 100 disorder configurations to reduce fluctuations . energies @xmath248 were chosen close to the critical energy and measured in units of @xmath251 similar to ref . . our program is also validated by the case @xmath252 , @xmath253 , and @xmath32 . in this case , the two types of fermions are independent . because the lll wave function is identical to the non - relativistic one , the properties should be the same as in ref . . numerical computations show a single peak in the density of states and a localization length exponent of @xmath254 ; both agree well with the previous results . consider first the cases where only one type of disorder has nonzero strength . numerically , we considered ( 1 ) @xmath255 , @xmath256 , ( 2 ) @xmath257 , @xmath258 , and ( 3)@xmath259 , @xmath260 . in all of these cases , the delta function density of states in the pure system are broadened into a simple bell shape function due to disorder . the results of successful data collapse , not shown here , yield critical exponents @xmath261 , @xmath262 and @xmath263 for cases ( 1 ) , ( 2 ) , and ( 3 ) , respectively , that is , they are the same within the error bars . the critical exponents are all equal to that of single type of fermions subject to potential disorder . this can be understood as follows . the original hamiltonian matrix is in the basis @xmath264 , where @xmath265 and @xmath266 label the type of fermions . if we reorder the basis as latexmath:[$\{|k_1,1\rangle , |k_2,1\rangle , \ldots , @xmath23 block matrix with diagonal blocks representing intranode parts , and the off - diagonal blocks representing internode scatterings . explicitly , it is in the form : @xmath268 where @xmath269 , @xmath270 are statistically independent random landau matrices , and @xmath132 is the identity matrix . the case with only @xmath271 nonzero , has the same structure , and hence same statistical properties , as the case when only @xmath272 is nonzero . when only @xmath273 is nonzero , we can , by a unitary transformation given by , @xmath274 where @xmath275 is a normalization constant , bring the hamiltonian back to the block diagonal form , resulting in a structure corresponding to two types of independent fermions in the presence of mass disorder . thus , the critical exponent is the same as the case when only @xmath272 is non - zero . the same argument also applies to the case when only @xmath276 is non - zero . if one of the two disorder fields is @xmath277 , and another is @xmath278 , or @xmath279 , or @xmath280 , an appropriate unitary transformation about an axis by @xmath281 will map one possible case to another . for instance , the transformation in eq . ( [ rot_y ] ) will transform the case with @xmath277 and @xmath279 to @xmath282 and @xmath280 . it is therefore sufficient to consider only the case with just @xmath277 and @xmath280 . however , from eq . ( [ hamiltonian ] ) , the hamiltonian is block diagonal , and the blocks @xmath283 and @xmath284 are statistically equivalent to a new block @xmath285 , with @xmath286 a new random matrix satisfying the same statistical properties as @xmath269 s ; see eq . ( [ correlation ] ) . that is , the hamiltonian for @xmath287 and @xmath288 is statistically the same as that corresponding to a potential disorder @xmath289 . our numerical computations confirm this argument . the data collapse was found to be successful , assuming @xmath290 , and the critical exponents are @xmath291 for @xmath292 , @xmath293 , and @xmath294 for @xmath295 , @xmath296 . next , we choose two disorder fields from @xmath297 , @xmath279 , and @xmath280 . there are three possible combinations . in fact , these three cases are not independent ; we can map one case to another by an appropriate unitary transformation corresponding to a rotation by @xmath281 about a certain axis . therefore , it is sufficient to consider only one of the three cases ; for example , let us choose @xmath280 ( mass disorder ) , and @xmath297 ( internode coupling ) . we set @xmath298 , @xmath299 . the density of states and the localization lengths are plotted in fig . [ g_1g_3detail ] for @xmath300 ; there is symmetry under @xmath301 . the extended states are no longer at @xmath302 but shifted to @xmath303 . at @xmath304 , we study the localization properties using the data in the range of @xmath305 , since data in the range @xmath306 are close to both critical points and are likely to result in inaccurate results . the maximum system size used is @xmath307 . because we do not have _ a priori _ knowledge of @xmath308 , the statistical procedure discussed in appendix [ appendixc ] is employed to determine @xmath247 , hence the critical exponent @xmath59 . the data collapse is shown in fig . [ g_1g_3 ] . the critical exponent for this parameter set is found to be @xmath309 , distinct from the nonrelativistic case of @xmath310 . , @xmath299 . the dashed line shows the ld transition . ] , and @xmath299 . insert : the dependence of @xmath311 on the energy @xmath312 for different system sizes @xmath231 . the shaded area is used for scaling and the critical exponent is found to be @xmath309 ] the present problem can be exactly mapped onto the the spin - orbit scattering involving the two - state landau level problem discussed in ref . ; our results are in full agreement . from fig . [ g_1g_3detail ] , there appears to be a divergence in the dos at the band center , corresponding to a possible ld transition at @xmath302 . as shown above , the cauchy distribution does lead to a @xmath314 divergence , but such a weak divergence is difficult to detect numerically ; note , however , that the numerical calculation involves gaussian disorder . a semiclassical explanation@xcite of the existence of an extended state at the band center for the two - state landau level problem is known . however , such an argument is delicate and fails if a the third kind of disorder is present , which is likely in graphene , where potential disorder can not be avoided . thus , we shall not consider further the possible extended state at the band center . it is interesting to study the behavior as the ratio @xmath315 is varied . the result for the dos is shown in fig . [ dosg_1g_3 ] . note that the energy @xmath312 is in the unit of @xmath316 . so the extent of the band increases , as @xmath273 increases . the divergence of the dos at the band center is a unique feature when @xmath273 and @xmath271 are equal , while in the extreme limits there may be a slight dip at @xmath302 . and the internode coupling @xmath273 are both present . parameters are : top : @xmath317 ; middle : @xmath318 ; bottom : @xmath319 . ] as to @xmath59 , a continuously exponent is suggested in fig . [ g_1g_3nu ] . in the limits @xmath320 or @xmath321 , only one type of disorder dominates , hence the value @xmath310 is plausible . the deviation from this value is the largest when @xmath322 , although the data collapse becomes insensitive to the value of @xmath59 in the same regime , resulting in larger error . nonetheless , the results are suggestive of a continuously varying critical exponent . on @xmath323 . the parameters are @xmath299 , @xmath324 , and @xmath32 . the dashed line corresponds to @xmath325 . ] in the hamiltonian , the mass disorder @xmath280 and the internode scattering disorder @xmath297 are accompanied by the pauli matrices @xmath326 and @xmath327 . if we apply a unitary transformation @xmath328 corresponding to a rotation of @xmath281 about @xmath218 axis , where @xmath329 is a normalization factor , the disorder hamiltonian ( [ hamiltonian ] ) will be transformed such that @xmath330 , @xmath331 . because we are studying statistical properties of the system , and all distribution functions are symmetric about zero , the negative sign in front of the @xmath271 is of no importance . this means that this unitary transformation effectively interchanges @xmath273 and @xmath271 , hence map the regime @xmath332 to the regime @xmath333 . note the symmetry between the two regimes in fig . [ dosg_1g_3 ] and fig . [ g_1g_3nu ] . the important case in this category is when @xmath334 ; other cases can be roughly understood in terms of the cases discussed above . for numerical computation , we take @xmath335 . the dos is shown in fig . [ dosg_1g_2g_3 ] . compared to the case when only @xmath298 , discussed above , the divergence of the dos at @xmath302 is missing , but the two peaks at @xmath336 survive . this is suggestive of nonexistence of extended states at the band center , but a ld transition at @xmath337 , which is confirmed by the scaling curve shown in fig . [ g_1g_2g_3 ] , and a critical exponent of @xmath338 is obtained . the error bar is large due to a substantial degree of disorder , but the exponent is distinctly different from the value @xmath325 , indicating a new universality class . in the case @xmath335 , @xmath339 . ] , and @xmath335 . insert : the dependence of @xmath311 on the energy @xmath312 for different system sizes @xmath231 . the shaded area is used for scaling . the critical exponent @xmath59 is found to be @xmath340 . [ g_1g_2g_3 ] finally , we have also examined the case @xmath341 , @xmath342 . @xmath343 , @xmath344 . we have found that when @xmath272 is small , the potential disorder simply broadens the density of states , and results in a value of @xmath59 as though it did not exist ; on the other hand a large value of @xmath272 drives @xmath59 to a value close to @xmath345 . the constant mass term results in new physics when combined with the internode coupling , and the resulting phenomena are different from the case when the mass disorder and the internode coupling are combined , as in the previous section . consider first the case @xmath62 , and either @xmath272 or @xmath271 is nonzero . from eq . ( [ hamiltonian ] ) , the random hamiltonian matrix are block diagonal , hence the two types of nodal fermions are uncoupled . the energy of the two fermions are shifted by the amount of @xmath65 , and the dos is simply a superposition of two bell shaped functions centered at @xmath65 . as to localization properties , because the two fermions will have different critical energy , namely @xmath346 , at a particular energy @xmath312 these localization lengths will be different from each other . the data collapse for the fermion with @xmath347 once again gives a critical exponent of @xmath310 . consider now finite mass @xmath62 and only one internode coupling , for example @xmath348 . the calculated dos with @xmath300 region is shown in fig . [ dosg_1 m ] for the parameter set @xmath258 , @xmath257 , and @xmath349 . qualitatively the results are similar to the analytical calculations involving the cauchy distribution , even though the numerical computation is for the gaussian distribution of disorder . first , the dos vanishes in the region @xmath66 . second , the analytical calculation shows that @xmath350 , @xmath351 , independent of the value of @xmath273 . the insert of fig . [ dosg_1 m ] yields an exponent of @xmath352 . we have checked that this value does not vary with @xmath273 , within our numerical accuracy . . in the case @xmath258 , @xmath257 , and @xmath349 , the dos vanishes when @xmath66 . insert : the logarithmic plot of the dos around @xmath353 , and @xmath349 . the best fit gives a slope of @xmath352 . ] as to ld transition , we perform data collapse with @xmath347 . the critical exponent turns out to be @xmath354 for the parameters @xmath258 , @xmath355 , and @xmath356 , see fig . [ g_1 m ] , which is significantly different from the usual case corresponding to @xmath15 . this striking result implies that the system belongs to a new universality class . we now vary @xmath273 , keeping @xmath22 fixed to @xmath357 , and the result for the exponent @xmath59 is shown in fig . [ g_1mnu ] . it appears that the exponents continuously vary with the ratio @xmath358 . , @xmath355 , and @xmath356 . insert : dependence of @xmath311 on energy @xmath312 for different system sizes @xmath231 . the shaded area is used for scaling . critical exponent is found to be @xmath354 , with the choice of @xmath347 . ] on @xmath273 ( normalized by the mass @xmath22 ) . the parameters are @xmath258 , and @xmath356 . the dashed line indicates the level of @xmath359 . ] the most relevant case corresponding to graphene is the one with two types of internode scattering of comparable magnitude . therefore , we choose @xmath360 . the dos is shown in fig . [ g_1g_2mdetail ] . note that the energy is now measured in units of @xmath361 , so that the divergence is located at @xmath353 , which is @xmath362 . from the insert in fig . [ g_1g_2mdetail ] , we find a slope of @xmath363 , which is consistent with the analytical result for the cauchy distribution , namely , @xmath364 . also note the gap in the dos . the crossing point in fig . [ g_1g_2mdetail ] indicates @xmath365 instead of @xmath347 . the data collapse is shown in fig . [ g_1g_2 m ] with a critical exponent of @xmath366 . this critical exponent is not close to any of the values found for a finite mass with a single internode coupling , as in fig . [ g_1mnu ] . however , it is reasonably close to the exponent for @xmath367 , and @xmath322 ( see fig . [ g_1g_3nu ] ) , which is also equivalent to the case @xmath368 , and @xmath369 , as discussed above . note that @xmath349 is much smaller than the bandwidth @xmath370 , and this critical exponent indicates that the presence of small finite mass will have little effect on the critical exponent as long as two internode couplings are finite . at the other limit , when @xmath22 is large enough compared to the band width , the critical exponent @xmath371 . thus , it is reasonable to believe that the exponent also varies continuously , as a function of @xmath372 , provided that @xmath373 , and the behavior is similar to in fig . [ g_1 m ] , except that @xmath374 in the limit @xmath375 . , @xmath376 . insert in the top : the logarithmic plot of the dos around @xmath353 . the best fit gives a slope of @xmath363 . ] , @xmath376 . insert : the dependence of @xmath311 on the energy @xmath312 for different system sizes @xmath231 . the shaded area is used for scaling . ] because potential disorder is always present in experiments on graphene , we would like to discuss the case when @xmath377 , @xmath63 , and @xmath378 . first , when @xmath22 is the smallest parameter , it can be neglected , and hence the massless case discussed above is recovered . our numerical computations gave a critical exponent of @xmath70 when @xmath379 , and @xmath380 in the limit @xmath381 , because now the potential disorder is more important than the rest . when @xmath272 is the smallest , it will have little effect . therefore , as discussed in the previous subsection , there will be a continuously varying exponent from @xmath382 for @xmath383 to @xmath384 for @xmath385 . finally , if @xmath373 are smaller than the rest , the internode scattering is no longer important , and hence the two nodal fermions will be decoupled . the exponent will therefore be always @xmath380 regardless of the relationship between @xmath22 and @xmath272 . we have analyzed the effects of disorder on the ld transition in the lll of graphene . because both types of internode scattering , present in the lll , arise from the random hopping , they will have roughly the same strength . because the sources of the mass disorder and the internode scattering are different , their strengths will be generically different . in some special cases of disorder combinations we have found new universality classes of ld transition in contrast to the conventional iqh . our results for the ld transitions in the lll have direct experimental relevance for the plateau transitions in graphene . consider first the cases where both the spin and the nodal degeneracies are completely removed . a number of authors have shown that the inclusion of a finite mass and zeeman energy can explain the appearance of plateaus at @xmath386 . because experiments resolve the spin and nodal splitting , intranode scattering which always broadens the landau levels is weak compared to @xmath94 and @xmath22 . if the internode scattering strength is larger than the intranode scattering strength , we expect that in the lowest landau level @xmath387 , @xmath388 and @xmath389 plateau transitions can have different universality classes , in contrast to the conventional iqh effect . in the opposite limit , when the intranode scattering is considerably stronger , the plateau transitions will fall into the conventional iqh universality class with @xmath390 . for the spin and nodally degenerate plateaus , the potential scattering is strong compared to the zeeman energy and the mass gap . theoretically , from our analysis of the massless cases we can infer the plateau transitions to be of the conventional iqh type . however , in experiments , if scaling with respect to the band center is invoked , an effective exponent for these plateau transitions will be observed . plateaus at @xmath391 involve higher landau levels . in the higher landau levels both the random potential at the lattice scale and the random hopping will have nonzero intranode as well as internode scattering contributions . this will complicate the analysis of ld transitions in these levels . because inclusion of a finite mass does not lift the nodal degeneracies of higher landau levels , the effect of internode scattering can be strong . in view of the degeneracy factor , there is the possibility of observing an effective exponent for these higher plateau transitions . this work was was supported by the nsf under the grant no . for the cauchy distribution , in the absence of mass disorder , the calculation of the dos in eq . ( [ eq : massdencauch ] ) involves the integral @xmath392,\ ] ] where @xmath393 and @xmath394 . in the massless case @xmath395 . after performing the integrals over one of the angles , the double integral over the angles is reduced to @xmath396,\ ] ] where @xmath397 and @xmath398 . now , expanding the exponential in a power series , integrals over @xmath225 and @xmath218 can be easily performed . for the angular integral we use the relation @xmath399,\ ] ] where @xmath400 is gauss s hypergeometric function and obtain @xmath401 the summation over @xmath224 can be performed by the following trick : ( 1 ) use the integral representation of @xmath400 , ( 2 ) perform a power series summation which is simple in these cases , and ( 3 ) complete the integration over the auxiliary variable introduced for the integral representation . for simplicity we will specialize to the cases : ( i ) @xmath402 and ( ii ) @xmath194 . \(i ) @xmath403 : in this case we have @xmath404 we now use the following two relations @xmath405 @xmath406 to obtain eq . ( [ eq : int1 ] ) . \(ii ) @xmath194 : in this case we have @xmath407 using the integral representation @xmath408 and eq . ( [ eq : sum])we get @xmath409 after performing the integral over @xmath410 we get eq . ( [ eq : int2 ] ) . after setting @xmath32 one recovers eq . ( [ eq : integral1 ] ) because there are two types of fermions in our problem , the recursive green s function technique is a little more complicated than that introduced by huckenstein@xcite . all the matrix elements , such as those in eqn . ( 6.9 ) in ref . , become @xmath23 matrices ; hence , all the operations , such as the multiplication and inversion are matrix operations . denote the @xmath23 matrix @xmath411 simply by @xmath412 . suppose @xmath413 , @xmath414 , the green s function containing @xmath415 momentum states is available , then as we add another momentum state , the recursion relations for @xmath416 are : @xmath417g^{(k+1)}(k+1,k+1)\mbox{\quad:\quad}i\le k\\ g^{(k+1)}(i , j)=g^{(k)}(i , j)&+g^{(k+1)}(i , k+1)g^{(k+1)}(k+1,k+1)^{-1}g^{(k+1)}(k+1,j)\mbox{\quad:\quad}i , j\le k \end{split}\ ] ] these matrix inversions here can be accurately computed because the sizes are small . corresponding to the two types of fermions , we are interested in @xmath418 , @xmath419 , because the localization length of a system of width @xmath231 for the @xmath420 fermion , @xmath421 , is related to this quantity by , @xmath422 where @xmath423 . furthermore , if we define a set of @xmath23 matrices @xmath424 , @xmath425 , such that their elements @xmath426 , from the recursion relations ( [ recursion ] ) , we obtain : @xmath427_{n , n}\\ g^{(k+1)}(i)_{m , n}&=\frac{1}{q^{(k+1)}_m}\left[g^{(k)}(i)_{m , n}+\left(\sum_j g^{(k)}(j)v(j , k+1)g^{(k+1)}(k+1,i)\right)_{m , n}\right]\mbox{\quad:\quad}i\le k\\ g^{(k+1)}(i)_{m , n}&=\frac{1}{q^{(k+1)}_m}\left(\sum_j g^{(k)}(j)v(j , k+1)g^{(k+1)}(k+1,i)\right)_{m , n}\mbox{\quad:\quad}i = k+1\\ \end{split}\ ] ] in this appendix , we describe how we can extract the the exponent @xmath59 , and the critical energy @xmath247 , if necessary , based on the computed localization lengths in finite systems @xmath428 assuming a single parameter scaling assumption . suppose we have obtained @xmath429 in systems with @xmath430 for @xmath431 , each with a standard deviation @xmath432 . our goal is to find out the proper values of @xmath59 and @xmath247 such that all the @xmath433 data points collapse on to a single curve : @xmath434 since @xmath246 is unknown , it is difficult to characterize the quality of the data collapse . to overcome this difficulty , we proceed as follows : suppose that we are given a pair of values @xmath435 , we can attempt to represent the unknown function @xmath436 by a polynomial of degree @xmath437 by simply performing a general fit to eq . ( [ fitlambda ] ) given below , based on a total of @xmath433 data points @xmath438,\log[\lambda_m(e_i)/m_j],\sigma_{m_j , e_i})\}$ ] : @xmath439^k , \label{fitlambda}\ ] ] where @xmath440 are the coefficients to be fitted . in the computer implementation , the order of polynomials was chosen to be @xmath441 , since no significant changes were noted by increasing @xmath437 to 9 . the quality of this fit @xcite is represented by the variable @xmath442 defined as : @xmath443 ^ 2\ ] ] if the preset values @xmath435 are not correct , the data points will be scattered , resulting in a large value of @xmath442 , which in turn indicates a poor data collapse . however , when @xmath435 attain the correct localization length exponent and the correct critical energy , respectively , @xmath442 will be minimized . following this procedure , by minimizing @xmath442 with the standard gradient descent method , we are able to determine correctly both the critical energy @xmath247 and the localization length exponent @xmath59 . because the scaling law is only valid in the close vicinity of the critical energy , once @xmath247 is obtained from above procedure , we have to check , for the purpose of self consistency , that all energies used in the data collapse are indeed close to @xmath247 . as for the statistical error of @xmath59 , the usual procedure is to assume that the minimized @xmath444 follows a @xmath445 distribution , and hence the error bar can be drawn corresponding to a certain confidence probability . however , this is not the case in this problem , since @xmath444 does not follow the @xmath445 distribution due to the nonlinear form of the estimated parameters @xmath59 and @xmath247 in ( [ fitlambda]).@xcite to draw an error bar for @xmath59 statistically correctly , recall that we have the original data @xmath446 . we generate a large number of data sets synthetically @xmath447 , for @xmath448 , such that @xmath449 , @xmath450 , and @xmath451 a variable randomly distributed in the gaussian form with a mean of @xmath452 and a standard deviation of @xmath453 . next , we perform exactly the same procedure to get @xmath454 for each synthetic data set @xmath447 , as was performed in actual data set @xmath446 for estimating the @xmath59 and @xmath247 . finally , the error bar for @xmath59 is drawn as the estimated standard deviation : @xmath455^{1/2}\ ] ] where @xmath456 is the number of synthetic data sets , and @xmath457 is the average of @xmath454 . @xmath458 in computer implementation .

we report measurements of the carbon monoxide ground state rotational transition (co ) with the zpectrometer ultra - wideband spectrometer on the 100 m diameter green bank telescope . the sample comprises 11 galaxies with redshifts between and 3.5 from a total sample of 24 targets identified by _ herschel_-atlas photometric colors from the spire instrument . nine of the co measurements are new redshift determinations , substantially adding to the number of detections of galaxies with rest - frame peak submillimeter emission near 100 m . the co detections confirm the existence of massive gas reservoirs within these luminous dusty star - forming galaxies ( dsfgs ) . the co redshift distribution of the 350m - selected galaxies is strikingly similar to the optical redshifts of 850m - selected submillimeter galaxies ( smgs ) in . spectroscopic redshifts break a temperature - redshift degeneracy ; optically thin dust models fit to the far - infrared photometry indicate characteristic dust temperatures near 34k for most of the galaxies we detect in co. detections of two warmer galaxies and statistically significant nondetections hint at warmer or molecule - poor dsfgs with redshifts difficult determine from from _ herschel_-spire photometric colors alone . many of the galaxies identified by h - atlas photometry are expected to be amplified by foreground gravitational lenses . analysis of co linewidths and luminosities provides a method for finding approximate gravitational lens magnifications from spectroscopic data alone , yielding . corrected for magnification , most galaxy luminosities are consistent with an ultra - luminous infrared galaxy ( ulirg ) classification , but three are candidate hyper - lirgs with luminosities greater than .

1204.4706

observations of the far - ir / submillimeter background with _ cobe _ demonstrated that a substantial fraction of the universe s star formation took place behind a veil of dust @xcite . because these were integrated measurements , however , they could not identify which populations of dust - obscured galaxies contained this vigorous star formation . a breakthrough in resolving the background came in the late 1990s , with imaging by the submillimeter common - user bolometer array ( scuba ) on the james clerk maxwell telescope ( jcmt ) . its initial surveys at 850@xmath4 m @xcite detected a new population of bright ( @xmath8mjy ) galaxies . named after the wavelengths where they are most visible , these submillimeter galaxies ( smgs ) are systems with apparently vast ( @xmath9 ) bolometric luminosities but with such high obscurations that their optical counterparts are faint or absent ( see * ? ? ? * and references therein ) . smgs brighter than scuba s confusion limits could not account for all of the 850@xmath4 m background , but clearly made a substantial contribution to it . over the last fifteen years , much of the effort to understand the origins of the far - ir / submillimeter background has focused on bright smgs selected from 8501200@xmath4 m surveys . smgs are sometimes treated as representatives of a more general category of high - redshift galaxies , dusty star - forming galaxies ( dsfgs ) whose luminosities are dominated by obscured star formation . a major initial hurdle was verifying that bright smgs actually do lie at high redshifts . while two early scuba detections had optical redshifts @xcite , quickly confirmed with co spectroscopy @xcite , progress in measuring redshifts of other smgs foundered due to their very high obscurations . only after @xcite took advantage of radio continuum mapping to determine precise positions for blind optical spectroscopy did it become possible to obtain co detections in large numbers @xcite , and to derive an smg redshift distribution peaking around @xmath102.5 @xcite . heroic efforts to explore the high-@xmath11 tail that radio pre - selection misses when radio counterparts fall below typical survey flux limits have identified a handful of sources at @xmath12 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but detailed analysis limits the possible significance of this tail in the @xmath13 mm population @xcite . we now know that bright smgs have large stellar @xcite , molecular gas @xcite , and dynamical @xcite masses , that many of them are mergers @xcite , and that their large luminosities are powered principally by star formation @xcite . explaining the observed properties of bright smgs , their evolutionary states , and their relationships to populations of galaxies selected at other wavelengths is a current major challenge for galaxy evolution models ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in parallel with the growth in our understanding of bright smgs , it is also becoming clear that current samples give an incomplete picture of the full variety of dsfgs . first , 850@xmath4 m sources fainter than scuba s nominal confusion limit of about 2 mjy , although accessible via gravitational lensing ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , must at some point start to resemble optically selected galaxies more than heavily obscured smgs ( e.g. , * ? ? ? * ; * ? ? ? second , even among bright dsfgs , 850@xmath4m - bright smgs have distinct selection biases . dsfg samples selected at longer wavelengths appear to have cooler dust temperatures and higher median redshifts @xcite . dsfg samples selected at shorter wavelengths , conversely , include populations with warmer dust that 850@xmath4m - selected samples can miss , @xcite , and that tend to have both lower redshifts and more bolometrically significant active galactic nuclei ( agns ; e.g. , * ? ? ? * ; * ? ? ? third , existing samples of smgs with spectroscopic redshifts often suffer from biases associated with the steps used to determine those redshifts . for example , precise localization of a submillimeter source usually relies on radio continuum mapping , and while the deepest vla imaging yields counterparts for a high fraction of dsfgs @xcite , more typical vla maps tend to deliver counterparts for only 6070% of smgs . subsequent optical spectroscopy based on these positions fails to yield redshifts for a modest fraction of candidates @xcite , and even when apparently successful , attempts to obtain co detections of the gas reservoirs associated with these massive starbursts can sometimes fail to yield conclusive confirmation , raising questions over either the identification or redshift @xcite . mid - ir spectroscopy can avoid some of these difficulties ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) but suffers from its own problems for sources with power - law spectra or confusion from multiple sources within large beams or slits . finally , many of the seminal studies of smgs mapped relatively small areas on the sky . notwithstanding the large line - of - sight interval probed by 850@xmath4mselection , which can partly compensate for a small area , smgs are such rare sources ( mergers caught at special moments , with the most luminous caught at the most special of moments ) that cosmic variance remains a concern for the derived redshift distributions . new instruments capable of producing deep images of large regions of the sky , and of conducting efficient spectral surveys over wide bandwidths , have accelerated the discovery of high - redshift dsfgs with a broader range of physical properties than could be probed by previous efforts . survey areas at @xmath14 mm have increased substantially ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , with extremely wide - area surveys possible both from the ground @xcite , and from space with the _ herschel space observatory _ @xcite . large - area surveys are identifying many very bright dsfgs whose fluxes are rivaled by those of only a few extreme , serendipitously discovered objects that have been confirmed to be gravitationally lensed ( e.g. , * ? ? ? regardless of the balance between intrinsically hyperluminous systems and less extreme but gravitationally lensed galaxies within these samples , we are no longer missing the rarest dsfgs because of limited sky coverage . _ herschel _ is playing a particularly important role because the wavelength coverage of its spire instrument @xcite allows selection of dsfg samples that are relatively free of dust temperature biases @xcite , although they are limited by confusion to only the most extreme luminosity systems at @xmath15 @xcite . specialized instruments designed for wide - band spectral line surveys now enable the determination of blind co redshifts ( see , e.g. , * ? ? ? * ; * ? ? ? * ) for bright dsfgs , without intermediate radio continuum mapping or optical spectroscopy . an example of the combination of these new developments is the recent use of two wide - bandwidth instruments to obtain co redshifts for a complete sample of five bright _ herschel _ sources @xcite , an essential step in confirming their status as galaxy - galaxy lenses @xcite . expanding on this initial work , we here report on @xmath14 cm spectroscopy of the @xmath0c@xmath1o @xmath2 ground - state rotational transition toward two dozen of the brightest dsfgs in catalogs from the _ herschel _ astrophysical terahertz large area survey ( h - atlas ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) program , using the zpectrometer ultra - wideband spectrometer on the national radio astronomy observatory s 100-meter diameter robert c.byrd green bank telescope ( gbt ) . submillimeter continuum flux ratios provide some coarse redshift information for many sources , but the precise redshifts needed to enable additional observations with narrow - band instruments require spectroscopy of atomic or molecular lines . the zpectrometer is one of several ultra - wideband spectrometers built for this purpose , and is the first instrument to make routine measurements of the co @xmath16 rotational transition from high - redshift galaxies ( e.g. , * ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . with the co molecule s @xmath17 level only 5.4k above the ground state , its low but nonzero permanent dipole moment , and its strong c o bond , the @xmath2 transition is the best tracer of molecular gas over a wide range of conditions in molecular clouds . in addition to giving a spectroscopic marker for redshift measurements , velocity - resolved spectroscopy yields dynamical information , which together with gas masses derived from co intensities provides key inputs to understanding dsfgs star formation efficiencies and overall evolutionary states . subsequent sections of this paper describe the observations and the initial results from this sample . section [ sec : obs ] describes target selection and observations , section [ sec : res ] contains observational results , and section [ sec : disc ] provides further analysis and discussion . calculations use a @xmath18cdm cosmology with @xmath19 , @xmath20 , and @xmath21 @xcite . we drew our targets from early _ herschel_-atlas catalogs ( h - atlas collaboration , priv . 2010 ) of continuum detections in the _ herschel _ spire instrument s 250 , 350 , and 500@xmath4 m wavelength photometric channels . for this initial study , we selected bright `` 350@xmath4 m peaker '' galaxies with flux densities @xmath22mjy and observed spectral energy distributions ( seds ) peaking in the spire 350@xmath4 m band , within errors . the catalogs screened out local spiral galaxies and high-@xmath11 blazars with the methods described in @xcite and @xcite . this provided a set of bright targets with peak far - infrared emission near 100@xmath4 m wavelength in the rest frame for galaxies with @xmath23 to 3 . most of the targets were in the equatorial multi - wavelength galaxy and mass assembly ( gama ) survey @xcite fields near right ascension 9 , 12 , and 15 hours , with a few targets from the north galactic pole ( ngp ) field near @xmath24 hours and @xmath25 . table [ tab : obs ] is a list of target positions and integration time information . lrrrrr j083051.0 + 013224 & gama09 & 7.87 & 3 & 1 + j084933.4 + 021443 & & & & 2 + j083929.5 + 023536 & gama09 & 7.08 & 3 & 3 + j084259.9 + 024958 & & & & 4 + j090302.9 - 014127 & gama09 & 5.46 & 2 & @xmath26 + j091305.0 - 005343 & & & & @xmath27 + j091840.8 + 023047 & gama09 & 4.59 & 2 & 5 + j085111.7 + 004933 & & & & 6 + j091948.8 - 005036 & gama09 & 3.67 & 2 & 7 + j092135.6 + 000131 & & & & 8 + j113526.3 - 014605 & gama12 & 5.64 & 2 & 9 + j113243.1 - 005108 & & & & 10 + j113833.3 + 004909 & gama12 & 3.67 & 2 & 11 + j113803.5 - 011735 & & & & 12 + j114637.9 - 001132 & gama12 & 7.35 & 3 & 13 + j115112.3 - 012638 & & & & 14 + j115820.2 - 013753 & gama12 & 5.25 & 2 & 15 + j114752.7 - 005832 & & & & 16 + j132426.9 + 284452 & ngp & 2.89 & 2 & 17 + j133008.3 + 245860 & & & & 18 + j134429.4 + 303036 & ngp & 3.94 & 2 & 19 + j133649.9 + 291801 & & & & 20 + j141351.9 - 000026 & gama15 & 10.23 & 4 & 21 + j142751.0 + 004233 & & & & 22 we observed the targets with the zpectrometer analog lag cross - correlator spectrometer @xcite connected to the gbt s facility ka - band receiver , which was configured as a correlation receiver . receiver improvements in fall 2010 extended its high frequency performance , allowing spectroscopy from 25.6 to 37.7ghz , corresponding to the co molecule s 115.27ghz @xmath2 transition at redshifts of @xmath28 . the zpectrometer s spectral resolution is a sinc function with full - width at half maximum ( fwhm ) of 20mhz , sufficient to provide a few resolution elements across typical galaxy lineshapes . given the instrument s wide overall bandwidth , the spectral resolution varied with frequency across the spectra from 234 to 157kms@xmath29 , and the fwhm beamsize from 27 to 16 arcsec . a combination of the correlation receiver s ability to difference power between two beams on the sky and the zpectrometer s large bandwidth allows different observing techniques from those common in narrowband total - power radio astronomy . our instrument , observing technique , data reduction , and calibration methods are fully described in @xcite . briefly , the correlation receiver implementation electrically differences power between the receiver s two input horns , which are separated by 78arcsec on the sky . we switched the source between the two horns by moving the gbt s secondary mirror 78arcsec in a 10s cycle , then differencing spectra from the two positions to obtain a source spectrum . to eliminate the residual tens of mjy of spectral baseline structure from optical offsets , we observed pairs of targets close in position on the sky , cycling between sources every 4min , again differencing this pair . in this difference spectrum of the two positions an emission line from the first source would appear in the positive sense , while one from the second source would appear in the negative sense . differencing left little optical offset and baseline structure , at the cost of eliminating information about individual source continua . observations were conducted in sessions of 3 to 8 hours duration ( see table [ tab : obs ] ) on dates from 2010 november through 2011 april for a total of 64.5hr of on - sky observing time . counting observing overheads , the actual elapsed observing time was about 130hr . we observed in a variety of weather conditions , mostly with reasonable to good ka - band atmospheric transmission and low wind . we established absolute flux scales across the spectra by dividing the astronomical source difference spectra by the spectrum of a bright ( few jy ) quasar suitable as a pointing reference near each target pair . in some cases we could use one of the cm - wave flux standards 3c48 , 3c286 , or 3c147 directly ( 0.80 , 1.83 , 1.47jy at 32ghz , respectively ; @xcite 2011 ) , but for most of our sources we cross - calibrated spectra of the nearby pointing source with one or more of the standards . we cross - checked 3c48 s flux density from @xcite against a mars flux density model by b. butlerbbutler / work / mars / model/ ] and found agreement within 2% . comparison with the recommended 2012 january jansky - vla flux density for 3c286 is 1.96jy ( 7% higher ) than the value we use , and we find a ka - band spectral index of @xmath30 instead of the jansky - vla s @xmath31 . cross - calibrations of the flux density for the quasar we used for the gama09 sources against all three of the standards agreed within 10% . we determine calibrator spectral indices across the ka - band , which can be important in transferring the 32ghz standard fluxes to other frequencies in the band , by comparison with mars blackbody @xmath32 . we pointed and focused every hour on the bright quasars we had selected as secondary flux calibrators near our sources . pointing offsets were always within a third of a beam near band center . we took spectra of the pointing source at the beginning and end of each pointing cycle to measure changes in optical gain and atmospheric transmission . dividing the difference spectrum of an astronomical source pair by the average spectrum of the pointing source not only corrected for bandpass gain but also for the effects of pointing errors and changing atmospheric transmission with timescales comparable to an hour , or longer . session to session repeatability for the quasar flux densities was generally within 10% , and we take 20% as representative of the overall amplitude calibration uncertainty to include gain effects from pointing drift . version d of the zpectrometer s standard gbtidl scripts provided quick - look assessment during observations and produced files for combining data from different sessions . we used version 5.4 of our zred package , which is written in the r language @xcite , for further analysis . the receiver produced systematic spectral structure with fluctuating amplitude across many of the spectra . spectral lines from galaxies were much narrower than features in the systematic structures , which enabled us to remove the instrumental artifacts ( * ? ? ? * contains further information ) . we removed a systematic ripple with fourier filtering and a complex but fixed - pattern structure by fitting and subtracting median - filtered session - average templates of the structure . we removed no further baseline structure , and with these corrections we could keep nearly all of the data . in this exploratory phase of a full survey , we set integration times to identify bright co sources rather than to get high signal - to - noise ratios on individual sources . since system noise changes across the zpectrometer s 12.1ghz band and nonideal noise is clearly present , attempting to estimate noise for signal - to - noise ratios by computing the channel fluctuations across the spectrum s entire frequency range would overestimate noise in some parts of the spectrum and underestimate it in others . instead , we harness time - series information from sub - integrations to make individual channel noise estimates as part of a detection confidence algorithm . @xcite has a complete discussion of this algorithm , which analyzes whether the amplitude in each frequency channel is statistically higher than the average of its neighbors , based on a generalized student-@xmath33 test of many bootstrapped realizations of each spectrum . the algorithm s main advantage over traditional methods is that it treats noise in individual frequency channels . its main weakness stems from its assumption that the underlying spectral baselines are relatively smooth on small scales so that local averages are representative . failure of this assumption keeps the computed confidence level from being absolute , as discontinuities in spectral baselines and small fluctuations near bright features can all register as potential detections . our experience with the algorithm on zpectrometer data is very good , based on comparisons of tentative detections within subsets of data and from detections of different lines from the same galaxies with other instruments . given deviations from the assumptions caused by nonideal noise of a few hundred @xmath4jy , the algorithm remains a powerful guide rather than an exact indicator for identifying weak lines . experience gained from viewing many spectra toward different sources under different weather conditions , but all with the same frequency coverage , provides valuable impressions of typical baseline structures and noise behavior , further aiding in screening against spurious detections . all detections here come from at least two independent observing sessions , and most have at least tentative detections in individual sessions . by generating 1000 bootstrapped spectra as part of the detection algorithm , the data are thoroughly mixed by time and session . sweeping through a wide range of binnings ensures that a detection is not based on a single favorable choice . we have been conservative in our decisions , while recognizing that a small number of false positives are more beneficial than false negatives in observations designed to spur followup work . in practice , independent observations of other lines from the same targets have justified our detection selections . panels in figure [ fig : spectra ] show the spectra and confidence plots for all of the source pairs we observed . the upper panel for each source pair is the spectrum across the zpectrometer band . vertical dashed lines mark line frequencies of detected galaxies . the lower panel shows the detection probabilities @xmath34 from our detection algorithm , versus frequency , given as confidence@xmath35 ( numerically , the scale is equivalent to the number of nines in confidence : 0.9 , 0.99 , 0.999 , etc . for @xmath36 ) . each dot in the plot is an individual channel s confidence measure ( within the algorithm s assumptions ) for a given combination of binning width and starting point . columns of dots at specific frequencies show where a potential detection is relatively immune to exact binning parameters , indicating that a real line is present rather than a favorable binning for a chance fluctuation . our brief comments on individual spectra are : _ j083051.0 + 013224 _ and _ j084933.4 + 021443 _ ( fig . [ fig : spectra ] , 1 - 2 ) : only the second target in the pair ( emission appears in the negative sense ) is detected . riechers ( priv . comm . 2011 ) has detected a single strong line with the carma observatory toward the first in the pair . if the line were co @xmath37 , j083051.0 + 013224 would lie at a redshift with the @xmath38 line in a low - noise region of the zpectrometer band . this is a statistically significant nondetection that we discuss in section [ sec : completeness ] . _ j083929.5 + 023536 _ and _ j084259.9 + 024958 _ ( fig . [ fig : spectra ] , 3 - 4 ) : neither target is clearly detected in this spectrum . _ j090302.9@xmath39014127 _ and _ j091305.0@xmath39005343 _ ( fig . [ fig : spectra ] , a - c ) : clear detections of both targets . the high confidence at slightly lower frequency than the strong positive line is likely an artifact , as even a modest dip can be far from the local amplitude mean , a signature the detection algorithm interprets as a line . _ j091840.8 + 023047 _ and _ j085111.7 + 004933 _ ( fig . [ fig : spectra ] , 5 - 6 ) : clear detection of the first target in the pair . the continuum offset from zero shows that the first target has a higher continuum flux than the second target . _ j091948.8@xmath39005036 _ and _ j092135.6 + 000131 _ ( fig . [ fig : spectra ] , 7 - 8 ) : neither target is clearly detected in this spectrum . residual large - scale structure may obscure what could be tentative detections . even small noise fluctuations at the tops of large - scale positive and negative structures are far from local means , so they register strongly in the confidence plot . _ j113526.3@xmath39014605 _ and _ j113243.1@xmath39005108 _ ( fig . [ fig : spectra ] , 9 - 10 ) : clear detections of both targets . some spurious high confidence peaks are associated with each of the bright lines . _ j113833.3 + 004909 _ and _ j113803.5@xmath39011735 _ ( fig . [ fig : spectra ] , 11 - 12 ) : neither target is clearly detected in this spectrum . _ j114637.9@xmath39001132 _ and _ j115112.3@xmath39012638 _ ( fig . [ fig : spectra ] , 13 - 14 ) : clear detection of the first target . _ j115820.2@xmath39013753 _ and _ j114752.7@xmath39005832 _ ( fig . [ fig : spectra ] , 15 - 16 ) : clear detection of the first target . _ j132426.9 + 284452 _ and _ j133008.3 + 245860 _ ( fig . [ fig : spectra ] , 17 - 18 ) : tentative detection of the first target . baseline structure to slightly higher frequencies makes it difficult to find a local baseline , so the line parameters are uncertain . _ j134429.4 + 303036 _ and _ j133649.9 + 291801 _ ( fig . [ fig : spectra ] , 19 - 20 ) : clear detections of both targets , with one line in the positive sense and two in the negative sense . we attribute the stronger negative line with the spire source . for completeness , we assign the other negative line a tentative formal detection because it is uncharacteristically broad for a spurious signal , but it is close to the positive - sense line where the local mean changes rapidly and may throw off the detection algorithm . _ j141351.9@xmath39000026 _ and _ j142751.0 + 004233 _ ( fig . [ fig : spectra ] , 21 - 22 ) : a strong detection of the first target . the high confidence measures for dips to either side are most likely due to a high local mean in the region . the dips are rather wide to be due to a galaxy , so it is unlikely that they represent detections . the continuum slope indicates a difference in spectral index between the two targets . overall , we detected 11 of the 24 targets in our sample . two of the detections were blind independent confirmations of sources with established redshifts ( id.17b in * ? ? ? * and id.130 in ; * ? ? ? this success rate is similar to that of detections from the plateau de bure millimeter - wave interferometer starting from optical redshift catalogs ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but without the complications associated with finding optical redshifts for submillimeter sources ( see , e.g. , * ? ? ? section [ sec : completeness ] contains a more extensive discussion of detection completeness . in addition to the 11 detections , we also list two tentative detections , denoted by italics in table [ tab : summary ] and ( for j132426.9 + 284452 ) with open circles in the figures . line parameters for tentative detections were too uncertain for robust error estimates . based on the submillimeter photometric selection and line strength , we initially derived redshifts assuming that the lines were the redshifted co @xmath2 transition rather than co @xmath40 from a @xmath41 galaxy or a line from a species other than co. this assumption has proved correct for all 11 sources , which have other observed lines , most starting with redshifts from the zpectrometer observations ( d.a . riechers , priv . 2011 ; p.p . van der werf , priv . comm . 2011 ; table [ tab : summary ] summarizes observed and derived source parameters . for detected lines , the parameters are from single - component fits to gaussian lineshapes , with errors given by the statistical uncertainties in the fit at the 68% ( `` @xmath42 '' ) confidence level . gaussian fits to the convolutions of gaussian line shapes and the correlator s sinc instrumental profile shows that linewidth corrections are unimportant for linewidths above 200kms@xmath29fwhm ( the correction is 12% when the gaussian line and sinc fwhms are equal , falling below 1% when the gaussian s fwhm is 1.5 times the sinc s fwhm or wider ) . all lines are broader than this , so we make no corrections . the table also contains estimates of the total infrared ( 81000@xmath4 m ) fluxes and dust temperatures obtained from fits to the spire photometry ( h - atlas collaboration , priv . 2010 ) , as discussed later . in addition to the detections from this program , table [ tab : summary ] also includes co @xmath38 data from a previous zpectrometer detection of h - atlas j090311.6 + 003906 ( * ? ? ? * id.81 ) . [ tab : summary ] we have firm detections of 11 of the 24 targets in our program , a number large enough to draw some sample conclusions . here we use the molecular and continuum flux information to explore the detection completeness and to evaluate possible reasons for co nondetections . figure [ fig : scovss350 ] explores the relationship between 350@xmath4mflux density , @xmath43 , and the co @xmath2 integrated flux , @xmath44 . sources without co detections have zero amplitude in this plot , and the open circle denotes a tentative co detection . line nondetections generally fall to lower 350@xmath4 m flux densities , but several galaxies are bright in continuum but not detected in co. we can take the third - brightest @xmath43 source in our sample ( j083051.0 + 013224 , target number 1 in figures and tables ) as an example of a nondetection that implies a galaxy with a redshift outside the zpectrometer band or an abnormally low co to continuum flux ratio . the former seems more likely in this case : as noted above , carma has a clear detection of a mid-@xmath45 co line from this source , but deep integrations by the zpectrometer and other instruments have not found other lines corresponding to a redshift within the zpectrometer s @xmath463.5 band . versus 350@xmath4 m flux density @xmath43 for co - detected and co - undetected sources from our sample . co nondetections have been assigned zero flux . horizontal dashed lines show completeness levels for line detections derived from our 400kms@xmath29 linewidth simulations , with corresponding @xmath43 derived from the power - law fit shown by the smooth curve . all galaxies with co nondetections have 350@xmath4 m flux densities above the 70% completeness limit for detections if a simple co - continuum flux scaling law holds ( see text ) . this implies that chance alone can not explain the fraction of galaxies in the sample that we do not detect , but that some galaxies have redshifts outside the zpectrometer band or contain relatively little molecular gas . points are labeled by target numbers given in the tables . [ fig : scovss350],width=307 ] most likely , some fraction of the nondetected galaxies have redshifts that are not well predicted by continuum properties , while others have co @xmath38 lines that are fainter than our detection threshold . before we simply ascribed co nondetections to galaxies with observed luminosities below some threshold , we first examined the line detection completeness and the relationship between line and continuum fluxes . estimating line detection completeness in broadband spectra is complicated by system temperature and nonideal noise that vary with frequency ; this frustrates any attempt to define a quantity such as the baseline rms in narrowband spectra that could specify a simple detection limit . we estimated completeness levels for nondetections by simulation , adding sets of synthetic lines as frequency combs across subscans for all sources , running the modified spectra through the data reduction pipeline , and inspecting the spectra to see which synthetic lines we could clearly identify . with a comb of seven 400kms@xmath29 wide lines ( a width close to the median of the astronomical source linewidths ) across each spectrum , we recovered 90% of the synthetic lines with co integrated intensity @xmath44 = 800 mjykms@xmath29 , 70% of the lines with @xmath44 = 600 mjykms@xmath29 , and 35% of those with @xmath44 = 400 mjykms@xmath29 . taking @xmath44 = 600 mjykms@xmath29as the typical lower limit for our detections , the simulation gives 70% completeness for 400kms@xmath29 lines , 90% completeness for 200kms@xmath29 lines , and 40% completeness for 800kms@xmath29 lines . this shows that the statistical algorithm is most sensitive to peak intensity for lines near detection thresholds . the average completeness for these three widths is 67% , so 70% is representative for a set of lines with various widths and 600 mjykms@xmath29 . a somewhat monotonic relationship between the continuum and line fluxes in dsfgs must exist : energy balance requires that galaxies with little far - ir continuum luminosity will have weak molecular emission . the exact relationship is unknown , but a power - law fit established a representative correspondence between @xmath44 and 350@xmath4 m flux densities @xmath43 for sources with co detections . this fit yielded equivalent limits of @xmath43 = 150mjy , 90mjy , and 50mjy for 90 , 70 , and 35% completeness for the median ( 400kms@xmath29 ) linewidth , as shown in figure [ fig : scovss350 ] . given the @xmath44@xmath43 distribution , the exact form of the continuum line relationship is unimportant over the relatively small range at low flux densities , and a linear fit gave essentially the same values . excluding the extreme high - flux point from the fit flattens the @xmath44@xmath43 relationship , pushing the completeness limits to lower continuum flux densities . pushing the completeness limits to higher flux densities would require a steeper relationship than could be supported by these data with a simple model . whatever the exact form of the correspondence between @xmath44 and @xmath43 , all of our targets have @xmath43 falling above the 70% completeness level ( probability of detection ) if their co @xmath16 emission falls within the zpectrometer frequency range . we estimated the number of targets we might expect to have missed due to faint co flux alone by considering the targets below @xmath43 = 150mjy , the approximate 90% completeness level . there are 10 sources in our target list below this limit , of which we detect only two . if chance alone dominates , the detection rate will be given by the binomial distribution . taking a lower limit of a 70% detection probability , the distribution finds 7 detections as most likely , with @xmath47 detections accounting for 99% of the total probability . the probability of detecting just two sources is 0.1% . increasing either the detection probability or the flux limit corresponding to a given detection probability reduces the probability of detecting just two of ten sources . if chance alone ruled , we should therefore have detected some 2 to 5 more weak sources . considering the additional four nondetections with 350@xmath4 m flux densities significantly above the nominal 90% completeness limit , this analysis points to a strong disparity between actual and expected detections . we conclude that chance is not the only reason for nondetections , but that systematic effects are also important : some of the sources could not be detected because their redshifts are outside of the zpectrometer s band , some because they have lower @xmath44/@xmath43 ratios than the detected sources , some because their lines are weak and broad ( although there is no sign of these even though the eye is good at picking out correlated channels ) , or some combination . zpectrometer redshifts for the h - atlas sources provide an independent sample for comparison with previous spectroscopic redshift surveys . comparing surveys , we find that the redshift distributions of the `` 350@xmath4 m peaker '' galaxies with zpectrometer detections and those of the radio - preselected smgs @xcite that fall within the zpectrometer s redshift range are strikingly similar , although the source selection and lines used in the redshift measurements were quite different : our galaxies are 350@xmath4m - bright targets from a wide - area survey that highlights lensed sources , while the @xcite galaxies must be bright at 850@xmath4 m and 20 cm radio continuum . figure [ fig : zdistribs]a gives the binned distribution for the galaxies with zpectrometer co detections . the median of the zpectrometer co redshifts is @xmath48 , and the mean is @xmath49 ( 68% confidence levels by bootstrap analysis ) , both well below the band center at @xmath50 . figure [ fig : zdistribs]a shows that the peak of the observed density function is near @xmath51 , agreeing well with the peaks found for optical spectroscopic redshifts of scuba - selected sources @xcite , and for photometric redshifts of both _ herschel_-selected sources bright at 350@xmath4 m @xcite and 870@xmath4m - selected sources identified by laboca @xcite . m - selected galaxies . [ fig : zdistribs],width=307 ] figure [ fig : zdistribs]b gives the cumulative distribution functions of the zpectrometer and the @xcite sample redshifts , allowing a clean comparison free of binning effects . it shows that the distributions are indistinguishable ; a kolmogorov - smirnov test gives a probability of 0.995 that the two underlying distributions are the same . such good agreement must in part reflect random chance in a statistically small sample , but nevertheless it is clear that the two distributions are very similar . there is otherwise no _ a priori _ reason that the distributions should be so similar : for instance , the @xcite sample could be concentrated to lower redshifts because of the radio pre - selection , while the zpectrometer detections could highlight a population of galaxies with strong molecular but very little rest - frame uv line emission . another similarity between the zpectrometer sample and galaxies detected in millimeter - wave co followup from optical redshift catalogs is the distribution of co linewidths . linewidths trace the dynamics , and thus to some extent masses , of the galaxies , and are not strongly affected by lensing . full width at half maximum ( fwhm ) linewidths in the zpectrometer sample range from 210 to 1180kms@xmath29 , with a median width of @xmath52kms@xmath29 ( 68% confidence levels by bootstrap analysis ) . in general this is somewhat narrower than the widths of mid-@xmath45 lines in the millimeter - wave studies of @xcite and @xcite , but a permutation test shows that the difference in median widths between the studies is significant at only the 70% level , and could easily be from small number statistics . combining the samples , a characteristic width of about 500kms@xmath29 is representative for dsfgs . in absolute terms , the widest @xmath38 lines are very broad , however , including two with fwhm linewidths greater than 1100kms@xmath29 . if tracing virialized matter , these widths correspond to emission from very massive galaxies or interacting massive galaxies . spectroscopic redshift measurements unambiguously break the @xmath53-@xmath11 degeneracy ( e.g. , * ? ? ? * ) that renders dust temperature estimates uncertain for galaxies without firm redshift measurements . we fit a simple optically thin single - temperature dust model with dust emissivity @xmath54 to the 250 , 350 , and 500@xmath4 m spire flux densities . to estimate the observed total infrared ( rest - frame 81000@xmath4 m ) luminosity , we joined the far - infrared fit smoothly in slope to a power - law spectrum with form @xmath55 at short wavelengths @xcite . assuming that the same gravitational magnification @xmath4 applies to emission in all far - ir wavebands , lensing should not affect temperature estimates , but it will scale the intrinsic luminosity @xmath56 so the observed infrared luminosity is @xmath57 . table [ tab : summary ] contains the model results and monte - carlo error estimates for individual galaxies . excluding the two @xmath58 galaxies , the dust temperature is @xmath59k averaged over all of the detected sources , with a slight dependence on the observed 350@xmath4mflux density ( fig . [ fig : dustvsalla]b ) or , equivalently , @xmath60 . little temperature scatter is to be expected among the detected galaxies because the source selection criteria favored similar spire - band seds . temperatures near 35k are similar to those derived from other surveys of smgs . the two highest-@xmath11 galaxies in our sample have temperatures of about 40k , a temperature higher than scatter alone can explain . varying the dust emissivity parameter @xmath61 from 1.2 to 1.7 did not change the dust temperatures by more than 3k . dust emission models that allow for dust emission optical depths yield temperatures about 15k higher than those from the optically thin model ; we quote the optically thin results to facilitate comparisons with previous work , most of which uses the same formalism . m flux density @xmath43 , showing that the luminosity scales reasonably well with flux density , even in the presence of a range of redshifts and dust temperatures . panel ( b ) shows that the derived characteristic dust temperatures are nearly constant with @xmath43 , possibly at least in part from selection effects . the highest two points in both panels are from the two galaxies in our sample with @xmath62 . points are labeled by target numbers given in the tables . [ fig : dustvsalla],width=307 ] figure [ fig : dustvsalla]a shows that the 350@xmath4 m flux density is an excellent proxy for the observed infrared luminosity @xmath57 at @xmath103 . the two points falling above the general trend are the two galaxies with the highest redshifts , @xmath58 , in our sample . excluding these two galaxies , we derive @xmath63 . apart from the two galaxies at @xmath58 there is no dependence of @xmath57 on redshift within errors . one of the goals of this project was to provide a data set suitable for evaluating the precision of photometric redshifts against spectroscopic measurements . here we make a first - cut comparison of techniques based on simple spire 3-band colors and on fitting to template galaxy seds . with galaxies chosen as `` 350@xmath4 m peakers , '' the simplest color selection is a flux ratio that estimates the sed s peak wavelength . since the 250@xmath4 m and 500@xmath4 m bands straddle the emission peak their ratio provides this estimate , reflecting some combination of temperature and redshift . as section [ sec : continuum ] shows , our sample of galaxies has a small range of dust temperatures , so we can test whether there is a simple relationship between the observed sed peak wavelength and redshift . figure [ fig : ratiodetz ] shows that there is a relationship between the peak wavelength , as given by the 250@xmath4m/500@xmath4 m flux density ratio , and spectroscopic redshift for galaxies with co detections . for these galaxies the 250@xmath4m/500@xmath4mflux density ratio predicts redshifts within @xmath64 ( standard deviation of the redshift error ) . other combinations of continuum flux densities and ratios are less successful at predicting redshifts than the peak wavelength . for example , both the co - detected and nondetected target galaxies fall along a common locus in a 500@xmath4m/350@xmath4m350@xmath4m/250@xmath4 m color - color diagram , but the galaxies are well mixed in redshift along that locus . a more general method of estimating redshifts is to take seds of galaxies with known redshift and temperature as templates , then find which redshifted template best matches the observed sed . template fits are not very tightly constrained by spire data alone because the 250 , 350 , and 500@xmath4 m bands lie near the peak of the observed - frame sed , however . flux ratios near the peak have little dynamic range and consequently can not provide strong constraints : the maximum @xmath65 and @xmath66 ratios are about 1.6 . figure [ fig : templcmp ] shows the redshift error @xmath67 versus redshift @xmath11 for the detected galaxies comparing results from the simple @xmath68 ratio and those from independent template fitting codes by co - authors clements , gonzlez - nuevo , and wardlow . each estimates the redshift by minimizing the chi - squared deviation of the data points compared with one or more observed seds of galaxies with known redshifts . minimizing chi - squared by shifting the templates in wavelength gives the redshift , while the widths of the chi - squared values versus shift provide redshift error estimates . comparing the template fitting with the linear fit between @xmath11 and @xmath68 shown in fig . [ fig : ratiodetz ] , shows that the linear fit gives the lowest dispersion in @xmath67 for most of the detected galaxies . redshifts from an sed corresponding to the cosmic eyelash galaxy smmj2135@xmath390102 ( * ? ? ? * , with sed template from ; * ? ? ? * and ; * ? ? ? * ) are nearly as accurate as those from the linear fit . comparison with models that choose from libraries of extragalactic seds shows these have generally poorer agreement than constrained models for most of the detected galaxies , with @xmath691 and internal error estimates smaller than the actual redshift deviation between observation and model . their errors are lower than those of the @xmath68 ratio or eyelash fits for the two galaxies at @xmath70 , however . this result emphasizes the difficulty of fitting curves with only a few samples near the peak ; photometric redshifts incorporating substantially longer - wavelength data are substantially more accurate ( see , e.g. , photometric redshifts including mambo 1200@xmath4 m data in * ? ? ? * ) . versus 250@xmath4m/500@xmath4 m continuum flux density ratio ( solid points ; open point for a tentative detection ) . dashed horizontal lines mark the zpectrometer s band edges . co nondetections ( triangles ) are plotted outside the band at an arbitrary @xmath46 . for galaxies with co detections , the figure shows that spire colors give @xmath71 for this carefully filtered homogeneous sample . points are labeled by target numbers given in the tables . [ fig : ratiodetz],width=307 ] redshifts , either the linear fit ( triangles ; fig . [ fig : ratiodetz ] ) or template fitting against the eyelash sed ( crosses ) provides redshift estimates within @xmath72 for @xmath73 . dashed lines highlight grouping by method only . template fits that automatically select from a library of seds ( squares ) generally predict redshift to @xmath69 - 1 ; the grey background shows the typical internal uncertainty for the template fits . models that choose between multiple seds predict redshifts outside the zpectrometer band for several of the nondetected sources , even though the linear and eyelash templates place them within the band , and have lower errors for the @xmath74 sources . as discussed in the text , this indicates that selection effects are important in the agreement between eyelash sed and co detections . [ fig : templcmp],width=307 ] we emphasize that the correspondence between co @xmath2 detections and spire 250@xmath4m/500@xmath4 m color or eyelash template redshifts yields no redshift information about the galaxies that we do _ not _ detect in the co @xmath2 line . galaxies with and without detections span nearly the same 250@xmath4m/500@xmath4 m flux density ratios ( fig . [ fig : ratiodetz ] ) , and the success of the eyelash template may simply arise from its ability to identify the @xmath75k galaxies we find in co. as we discussed in section [ sec : completeness ] , some fraction of galaxies with continuum properties similar to galaxies with co detections very likely fall outside the zpectrometer s band . lens magnifications ( @xmath4 ) are needed to convert the observed fluxes to intrinsic luminosities . determining magnification is generally done by constraining a model of the source - lens pair with an image of the gravitationally - distorted source galaxy . detailed lens models are not yet available for most of our sources . we do have velocity - resolved spectra , however , and can take an alternative approach of comparing observed and intrinsic luminosities established by an empirical luminosity - linewidth relationship similar in spirit to the tully - fisher relation @xcite . without correction for galaxy inclination or dispersion in intrinsic galaxy properties , such a relationship can not be exact , but it can provide approximate estimates of lensing magnifications . since the galaxies with zpectrometer co detections seem to be quite typical of the general smg population , as we discussed above , we assume that only lens magnifications modify the observed typical luminosity distributions of galaxies detected in co @xmath2 . following the method of @xcite , as outlined below , we find an empirical intrinsic integrated line luminosity - linewidth relationship @xmath76 from galaxies with published co @xmath2 intensities , widths , and magnifications . this relationship forms the basis to solve for the unknown magnification @xmath4 that scales the true line luminosity @xmath77 to an apparent luminosity @xmath78 , or @xmath79 figure [ fig : co_tf ] shows @xmath80 for 15 @xmath814 smgs from the literature versus linewidth , with corrections for lens magnification when needed ( co data from * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; magnifications as needed from * ? ? ? points for these galaxies fall close to a power law fit with relatively small scatter . inserting the fit parameters into equation ( [ eq : mu1 ] ) , we obtain @xmath82 with the units for line luminosity @xmath77 as defined in @xcite . apparent line luminosities for the h - atlas galaxies with zpectrometer co measurements , uncorrected for lens magnification , are also plotted in figure [ fig : co_tf ] . as expected for a flux - limited sample , these galaxies have a narrower range of luminosities than the comparison sample from the literature . galaxies with narrower linewidths , which will generally have lower luminosities , require more magnification to reach the observational detection threshold . versus co @xmath2 fwhm linewidths for 15 sources with redshifts , fluxes , and magnifications as needed from the literature ( co data from * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; magnifications as needed from * ? ? ? the solid line shows a power - law fit to these data , representing an estimate of the intrinsic line luminosity versus linewidth for typical galaxies . round points show the observed line luminosities versus widths for the zpectrometer sample ( values from table [ tab : summary ] ) , without correction for lens magnifications . the literature sample has a steep power - law relation with modest scatter , while the zpectrometer sample shows a much flatter trend . intrinsically weaker sources require higher magnification to rise above the observational detection threshold , which is approximately constant in flux . we scaled the observed line luminosity for each of the zpectrometer sources by the prediction from the power - law estimate of its likely magnification @xmath4 , as given in table [ tab : galprops ] . this method yielded a range of @xmath8320 and a median of @xmath84 for the zpectrometer sample . [ fig : co_tf],width=307 ] table [ tab : galprops ] lists the lens magnifications derived from equation ( [ eq : mu ] ) . the empirical magnification distribution is approximately uniform over its range of 3 to 20 , with both median and mean equal to 10 . simple uncertainty estimates for the scaling term and power law index are not available because the two are not independent , but a bootstrap analysis provides @xmath8568% confidence bounds on @xmath4 for each galaxy in table [ tab : galprops ] . fractional uncertainties range from about 30% at a minimum near 500kms@xmath29 , climbing to about 50% at 300 and 1200kms@xmath29 , and about 100% at 200kms@xmath29 . the divergence toward low @xmath86 is also a reminder that the ( unknown ) inclination corrections become large for galaxies with observed low linewidths . it is possible that some of the sources with broad co lines and low inferred amplifications from equation ( [ eq : mu ] ) are intrinsically hyper - luminous systems which are not significantly magnified at all ; the magnification uncertainty encompasses that possibility . lrrrrrrrr j084933.4 + 021443 & @xmath87 & @xmath88 & @xmath89 & @xmath90 & @xmath91 & @xmath92 & @xmath93 + j090302.9@xmath39014127 & @xmath94 & @xmath95 & @xmath96 & @xmath97 & @xmath98 & @xmath99 & @xmath26 + j090311.6 + 003906 & @xmath100 & @xmath101 & @xmath102 & @xmath103 & @xmath104 & @xmath96 & @xmath105 + j091305.0@xmath39005343 & @xmath106 & @xmath107 & @xmath108 & @xmath109 & @xmath110 & @xmath111 & @xmath27 + j091840.8 + 023047 & @xmath112 & @xmath113 & @xmath114 & @xmath115 & @xmath116 & @xmath117 & @xmath118 + j113243.1@xmath39005108 & @xmath119 & @xmath120 & @xmath121 & @xmath122 & @xmath123 & @xmath124 & @xmath125 + j113526.3@xmath39014605 & @xmath126 & @xmath127 & @xmath128 & @xmath129 & @xmath130 & @xmath131 & @xmath132 + j114637.9@xmath39001132 & @xmath133 & @xmath134 & @xmath135 & @xmath136 & @xmath137 & @xmath138 & @xmath139 + j115820.2@xmath39013753 & @xmath140 & @xmath141 & @xmath142 & @xmath143 & @xmath144 & @xmath145 & @xmath146 + j133649.9 + 291801 & @xmath147 & @xmath148 & @xmath149 & @xmath150 & @xmath151 & @xmath152 & @xmath153 + j134429.4 + 303036 & @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & @xmath159 & @xmath160 + j141351.9@xmath39000026 & @xmath161 & @xmath162 & @xmath163 & @xmath164 & @xmath165 & @xmath166 & @xmath167 magnifications from the luminosity - linewidth relation ( [ eq : mu ] ) compare well with values from the two galaxies in the h - atlas survey that have both zpectrometer detections and preliminary lens models from image reconstruction with lens and source galaxies at known redshifts . for j090311.6 + 003906 ( id.81 ; co data from * ? ? ? * ) , the co linewidth - luminosity relationship in equation ( [ eq : mu ] ) predicts @xmath168 , while @xcite quote @xmath169 ( range 1831 ) from their modeling . for j091305.0@xmath39005343 ( id.130 ) , equation ( [ eq : mu ] ) predicts @xmath170 while @xcite quote @xmath171 ( range 57 ) . table [ tab : galprops ] provides estimates of the intrinsic luminosities and gas masses corrected by the source - specific magnification factors , and also by the median @xmath84 . most of the galaxies appear to have intrinsic luminosities typical of ulirgs , although three are candidate hyper - lirgs , with luminosities greater than @xmath172 . gas mass estimates use the same co luminosity - to - mass conversion factor as in @xcite and @xcite for ease of comparison , @xmath173 m@xmath174 ( kkms@xmath29pc@xmath175)@xmath29 @xcite . although absolute mass estimates carry double uncertainties from both conversion and magnification factors , they are similar to estimates for smgs selected from 850@xmath4 m observations that use the same conversion factor . demonstrating an ability to rapidly establish spectroscopic redshifts and molecular line parameters for luminous high - redshift galaxies is a major step forward in exploring the properties of dsfgs . in a single observing season we have added a dozen galaxies with co @xmath2 line detections to the pool of sources for study among the bright `` 350@xmath4mpeakers '' from the _ herschel_-atlas survey . we established accurate redshifts for sufficient galaxies to compare with an independent survey of smgs with observed optical redshifts . that survey started with submillimeter source continuum detections in @xmath176 arcsecond beams at 850@xmath4 m , associated those with 20 cm radio continuum sources to obtain positions to @xmath177 arcsecond , and then used optical spectroscopy to determine redshifts ( see e.g. , * ? ? ? subsequent observations with a millimeter - wave interferometer used the optical redshifts as starting points to determine molecular redshifts from the star - forming gas itself ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , conclusively connecting the 850@xmath4 m sources with massive molecular clouds in the target galaxies in about half of the attempts . the zpectrometer co redshifts , in contrast , started with bright 350@xmath4 m sources with submillimeter photometry indicating rough photometric redshifts , but then made a single step from continuum to molecular line observations with this wideband spectrometer . we find nearly identical redshift distributions for the two surveys , indicating that each sample is representative of a general dsfg population . coupled with mm - wave molecular spectroscopy of many of the brighter 850@xmath4 m sources ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , the agreement in distributions between independent samples lays to rest some of the concern that smg redshift distributions from optical spectroscopy may be influenced by misidentifications between the submillimeter sources and counterpart optical galaxies . dust temperatures and linewidths of the galaxies with co 10 detections appear to be similar to smgs . with spectroscopic redshifts , we are able to break the temperature - redshift degeneracy to find a characteristic temperature of 34k ( in an optically thin formalism ) , similar to those typically reported for smgs @xcite . galaxies with redshifts @xmath74 may be somewhat warmer than those at lower redshift . the co linewidths have very similar distributions for these h - atlas co @xmath2 detections and for smgs observed in mid-@xmath45 lines . both of these comparisons add further to an argument that the bulk of samples selected by bright 850@xmath4 m and spire emission draw from the same parent population . most , if not all , of the galaxies we observed appear to be gravitationally lensed . we estimate lens magnifications with an empirical co linewidth - luminosity relationship made possible with velocity - resolved spectroscopy . we verified magnification agreement within factors close to two for two galaxies with published lens modeling . we find magnifications from 3 to 20 , a range matching those for lenses with detailed models , @xmath17830 @xcite . in the absence of other information , a typical value of @xmath84 is likely to be correct within a factor of two for the bright h - atlas galaxies . with even moderately accurate lens magnification estimates , we can make reasonable estimates of intrinsic luminosities and gas masses of the galaxies in our sample ( table [ tab : galprops ] ) . most have luminosities characteristic of ulirgs , but there are three targets that could be hyper - lirgs with luminosities @xmath179 . the most luminous of these could well be multiple galaxies within the beam , as their linewidths are broad . even accounting for lensing magnification , however , the bright h - atlas galaxies seem to be drawn from the high - luminosity , high - mass end of the dsfg distribution . the overall agreement in properties between the zpectrometer and 8501200@xmath4m - selected smg samples indicates that the two populations are drawn from the same distribution . in principle , the additional selection introduced by lens magnification could have emphasized different properties from the smg population , but it does not appear to have done so . lensed sources will be valuable in understanding the properties of the underlying dsfg population , including the smgs : for observations , magnification by even a factor of a few translates to an order of magnitude reduction in integration time . detailed studies of the galaxies interstellar media in multiple molecular and atomic lines becomes feasible in the bright lensed sample ( e.g. , * ? ? ? * ) , with magnification also increasing spatial resolution in the source plane ( e.g. , * ? ? ? overall , the detection fraction of about half of the targets compares well with other programs that observed molecular lines from high redshift galaxies . a conservative extrapolation of a co - to - continuum intensity relationship indicates that the nondetection rate is substantially higher than expected from chance alone , evidence that some of the simple photometric redshifts are incorrect . our selection is from spire photometry identification of `` 350@xmath4 m peakers . '' comparison of the spire data and co redshifts indicates that the continuum peak wavelength , deduced from the 250@xmath4m/500@xmath4 m flux density ratio , predicts the redshift of the galaxies from our carefully - selected sample with @xmath180 . the spire colors do not necessarily provide accurate estimates for galaxies we do not detect , however . a simple explanation for some of the nondetections is emission from dust warmer than the @xmath181k characteristic for most of the galaxies with co detections . the redshift - temperature degeneracy for the observed peak emission wavelength , @xmath182 ( e.g. , * ? ? ? * ) causes problems for photometric selection in the spire bands alone : warmer galaxies with redshift above the zpectrometer s upper limit will appear in the target list . it is very likely that this is the reason for the nondetection of the target with third - brightest 350@xmath4 m flux density in this study . reasons other than warm galaxies or co intensities that fall below our observational limit can also contribute to nondetections . emission from multiple galaxies within the beam at one or more wavelengths can contribute flux to distort the apparent sed , leading to erroneous redshifts from single - component models . this is likely to affect weaker sources preferentially , both because smaller absolute amounts of contaminating flux are increasingly important and because the density of contaminating sources increases with decreasing flux . multiple distinct temperature components within a source , for instance a central agn - heated dust region surrounded by less excited material , will also distort the sed . depending upon the temperature structure within the source and the size and position of the source relative to any lens caustic , differential amplification could distort observed seds . seds in all these cases will be poorly represented by the simple single - component model we have used in target selection , and photometric redshifts based on spire data alone will fail to some extent . the observed redshift distribution and the dust sed properties hint at some change in dsfg properties near @xmath183 . the redshift cumulative distribution functions for both the zpectrometer and @xcite samples show a slope break at @xmath184 although they are selected in different ways . this indicates that dsfgs become less common , or are selected less efficiently by continuum properties , above that redshift . an increase in the number of warm galaxies with redshift would both explain the tentative trend in temperature we see and explain the lower co detection rate . temperature alone may not suffice to explain the break , and explanations involving more complex seds with increasing redshift are attractive as well . @xcite use extensive optical and infrared photometry on a sample of 74 870@xmath4m - selected smgs , also finding a precipitous drop in submillimeter source counts above @xmath185 in spite of their sensitivity to emission from warm dust . increased sample sizes are necessary to establish whether there is a real change in dsfg properties near @xmath183 or not . the co data presented here are among the earliest spectroscopic explorations of _ herschel_-selected galaxies . continuing observations of the co @xmath2 line by the gbt and jansky - vla will provide a more diverse sample of galaxies : some from broadened continuum selection criteria , and some with deeper integrations to reach weaker line emission . wideband spectrometers now coming on line for the 3 mm band at the lmt and iram 30 m will also add to the number of galaxies with known molecular redshifts . an expanded sample will strengthen our understanding of the relationship between molecular and continuum emission in dsfgs . we thank l. leeuw , m. michalowski , and i. valtchanov for their comments on aspects of this work . we acknowledge support from the national science foundation under grant numbers ast-0503946 to the university of maryland , ast-0708653 to rutgers university . dtf acknowledges support by nasa through an award issued by jpl / caltech ; dr acknowledges support from nasa through a spitzer space telescope grant ; irs and ams acknowledge support from stfc ; sb acknowledges financial contribution from the agreement asi - inaf i/009/10/0 ; jgn acknowledges financial support from spanish csic for a jae - doc fellowship and partial financial support from the spanish ministerio de ciencia e innovacion project aya2010 - 21766-c03 - 01 . results here came from gbt programs 8c-09 ( pi smail ) , 9a-40 ( pi swinbank ) , 10c-29 ( pi frayer ) , and 11a-27 ( pi frayer ) . we thank the gbt staff for their support and contributions . the national radio astronomy observatory is a facility of the national science foundation operated under cooperative agreement by associated universities , inc . herschel_-atlas is a project with _ herschel _ , which is an esa space observatory with science instruments provided by european - led principal investigator consortia and with important participation from nasa . the h - atlas website is http://www.h - atlas.org/. , a. , stacey , g. , glenn , j. , erickson , n. , harris , a. , & wootten , a. 2007 , in astronomical society of the pacific conference series , vol . 375 , from z - machines to alma : ( sub)millimeter spectroscopy of galaxies , ed . a. j. baker , j. glenn , a. i. harris , j. g. mangum , & m. s. yun , 3 , a. i. 2007 , in astronomical society of the pacific conference series , vol . 375 , from z - machines to alma : ( sub)millimeter spectroscopy of galaxies , ed . a. j. baker , j. glenn , a. i. harris , j. g. mangum , & m. s. yun , 8292

this paper describes some interesting properties of waves in , and oscillations of , the interstellar medium in the direction normal to the plane of the galaxy . our purpose is to examine possible reasons for four observed phenomena : the falling sky in the northern hemisphere ; the apparent presence of clouds in absorption spectra when a sightline is occupied primarily only by warm intercloud gas ; the peculiar structuring of spiral arms involving clumps , spurs , and feathering ; and the existence of an abundance of high stage ions far off the plane of the galaxy . we explored the reaction of the interstellar medium - in the vertical direction only - to large imposed disturbances ( initial displacements , expansive velocities , and compressions ) , and to the introduction of small amplitude waves via oscillation of the midplane . our findings included : 1 ) the anticipated growth in amplitude of high frequency waves with height ; 2 ) the four lowest normal modes for the oscillation of the atmosphere as a whole , as functions of the height of the outer boundary ; 3 ) the time for material to ` bounce ' from one unusually dense state to the next as a function of height ; and 4 ) the tendency for the disk to develop a hot outer halo , either after the passage of a single shock from a large event , or in response to a continuous stream of small amplitude waves . we discovered that three of the four observed phenomena targeted are likely to be closely connected . following a large expansion , material near the plane falls back first , with material at higher then falling in upon it . this provides precisely the sort of velocity segregation observed in the northern sky , at about 50 myr after the event . in addition , this bounce time ( and/or the period of the subsequent smaller oscillations , which is about twice the bounce time ) may be linked to structure in the spiral arms , with vertical oscillations having been provoked by initial compressions in the arms . oscillations of the fundamental symmetric ( breathing ) mode of the ism also produce substantial disturbances in the outer atmosphere . this can result in the production of an extensive layer of hot gas overlying the cooler disk material , i.e. a hot galactic halo with a significant population of high stage ions . hence , three of the four phenomena may be natural results of the simple existence of strong local compressions at the spiral arms , and the associated vertical motions in a thick galactic disk . finally , the somewhat mysterious appearance of clouds in some absorption spectra can be produced by small amplitude waves in the ism . under the right conditions , clouds will seem to appear through `` velocity crowding , '' when in fact there are no density concentrations in space .

astro-ph0010199

in this paper , we discuss the results of a range of experiments performed with a 1-dimensional hydrocode to determine the characteristics of vertical oscillations of the interstellar medium . we were interested in the dynamics of the ism in the direction normal to the plane of the galaxy ( i.e. the vertical direction ) . because our model treats the distribution of the ism as plane parallel , we sometimes refer to this as the `` galactic atmosphere problem . '' our experiments were aimed toward shedding some light on four observed aspects of the interstellar medium . the first of these aspects is the fact that in roughly half of the northern sky , 21 cm observations indicate that there are two primary components to the material distribution , a low velocity component and another , comparable in column density , which is falling toward the plane with a velocity on the order of 50 km s@xmath1 ( e.g. stark et al . 1992 ; see figure [ fig20 ] for a typical profile ) . this infalling component has been studied extensively ( e.g. danly et al . 1992 ) and named the iv ( intermediate velocity ) arch . our experiments explored dynamical situations that could lead to such velocity segregation , a circumstance that we refer to loosely as the `` falling sky . '' the second aspect is revealed by the uv absorption line spectrum of hd 93521 ( spitzer & fitzpatrick 1993 ) , a star which has a @xmath0 distance of about 1.5 kpc and lies behind the iv arch . the h i spectrum in that direction shows the usual two components , with considerably more than thermal broadening , and one rather narrower component at roughly -10 km s@xmath1 heliocentric velocity . on the other hand , the uv spectrum ( with the reduced thermal broadening of the lines of the heavier elements ) shows a total of about 10 components , which have been interpreted as localized regions of enhanced density . however , it is well known that velocity crowding ( caustics in the velocity distribution ) can produce spectral features in an otherwise uniform medium . hence we performed experiments to explore whether the features observed toward this star might in part derive from typical interstellar velocity irregularities rather than density concentrations in space . the third aspect involves the structure of spiral arms in galaxies . such arms are generally not entirely smooth and continuous , having localized regions of enhanced contrast , in a pattern that could be called feathering . one gets the impression that ( as in the wake of a boat ) there are wavelets that are not traveling normal to the locus of the arms . in addition , there are larger irregularities - pieces of arms or spurs - that branch off of the smooth locus of the more regular pattern . martos & cox ( 1998 ) began to explore the vertical structure anticipated for a spiral arm , including the thick disk and the significant nonthermal pressure . they used only a two - dimensional model , but speculated on some of the potential three - dimensional characteristics , including the possibility that arm substructure might be related to resonances in the induced vertical motions . in this paper we have performed two sets of experiments useful in exploring the relevance of this idea , determining the normal modes of vertical oscillations and the timescale for interstellar material to return to the galactic plane after being set into motion away from it . the fourth aspect is related to the substantial quantities of high stage ions ( in particular c iv , n v , and o vi ) found well off the plane of the galaxy , with apparent scale heights up to several kpc ( savage & massa 1987 ) . a variety of possible sources for these ions have been suggested . here we are exploring only one , a possibility related to the suggestion by sciama ( 1972 ) that the outer boundary of the cool interstellar medium might be chromospheric , with a layer of hot gas beyond that . our final set of experiments was aimed at examining the structure of the outer edge of the disk resulting from mechanical disturbances within it , to learn whether the transition to a hot outer layer is a natural consequence . the simulations in this paper were performed with a 1-d plane parallel lagrangian hydrocode . unlike some other lagrangian codes ( e.g. bowers & wilson 1991 ) , ours did not make use of a mass coordinate for simplifying the computation of derivatives . instead , our code was constructed to maintain second order accuracy by differentiating ( and interpolating ) explicitly in space with piece - wise parabolic fitting ( walters 1993 ) . this method allowed us to maintain good resolution in all parts of the grid by using zones of initially uniform spacing . it is well suited for cases in which the grid spans several orders of magnitude in density , such as the galactic atmosphere problem investigated in this paper . the code was otherwise similar to several others constructed in our group for the study of supernova remnants ( e.g. slavin & cox 1992 ; cui & cox 1992 ) , based on the methods of richtmyer & morton ( 1967 ) . several additional features of the code are worth mentioning . although this was a one fluid model , it allowed for the possibility of different electron and ion temperatures via the methods described in cui & cox ( 1992 ) and cox & anderson ( 1982 ) . it followed the ionization of hydrogen and helium explicitly , including collisional ionization , radiative recombination , and an additional low level of ionization ( and heating ) nominally due to cosmic rays . the code also followed the heating and cooling rates for each parcel ( with heating primarily due to compressions and shock waves ) . for hydrogen , cooling terms from bremsstrahlung , radiative recombination , and collisional excitation were computed explicitly . all other elements were assumed to radiate at their collisional equilibrium rates , using a rate table produced from the raymond and smith code ( raymond , cox , & smith 1976 ) with standard cosmic abundances . in order to avoid excessively low temperatures , all cooling except that due to recombination of hydrogen was turned off below 10@xmath2 k. electron thermal conduction was also included in the code . most of these features had little effect on the overall behavior of the atmosphere , but allowed for a more realistic study of the outer layers of the galactic disk . our model for the vertical distributions of density and pressure was similar to that of model a of martos & cox ( 1998 ) , based on the hydrostatic solution of boulares & cox ( 1990 ) . the density distribution consisted of the smoothed contributions from molecular and cold atomic clouds as well as the warm neutral and the warm ionized components . each component had its own total column density , scale height , and form of distribution . the total number density of nuclei was given by : @xmath3 + 0.37\exp \left [ -\frac{z^2}{2\left ( 135\textrm { pc } \right ) ^2}\right ] \right\ } \textrm { cm}^{-3 } \nonumber \\ & + & \left\ { 0.1\exp \left ( -\frac{\left| z\right| } { 400\textrm { pc}}\right ) + 0.03\exp \left ( -\frac{\left| z\right| } { 900\textrm { pc}}\right ) \right\ } \textrm { cm}^{-3 } . \label{eqn1}\end{aligned}\ ] ] the gas consisted of hydrogen and helium atoms , in the ratio of 10:1 . ( the presence of other elements was neglected when computing the bulk properties of the gas such as density and pressure . ) the electron density was inferred from the density of nuclei and the ionization level . in the initialization , a constant cosmic ray ionization rate was included which was sufficient to maintain 25% ionization of hydrogen at a @xmath0 distance of 5 kpc from the galactic plane ( assuming the equilibrium density at that height ) . we used the gravity given by bienaym , robin , & crz ( 1987 ) , as fit by martos & cox ( 1998 ) : @xmath4 \textrm { cm s}^{-2 } . \label{eqn2}\ ] ] given the density and gravity distributions , one can derive the total pressure as a function of height , the result depending somewhat on the assumptions about the outer boundary . our assumption was that the density followed the form described above until the outer boundary was reached , at which point the pressure was taken as purely thermal and then integrated downward . beyond the boundary , the density and pressure were taken to be zero , making the outer boundary a free surface . the location of the boundary was most often 5 kpc , but that height was varied somewhat for some of the experiments described later . ( note : having the pressure drop rapidly to thermal near the outer boundary while the density remained finite had the effect of terminating the disk with a layer of low signal speed . however , this oddity at very high @xmath0 appears to have had little effect on the results of our experiments . ) the pressure forms assumed for the calculation were thermal pressure and a pseudo - magnetic pressure . the thermal pressure was intended to include the internal dynamical pressures of the denser components , but not the large - scale motions actually being modeled . this was accomplished by assuming a constant temperature of 10@xmath2 k for all components of the static solution . there was no explicit inclusion of cosmic ray pressure , known to be a very important component ( boulares & cox 1990 ) . instead , the entire nonthermal pressure was represented by the pseudo - magnetic term . given the total pressure from the weight integration and the thermal pressure from the density , ionization , and temperature , the magnetic pressure was found by simple subtraction . our form for the equilibrium thermal pressure was : @xmath5 where @xmath6 is the ratio of the total number of particles to the number of atoms or nuclei . it then followed that the equilibrium magnetic field would be given by : @xmath7 } , \label{eqn4}\ ] ] where @xmath8 is the total initial pressure ( thermal plus magnetic ) . when evolving the model over time , it was assumed that the magnetic field was parallel to the galactic plane and the flux was frozen in the flow , so that : @xmath9 in this equation , the functions @xmath10 and @xmath11 are the hydrostatic values for the lagrangian parcel initially located at @xmath12 , and now found at @xmath0 . we employed both one - sided and two - sided models of the atmosphere . the one - sided model was used for studying symmetric modes and wave propagation into the atmosphere . in this model , the lower boundary represented the galactic midplane and was rigid , meaning that there was no inflow or outflow of gas . the two - sided model was used to study asymmetric modes of oscillation . figure [ fig1 ] shows the equilibrium pressure , density , magnetic field , and ionization distributions of the two - sided hydrostatic model out to @xmath135 kpc . the one - sided atmosphere was the same except that it was limited to that shown in the right half of each graph ( @xmath14 ) . we tested the hydrocode with several problems that have analytical solutions . one such problem was that of a piston moving at a constant speed down a 1-d tube containing a uniform medium . in this case , we found that the speed of the shock produced by the piston was in excellent agreement with theory , with differences usually less than 1% . our results were equally good when magnetic pressure was included in the calculation . another interesting test was to observe the growth of small waves as they propagated upward in our model atmosphere . using a one - sided disk , we oscillated the midplane boundary up and down sinusoidally with a very small amplitude . the amplitudes and positions of the extrema in velocity were tracked as they propagated , and the results were compared with the wkb approximation , which tells us that @xmath15 is constant vs. @xmath0 ( see appendix ) , where @xmath16 is the undisturbed density distribution , @xmath17 , and @xmath18 is the amplitude . in our case , the signal speed is given by @xmath19 . again , we found good agreement between theory and the results of our program . for example , one test case had oscillations with a period of 5 myr and an amplitude of 0.01 km s@xmath1 at the midplane . at this period , the wavelengths are roughly one third of a scale height , and the behavior should not necessarily correspond precisely to the short wavelength limit . nevertheless , we found that the growth in amplitude closely approximated the wkb result . ( the amplitudes grew by factors of 10 , 20 and about 60 at heights of 2 , 3.6 , and 5 kpc . ) however , there was a small asymmetry between the positive and negative velocities that appeared to decrease with the passage of successive waves . in this section , we present the results of our investigation ( via numerical experiments ) into the normal modes of vertical oscillation of our model ism . we attempted to find these modes , along with the corresponding resonant frequencies , by introducing a large disturbance into our model ism . we succeeded in discovering both antisymmetric modes ( which we call sloshing modes , for reasons discussed later ) and symmetric ( breathing ) modes . we found two distinct frequencies for each type , and also discovered how those frequencies depend on the assumed height of our outer boundary . for comparison , we undertook a separate linear normal modes analysis ( see appendix ) assuming a pressure node at the top of the atmosphere , and found almost exactly the same frequencies as reported below from the numerical experiments . our interest in these frequencies arises from the hypothesis that resonances between the vertical modes and spiral density waves might lead to reinforcement of particular patterns within the arms , as suggested in martos & cox ( 1998 ) . this could produce substructure in the spiral arms , such as feathering , spurs , or `` beads on a string . '' in addition to the periods of normal modes , we also examined the response of the ism to a large expansive event , provoked by an initial strong compression of the central portion of the disk . we explored the time required for the gas to bounce from one compressed state to the next as a function of the strength of the event and of the distance from the midplane . the motivation was similar , to understand this characteristic bounce time as a possible indicator of the distance between compressed regions along the length of a spiral arm , or between the arms themselves . it also gives an idea of the recovery time of the disk after any major perturbation , such as impact by a high velocity cloud . in our search for antisymmetric modes , the entire two - sided atmosphere was initialized as if it were in hydrostatic equilibrium , then shifted in height by a constant amount and released . for our first simulation , we chose an offset of 100 pc and atmospheric boundaries at @xmath133 kpc . our program keeps track of the properties of certain designated parcels , chosen to show what different regions of the atmosphere are doing . as a rule , the parcels examined have equilibrium heights ( before the displacement ) of @xmath12 = 0 , 10% , 20% , 50% , and 100% of the upper boundary height ( i.e. the top of the atmosphere , @xmath20 ) . in the results below , we show the locations of these parcels upper boundaries versus time . we also followed changes in the total energy in the grid . these results were then fourier transformed to find the periods of oscillation . figure [ fig2 ] shows the history of the locations of these tracer parcels over the course of a 3 billion year simulation . as the atmosphere oscillates vertically , the motions of the individual parcels appear to be sinusoidal with beats of roughly five oscillations each , implying the superposition of two frequencies differing by about 20% . note also that the modulation due to the beats is relatively small for the parcels at intermediate distances from the midplane , implying that one of the frequencies has a node in the displacement in that vicinity . this is just what would be expected for a mixture of the fundamental antisymmetric mode and its first overtone . the amplitude of the modulation is quite large at the midplane and the upper boundary , requiring comparable wave amplitudes there . the midplane is a displacement anti - node for both frequencies . the outer boundary is a pressure node , corresponding to a velocity and displacement anti - node . also shown in figure [ fig2 ] is the history of the total energy ( per unit area ) in the grid . note that this is the total energy of _ both _ sides of the atmosphere . the oscillations in this plot arise from the weak heating term that acts when expansion cools the gas below 10@xmath2 k. because this happens with opposite phase on the two sides , the oscillations in energy have twice the frequency of those in displacement . we performed fourier transforms on these histories , using the algorithm given by press et al . ( 1989 ) . to determine the periods accurately , we eliminated the early complications of the transient response by examining only the last 2 billion years of the simulation . figure [ fig3 ] shows the power spectra resulting from the fourier analysis . as expected , they are dominated by two large peaks separated by about 20% in frequency . the corresponding periods are 117.6 and 93.0 myr , independent of height above the midplane . ( for comparison , the periods from our linear normal mode analysis are 118.5 and 93 myr . ) note that the relative amplitude of the higher frequency peak is low for tracer parcels at intermediate distances from the plane , in agreement with the expectation that it has a node in that vicinity . figure [ fig4 ] illustrates the structure in the atmosphere at 1.5 , 2.0 , and 2.5 billion years . the `` sloshing '' back and forth ( across the midplane ) is evident , though complicated by the contribution from the second mode that tends to move material at high @xmath0 opposite in direction from that at low @xmath0 . this sloshing motion ( which is why we call this the sloshing mode ) can be seen by observing changes in the pressure distribution between the three times . in a linear approximation , the velocity distributions would be entirely symmetric , with the velocity of the center occasionally opposite in direction from that at great distances from the plane , depending on the amplitudes and phases of the two modes . for the particular times shown , the velocities are everywhere of the same sign ( an accident of the phasing ) , but there is an obvious lack of complete symmetry in the velocity field . this asymmetry could be due to any of several factors : the anticipated symmetry is with respect to the unperturbed positions of the parcels ; the flow includes the full nonlinearities which should be most significant far from the plane where the velocities are large ; and the power spectrum shows very weak features at other frequencies that may include contributions excited at the opposite symmetry through round - off errors or asymmetries in the code . in spite of these complications , however , the atmosphere oscillates at moderate amplitude at just those frequencies expected from the linear analysis . in addition , the oscillations persist for an extremely long time . repeating this simulation using atmospheres with various maximum heights , @xmath21 , ranging from 1 to 5 kpc , we found that the periods increase as the height increases , still in excellent agreement with the linear analysis . the results are shown as the two solid lines in figure [ fig5 ] . note that while the height of the atmosphere increases from 1 to 5 kpc , the period of the fundamental mode goes from 104 to 133.3 myr , an increase of only 30% . in fact , our linear analysis tells us that the period should approach an asymptotic value as @xmath21 increases , as the forces inversely proportional to the wavelength squared fade in importance relative to those inversely proportional to the scale height squared . to excite modes that are symmetric about the galactic midplane , we initialized the model with a symmetric compression of the lower atmosphere , reminiscent of the compression observed in spiral arms . in the first experiment , all parcels within 500 pc of the midplane in equilibrium were compressed by a factor of 1.5 , so that their heights decreased and their densities increased by a factor of 1.5 . above ( or below ) this compressed layer , all of the parcels retained their equilibrium properties , although they were shifted downward ( or upward ) due to the smaller thickness of the compressed layer . for comparison with our previous simulation , we again chose an initial boundary height for the atmosphere of 3 kpc . the results for a 3 billion year run are shown in figure [ fig6 ] . again there is periodic motion with amplitude modulation . the fourier transforms of these histories are shown in figure [ fig7 ] . notice that in this case not all parcels have the same oscillation frequencies . the motion of the galactic midplane ( @xmath22 ) corresponds to periods of 117.6 and 93.0 myr , just those of our previous asymmetric disturbance . however the rest of the parcels and the energy show periods of 109.6 and 70.2 myr ( compared with 109.5 and 69.5 myr for the two lowest symmetric modes from the linear analysis ) . and once again , the relative amplitude of the higher frequency component is low at intermediate heights , close to its node . ideally , one would expect the midplane to remain perfectly still in this experiment , but we find that a very slight amount of sloshing has been excited , which is probably due to round - off errors . these oscillations are much , much smaller than any of the other oscillations , and are only visible because the midplane is a velocity node for all symmetric modes . any spurious sloshing mode frequencies are totally invisible ( below noise level ) in the oscillation spectra taken at other heights . figure [ fig8 ] shows the structure of the atmosphere at 1 , 2 , and 3 gyr . as expected the density , pressure , and temperature distributions are symmetric about the midplane , while the velocity profiles are antisymmetric . the velocity profiles at 1 and 3 gyr show the entire disk ( both above and below the midplane ) to be either expanding or contracting , in other words `` breathing . '' however , at 2 gyr the motion changes from expansion to contraction at about 2.3 kpc off the plane , revealing the presence of the overtone . repeating the simulation using several different boundary heights for the atmosphere , as we did in the previous section , yielded the results shown as dotted lines in figure [ fig5 ] . figure [ fig5 ] in fact illustrates all of the anticipated characteristics of normal modes . the longest period belongs to the sloshing mode with no displacement nodes . from there , the periods decrease with increasing node number , alternating between symmetric ( breathing ) and antisymmetric ( sloshing ) forms . the periods also increase gradually with increasing atmospheric thickness , the specific rate depending on the detailed structure of the atmosphere and in particular the @xmath0 dependences of gravity , density , and the signal speed . as @xmath20 goes to infinity , these periods should approach asymptotic values . a final experiment was performed to excite both sloshing and breathing modes with comparable amplitudes for an atmosphere with boundaries at @xmath132.5 kpc . in this setup , a one - sided expansion was initialized by imposing a uniform velocity gradient equal to 25 km s@xmath1 kpc@xmath1 on the upper half of the atmosphere ( @xmath14 ) and leaving the other half ( initially ) at rest . the resulting oscillations of the tracer parcels are shown in figure [ fig9 ] . in the fourier analysis of these motions , shown in figure [ fig10 ] , this modulation pattern is found to correspond to periods of 114.3 and 102.6 myr . the 114.3 myr mode is exactly that of the first sloshing mode for an atmosphere of this height , while the shorter period is very close to the 103.9 myr value for the first breathing mode . ( the linear analysis yielded 114.7 and 103 myr , respectively . ) the midplane , which is a displacement node for the breathing modes , shows no beats and its spectrum contains only the longer period . the amplitude modulation shown in figure [ fig9 ] increases monotonically with height until it is 100% at the outer boundary . this is reflected in figure [ fig10 ] as an increase in the higher frequency component until the two are equal at the outer boundary . comparison of figures [ fig2 ] , [ fig6 ] , and [ fig9 ] shows that our experiments regularly produced large amplitude modulation at the outer boundary . we do not know whether this is due to an accident of the initializations , nonlinearities in the early more dramatic motions , or a natural property of this atmosphere . note also that all experiments led to spectra dominated by only two frequencies . in the mixed mode experiment , we did not obtain all four of the previous frequencies , only the two lowest ones . this pair of curiosities may indicate something fundamental about how the atmosphere responds to major disturbances . in the previous section , we were only interested in the long - term behavior of the atmosphere , disregarding the atmosphere s initial ( transient ) response to disturbances . as indicated previously , the initial response to the disturbance is also of interest and we examine that next . in the experiments on symmetric modes , one of our simulations had a compression factor of 1.5 , a total atmosphere height of 5 kpc , and a run time of 3 gyr . figure [ fig11 ] shows the tracking of our tracer parcels for the first 1 gyr of this simulation . when the parcels first rebound from the compression imposed on them , they are propelled to relatively large heights . they then fall back toward the midplane , slowing down only as they come close to their equilibrium positions . oscillations subsequent to this first bounce are smaller and rapidly approach the periodic pattern discussed previously . the parcels very close to the outer boundary require more than one bounce to settle . the midplane oscillation amplitude should in principle be zero , but in actuality has a small asymmetric component . all other regions seem to settle after only one return . a more detailed view of these transients can be seen in figure [ fig12 ] , which shows the first 250 myr of this simulation . note that for this illustration , the simulation has been redone with the first tracer parcel set at 5% of the peak height of the atmosphere , rather than at the nearly stationary midplane . we repeated the experiment for a range of initial compression factors and two different boundary heights ( 5 kpc and 3 kpc ) . defining the bounce time , @xmath23 , for a parcel as the time from the beginning of the experiment to time of its next minimum in height , the results for the atmosphere with the 5 kpc boundary height are shown in figure [ fig14 ] . as we can see in figure [ fig14 ] , the bounce times for the parcels of the inner disk depend very little on compression factor , which is a measure of the strength of the disturbance . there is a slight decrease in bounce times with increasing compression factor , perhaps because these parcels return more rapidly when they hit the upper atmosphere harder . also , these times are only weakly dependent on the equilibrium heights of the parcels , with a slight linear gradient that arises from the flattening of gravity and the increase in the amplitude of the disturbance with increasing @xmath0 . conversely , parcels very high in the atmosphere have significantly longer bounce times , which is partly due to the longer time it takes the initial disturbance to reach them , partly to the nearly constant gravity there , and partly to the rapid growth in the amplitude of the disturbance with @xmath0 . they also take progressively longer to return with increasing compression factor . much the same behavior is found when the boundary height is reduced to 3 kpc , though the initial decrease in bounce time with increasing compression factor for parcels near the plane is more pronounced . the overall result for moderate strength disturbances and either boundary height is that the bounce times near the plane are about 40 myr , increasing slowly with parcel height by roughly 20 myr kpc@xmath1 . for weaker disturbances , the bounce time is somewhat longer , closer to the half period of the first symmetric mode , but increases more slowly with height . on the whole , bounce times for material near the midplane lie between 40 and 50 myr , increasing to about 70 myr for material 1.5 kpc off the plane , and to much longer times for the outer boundary . a fundamental consequence of these results is explored in the next section , but a discussion of additional possible significance is deferred to the conclusions . as discussed in the introduction , spitzer & fitzpatrick ( 1993 ) have observed the uv absorption spectrum of the galactic halo star hd 93521 , identifying at least nine separate velocity components that they interpret as interstellar clouds . these components congregate in two velocity regions , one approximately stationary and the other moving towards the midplane at roughly 50 km s@xmath1 . this general structuring is typical of roughly half of the northern sky , which is covered by the iv arch . the numerical experiments in this section are designed to show how a large disturbance in the galactic atmosphere can naturally create such velocity segregation , and that small velocity perturbations can produce the multiple velocity components seen without any actual clouds ( via velocity crowding , i.e. caustics in the velocity field ) . as explored in the previous section , material may move away from the plane following a large disturbance within the disk , returning after a characteristic bounce time . but , because that return time increases with height , material nearer the plane falls back sooner and accumulates near the midplane at low velocity , with material further up continuing to rain down upon it . if the velocities involved are large , the infalling outer atmosphere will be bounded on the bottom by a shock between it and the already decelerated material close to the plane . owing to the considerable density gradient in the infalling material , the spectrum during such infall will show two fairly sharp velocity components , one at low velocity due to the nearly stationary material near the disk , and one at an `` intermediate '' negative velocity due to the material just about to pass through the shock . ( for smaller disturbances with lower velocities and no shock , the second feature will appear at the extremum velocity of the infall as it just begins to slow . ) thus , the observed velocity segregation is a natural consequence of a large disturbance and should be found at a time roughly equal to the return time for material a few hundred parsecs off the plane , which from our previous discussion is expected to be 50 to 60 myr , depending on the thickness of the disk . in our experiment to explore this phenomenon , we initialized the program with the two - sided equilibrium atmosphere . a velocity profile was then imposed such that the whole atmosphere was expanding . the velocity rose / dropped linearly from @xmath24 at @xmath25 to a maximum / minimum of @xmath26 km s@xmath1 at @xmath27 kpc . for @xmath28 kpc , the velocity was constant at @xmath29 km s@xmath1 . the equilibrium height of the atmosphere s boundary was 5 kpc . figure [ fig18 ] illustrates the evolution of one side of this symmetric model , showing the density and velocity profiles at several times . also shown is the predicted differential column density vs. velocity ( i.e. the velocity spectrum ) that would be seen in the northern sky by an observer at the midplane . for comparison with hd 93521 , we integrate the column density out to 1.5 kpc , the approximate distance to that star . we also use a lower limit of 235 pc for the integration to account for the presence of the local interstellar cavity . this particular limit was chosen to make the total column density equal to 1.25@xmath3010@xmath31 @xmath32 , at least initially , which is the value observed in that direction . the solid line in the left column of figure [ fig18 ] illustrates the initialization . the absence of the lowest velocities in the spectrum is due to the lower limit of the integration ( at 235 pc ) . the dotted line , 20 myr later , shows the substantial amount of expansion involved , and that a significant amount of material close to the plane has reached zero velocity . the dashed line at 40 myr shows that the velocities have reversed out to about 3 kpc . the spectrum at that time has a peak at the extremum of negative velocity , and again has a hole at zero velocity due to the lower integration limit . the right column of figure [ fig18 ] shows times of 50 , 55 , and 60 myr , during which a region of nearly zero velocity expands out to about 1 kpc from the midplane . outside this region there is substantial infall . as time progresses , the velocity of infall increases , while the column density of infalling material decreases . at the latest time shown , the structure is forming a shock which appears as an abrupt change in the velocity and as a step in the density profile . ( note also that the density profile interior to the shock differs very little from the equilibrium distribution . ) at later times the shock strengthens and passes out through the atmosphere . this shock substantially heats the tenuous outer parts of the atmosphere ( a subject we will return to in the next section ) and drives it back into expansion , leading to the subsequent smaller amplitude oscillations explored in the previous sections . figure [ fig19 ] provides an enlarged view of the velocity spectrum at 55 myr . the solid line is the actual spectrum , with strong peaks near -3 and -48 km s@xmath1 ; the dotted and dashed lines show convolutions with the thermal profiles for oxygen and hydrogen , respectively . the synthetic spectrum for hydrogen closely resembles the 21 cm observations for many regions in the northern sky in the general direction of the iv arch ( danly et al . 1992 ) , in particular that shown by spitzer & fitzpatrick ( 1993 ) in the direction of hd 93521 , reprinted as figure [ fig20 ] . the details of these spectral results depend somewhat on the initial form of the disturbance . roughly speaking , a steeper initial gradient promotes more mass to higher velocity . this results in a larger column density in the infalling component when it is centered at a given velocity , or a higher infall velocity for a given column density in the infalling component . our initialization was selected to produce comparable column densities in the two components when the infalling component had a speed of roughly 50 km s@xmath1 . we have shown that the bi - modal velocity distribution observed in the northern sky can be produced by vertical motions in the galaxy . next , we show how features resembling clouds can be produced . in the previous section , we initialized our model with a velocity profile which has a uniform gradient near the midplane and is flat away from the midplane . but the interstellar medium has a considerable amount of velocity structure on scales of a few kilometers per second , which needs to be included for a more realistic assessment of the anticipated spectrum . rather than trying to introduce these small - scale structures in the initialization , however , we chose to produce them by mechanical disturbances at the midplane , which then propagated into the outer disk as waves . the experiment in this section had exactly the same initial configuration as that in the previous section , with one important difference . to produce small waves , we applied a small sinusoidal motion to the midplane boundary . the period for these oscillations was 10 myr and the velocity amplitude was 5 km s@xmath1 . there were no waves present prior to the beginning of the expansion . although this method does not reproduce the randomness of interstellar turbulence , it nevertheless provides disturbances of the about the right amplitude and spacing . it also gives us a clear example to illustrate our point about caustics . figure [ fig22 ] illustrates the evolution of this model . comparing this to figure [ fig18 ] , one can see that the behavior of the atmosphere as a whole is very similar to the case without waves . the waves are of such low amplitude that there is scarcely any density or pressure variation visible , but they can be seen as small wiggles in the velocity profile near the midplane , moving outward with time ( until they are swept back by the infall ) . at late times , these variations also appear as small inclined steps in the density plot . on the other hand , the effect on the velocity spectrum is quite pronounced , even at early times . at later times we can discern at least 5 separate peaks , as can be seen in the spectrum at 55 myr shown in figure [ fig23 ] . these additional components arise from the convolution of the monotonic density profile with the steps introduced into the velocity structure by the small amplitude waves . the number of such steps increases with the oscillation frequency , but at a significantly higher frequency they would smear together to return to a nearly smooth distribution . in this particular experiment , only 5 oscillations have occurred by the time the infall has reached the desired state , and they have not quite managed to propagate as far as the velocity minimum . had there been waves in the experiment prior to the expansion , there might have been even more high velocity components . comparing the spectrum of figure [ fig23 ] with the 21 cm profile and uv components toward hd 93521 , shown in figure [ fig20 ] , there are several qualitative similarities . both show two large concentrations of mass in velocity space : one near 0 km s@xmath1 and the other near -50 km s@xmath1 . furthermore , each of these broad features consists of multiple components . the inclusion of thermal broadening for hydrogen ( dashed line ) , however , obliterates the small scale structure , making the spectrum into a pair of bell - shaped features . this is very much like the observed profile of figure [ fig20 ] , the only significant difference being that the latter contains a single much cooler component with low broadening . thus the combination of the increasing bounce times with increasing @xmath0 and the presence of small amplitude waves ( or , more realistically , the observed random motions ) in the interstellar medium naturally produces a spectrum with the form and complexity of that of hd 93521 , but with no density maxima actually associated with the velocity features ! our model includes no clouds whatsoever . all it requires is that our region of the galaxy have undergone a substantial amount of vertical expansion roughly 50 myr ago in order to produce the downfalling iv arch that is presently observed over roughly half of the northern sky . one caveat to our model is that the distribution of material in the iv arch is observed to have column density variations from direction to direction over the sky , almost surely implying that there are density concentrations within it . it would be difficult to imagine that its return to the plane of the galaxy would not have led to density structuring , even if it were not already present . however , velocity structuring within the ism is just as real and should frequently contribute to features in the spectra in low column density directions . the fact that one feature in the spectrum of hd 93521 is noticeably cooler than all the others suggests that it might be the one real cloud seen in the spectrum , the others being due to velocity crowding in the diffuse intercloud component . it is worth mentioning that these results are not peculiar to a one - dimensional model . martos has conducted similar eexperiments in two dimensions , and has achieved similar results ( personal communication ; see also fig . 16 in martos & cox 1993 ) . as described in section [ hydrostatic ] , our equilibrium atmosphere terminates at its outer boundary with a sharp cutoff in density , with material just interior to the boundary having a temperature of 10@xmath2 k. this is an unrealistic termination but was meant to be one that included no bias concerning what actually occurs there . our intent was to investigate what does happen there , keeping in mind the hypothesis of sciama ( 1972 ) . he suggested that the galactic atmosphere should pass through a chromospheric state far off the plane and have a high temperature corona beyond that . this transition could be produced by a variety of influences including the dissipation of mechanical energy arriving from below , the eventual significance of cosmic ray heating at very low densities , heating due to flare - like activity , or dissipation of infall energy . in this final set of experiments , we have explored how outward fluxes of mechanical energy ( like those invoked in our earlier experiments ) restructure the outer parts of the atmosphere , moving it away from our initially assumed equilibrium . the basic result is that when the disturbance produces a strong shock , or a series of shocks that are not too far apart , there is a sudden transition at some height to a much higher temperature regime . when the disturbance leads to a only a single strong shock , its passage generates a hot outer layer which then eventually cools back to the initial equilibrium . however , the cooling time is extremely long owing to the low densities , and is longer still if the temperature rises beyond the peak in the cooling curve . for small periodic perturbations , a `` hot halo '' can be built up by the successive passage of many waves or shocks . also , there is a significant resonance at which the halo produced is hotter and more distended than it is otherwise . the power required in the waves to maintain an appreciable hot halo is extremely small . as a first example , we return to the simulation from the previous section that has an expanding atmosphere with a maximum velocity of 60 km s@xmath1 . it produces a large segregation in velocity after @xmath33 50 myr , with an outward facing shock whose velocity jump approaches the original expansion speed , 60 km s@xmath1 . following this simulation for a much longer time , we see that the shock moves outward through the infalling material , remaining radiative but generating long lived ionization until it reaches a region of the atmosphere with sufficiently low density that the cooling time exceeds the shock transit time through the remainder of the atmosphere . the material heated during that final phase is the transient hot halo in this case . subsequently , the inner denser parts of the halo cool faster than the outer regions . both the amount of material remaining hot and the temperature of the hot material gradually decrease over time . this description is complicated slightly by the further reverberations of the atmosphere , which send additional weaker shocks into the halo . these effects are illustrated in figure [ fig24 ] , which shows the density , temperature , and ionization structure of the model at very late times : 400 and 600 myr after the start of the simulation . the sudden drop in density and rise in temperature at height of 5 kpc ( the nominal boundary height ) are apparent . the temperature in the halo is roughly 10@xmath34 k , decreasing with time . note , however , that the halo persists for hundreds of millions of years . the density in the halo is roughly 10@xmath35 @xmath36 , at which the anticipated cooling time is consistently expected to be around 500 myr . next we examined the effect a series of small waves or perturbations , with particular interest in what would happen at the resonant frequencies found in section [ large ] . in this set of experiments , we considered a one - sided atmosphere with an outer boundary initially at 5 kpc . we forced the midplane to oscillate with a velocity amplitude of 2 km s@xmath1 and explored a range of oscillation periods between 100 to 150 myr , which includes the lowest frequency resonant modes ( which have periods of 129 and 133 myr for this atmosphere ) . the perturbations were made small enough so that an individual oscillation would not cause a dramatic change in the atmosphere , but would still be on the scale of the velocity dispersion of the interstellar medium . in each case , we followed the evolution for 3 gyr , and the structures discussed below are those at the end of that time . figure [ fig25 ] shows the final structure of the atmosphere for an oscillation period of 100 myr ( relatively far from the resonances ) , as well as the net change in energy in the atmosphere during the evolution . a hot ionized halo with a temperature rising to roughly 2.5@xmath3010@xmath34 k has formed above the cool disk boundary . the density in the halo drops from @xmath37 to @xmath38 @xmath36 over the range 5.3 to 6.4 kpc . examining the structure at other times , we found that this hot layer remains approximately constant in size , position , and temperature . the atmosphere has therefore reached a new equilibrium state in response to the energy input from the oscillations . this can also be seen by looking at the net total energy in the atmosphere vs. time , shown in the lower right graph in figure [ fig25 ] . the large oscillations are just the insertion and removal of @xmath39 work as the lower boundary pushes and pulls against the midplane pressure . if one looks past these oscillations for an overall trend , one can see that the average energy is nearly constant . for an oscillation period of 133 myr , the first sloshing mode resonance found in section [ large ] , we again find a hot halo , similar to but slightly hotter than in the previous experiment . interestingly , the envelope of the energy curve appears to oscillate in this case , suggesting that the driving frequency may be beating against a nearby resonance . figure [ fig27 ] shows the results for a period of 129 myr , the first breathing mode resonance . they are dramatically different from the previous cases . the halo has become very distended , extending from 4 to 20 kpc , and is significantly hotter , with a temperature of 10@xmath40 k. the lower right panel shows the time evolution of the net energy per unit area , relative to the initial equilibrium state . in this case , the mean total energy has a significant increase , rising by about 2@xmath3010@xmath41 erg @xmath32 over the 3 gyr run . interestingly , the average energy seems to level off after 1.5 gyr , but then turns up again near the end . the results for 133 and 129 myr are remarkably different for having been produced by only a relatively small shift in period . both cases used resonant frequencies for the unperturbed atmosphere , but those frequencies may not correspond exactly to resonances in the new atmosphere with its lowered boundary for the cool gas and its hot outer layer . in order to explore the possibility that there might actually be a strong resonance for hot gas generation , suggested by the great differences between these two results , we explored a wide range of driving periods . figure [ fig28 ] illustrates the change in the mean energy column density over 3 gyr runs for a wide range of periods of oscillation of the midplane . the mean energy is taken as the average of the maximum and minimum in one cycle , and the change as the difference between the values for the first and last full cycles in the simulation . the curve shows a strong resonance with peak response at 127 myr . ( a quick investigation for smaller periods revealed the presence of additional resonances in the neighborhoods of 60 and 80 myr . ) the simulation using the peak period of 127 myr is shown in figure [ fig29 ] . the hot halo is even more dramatic than for the 129 myr case . the temperature now jumps up sharply and continues to rise , reaching 2@xmath4210@xmath40 k. because material formerly in the cool part of the atmosphere has been promoted into the halo , supplying a density in excess of @xmath38 @xmath36 out to 11 kpc , the boundary between the two regimes has retreated downward slightly . in this case , the average energy does not level off , but continues to increase and even to increase in rate toward the end of the run , possibly due to a decrease in the radiative losses of the hot gas as it is heated further . let us now consider the power involved in generating this remarkable structure . the only input into the model comes from the ( sound ) waves generated by the midplane , which acts like a piston . if the midplane were a source of ordinary outward moving sound waves then the average power per unit area , or intensity @xmath43 , would be : @xmath44 we have used the initial midplane values for the density ( @xmath45 ) and signal speed ( @xmath46 ) . note that this result does not depend on the period of the piston motion , given the constant velocity amplitude . in equilibrium there are nearly equal amounts of power propagating upward and downward , so the above estimate is an extreme upper limit . in reality , the pressure and velocity variations are nearly a quarter period out of phase , making their product average to a much smaller value . the difference between the amounts of upward and downward propagating power must be just that delivered to the halo across the upper boundary . if that power is all outgoing , the corresponding result should be : @xmath47 where the next to the last step is taken from the results of our normal mode analysis and is very approximate . quantities with subscript _ hot _ are evaluated at the lower boundary of the hot region . this result differs from our upper limit in eq . [ eqn6 ] by approximately the ratio of the signal speed at the midplane to that in the halo . for comparison , the power per unit area actually retained by the atmosphere during the simulation ( after losses due to radiation by the halo , which we estimate to be comparable ) is : @xmath48 let us compare this with the amount of power released into the galaxy from supernovae and spiral arm disturbances . if there is one supernova of energy 10@xmath49 ergs every 100 years , the average power input into the upper half ( @xmath50 ) of the galaxy ( assuming a radius of 10 kpc ) is roughly : @xmath51 this is several orders of magnitude greater than the power input of our perturbations , though it is certainly not provided coherently , or at the resonant frequency . however spiral wave disturbances , occurring with a period of perhaps 120 myr , put at least 3@xmath3010@xmath52 ergs per atom into compression , in a column density of order 3@xmath3010@xmath31 @xmath32per side . the corresponding average power density per side is roughly : @xmath53 which is four times the nominal power of our waves in equation [ eqn6 ] and sixty times that actually transmitted into our most extreme halo . therefore the energy requirement for the resonant production of a hot , ionized halo above 5 kpc is quite small compared to the energy potentially available . the essentially resonant spiral arm period chosen above corresponds to a four arm spiral at the solar circle , with a differential speed between gas and pattern of 100 km s@xmath1 . we found several potentially important properties of the interstellar medium in our experiments . probably the most intriguing of these is that segregation of infalling material into two dominant components occurs naturally after any large expansive event in the ism as a whole . the material closest to the midplane returns first and stops . after roughly 50 myr a substantial low velocity column density has collected , with outer material continuing to fall in upon it . with a sufficiently large velocity differential , the two components are separated by a nearly stationary shock . the large structure known as the iv arch , covering about half of the northern sky , could well be the consequence of such a process . the concentration of material in velocity space ( rather than in real space ) is a natural occurrence in the ism , and can easily lead to apparent structures in the diffuse gas which mimic low - density clouds . experiments in which we added small velocity perturbations ( via oscillation of the galactic midplane ) to cases in which a velocity segregated infall was already present ( from a general expansion , as above ) provided velocity spectra closely resembling those of hd 93521 , even though there were no density concentrations , i.e. clouds , whatsoever . we suggest that perhaps only the one colder component in the spectrum of that star is actually a localized density feature . we found the periods of the four lowest normal modes for our model ism as functions of the height of the outer boundary , both through fourier analysis of the oscillations in the aftermath of a major perturbation , and via linear analysis , with the two methods giving remarkably similar results . our hydrodynamics experiments further revealed two curious tendencies : first that the oscillating structure appears to concentrate energy into two modes , which two depending on the nature of the initial perturbation ; and second that the relative amplitudes of the two modes are comparable at the outer boundary , which therefore shows substantial amplitude modulation in the beat pattern . we also explored a timescale that we refer to as the bounce time . starting with a large compression of material near the midplane , the bounce time at a given height is the length of time for the material to return to its next minimum height . for material in the general vicinity of the plane , the bounce time tended to lie between the half period of the lowest frequency breathing mode and the half period of the first overtone , depending on the magnitude of the initial compression , and therefore perhaps on the relative amplitudes of the excited modes . in general , the bounce time for material near the plane lay in the range 40 to 60 myr , while material beyond 1.5 kpc took substantially longer to return . it is this characteristic difference that leads to the velocity segregation discussed above . it is our impression that the approximate equality between the bounce time and the half period of the most strongly excited breathing mode derives from the fact that the amplitude is very nonlinear in the initial and secondary compressions , with the result that very little time is spent in the compressed state . most of the bounce time is spent in the expanded state , which lasts roughly half the period of the oscillation itself . for atmospheric boundary heights , @xmath54 , in the range 2 to 4 kpc , we found the periods of the fundamental breathing mode to be between 100 and 120 myr , and bounce times to be about half those values . following martos & cox ( 1998 ) , we should then expect to find dense structures within or along spiral arms , separated in space by distances proportional to relative velocity and timescales of 50 to 60 myr , as the disturbed structure bounces . with a relative velocity within the arm of 20 km s@xmath1 , the spacing would be roughly 1 kpc . with an average relative speed between the pattern and rotation of 100 km s@xmath1 , the scale would be 5 - 6 kpc . if the first overtone in the breathing mode is excited , the half periods drop to the range of 30 to 40 myr , and the characteristic feature spacings for the assumed velocities reduce to 0.6 - 0.8 kpc and 3 - 4 kpc ( for relative speeds of 20 km s@xmath1 and 100 km s@xmath1 , respectively ) . for comparison , the circumference at the solar circle is roughly 50 kpc , and the timescale for material to circumnavigate it , relative to the pattern , is of order 500 myr . with a bounce timescale of roughly one tenth that , one might expect to encounter roughly ten unusually dense regions around the circle , some spaced closer than 1 kpc ( feathers ? ) with others several kpc apart along the circle ( arms and spurs ? ) . if the full oscillations of the normal modes ( rather than strong bounces ) are responsible for the structuring , the feature separations will be roughly twice these , their number on the solar circle reduced to roughly four or five , possibly reinforcing a four arm spiral . whether these expectations are borne out in some form by 3-dimensional modeling of the arm pattern remains to be seen . some alteration is expected as waves which are not entirely vertical will have somewhat different periods and finite wavelengths parallel to the plane . the basic idea , however , is that the spiral arm `` shock '' is fragmented , feeding energy preferentially into over dense regions resulting from waves provoked by other spiral arm fragments upstream ( and slightly off to the side ) . that energy input drives further oscillations to provide dense regions further downstream yet , where the next shock fragments occur . as these shock fragments are not oriented parallel to the arm locus , significantly stronger interactions can occur than if their strength were determined only by the component of the entry velocity normal to that more oblique locus . an individual large disturbance in the disk leads eventually to a shock wave propagating up into the low - density outer parts of the atmosphere . this shock strengthens as it proceeds , finally reaching such low density that it is no longer radiative on the timescale for propagating through the rest of the atmosphere . a hot low - density layer is produced , on top of the cool disk material . with no further disturbance , this layer slowly cools and rejoins the disk , but the timescale is extraordinarily long . if the time between such disturbances is longer than this cooling timescale , then the hot halo may become a more or less permanent feature of the atmosphere . similarly , when low amplitude waves are generated continuously within the disk and propagate into its outer reaches , they too can heat an outer layer to high temperatures , eventually reaching a new equilibrium with the unreflected portion of the wave power . we found that there is a strong resonance at which the temperature and size of the outer layer are greatly enhanced , leading to temperatures in excess of 10@xmath40 k and a thickness of many kiloparsecs . in our experiments , the period of the resonance was that of the fundamental breathing mode for the slightly altered disk , essentially the same oscillation invoked to provide structuring in the spiral arm pattern . it appears that the two situations are linked . if the spiral waves are sufficiently large in amplitude to cause substantial vertical motion in rebounding from their compressions , vertical bounces and subsequent oscillations will be produced that can both structure the details of the arms themselves and lead to substantial heating of material at the outer disk . the latter effect produces a large and very hot gas layer , potentially contributing to the populations of high stage ions ( o vi in particular ) and to the soft x - ray background . the details of the hot halo depend on the flux of driving mechanical energy , but quite an impressive halo was formed in our simulation with a very small power , on the order of 4@xmath3010@xmath55 erg @xmath32 s@xmath1 , which is less that 2% of that available from spiral wave disturbances . our experiments on hot halo formation were all performed with a nominal disk boundary height of 5 kpc . the resulting haloes have very low densities , with thermal pressures more than two orders of magnitude below the total midplane pressure . we estimate that their o vi column densities would be of order 10@xmath56 @xmath32 . with a lower boundary height , the halo pressure and o vi column densities could be significantly higher , though much more power would be required to sustain them . surely the details of these results will change when the studies are repeated in two and three dimensions , when mhd effects are included explicitly rather than just through an effective pressure term , when self gravity and spiral wave behavior are included , and perhaps again when the driving terms associated with correlated star formation are incorporated . but the essential effects are likely to persist . what goes up must come down . material nearer the plane returns sooner , hence producing the `` falling sky . '' velocities associated with small disturbances in the intercloud medium will produce spectral features , appearing to be clouds which in reality do not exist . spiral arm shocks will provoke vertical motions with periods that will lead to structure within the spiral waves themselves , and at periods that will tend to promote the formation of a hot halo overlying the cool disk . it is conceivable that the spiral waves might also , on occasion , produce such large upflows that the falling sky is an associated phenomenon . we refer the reader once again to martos & cox ( 1998 ) for further discussion . * * consider a small parcel initially with position @xmath12 , thickness @xmath57 , and density @xmath58 . after a small disturbance , it has a displacement @xmath59 , new position @xmath60 and new thickness @xmath61 where we have introduced the notation @xmath62 . from continuity , we have @xmath63 , so : @xmath64 the ( comoving ) velocity of the parcel is : @xmath65 with @xmath66 . its gravitational acceleration at its perturbed location is : @xmath67 where @xmath68 . assuming that the gas remains nearly isothermal ( as in our hydrodynamic simulations ) , the new pressure of the parcel is : @xmath69 where @xmath70 is the initial thermal pressure , @xmath71 is the initial magnetic pressure , @xmath72 is the total initial pressure , @xmath73 , and the signal speed ( i.e. magnetosonic velocity ) is given by : @xmath74 so the equilibrium signal speed , @xmath75 , is given by : @xmath76 the pressure gradient is then : @xmath77 substituting these results into euler s equation and then canceling the equilibrium terms , we get : @xmath78 equation [ eqn26 ] is the general equation of motion for linear waves and oscillations . the input information required includes @xmath79 , @xmath80 , @xmath81 , @xmath82 , and the location of and pressure at the outer boundary . from those @xmath83 , @xmath84 , @xmath85 , and then @xmath86 can be found as per the prescription in the text . in the next two sections , we consider first high frequency waves to learn their amplitude dependence on @xmath0 , and then normal modes . assuming a harmonic disturbance of the form : @xmath87 , \label{eqn27}\ ] ] we can force most of the @xmath0 dependence into the wave number @xmath88 , and solve for the gradual change in the amplitude @xmath89 . equation [ eqn26 ] becomes : @xmath90 . \label{eqn29}\ ] ] in order to satisfy this equation for all time and space , the real and imaginary parts must each be satisfied . from the imaginary part , we have : @xmath91 the real part yields : @xmath92 using the results from the imaginary part , we can write this as : @xmath93 \right\ } + c_o^2\frac { a^{\prime \prime } } a - c_o^2k^2 . \label{eqn32}\ ] ] by high frequency waves , we mean those whose wavelength ( @xmath94 ) is sufficiently small that @xmath95 where @xmath96 is the functional scale height of the atmosphere . with this assumption the middle two terms of equation [ eqn32 ] , which are second order , can be neglected , giving : @xmath97 for high frequencies , we can also neglect the @xmath98 @xmath99 term , giving us : @xmath100 thus equation [ eqn30 ] becomes : @xmath101 since the frequency is constant , we have finally : @xmath102 equation [ eqn37 ] is the standard result introduced in section [ secsmallosc ] for the growth in amplitude of short wavelength waves . the amplitude is proportional to @xmath103 , where @xmath104 and @xmath75 are the local unperturbed density and signal speed as functions of @xmath0 . equation [ eqn26 ] can be simplified for use in searching for the normal modes of the disk by assuming that @xmath105 is a product of two functions : @xmath106 then equation [ eqn26 ] becomes : @xmath107 + \left [ \left ( \frac{f^{\prime } } f\right ) ^2+\left ( \frac { f^{\prime } } f\right ) ^{\prime } \right ] \epsilon + 2\left ( \frac{f^{\prime } } f\right ) \epsilon ^{\prime } + \epsilon ^{\prime \prime } -\frac{g^{\prime } } { c_o^2}\epsilon . \label{eqn42}\ ] ] we can make the @xmath108 terms cancel if we define : @xmath109 so equation [ eqn42 ] becomes : @xmath110 further defining @xmath111 such that : @xmath112 and invoking @xmath113 in our search for normal modes , we have @xmath114 \epsilon . \label{eqn48}\ ] ] this is a second order differential equation with no first derivative term , similar to the time independent schrdinger equation . it can easily be solved numerically , and the eigenvalues @xmath115 searched for such that the boundary conditions are satisfied . the outer boundary condition , however , is most easily described in terms of the original variable @xmath116 , to which we return next . ( for an isothermal atmosphere with constant gravity , the quantity in brackets above is constant and the solutions sinusoids , providing a nice analytic test of numerical methods . similarly , an infinite isothermal atmosphere with linear gravity has polynomial solutions like the quantum mechanical harmonic oscillator , the eigenvalues of which provide the mode frequencies in the limit of large - scale height , another useful test . ) identification of the appropriate outer boundary condition is difficult , but we have designed the last parcel of the hydrocode to have only thermal pressure within it , and to have effectively zero pressure on its outer boundary . we therefore suspected that the appropriate boundary condition for the linear analysis might be that of a constant zero pressure . this would then require having zero change in the pressure , a pressure node , at the outer edge . the comoving pressure change is due to a change in density , such that @xmath122 , or with equation [ eqn15 ] : @xmath123 hence the outer boundary would be an antinode in @xmath105 , with no variation in pressure or density , and an extremum in the velocity . referring to figures [ fig4 ] and [ fig8 ] , it is easy to see why we were discouraged with this assumption . the pressures and densities do vary considerably in the outer parcels of the code . sometimes , but not frequently , the velocity has an extremum there . and yet , continuing with the assumption that @xmath105 should have an antinode at the boundary gave the same set of frequency eigenvalues found hydrodynamically , while no other assumption came close . the boundary condition at the midplane depends on whether the velocity is an even or odd function of @xmath0 . in terms of our other functions , the velocity is : @xmath124 because @xmath125 is an even function of @xmath0 , even solutions in @xmath126 require : @xmath127 these are the `` sloshing '' modes , symmetric in @xmath126 and @xmath105 and antisymmetric in @xmath45 . our procedure for solving for the normal modes involved , for a chosen value of @xmath129 , integrating equation [ eqn48 ] upward from @xmath130 where the boundary condition was taken to be either @xmath131 and @xmath132 for the sloshing modes , or @xmath133 and @xmath134for the breathing modes , with 1 being chosen as an arbitrary nonzero value . we also evaluated equation [ eqn55 ] for @xmath135 , and then scanned in @xmath129 until the outer boundary condition of equation [ eqn56 ] was satisfied with either zero or one node between the midplane and the outer boundary , finding in this way the fundamental and first overtones for that symmetry . as reported in the text , our results for the allowed periods were nearly identical to those found from the hydrodynamics . bienaym , o. , robin , a. c. , and crz , m. 1987 , , 180 , 94 boulares , a. , and cox , d. p. 1990 , , 365 , 544 bowers , r. l. , and wilson , j. r. 1991 , numerical modeling in applied physics and astrophysics ( boston : jones and bartlett ) cox , d. p. , and anderson , p. r. 1982 , , 253 , 268 cui , w. , and cox , d. p. 1992 , , 401 , 206 danly , l. , lockman , f. j. , meade , m. r. , and savage , b. d. 1992 , , 81 , 125 martos , m. a. and cox , d. p. 1998 , , 509 , 703 press , w. h. ( et al . ) 1989 , numerical recipes : the art of scientific computing ( new york : cambridge university press ) raymond , j. c. , cox , d. p. , and smith , b. w. 1976 , , 204 , 290 richtmyer , r. d. , and morton , k. w. 1967 , difference methods for initial - value problems ( new york : interscience publishers ) savage , b.d . and massa , d. 1987 , apj , 314 , 380 sciama , d.w . 1972 , nature , 240 , 546 slavin , j. d. , and cox , d. p. 1992 , , 392 , 131 spitzer , l. , and fitzpatrick , e. l. 1993 , , 409 , 299 stark , a. a. , gammie , c. f. , wilson , r. w. , bally , j. , linke , r. a. , heiles , c. , and hurwitz , m. 1992 , , 79 , 77 walters , m. a. 1993 , b.s . thesis , university of wisconsin - madison

for legendrian links in the-jet space of we show that the-graded ruling polynomial may be recovered from the kauffman skein module . for such links a generalization of the notion of normal ruling is introduced . we show that the existence of such a generalized normal ruling is equivalent to sharpness of the kauffman polynomial estimate for the thurston - bennequin number as well as to the existence of an ungraded augmentation of the chekanov - eliashberg dga . parallel results involving the homfly - pt polynomial and-graded generalized normal rulings are established .

1109.1319

in @xmath3 interesting connections exist between the @xmath2-variable knot polynomials and invariants of legendrian knots . with respect to the standard contact structure on @xmath3 , fuchs and tabachnikov @xcite showed that an upper bound for the thurston - bennequin number arises from the kauffman and homfly - pt knot polynomials . furthermore , when this estimate is sharp some non - classical invariants exhibit nice properties . specifically , combining results from @xcite , @xcite , @xcite , and @xcite we have : [ thm : review ] for a legendrian link @xmath4 the following three statements are all equivalent : 1 . the estimate @xmath5 ( resp . @xmath6 ) is sharp where @xmath7 $ ] denote the kauffman and homfly - pt polynomials . 2 . a front diagram for @xmath8 has a @xmath0-graded ( resp . @xmath2-graded ) normal ruling . 3 . the chekanov - eliashberg dga of @xmath8 has a @xmath0-graded ( resp . @xmath2-graded ) augmentation . in addition , the conditions of theorem [ thm : review ] are necessary in order for @xmath8 to have a linear at infinity generating family . in this article , we establish analogous results for legendrian knots in the @xmath0-jet space of the circle , @xmath9 . the manifold @xmath9 is topologically an open solid torus and carries a standard contact structure . legendrian knots in @xmath9 have attracted a fair amount of attention in the literature ; see for instance @xcite , @xcite , @xcite . the @xmath0-jet space setting comes with convenient projections from which legendrian knots may be presented via front or lagrangian diagrams and legendrian isotopy may be described in a combinatorial manner . in addition , @xmath0-jet spaces provide a natural setting for the use of generating families . a convenient formal way to define a normal ruling , @xmath10 , of @xmath8 is as a family of fixed point free involutions of the strands of the front diagram of @xmath8 subject to many restrictions . at least locally , this may be viewed as a decomposition of the front diagram into pairs of paths . chekanov and pushkar introduced normal rulings in @xcite albeit with different terminology as well as related legendrian isotopy invariants which have become known as ruling polynomials . in connection with augmentations , fuchs independently defined normal rulings of knots in @xmath3 and , in the case of the kauffman polynomial , already conjectured the equivalence of ( 1 ) and ( 2 ) in @xcite . this conjecture was verified in @xcite where it was shown that in fact the @xmath0-graded and @xmath2-graded ruling polynomials appear as coefficients of the kauffman and homfly - pt polynomials respectively . relationships between the kauffman / homfly - pt invariants and legendrians knots in @xmath9 have already begun to be studied , and several factors make the situation more interesting . for instance , the homfly - pt polynomial , @xmath11 , of a solid torus link , @xmath8 , belongs to a polynomial algebra over @xmath12 $ ] with a countably infinite number of generators @xmath13 ; the kauffman polynomial has a similar form . chmutov and goryunov @xcite proved thurston - bennequin number estimates analogous to those appearing in ( 1 ) of theorem [ thm : review ] using these many variable kauffman and homfly - pt polynomials . in the case of the homfly - pt polynomial , it was shown in @xcite that the @xmath2-graded ruling polynomial can be recovered from the homfly - pt polynomial , but this requires first specializing via an @xmath14-module homomorphism @xmath15 \rightarrow r$ ] . in the present work we develop analogous results involving the @xmath0-graded ruling polynomial and the kauffman skein module . ( see theorems [ thm : front ] and [ thm : spec ] . ) the need to specialize the kauffman and homfly - pt invariants in order to recover the ruling polynomials has an interesting consequence . there are many solid torus links where the kauffman or homfly - pt polynomial estimate is sharp , yet the corresponding ruling polynomial vanishes . as a result , for legendrians in @xmath9 some adjustment is required to statement ( 2 ) of theorem [ thm : review ] . for this purpose , we introduce a quite natural notion of _ generalized normal ruling _ where the fixed point free condition is relaxed . our main result is the following analog of theorem [ thm : review ] : [ the : main ] let @xmath16 be a legendrian link . 1 . then , the estimate @xmath5 ( resp . @xmath6 ) is sharp if and only if @xmath8 has a @xmath0-graded ( resp . @xmath2-graded ) generalized normal ruling . 2 . suppose @xmath8 has been assigned a @xmath17-valued maslov potential . then , the chekanov - eliashberg dga of @xmath8 has a @xmath18-graded augmentation if and only if a front diagram for @xmath8 admits a @xmath18-graded generalized normal ruling . aside from allowing the more general @xmath18-graded condition in ( 2 ) , it is natural to organize the three statements into these two equivalences . even in @xmath3 , the authors do not know of any proof of an implication between the statements about the knot polynomial estimates and existence of augmentations which is able to avoid using normal rulings . there are settings , for instance certain contact lens spaces , where legendrian contact homology @xcite and homfly - pt polynomial estimates ( @xcite , @xcite ) for @xmath19 have been established while an appropriate notion of normal ruling has yet to be formulated . for this reason , establishing a more direct link between bennequin type inequalities and augmentations could prove interesting . the article is arranged as follows : in section 2 , we provide the necessary background about normal rulings and the kauffman and homfly - pt invariants and also introduce generalized normal rulings . section 3 runs parallel to the results on the homfly - pt skein module and @xmath2-graded rulings from @xcite . we show how to recover the @xmath0-graded ruling polynomial from an appropriate specialization of the kauffman skein module . a natural basis for the kauffman skein module is indexed by partitions , and for this basis we provide an explicit formula for the specialization . in section 4 we prove ( 1 ) of theorem [ the : main ] by combining the results of section 3 ( and @xcite for the homfly - pt case ) with a linear independence argument . the final section of the article deals with part ( 2 ) of theorem [ the : main ] . for the forward implication we base all of our arguments on the linear algebraic results of barannikov @xcite from which the reason behind the normality conditions , with or without fixed points , becomes clear . this work was initiated through the pruv program at duke university . we thank david kraines for supervising the program and encouraging our participation . also , we thank lenny ng for his interest in the project . a portion of the writing was carried out while the second author was a visitor at the max planck institute for mathematics in bonn , and it is a pleasure to acknowledge mpim for their hospitality . the first author received support from nsf career grant dms-0846346 . we assume familiarity with basic concepts about legendrian knots such as _ front projections _ , _ legendrian reidemeister moves _ , _ thurston bennequin number _ , and _ rotation number _ at least for knots in @xmath3 . see , for instance , @xcite , and also note that @xcite contains an alternate discussion of the case of legendrian knots in @xmath9 . we view the @xmath0-jet space of the circle , @xmath9 , as @xmath20 equipped with the contact structure @xmath21 where @xmath22 is a circle - valued coordinate . we occasionally refer to a ( legendrian ) link @xmath23 as a _ ( legendrian ) solid torus link_. the front projection of a legendrian solid torus link consists of some number of closed curves in the @xmath24-annulus which we view as @xmath25\times { \mathbb{r}}$ ] with the identification @xmath26 . generically , front projections are immersed and embedded except at semi - cubical cusps and transverse double points , and two such projections represent legendrian isotopic links if and only if they are related by a sequence of legendrian reidemeister moves . we make the convention of extending the thurston - bennequin number to homologically non - trivial links by using the front projection formula @xmath27 where @xmath28 denotes the writhe of @xmath8 ( a signed sum of crossings ) and @xmath29 is half the number of cusps of @xmath8 . similarly , for a legendrian knot @xmath23 we will define the rotation number as @xmath30 where @xmath31 denotes the number of downward oriented cusps and @xmath32 the number of upward oriented cusps . given two annular front diagrams , @xmath33 and @xmath8 , we define the product , @xmath34 , by stacking @xmath33 above @xmath8 . in contrast to the case of smooth knot diagrams , this product is non - commutative as the legendrian isotopy types of @xmath34 and @xmath35 will not agree in general ; see @xcite and @xcite . in this article the _ basic fronts _ , @xmath36 , will play an important role . given @xmath37 , @xmath36 is the front diagram that winds @xmath38 times around the annulus with @xmath39 crossings and no cusps ; see figure [ fig : bfront ] . when it is necessary to pay attention to orientations , for @xmath40 , we will use @xmath36 ( resp . @xmath41 ) for the basic front oriented in the direction of the positive ( resp . negative ) @xmath22-axis . . ] given an @xmath42-tuple of positive integers @xmath43 we write @xmath44 for the product of basic fronts and @xmath45 for the product with all orientations reversed . we now describe a generalization from @xcite of the kauffman polynomial to smooth links ( not necessarily legendrian ) in the solid torus . in practice , this invariant is computed by reducing a link diagram to products of basic fronts via skein relations . whenever appropriate , we will view a front diagram of a legendrian link as a smooth link diagram by placing the strand with lesser slope on top at crossings and smoothing cusps . let @xmath46 denote the set of regular isotopy classes of unoriented link diagrams in the annulus . that is , we consider link diagrams up to the equivalence generated by type ii and type iii reidemeister moves . using the coefficient ring @xmath12 $ ] we define the kauffman skein module @xmath47 as the quotient of the free @xmath14-module , @xmath48 , by the sub - module generated by the kauffman skein relations , @xmath49 @xmath50 @xmath51 the product of diagrams gives a well defined product on @xmath47 which is commutative as we now consider diagrams of smooth links rather than front diagrams of legendrian links . turaev showed in @xcite that @xmath47 is a polynomial @xmath14-algebra in the basic fronts . thus , to a link diagram @xmath8 we may associate a polynomial @xmath52 according to @xmath53 \\ [ l ] & \leftrightarrow & d_l \end{array}.\ ] ] the _ kauffman polynomial _ of an oriented link @xmath23 is then defined by the normalization @xmath54 where @xmath28 denotes the writhe of @xmath8 . chmutov and goryunov @xcite proved that for any legendrian link @xmath23 , @xmath55 while @xcite uses a different projection annulus for computing @xmath56 , a proof of ( [ eq : cgest ] ) matching our conventions for @xmath56 may be given precisely as in the case of the homfly - pt polynomial addressed in section 6.2 of @xcite . [ rem : basis ] ( i ) recall that a possibly empty sequence of positive integers , @xmath43 , is called a _ partition _ if @xmath57 . the integers @xmath58 are called the _ parts _ of @xmath59 and we sometimes use the notation @xmath60 to indicate that @xmath59 is the partition with @xmath61 parts equal to @xmath62 , @xmath63 . as it will be useful later , we note that the collection of products @xmath64 with @xmath59 a partition forms an @xmath14-module basis for @xmath47 . \(ii ) the homfly - pt skein module is defined in a similar manner using oriented link diagrams and an appropriate modification of the skein relations ( [ eq : sr1])-([eq : sr3 ] ) ( see , for instance @xcite ) . the result is a polynomial algebra generated by the oriented basic fronts @xcite . for a given oriented link @xmath23 we denote the corresponding homfly - pt polynomial as @xmath65.\ ] ] let @xmath66 be the front projection of a legendrian link in the solid torus satisfying the additional assumption that all crossings and cusps have distinct @xmath22-coordinates none of which equals @xmath67 . a normal ruling can be viewed locally as a decomposition of @xmath8 into pairs of paths . we make some notational preparation before giving the formal definition . denote by @xmath68 those @xmath22-coordinates which coincide with a crossing or cusp of @xmath8 . we can write , @xmath69 with each @xmath70 an open interval ( or all of @xmath1 if @xmath71 ) . making the convention that @xmath72 , we assume that the @xmath70 are ordered so that @xmath73 appears immediately to the left of @xmath74 and @xmath75 contains @xmath76 . on subsets of the form @xmath77 the front projection @xmath8 consists of some number of non - intersecting components which project homeomorphically onto @xmath70 . we refer to these components as the _ strands _ of @xmath8 above @xmath70 , and we number them from _ top to bottom _ as @xmath78 . finally , for each @xmath79 we choose a point @xmath80 . [ def : nr ] a _ normal ruling _ of the front diagram @xmath8 is a sequence @xmath81 of involutions @xmath82 satisfying the following restrictions : 1 . each @xmath83 is fixed point free . 2 . if the strands above @xmath74 labeled @xmath84 and @xmath85 meet at a left cusp in the interval @xmath86 , then @xmath87 and when @xmath88 , @xmath89 3 . a condition symmetric to ( 2 ) at right cusps . 4 . if strands above @xmath70 labeled @xmath84 and @xmath85 meet at a crossing on the interval @xmath86 , then @xmath90 and either 1 . @xmath91 where @xmath92 denotes the transposition , or 2 . + in the second case we refer to the crossing as a _ switch _ of @xmath10 . finally , we have a requirement at switches that is known as the _ normality condition_. 5 . if there is a switch on the interval @xmath86 then one of the following three orderings holds : @xmath94 @xmath95 [ rem : altnr ] this definition is a slight variation on those found elsewhere in the literature . letting @xmath96 denote the projection , chekanov and pushkar defined a normal ruling as a continuous , fixed point free involution of @xmath97 which preserves the @xmath22-coordinate and is subject to some requirements for continuous extension near crossings or cusps as well as a normality condition at switches . such an involution is recovered from our definition by viewing the set @xmath98 that @xmath83 permutes as the set of strands above @xmath70 . from this perspective , the fixed point free condition causes the @xmath83 to divide the strands above @xmath70 into pairs , and in our figures we will present normal rulings by indicating this pairing . beginning at @xmath76 and working to the right , one may cover the front diagram with pairs of continuous paths with monotonically increasing @xmath22-coordinates , so that a given pair of paths corresponds to strands paired by the involutions . if a path proceeds all the way around the annulus , then it will not necessarily end up where it started . however , the division of the front diagram into pairs of points at @xmath76 and @xmath99 should match up . paired paths are only allowed to meet at common cusp endpoints . in particular , at any crossing the two paths of the ruling that meet should belong to different pairs and , for values of @xmath22 near the crossing , each will have a `` companion path '' located somewhere above or below the crossing . the two paths can either follow the link diagram and cross each other ( this corresponds to ( 4 ) ( a ) above ) or they may switch strands by each turning a corner at the crossing . the normality condition provides a restriction on the location of the companion paths near a switch ; out of six possible configurations for the switching strands and their companion strands only three are allowed . see figure [ fig : normc ] for the normality condition and the right half of figure [ fig : ex ] for an example of a normal ruling . further grading restrictions may be placed on a normal ruling after the introduction of a maslov potential for @xmath8 . let @xmath18 be a divisor of @xmath100 for each component @xmath101 of a legendrian link @xmath8 . a @xmath17-valued _ maslov potential _ , @xmath102 , for @xmath8 is a function from @xmath8 to @xmath17 which is constant except at cusp points where it increases by @xmath0 when moving from the lower strand to the upper strand . note that a chosen orientation provides @xmath8 with a @xmath103-valued maslov potential by following the convention that strands oriented to the right ( resp . left ) are assigned the value @xmath67 ( resp . @xmath0 ) mod @xmath2 . we say that a normal ruling @xmath10 is _ @xmath18-graded _ with respect to a @xmath17-valued maslov potential @xmath102 if whenever two strands @xmath104 and @xmath105 of @xmath8 are paired by one of the @xmath83 with @xmath104 above @xmath105 we have @xmath106 . suppose @xmath102 is a @xmath17-valued maslov potential for a legendrian link @xmath8 . the @xmath18-graded ruling polynomial of @xmath8 with respect to @xmath102 is given by @xmath107 where the sum is over all normal rulings of @xmath8 which are @xmath18-graded with respect to @xmath102 and @xmath108 the ruling polynomial does not depend on the choice of maslov potential when @xmath109 ; @xmath110 and @xmath8 is oriented ; or @xmath8 is connected . in any of these cases we denote the ruling polynomial simply as @xmath111 . the ruling polynomials are legendrian isotopy invariants @xcite . in the following definition the requirements from definition [ def : nr ] are relaxed in a manner which is appropriate for theorem [ the : main ] to hold . [ def : gnr ] a _ generalized normal ruling _ consists of a sequence of involutions @xmath81 as in definition [ def : nr ] subject to the following modifications : 1 . we remove the requirement that the @xmath83 be fixed point free . if a crossing occurs in the interval @xmath112 between the @xmath84 and @xmath85 strands above @xmath73 with exactly one of these two strands a fixed point of @xmath83 , then we decide if the crossing is a switch precisely as in ( 4 ) of definition [ def : nr ] . if the crossing is indeed a switch then we require the additional normality condition that either @xmath113 ( see figure [ fig : gennormc ] . ) . ] [ rem : gnr ] ( i ) if a crossing involving the @xmath84 and @xmath85 strands occurs on @xmath86 with both of the crossing strands fixed by the ruling , i.e. @xmath114 and @xmath115 , then @xmath116 . consequently , we will not consider such crossings to be switches . \(ii ) in the presence of an appropriate maslov potential , we can consider @xmath18-graded generalized normal rulings precisely as in section [ sec : mp ] . \(iii ) the number of generalized normal rulings of a legendrian link is not invariant under legendrian isotopy . however , in view of lemma [ lem : gnriff ] below , the polynomials @xmath117 serve as some form of substitute for a `` generalized ruling polynomial '' . for establishing ( 1 ) of theorem [ the : main ] we will use the following equivalent characterization of front diagrams that admit generalized rulings . [ lem : gnriff ] a front diagram @xmath8 has a @xmath0-graded ( resp . @xmath2-graded ) generalized normal ruling if and only if there exists partitions @xmath59 and @xmath102 so that @xmath118 ( resp . @xmath119 ) . for simplicity , we treat the @xmath0-graded case first . if @xmath118 , then @xmath120 has a normal ruling , @xmath10 . this produces a generalized normal ruling of @xmath8 by restricting @xmath10 to @xmath8 and treating any strands of @xmath8 which are paired with @xmath64 as fixed point strands . the normality condition from definition [ def : gnr ] follows from that of definition [ def : nr ] . now suppose that @xmath8 has a generalized normal ruling . if one of the @xmath83 has a fixed point strand , then we can continuously follow the fixed point strand around the diagram turning corners only at switches . the result is a portion of the front diagram , @xmath121 , without cusps that we suppose winds @xmath58 times around the annulus . there may be several fixed point components of this type . we may assume the @xmath58 are ordered so that they form a partition , @xmath59 . the product @xmath120 has a normal ruling where each @xmath121 is paired with the component @xmath122 of @xmath59 . such a ruling is completely determined once we specify the pairing between @xmath121 and @xmath122 at a single point of @xmath121 . now , the normality condition of definition [ def : nr ] follows from that of definition [ def : gnr ] , and the ordering of the factors of @xmath64 is not important here since we do not have switches between any of the @xmath121 ( remark [ rem : gnr ] ) . see figure [ fig : ex ] . for the @xmath2-graded case , observe that in a @xmath2-graded ruling the orientation of strands meeting at a switch must agree . therefore , the @xmath121 each have a consistent orientation , and we choose an orientation on the component @xmath122 accordingly . fixed point strands producing a normal ruling of @xmath123 with @xmath124.,title="fig : " ] fixed point strands producing a normal ruling of @xmath123 with @xmath124.,title="fig : " ] an analysis of how to compute @xmath2-graded ruling polynomials of legendrian solid torus links from the homfly - pt polynomial is done in @xcite . in this section , we will perform a similar analysis of the @xmath0-graded case . we will derive formulas for the @xmath0-graded ruling polynomial of @xmath64 , and then relate the general case to a coefficient of an appropriate specialization of the kauffman polynomial . given a front diagram @xmath8 with normal ruling @xmath10 we define the _ decomposition _ of @xmath8 with respect to @xmath10 as the legendrian link , @xmath125 , obtained by resolving the switches of @xmath8 into parallel horizontal strands as @xmath126 the involutions of the strands of @xmath8 piece together to provide an involution , which we also denote as @xmath10 , now defined on all of @xmath125 . the involution @xmath10 is continuous where we now view @xmath125 as a subset of @xmath9 rather than just a front diagram , and its only fixed points correspond to the cusps of the front projection of @xmath125 . ( compare with remark [ rem : altnr ] . ) the normal ruling of @xmath8 induces a normal ruling of @xmath125 where none of the crossings are switches . we record some observations about normal rulings of the products @xmath64 . [ lem : k ] suppose @xmath10 is a normal ruling of @xmath127 , 1 . the decomposition , @xmath125 , is also a product of basic fronts . 2 . the involution @xmath10 must take a component of @xmath125 isotopic to @xmath36 to another component isotopic to @xmath36 . 3 . if components @xmath128 and @xmath129 of @xmath125 share a common switch of @xmath8 , with @xmath128 above @xmath129 on the @xmath130-axis , then the vertical ordering of the four components @xmath128 , @xmath129 , @xmath131 , and @xmath132 must be one of : @xmath133 ; \quad [ \rho(c_1 ) , \dots , c_1 , c_2 , \dots , \rho(c_2 ) ] ; \mbox { or}\ ] ] @xmath134.\ ] ] 4 . the restriction of @xmath10 to a pair of components of @xmath125 , @xmath128 and @xmath135 , is completely determined by its value at a single point , @xmath136 . moreover , if @xmath137 then there are precisely @xmath38 choices for @xmath138 , and any one of them extends continuously to all of @xmath128 . two components of @xmath125 of the form @xmath128 and @xmath131 can not correspond to subsets of the same component of @xmath8 . item ( 1 ) is clear ; ( 2 ) follows from continuity of @xmath10 ; and ( 3 ) is a consequence of the normality condition . the first assertion of ( 4 ) follows from continuity of @xmath10 . the second follows since @xmath139 and @xmath140 must have the same @xmath22-coordinate and @xmath129 also consists of @xmath38 strands . that any such choice of @xmath139 extends to all of @xmath128 is easily seen . we prove ( 5 ) by contradiction . suppose @xmath128 and @xmath131 did come from the same component of @xmath8 , and without loss of generality assume @xmath131 is below @xmath128 . they can not meet at a switch as this would violate the normality condition . thus , there is some other component @xmath129 on the other end of the switch below @xmath128 . the only possible position of @xmath132 is then between @xmath129 and @xmath131 . then @xmath129 and @xmath132 also came from the same component of @xmath8 . they can not meet at a switch , so there is some further component @xmath141 immediately below @xmath129 , which is paired with a component @xmath142 between @xmath141 and @xmath132 . we can continue this argument to produce arbitrarily many components of @xmath125 between @xmath128 and @xmath131 . the results in the previous section are sufficient to compute the ruling polynomial for the simplest possible product , @xmath144 ( the ruling polynomial of a single basic front @xmath36 is @xmath67 by ( 5 ) of lemma [ lem : k ] ) . although this agrees with @xmath145 which is computed in lemma 4.1 of @xcite , the form of the answer given here is simplified and the proof is quite different . [ lem : am2 ] the ruling polynomial of @xmath146 is @xmath147 [ fig : bijection-2 ] with 2 switches , divisions of 5 objects into 2 parts with a marked object in each part , and compositions of 7 into 4 positive parts.,title="fig : " ] normal rulings of @xmath148 with @xmath149 switches are in bijection with subdivisions of @xmath38 ordered objects into @xmath85 consecutive parts , with a marked object chosen in each part . the subdivision corresponds to choosing the location of @xmath84 switches within the first @xmath36 factor . specifically , dividing @xmath38 into parts @xmath150 corresponds to choosing @xmath84 switches so that in the decomposition , @xmath125 , the first @xmath36 factor becomes @xmath151 . in @xmath125 , the @xmath122 must be paired with @xmath85 components of the same size from the second @xmath36 factor , by lemma [ lem : k ] ( 2 ) and ( 5 ) . then , lemma [ lem : k ] ( 3 ) determines the order of the components : they must be in the reverse order of the components from the first factor . the total number of switches is @xmath149 . the choice of marked object within a part @xmath58 corresponds to choosing which strand within the @xmath122 component is paired with the top strand of @xmath152 at @xmath76 . these choices may be arbitrary , and they uniquely determine a ruling by lemma [ lem : k ] ( 4 ) . see figure [ fig : bijection ] . to complete the proof observe that subdivisions of this type are in turn in bijection with compositions of @xmath153 into @xmath154 positive parts , @xmath155 : two consecutive parts of size @xmath156 and @xmath157 correspond to a part @xmath158 with the @xmath156-th object marked in a subdivision of @xmath38 . the number of ways to decompose @xmath159 objects into @xmath154 parts of positive size is well - known : it is @xmath160 or @xmath161 . this gives us the sum for the ruling polynomial . this formula will be used in the next section , so we will write @xmath162 for the ruling polynomial @xmath163 , following the convention in @xcite . we will use the formula for @xmath162 to calculate the ruling polynomial of @xmath64 for an arbitrary @xmath59 . given a normal ruling @xmath10 of @xmath127 , define the _ block _ @xmath164 to consist of those components of the decomposition @xmath125 which originated in the @xmath165-th component of @xmath8 , and are paired by @xmath10 with components that originated in the @xmath166-th component of @xmath8 . the size of the block , @xmath167 , is the number of points in @xmath164 with some fixed @xmath22-coordinate , away from crossings . [ lem : block ] given a normal ruling of @xmath168 , the blocks in the @xmath165-th component of @xmath8 consist of vertically consecutive components of @xmath125 , and are themselves vertically ordered as follows ( some blocks may be empty ) : @xmath169 suppose that when we resolve @xmath122 at switches , we get the components @xmath170 , in that vertical order . if , for some @xmath166 , @xmath171 is above @xmath172 , then the normality condition demands that @xmath173 is between @xmath171 and @xmath174 . similarly , if @xmath171 is below @xmath172 , then @xmath175 must be between @xmath176 and @xmath171 . as a result , if @xmath177 and @xmath178 come from the same component of @xmath8 , then @xmath171 for @xmath179 are between @xmath177 and @xmath178 . this implies each block is made up of some number of consecutive components . and due to the normality condition , the ordering of any two consecutive blocks must be either @xmath180 , with @xmath181 , or @xmath182 , with @xmath183 ( with the caveat that some of the blocks may be empty , if @xmath10 does not pair two components of @xmath8 at all ) . putting this together yields the block ordering above . this means that once we pick the sizes of the blocks @xmath184 , the locations of the blocks are determined . to complete the calculation of the ruling polynomial , observe that the choice of a normal ruling of the blocks @xmath164 and @xmath185 , with @xmath186 , is equivalent to the choice of a normal ruling of @xmath144 . [ thm : front ] let @xmath187 denote the ruling polynomial of @xmath144 , with @xmath188 taken to be @xmath189 . then the ruling polynomial of @xmath190 is given by @xmath191 where @xmath192 is the set of all symmetric matrices @xmath193 with nonnegative integer entries such that the row sums @xmath194 and the trace @xmath195 . the choice of a matrix in @xmath196 is equivalent to the choice of block sizes @xmath167 . by lemma [ lem : block ] , this also fixes the locations of the blocks . a normal ruling of @xmath64 is then completely determined by its restriction to pairs of blocks @xmath164 and @xmath185 . if the block size @xmath167 is nonzero , then @xmath197 describes the possible restrictions of the normal rulings to the union @xmath198 . we take the product to combine these normal rulings , but we have to account for the switches between the blocks . if all block sizes are nonzero , then there will be @xmath199 switches in each of the @xmath200 components of @xmath8 , giving us a factor of @xmath201 . any block @xmath164 of size @xmath67 will reduce this number by @xmath0 in component @xmath166 , but the corresponding block @xmath185 will reduce the number of switches by @xmath0 in component @xmath165 ; this gives a factor of @xmath189 which is accounted for by the convention of @xmath202 [ cor : comm ] the 1-graded ruling polynomial is commutative in front diagram products : that is , the ruling polynomials of @xmath203 and @xmath204 are equal . there is an easy bijection between the possibilities for the matrix @xmath192 and the new matrix @xmath205 : we simply exchange the @xmath165-th and @xmath206-th columns and rows ; the summands @xmath207 do not change . in @xmath3 , the 1-graded and 2-graded ruling polynomial of arbitrary legendrian links may be easily recovered from the kauffman and homfly - pt polynomials . the second author shows in @xcite that the @xmath0-graded ( resp . @xmath2-graded ) ruling polynomial of a link @xmath8 is the coefficient of @xmath208 in the kauffman polynomial ( resp . homfly - pt polynomial ) of @xmath8 . in the case of legendrian solid torus links we first need to specialize the extra variables in a non - multiplicative manner . using the notation of section [ sec : kp ] , consider the @xmath14-module homomorphism @xmath209 \rightarrow r$ ] determined by @xmath210 when @xmath59 is a partition . ( compare with remark [ rem : basis ] . ) given a link diagram @xmath8 , we let @xmath211 , and @xmath212 . [ thm : spec ] let @xmath23 be any legendrian solid torus link . then , the @xmath0-graded ruling polynomial @xmath213 is equal to the coefficient of @xmath214 in @xmath215 . this result is analogous to theorem 6.3 of @xcite , where it is shown that we can recover the @xmath2-graded ruling polynomial from such a specialization of the homfly - pt polynomial . the proof , via induction on a certain measure of complexity of a front diagram , carries through in the 1-graded case as well . the base case consists of all products of basic fronts where the result follows from the crucial corollary [ cor : comm ] . next , it is observed that the ruling polynomial and the coefficient of @xmath214 in @xmath216 share common skein relations which are legendrian analogs of equations ( [ eq : sr1])-([eq : sr3 ] ) ( see @xcite or @xcite ) . then , just as in @xcite , the inductive step is completed by an algorithm which uses these skein relations to evaluate the invariants in terms of front diagrams of lesser complexity . consider the legendrian knots @xmath217 and @xmath218 pictured in figure [ fig : ex ] , and suppose orientations are chosen so that all strands are oriented to the right when they pass through the vertical line @xmath76 . the kauffman polynomials are given by @xmath219 \\ & + & a_3 \times \left [ a^{-1}(z+z^3 ) + a^{-2}(-z^2-z^4 ) + a^{-3}(-z - z^3 ) \right ] \\ & + & a_2 a_1 \times \left [ a^{-1}(1+z^2 ) - a^{-2}z^3 - a^{-3}z^2 \right ] , \end{array}\ ] ] and @xmath220 . we have @xmath221 and @xmath222 , so in both cases the estimate ( [ eq : cgest ] ) is sharp . using theorem [ thm : front ] , one has @xmath223 ; @xmath224 ; and @xmath225 . this allows us to compute @xmath226 and theorem [ thm : spec ] gives @xmath227 which can be verified directly . in this section we establish the equivalence ( 1 ) of theorem [ the : main ] which follows from lemma [ lem : gnriff ] together with the following : [ thm : sharp ] let @xmath8 be a legendrian link in the solid torus . then the equality @xmath228 holds if and only if there exists a partition @xmath59 so that @xmath123 has a normal ruling . ( theorem [ thm : sharp ] ) one direction is straightforward . suppose that , for some @xmath59 , @xmath229 has a normal ruling . then the ruling polynomial of @xmath230 is nontrivial , so the coefficient of @xmath208 is nonzero . therefore @xmath231 which , combined with the inequality ( [ eq : cgest ] ) , gives us an equality @xmath232 . however , @xmath233 , since @xmath64 has no cusps . in addition , @xmath234 , so @xmath235 , and we compute @xmath236 now suppose @xmath237 . we will find a @xmath59 such that @xmath123 has a normal ruling . let @xmath238 be the coefficient of @xmath208 in @xmath56 , where the @xmath239 are polynomials in @xmath130 and @xmath240 . this coefficient is nonzero , or else the degree equality would not hold , so @xmath241 for at least one @xmath102 . let @xmath84 be the smallest integer such that at least one @xmath242 has a nonzero coefficient of @xmath243 . by theorem [ thm : spec ] , the ruling polynomial of @xmath123 is @xmath244 we will prove that for some @xmath59 , this polynomial is nonzero ( and therefore a normal ruling exists ) by looking at the @xmath243 coefficient of this polynomial . since @xmath245 is a polynomial in @xmath130 with no terms of @xmath240 or lower degree , the only way to get a @xmath243 coefficient is from the product of @xmath246 $ ] and @xmath247 $ ] for some @xmath102 ( here , @xmath248 $ ] denotes the coefficient of @xmath249 in @xmath250 ) . denote @xmath246 $ ] by @xmath251 , and @xmath252 $ ] ( which is the number of switchless rulings of @xmath253 ) by @xmath254 . the quantity @xmath254 is easy to calculate . without switches , each component of size @xmath84 must simply be paired with another component of size @xmath84 in one of @xmath84 ways . in particular , this is only possible if there is an even number of each component size . define the double factorial @xmath255 . this counts the number of ways to divide @xmath149 objects into pairs . it is clear that @xmath256 we wish to prove that for some @xmath59 , @xmath257 . here , if @xmath258 we will denote by @xmath259 the partition @xmath260 let @xmath261 be the collection of all partitions such that 1 . the parts of the partition are all no larger than @xmath200 , for some @xmath200 . parts of each size occur between @xmath67 and @xmath262 times , for some @xmath38 . we choose the parameters @xmath38 and @xmath200 such that we include all partitions @xmath102 with @xmath263 . let @xmath264 be a @xmath265-dimensional real vector space with basis vectors @xmath266 for @xmath267 . for each @xmath268 , consider the following vectors in @xmath264 : @xmath269 we will show that these vectors also form a basis of @xmath264 , and are therefore linearly independent . from there , observe that @xmath270 if the coefficients @xmath251 on the right are not all @xmath67 , then because the @xmath271 are linearly independent the resulting sum is a nonzero vector of @xmath264 . therefore the coefficients in terms of @xmath272 are not all @xmath67 as well i.e. for some @xmath59 , @xmath257 . so once we have the result of linear independence , we are done . from the formula for @xmath254 , it s easy to calculate that @xmath273 can be written as a product of @xmath274 , over all @xmath84 , where @xmath275 and @xmath276 are the number of parts of size @xmath84 in @xmath59 and @xmath102 respectively . suppose we write @xmath264 as the tensor product @xmath277 , identifying the basis vector @xmath278 on the left with the basis vector @xmath272 on the right , where @xmath60 . here we use a slightly non - standard basis of @xmath279 : it is @xmath67-indexed and consists of @xmath280 , for ease of notation . then if @xmath281 , @xmath282 therefore , rather than prove that the vectors @xmath271 are a basis of @xmath264 , it suffices to prove that the vectors @xmath283 , as @xmath84 goes from @xmath67 to @xmath262 , are a basis of @xmath279 . there are three simplifying observations to be made : 1 . @xmath284 if @xmath285 . therefore @xmath286 is a linear combination only of the odd - indexed @xmath287 if @xmath84 is odd , and only of the even - indexed @xmath287 if @xmath84 is even . furthermore , @xmath288 , so @xmath289 and @xmath290 have the same coefficients , just shifted over by one index . as a result , we will only show the independence of the vectors @xmath291 the result for @xmath292 is similar . 2 . by the first observation , we have @xmath293 this corresponds to starting in the case where @xmath294 , then scaling both the @xmath289 and the @xmath295 by powers of @xmath165 a scaling which does nt change the question of linear independence one way or the other . therefore it suffices to consider the case where @xmath294 . 3 . finally , we can scale each @xmath289 by @xmath296 ( which , too , does nt affect linear independence ) . now we want to look at @xmath297 if we put the coefficients of @xmath298 as columns of a matrix , ( i.e. @xmath166 indexes the rows and @xmath84 indexes the columns ) , we get @xmath299 here , the entries in the @xmath166-th row are given by @xmath300 , which is a degree @xmath166 polynomial function . in particular , @xmath301 can be written as @xmath302 plus lower - order terms ; these lower - order terms are necessarily a linear combination of @xmath303 . therefore , we can use row operations to eliminate the lower - order terms , so that the resulting matrix is : @xmath304 this is a vandermonde matrix whose determinant is @xmath305 , which is nonzero . therefore the vectors @xmath298 ( and @xmath289 ) form a basis of @xmath279 , which completes the proof . a similar approach applies in the case of the homfly - pt polynomial , @xmath11 . the proof of the reverse implication is identical . for the forward implication , we suppose @xmath306 and consider the coefficient of the lowest power @xmath243 that appears in the @xmath214 term of @xmath11 , @xmath307 fix parameters @xmath38 and @xmath200 so that the set @xmath308 contains all @xmath309 such that @xmath310 . using theorem 6.3 in @xcite , for any @xmath311 the coefficient of @xmath243 in the @xmath2-graded ruling polynomial of @xmath312 is given by @xmath313 it suffices to show that the coefficient matrix @xmath314 is non - singular . writing @xmath315 , @xmath316 , @xmath317 , @xmath318 , one has @xmath319 thus , @xmath320 is a tensor product ( kronecker product ) of matrices @xmath321 due to the kronecker delta , each @xmath322 is a direct sum ( block matrix ) of matrices @xmath323 @xmath324 $ ] obtained from keeping rows and columns satisfying @xmath325 . the proof is completed by showing that each @xmath326 is non - singular . we treat the case @xmath327 as @xmath328 is similar . then , @xmath329 and @xmath330 . dividing rows and columns by @xmath331 and @xmath332 respectively leaves @xmath333 where @xmath334 is a polynomial of degree @xmath335 . elementary row operations reduce this to a non - singular vandermonde matrix . in this final section we complete the proof of theorem [ the : main ] by establishing the following : we begin by briefly recalling the aspects of the chekanov - eliashberg dga that are important for the proof . the reader is refered to @xcite for the original , more detailed treatment of this dga in the @xmath9 setting . given a legendrian knot or link , @xmath23 , the _ lagrangian projection _ , @xmath337 , of @xmath8 to the @xmath338-annulus is an immersed curve . the chekanov - eliashberg dga , @xmath339 , is a graded algebra , @xmath340 , with a degree @xmath341 differential , @xmath342 , defined via a generic lagrangian projection of @xmath8 . after a small legendrian isotopy , we may assume @xmath337 to have only finitely many transverse double points which we label as @xmath343 . then , the algebra , @xmath340 , is the free associative @xmath103-algebra with unit generated by the double points @xmath343 . the set of monic non - commutative monomials in the @xmath344 forms a linear basis for @xmath340 . if @xmath8 is connected , then @xmath340 has a @xmath345 grading . in general , the grading depends on a choice of maslov potential for @xmath8 . the differential , @xmath342 , is defined by counting certain immersed discs in the @xmath338-annulus with boundary mapped to the lagrangian projection of @xmath8 . the existence of an augmentation of @xmath339 is a property that is invariant under legendrian isotopy . this follows from the fact that the `` stable tame isomorphism type '' ( see @xcite , @xcite ) of @xmath339 is unchanged by a legendrian isotopy . therefore , in establishing the equivalence of * ( a ) * and * ( b ) * we may work with the chekanov - eliashberg algebra of a legendrian isotopic link @xmath230 . the links @xmath230 which we will consider have a standard form so that @xmath352 may be described in a formulaic manner from the front projection of @xmath230 ( and this front projection is combinatorially the same as that of @xmath8 ) . for this reason we do not present the differential or the grading of the chekanov - eliashberg dga in full generality here . given a legendrian @xmath23 we begin by modifying the front diagram of @xmath8 via ( a slight variation of ) the resolution technique of @xcite . beginning near @xmath76 and working from left to right , we alter the front projection of @xmath8 by an isotopy in the @xmath24-annulus as follows . we arrange so that , except for intervals near @xmath99 or immediately prior to a crossing or right cusp , the slopes of the strands are constant and strictly decreasing as we move from the top to bottom . further , we will assume that all strands usually have non - positive slope . it is no problem to produce these conditions after a left cusp , but with crossings and right cusps the slopes of the two relevant strands will need to be interchanged prior to the crossing or cusp . as the @xmath353-coordinate is given by the slope @xmath354 , this has the effect of producing double points on the lagrangian projection corresponding to ( but located to the left of ) the crossings and right cusps of the front projection of @xmath8 . finally , when we near @xmath99 the strands have become very spread out and moved below their original @xmath130 values at @xmath76 . beginning with the top strand and then proceeding successively to the lowest strands , we return each strand back to its initial position via a steep upward step . this creates several new crossings on the lagrangian projection . see figure [ fig : res ] . next , we add `` splashes '' . recall that we view the @xmath1 factor of @xmath9 as @xmath355 $ ] with @xmath67 and @xmath0 identified . in a similar notation to section 2 , we let @xmath356 be a partitioning of the interval @xmath25 $ ] such that no @xmath357 coincides with the @xmath22-coordinate of a crossing or cusp and each interval @xmath86 contains exactly one crossing or cusp . for each @xmath358 , we add a miniature version of the steps appearing in the part of the resolution procedure near @xmath99 into a small interval centered at @xmath357 . that is , beginning at the top strand and then working downward add a brief but steep ( smooth ) upward step into the diagram . this has a minimal effect on the front projection but alters the lagrangian projection at each @xmath357 by replacing what had been several parallel lines with a collection of crossings similar to those pictured in the right half of figure [ fig : res ] . denote the legendrian link resulting from the combination of these two procedures as @xmath230 . we now give a complete description of the chekanov - eliashberg dga of @xmath230 . for each @xmath359 , let @xmath360 denote the number of intersection points of @xmath8 with the plane @xmath361 . the generators of @xmath362 come from two sources . first , we have generators corresponding to the crossings and right cusps of the front projection of @xmath8 via the resolution procedure . in addition , for each @xmath359 we have two upper triangular matrices worth of generators , @xmath363 and @xmath364 with @xmath365 . these correspond to the double points created by the splashes and the final step of the resolution procedure . if @xmath8 is equipped with a @xmath17-graded maslov potential , @xmath102 , then @xmath362 is @xmath17-graded . we will describe the degree @xmath366 assigned to the generators of @xmath362 ; degrees then extend additively as @xmath367 . in the following , @xmath368 denotes the value of the maslov potential on the @xmath165-th strand at @xmath357 . ( as in section 2 , we label strands from top to bottom . ) the generators of @xmath362 coming from splashes have the following degrees : @xmath369 between the @xmath84 and @xmath85 strands occurring in the interval @xmath86 has @xmath370 , and all right cusps have degree @xmath0 . formulas for the differential @xmath342 , are most efficiently provided by , for each @xmath38 , placing the generators @xmath363 and @xmath364 into strictly upper triangular matrices @xmath371 ( here , @xmath372 if @xmath373 . ) as the @xmath22-coordinate is @xmath1-valued , it is important to make the convention that @xmath374 and @xmath375 . then , applying the differential to each entry , we have the formulas @xmath376 with @xmath377 an identity matrix of the appropriate size . the precise form of @xmath378 depends on the tangle appearing on the interval @xmath86 and is described presently . suppose that @xmath86 contains a crossing , @xmath379 , between the strands labeled @xmath84 and @xmath85 . then , @xmath380 and @xmath381 where @xmath382 ( resp . @xmath383 ) agrees with the identity matrix except for a @xmath384 block @xmath385 $ ] ( resp . $ ] ) along the diagonal in rows @xmath84 and @xmath85 and @xmath387 is the matrix @xmath388 with the entry @xmath389 replaced with @xmath67 . next , we suppose @xmath86 contains a single left cusp between the strands labeled @xmath84 and @xmath85 at @xmath357 . then , @xmath390 where @xmath391 is the @xmath392 identity matrix with columns @xmath84 and @xmath85 removed and @xmath393 is a matrix with a single non - zero entry in the @xmath394 position . finally , we suppose @xmath86 contains a single right cusp , @xmath395 , between the strands labeled @xmath84 and @xmath85 at @xmath396 . then , @xmath397 and the matrix @xmath378 is most easily described entry by entry . let @xmath398 , be given by @xmath399 . the @xmath400 entry of @xmath378 is given by @xmath401 @xmath402 when @xmath403 and @xmath404 otherwise . the technique of adding some variation of splashes to simplify the differential has been used in several places in the literature . the version employed here is the same as that of @xcite where we refer the reader for more details . for an alternate approach , we expect that a dga of the same form would arise from iterating the `` bordered chekanov - eliashberg algebra '' construction introduced in @xcite . suppose that @xmath8 admits a generalized normal ruling , @xmath412 . an augmentation @xmath349 of the algebra @xmath362 is defined as follows : on all right cusps , @xmath395 , @xmath413 ; at crossings @xmath379 , @xmath414 is @xmath0 if @xmath379 is a switch and @xmath67 otherwise ; for all @xmath38 , @xmath415 ; and @xmath416 for all @xmath417 except when a switch occurs between @xmath396 and @xmath357 . assume the switch involves the @xmath84 and @xmath85 strands . if one of the switching strands is also a fixed point strand , then of the generators @xmath363 augment only @xmath418 . else , note that due to the normality condition , near the switch the intervals connecting the switching strands and their companion strands ( remark [ rem : altnr ] ) are either disjoint or nested . assume that the switch occurs between the strands labeled @xmath84 and @xmath85 . if the switch is disjoint , then augment only @xmath418 . if the switch is nested , then augment @xmath418 as well as @xmath419 ( resp . @xmath420 ) if @xmath421 ( resp . @xmath422 ) . it is straight forward to verify from the formulas of the previous section that @xmath349 is an augmentation . if @xmath10 is @xmath18-graded with respect to a maslov potential @xmath102 , then @xmath349 is as well . [ prop : specbasis ] if @xmath423 is an @xmath261-complex , then there exists a triangular change of basis @xmath429 , @xmath430 , and an involution @xmath431 such that @xmath432 moreover , the involution @xmath405 is unique . \(i ) suppose in addition that the basis elements @xmath433 are assigned degrees @xmath434 so that @xmath264 is @xmath17-graded and @xmath342 has degree @xmath341 . then , the change of basis may be assumed to preserve degree . hence , if @xmath435 , then @xmath436 . \(iii ) proposition [ prop : specbasis ] has the following matrix interpretation : there is a unique function , @xmath440 which assigns to every strictly upper - triangular @xmath441 matrix , @xmath442 , with @xmath443 an involution @xmath444 such that there exists an invertible upper - triangular matrix @xmath445 so that @xmath446 . notice that the uniqueness assertion implies that @xmath447 if @xmath448 is non - singular and upper triangular . [ prop : blemma ] suppose that @xmath423 is an @xmath261-complex , and @xmath449 is such that @xmath450 so that @xmath451 with @xmath452 is also an @xmath261-complex . then , the associated involutions @xmath405 and @xmath453 are related as follows . 1 . it is always possible to have @xmath454 where @xmath92 denotes the transposition . 2 . in the following cases , it is also possible to have @xmath455 : 1 . if @xmath456 , @xmath457 , or @xmath458 . 2 . if @xmath459 or @xmath460 . 3 . if @xmath461 . \(i ) from the matrix perspective , proposition [ prop : blemma ] puts restrictions on @xmath462 when @xmath463 is the permutation matrix of the transposition @xmath92 and the @xmath464-entry of @xmath442 is @xmath67 . for each @xmath38 , the matrix @xmath465 is strictly upper triangular and satisfies @xmath466 ^ 2 = \varepsilon \circ d ( y_m ) = 0.\ ] ] letting @xmath467 as in remark [ rem : matfunc ] produces a sequence , @xmath468 , with @xmath469 an involution of @xmath470 . we show that @xmath471 satisfies the requirements of a generalized normal ruling . this requires establishing that the restrictions provided by definitions [ def : nr ] and [ def : gnr ] on consecutive involutions , @xmath472 and @xmath469 , are satisfied . _ case _ @xmath86 contains a * left cusp * : then equation ( [ eq : diffe ] ) and the definition of augmentation allow us to compute @xmath473 using remark [ rem : matfunc ] we conclude that @xmath474 the @xmath261-complex associated with @xmath475 is related to that of @xmath476 by adding two new generators , @xmath477 and @xmath478 , to @xmath479 . the complex is the split extension of that of @xmath476 by @xmath480 with the differential @xmath481 . it can then be checked directly from the definition that the involutions @xmath472 and @xmath469 satisfy ( 2 ) of definition [ def : nr ] . _ case _ @xmath86 contains a * right cusp * : let @xmath482 denote the @xmath261-complex associated with the matrix @xmath476 by the formula @xmath483 note that @xmath472 is precisely the involution associated to @xmath484 by proposition [ prop : specbasis ] . from @xmath485 we deduce that @xmath486 , and it follows that @xmath487 . next , one observes that @xmath475 is the matrix of the @xmath261-complex @xmath488 \ , | \ , i \neq k , k+1\ } , \widetilde{d}\right)\ ] ] where @xmath489 is the quotient of @xmath490 by the subcomplex @xmath491 and @xmath492 is the differential induced by @xmath342 . if @xmath493 is a triangular change of basis for @xmath484 satisfying the conditions of proposition [ prop : specbasis ] , then @xmath494 \,|\ , i \neq k , k+1 \}$ ] will be such a basis for @xmath495 , so that the involution associated with @xmath475 is related to @xmath472 as required in ( iii ) of definition [ def : nr ] . finally , using equation ( [ eq : conjugate ] ) , @xmath496 . regardless of the value of @xmath414 , the @xmath464-entry of @xmath500 \varepsilon(y_{m-1 } ) [ i + \varepsilon(b_m ) e_{k , k+1}]\ ] ] is @xmath67 , so the matrix @xmath501 is strictly upper triangular and @xmath502 is related to @xmath503 as in proposition [ prop : blemma ] . it follows that @xmath504 and @xmath472 satisfy the requirements near crossings ( including the normality conditions ) of definition [ def : gnr ] . the statement that @xmath405 is @xmath18-graded if @xmath349 is @xmath18-graded follows from ( i ) of remark [ rem : matfunc ] . as in ( [ eq : mcmatrix ] ) , @xmath465 is the matrix of an @xmath261-complex with basis @xmath505 corresponding to the strands of @xmath8 at @xmath357 . if @xmath349 is @xmath18-graded with respect to @xmath102 , then we can assign a grading by @xmath506 and the differential will have degree @xmath341 . s. chmutov and v. goryunov , _ polynomial invariants of legendrian links and wave fronts _ , topics in singularity theory . v.i.arnolds 60th anniversary collection . ( a.khovanskii , a.varchenko and v.vassiliev , eds . ) , _ amer . math . translations _ * 2 * , vol . 180 , ( 1997 ) , ams , providence , ri , 25 - 44 . chekanov and p. pushkar , _ combinatorics of legendrian links and the arnold 4-conjectures , uspekhi mat . * 60 * ( 2005 ) , no . 1 , 99 - 154 , translated in _ russian math . surveys _ * 60 * ( 2005 ) , no . 1 , 95 - 149 . v. turaev , _ the conway and kauffman modules of a solid torus , zap . nauchn . sem . otdel . mat . inst . steklov . ( lomi ) _ * 167 * ( 1988 ) issled . * 6 * , 79 - 89 ( russian ) . english translation : _ j. soviet math . _ * 52 * ( 1990 ) , 2799 - 2805 .

neutron stars are expected to contain a significant number of hyperons in addition to protons and neutrons in the highest density portions of their cores . following the work of jones , we calculate the coefficient of bulk viscosity due to nonleptonic weak interactions involving hyperons in neutron - star cores , including new relativistic and superfluid effects . we evaluate the influence of this new bulk viscosity on the gravitational radiation driven instability in the-modes . we find that the instability is completely suppressed in stars with cores cooler than a few times k , but that stars rotating more rapidly than of maximum are unstable for temperatures around k. since neutron - star cores are expected to cool to a few times k within seconds ( much shorter than the-mode instability growth time ) due to direct urca processes , we conclude that the gravitational radiation instability will be suppressed in young neutron stars before it can significantly change the angular momentum of the star .

astro-ph0110558

the @xmath0-modes ( fluid oscillations whose dynamics is dominated by rotation ) of neutron stars have received considerable attention in the past few years because they appear to be subject to the chandrasekhar - friedman - schutz gravitational radiation instability in realistic astrophysical conditions ( see ref . @xcite for a recent review ) . if the @xmath0-modes are unstable , i.e. if the damping timescales due to viscous processes in neutron - star matter are longer than the gravitational - radiation driving timescale , a rapidly rotating neutron star could emit a significant fraction of its rotational energy and angular momentum as gravitational waves . with appropriate data analysis strategies , these waves could be detectable by interferometers comparable to enhanced ligo . the @xmath0-mode instability might also explain the relatively long spin periods observed in young pulsars and of older , accreting pulsars in low - mass x - ray binaries . recently jones @xcite has pointed out that long - neglected processes involving hyperons ( massive cousins of the nucleons ) can lead to an extremely high coefficient of bulk viscosity in the core of a neutron star . using simple scaling arguments he suggests that the viscous damping timescales associated with these processes may be short enough to suppress the @xmath0-mode instability altogether in realistic astrophysical circumstances . the purpose of this paper is to investigate this possibility more thoroughly . the hyperons only exist in the central core of a neutron star where the density is sufficiently high . the relevant effects of the @xmath0-modes however vanish as @xmath4 ( where @xmath0 is the distance from the center of the star ) . thus the overall effect of hyperon induced dissipation on the @xmath0-modes depends very sensitively on the size and structure of the core of a neutron star . jones initial estimates did not take properly into account either the structure of the @xmath0-mode or the detailed properties of the nuclear matter in the core of a neutron star . we improve on jones analysis in several ways : first we evaluate fully relativistic cross sections to determine the reaction rates of the relevant hyperon interactions . we find that these cross sections reduce to the results of jones @xcite in the low - momentum limit , but can be about an order of magnitude larger in some regimes of neutron - star matter . second we derive new expressions for the bulk viscosity coefficient that are appropriate even for a relativistic fluid such as neutron star matter . third we construct detailed neutron star models based on an equation of state that includes hyperons and the appropriate interactions among all of the particle species present . due to superfluid effects the temperature and density dependence of hyperon bulk viscosity turns out to be quite complicated : superfluidity increases the viscosity in some cases while reducing it in others . and fourth , we use a more accurate model of the structure of the @xmath0-mode eigenfunction in the cores of these stars to evaluate the effects of hyperon dissipation . our analysis shows that hyperon bulk viscosity completely suppresses the gravitational radiation instability in the @xmath0-modes of rotating neutron stars for temperatures below a few times @xmath1 k. we find that the gravitational radiation instability acts most strongly at temperatures around @xmath3 k where stars rotating more than @xmath2 of the maximum rotation rate ( depending on the details of the microphysics ) are driven unstable . our coefficient of bulk viscosity is actually several hundred times that of jones @xcite , who suggested that the instability was completely suppressed . however , our use of the proper @xmath0-mode eigenfunction reduces the dissipation by several orders of magnitude and we find that there is a window of instability . how long it lasts is another matter . if the core of the neutron star cools via the standard modified urca process , its temperature remains above a few times @xmath1 k for about a day @xcite . this is enough time for an unstable @xmath0-mode to grow and radiate away a substantial fraction of the star s rotational kinetic energy and angular momentum into gravitational waves @xcite . however if the core of the star cools too quickly the instability might not have enough time to grow before being suppressed by the hyperon bulk viscosity . the time needed for a neutron - star core to cool to a few times @xmath1 k is reduced to about a second when direct urca processes are able to act @xcite . modern equations of state have large enough proton densities in the core that direct urca cooling is now expected to act until the neutrons and protons condense into a superfluid state , i.e. above about @xmath1 k. the growth time for the gravitational radiation instability in the most rapidly rotating neutron stars is about 40 s @xcite . thus we conclude that the core of a neutron star will probably cool too quickly for the @xmath0-mode instability to grow significantly before being suppressed by the hyperon bulk viscosity . the organization of the rest of this paper is as follows . in sec . [ s : eos ] we provide details of the equation of state which we use , including numerical aspects of the evaluation of various thermodynamic variables and derivatives , and the model of hyperon superfluidity that we employ . in sec . [ s : bulk ] we present a new derivation of the coefficient of bulk viscosity for relativistic neutron - star matter ( including several interacting fluids ) in terms of the microscopic reaction rates and thermodynamic derivatives . in sec . [ s : rates ] , we compute the relevant cross sections in order to evaluate the reaction rates for hyperons in a dense medium . in sec . [ s : times ] we combine the thermodynamic expressions of sec . [ s : bulk ] with the microscopic reaction rates of sec . [ s : rates ] to obtain expressions for hyperon bulk viscosity as a function of density and temperature in neutron - star matter . in sec . [ s : rmodes ] we evaluate @xmath0-mode damping timescales for neutron stars containing ordinary fluid and superfluid hyperons . finally , in sec . [ s : end ] we discuss the implications of our results for the @xmath0-mode instability in real neutron stars , and also attempt to estimate how robust these conclusions are . neutron - star matter is a fermi liquid which at low densities is composed primarily of neutrons @xmath5 , protons @xmath6 and electrons @xmath7 . charge neutrality @xmath8 ( where @xmath9 is the number density of the @xmath10 species ) and @xmath11-equilibrium @xmath12 ( where @xmath13 is the chemical potential of the @xmath10 species ) determine the relative concentrations of these particles at each density . as the total baryon density increases however , it becomes energetically favorable for the equilibrium state to include additional particle species : first muons @xmath14 , and then a sequence of hyperons @xmath15 , @xmath16 , ... these additional particles appear as the density exceeds the threshold for the creation of each new species . the relative concentrations of the various species are determined at each density by imposing charge neutrality and @xmath11-equilibrium . at the highest densities of interest to us these equilibrium constraints are @xmath17 in order to solve these constraints and determine the equilibrium state of neutron - star matter , we need explicit expressions for the various chemical potentials @xmath13 as functions of the particle number densities @xmath18 . these functions have encoded within them the details of the interactions between the various particles in a dense fermi - liquid environment . in this paper we have adopted the expressions for these chemical potentials as given by glendenning s relativistic effective mean - field theory @xcite . figure [ f : p_fermi ] illustrates the fermi momenta of the various particle species as a function of the total energy density of the matter that we obtained with glendenning s ( case 2 @xcite ) expressions for the chemical potentials . glendenning also gives expressions for the total energy density @xmath19 and total pressure @xmath6 as functions of the particle densities @xmath9 . these quantities are illustrated and tabulated by glendenning @xcite , and we will not reproduce them here . our numerical code reproduces glendenning s numbers quite accurately . g/@xmath20 ) for glendinning s equation of state.[f : p_fermi],width=288 ] we are also interested here in some less familiar thermodynamic quantities that are relevant for calculating the bulk viscosity in neutron - star matter . these quantities are easily determined once the full description of the equilibrium state is known . in particular the partial derivatives of the chemical potentials with respect to the various particle number densities , @xmath21 , are needed in the expression for the relaxation time associated with bulk visocity as defined in eqs . ( [ tau ] ) , ( [ delta1 ] ) , and ( [ delta2 ] ) below . these @xmath22 are easily determined numerically ( or even analytically in some cases ) once the full equilibrium state is known . further the thermodynamic function , @xmath23 appears as a prefactor in the expression for the bulk viscosity , eq . ( [ zeta ] ) , that we derive below . here @xmath24 is just the partial derivative of the pressure with respect to the number density of neutrons ( keeping the other number densities fixed ) , and @xmath25 is the derivative of the fractional density of neutrons in the equilibrium state , @xmath26 , with respect to the total baryon density @xmath27 . the left side of eq . ( [ prefactor ] ) has been re - expressed in terms of @xmath28 the `` fast '' and @xmath29 the `` slow '' adiabatic indices defined in eqs . ( [ gammai ] ) and ( [ gamma0 ] ) below . figure [ f : thermo ] illustrates this function for the glendenning equation of state . for a non - relativistic fluid the pre - factor @xmath30 is identical to a commonly used alternative expression involving the sound speeds of the fluid : @xmath31 @xcite . however , this equality is not satisfied in neutron - star matter . consequently it is important to use the correct expression given in eq . ( [ prefactor ] ) . ( the difference between the `` fast '' and `` slow '' adiabatic indices ) that appears in our expression for the bulk viscosity.[f : thermo],width=288 ] we have solved the relativistic structure equations for the non - rotating stellar models based on this equation of state . figure [ f : density ] illustrates the total energy density as a function of radius for neutron - star models having a range of astrophysically relevant masses . this figure illustrates that these stars contain large central cores having material at densities that exceed the @xmath15 and @xmath16 threshold densities . the fact that it becomes energetically favorable to create hyperons ( or even free quarks ) above some threshold density is not really very controversial . however the expressions for the chemical potentials as functions of the particle densities are not well known , and so the detailed properties of nuclear matter at the densities where hyperons are likely to occur is not well determined at this time . this uncertainty translates then to an uncertainty about the sizes of the hyperon containing cores of real neutron stars . since the size of this hyperon core determines the strength of the bulk viscosity effects which we evaluate here , the implications for the stability of the @xmath0-modes are correspondingly uncertain as well . 0.4 cm g/@xmath20 ) vs. distance from the center of the star . threshold densities for @xmath15 and @xmath16 hyperon formation are also plotted . these equilibrium structures are computed using general relativity.[f : density],title="fig:",width=288 ] next , we must consider the possibility that the hyperons in neutron - star matter form cooper pairs and condense into a superfluid state at sufficiently low temperatures . various calculations are given in the literature of the @xmath16 superfluid gap function @xmath32 @xcite . the @xmath16 gap function is constrained by the experimental data on the energy levels of double @xmath16 hypernuclei such as @xmath33be and @xmath34b @xcite , however even so it is probably only known to within a factor of two or three . in our numerical analysis of the bulk viscosity timescales discussed in sec . [ s : times ] we use an analytical fit to the zero - temperature gap function @xmath32 as computed by balberg and barnea @xcite . their calculation produces a gap that depends on the total baryon density @xmath27 of the nuclear matter and on the fermi momentum @xmath35 of the @xmath16 itself . as illustrated in fig . [ f : lambda_gaps ] we find that the following empirical fit @xmath36,\label{lambda_gap}\end{aligned}\ ] ] matches their calculated values for the zero - temperature gap energies fairly well . in this expression the gap energy is measured in mev , @xmath35 is measured in @xmath37 , and @xmath27 is measured in @xmath38 . the @xmath15 superfluid gaps are not as well determined because comparable experimental data on double @xmath15 hypernuclei do not exist at present . calculations by takatsuka , et al . @xcite using several models of the nuclear interaction give values of @xmath39 in the range : @xmath40 . we perform two sets of calculations based on the extremes of this range . we either set @xmath41 or @xmath42 . by equality here we mean that the dependence of @xmath39 on @xmath27 and @xmath35 ( up to the overall factor of 10 ) is given by eq . ( [ lambda_gap ] ) . using these energy gaps , and the glendenning equation of state , we can evaluate then the density dependence of the superfluid gap functions . we illustrate these in the form of superfluid critical temperatures @xmath43 [ which are related to the zero - temperature gap functions by @xmath44 in fig . [ f : crit_temp ] . as calculated by balberg and barnea ( dots ) with the empirical analytical fit in eq . ( [ lambda_gap ] ) ( curves ) . the bottom and top curves correspond to @xmath45 and 0.8 @xmath38 respectively . [ f : lambda_gaps],width=288 ] 0.8 cm g/@xmath46 ) . the solid curve is for @xmath16 hyperons , while the dashed curves are for @xmath15 hyperons using either @xmath41 ( bottom ) or @xmath47 ( top).[f : crit_temp],title="fig:",width=288 ] the superfluid gap depends not only on the density of the superfluid material , as discussed above , but also on its temperature . this temperature dependence will be needed to determine the temperature dependence of the hyperon bulk viscosity below . the standard bcs model calculation @xcite of the temperature dependence of the gap is illustrated in fig . [ f : gap_temp ] . this figure compares the results of the exact calculation with a simple empirical fit to these data : @xmath48^{0.53}. \label{gap_temp}\ ] ] since this fit is quite good , we use it whenever the temperature dependence of the gap is needed . 0.4 cm ) . the critical temperature @xmath43 is related to the zero - temparature gap energy by @xmath49.[f : gap_temp],title="fig:",width=288 ] bulk viscosity is the dissipative process in which the macroscopic compression ( or expansion ) of a fluid element is converted to heat . the formalism for calculating the bulk viscosity coefficient in terms of the relaxation times of the microscopic processes which effect the conversion is well - known see , e.g. , landau and lifschitz @xcite . however , such calculations are generally performed for an ordinary fluid like air , in which the microscopic processes ( typically involving the transfer of energy between rotational and vibrational degrees of freedom of the molecules ) can be taken to be independent . in neutron - star matter there are several relevant microscopic processes ( involving weak interactions between the various particle species ) , but they are related by constraints ( such as conservation of baryon number)meaning that we can not simply use the standard formulas for either a single process or multiple processes . further , neutron - star matter is composed of some particles having relativistic energies . the standard expressions for bulk viscosity are not correct for such materials . in this section we present a modified and expanded derivation of the equations of bulk viscosity appropriate for neutron - star matter . bulk viscosity is due to an instantaneous difference between the total physical pressure @xmath6 of a fluid element and the thermodynamic pressure @xmath50 . the thermodynamic pressure is determined only by the equation of state for a fluid element of given particle number and entropy densities . it is the value toward which the microscopic processes are driving the physical pressure at any given time . the coefficient of bulk viscosity @xmath51 defines the proportionality of this pressure difference to the macroscopic expansion of the fluid : @xmath52 where @xmath53 is the velocity of the fluid element . consider now a fluid state that is an infinitesimal perturbation of a time - independent equilibrium state . let @xmath54 and @xmath55 denote the pressure and number density ( functions only of position ) that describe this equilibrium state . to calculate @xmath51 in terms of the microscopic reaction rates that drive the system toward equilibrium , we re - express both sides of the eq . ( [ zdef ] ) in terms of the lagrangian perturbation of the particle number density @xmath56 . using the particle conservation equation ( and keeping only terms linear in the deviation away from equilibrium ) , we express the right side of eq . ( [ zdef ] ) as @xmath57 where @xmath58 is the eulerian velocity perturbation , and we assume that the perturbation has time dependence @xmath59 in the comoving frame of the fluid . in order to analyze the left side of eq . ( [ zdef ] ) , we examine a fluid variable @xmath60 that characterizes the microscopic process which produces bulk viscosity . for small departures from equilibrium , the value of @xmath60 in the physical fluid state relaxes toward its value in thermodynamic equilibrium by @xmath61 where @xmath62 is defined as the relaxation time for this process we are interested in nearly equilibrium fluid states in which the physical values of the state variable @xmath60 ( and hence the thermodynamic state @xmath63 ) oscillate about the background equilibrium , so that @xmath64 in such a state it is straightforward to verify that @xmath65 now consider how the fluid variable @xmath63 changes as the particle number density of the state is varied slowly from one equilibrium state to another : @xmath66 ( since by definition @xmath67 ) . it follows then that the difference between @xmath50 and @xmath54 is given by @xmath68\delta n. \label{deltap1}\ ] ] a similar argument ( now using eq . [ amp ] to relate @xmath69 and @xmath70 ) gives the following expression for @xmath71 : @xmath72\delta n. \label{deltap2}\ ] ] combining eqs . ( [ deltap1 ] ) and ( [ deltap2 ] ) gives us an expression for the difference in the pressure @xmath73 that appears on the left side of eq . ( [ zdef ] ) : @xmath74 then equating this expression for the left side of eq . ( [ zdef ] ) with the expression for the right side from eq . ( [ rhs ] ) , we find the desired formula for the bulk viscosity : @xmath75 finally it is convenient to re - express the thermodynamic derivatives that appear in eq . ( [ zeta ] ) in terms of the more familiar @xmath76 the `` infinite '' frequency adiabatic index , and @xmath77,\label{gamma0}\ ] ] the `` zero '' frequency adiabatic index . in terms of these quantities , then the bulk viscosity may be written in the form : @xmath78 for a fluid composed of particles with non - relativistic energies this expression is equivalent to the conventional one @xcite written in terms of the sound speed @xmath79 , since @xmath80 for such fluids . however for a fluid containing particles with relativistic energies , the conventional form is wrong and eq . ( [ zeta2 ] ) is the appropriate form to use . note that this new form of the bulk viscosity is needed to describe any fluid containing relativistic internal particle energies , even if the bulk motion of the fluid itself has only non - relativistic velocities which are well approximated by newtonian hydrodynamics . we also see that for a fixed - frequency perturbation , the greatest bulk viscosity comes from processes with relaxation times @xmath81 . the importance of this fact will become apparent as we examine how the relaxation time @xmath62 varies inside a neutron star . the standard approach @xcite to treating multiple reactions is to repeat the preceding derivation for multiple degrees of freedom @xmath82 and relaxation times @xmath83 . however , this only works if ( as in air ) the @xmath82 can be chosen to be independent . this is not possible in neutron - star matter , since the degrees of freedom ( e.g. , concentrations of various baryons ) are related to each other even out of thermodynamic equilibrium by constraints such as conservation of baryon number . the reactions of interest to us here are the non - leptonic weak interactions @xmath84 given the microscopic reaction rates for these processes ( which are calculated in sec . [ s : rates ] ) , we can express all of the perturbed quantities in terms of a single one . since all the hyperon reactions that contribute to bulk viscosity involve neutrons , we choose as our primary variable the number density of neutrons @xmath85 . let @xmath86 be the fraction of baryons in a given fluid element that are neutrons . this variable changes only by internal reactions , not directly by changing the volume of the fluid element , and so we can write @xmath87 here @xmath88 is the production rate of neutrons per unit volume , which is proportional to the overall chemical potential imbalance @xmath89 ( see below ) . ( we normalize @xmath85 and @xmath88 to the baryon number density @xmath27 to remove the oscillating time dependence of the volume of the fluid element . ) assuming the reactions described in sec . [ s : rates ] , the relaxation time is then given by @xmath90 where @xmath91 . we obtain @xmath92 from the following constraints : @xmath93 where @xmath82 are the number densities of baryon species normalized to the total number density of baryons and the @xmath94 are defined below . the first constraint is conservation of baryon number , obeyed by all reactions . ( we note that since @xmath95 that @xmath96 . ) the second constraint is related to conservation of electric charge , but is stricter : we assume that all leptonic ( urca ) reaction rates are much smaller than those which produce hyperon bulk viscosity and so protons are only produced in reactions that produce a @xmath15 . the third constraint is that the non - leptonic reaction @xmath97 proceeds much faster than the weak interactions which produce hyperon bulk viscosity since it is mediated by the strong nuclear interaction . here we use the shorthand @xmath98 equilibrium with respect to reaction ( [ strong ] ) ensures that both processes described in sec . [ s : rates ] have the same chemical potential imbalance , @xmath99 it is straightforward then to express @xmath100 in terms of the @xmath101 , and then to eliminate all but @xmath102 using the constraints eqs . ( [ const1])([const3 ] ) . the result , @xmath103 then determines via eq . ( [ tau ] ) the relaxation time that appears in the bulk viscosity formula eq . ( [ zeta2 ] ) . for a certain range of densities there are @xmath15 hyperons present in glendinning s equation of state , but no @xmath16 . in that case the variable @xmath104 remains zero , and the constraint eq . ( [ const3 ] ) is no longer enforced . in this case the chemical potential imbalance can still be expressed in terms of @xmath102 with the somewhat simpler result : @xmath105 the equation of state of neutron - star matter is generally written in a form which gives the thermodynamic variables as functions of the various particle species present , e.g. the pressure would be specified as @xmath106 . in order to evaluate the thermodynamic derivatives @xmath107 and @xmath108 that are needed in eq . ( [ zeta ] ) , we note that @xmath109 for our choice of @xmath60 . thus the partial derivative needed in eq . ( [ zeta ] ) is given by @xmath110 . similarly , if needed , @xmath111 . the derivative @xmath112 that also appears in eq . ( [ zeta ] ) is determined by constructing a sequence of complete equilibrium models ( e.g. by imposing all of the necessary @xmath11-equilibrium constraints , etc . ) for different total baryon number densities @xmath27 . this complete model of the equilibrium states will include the functions @xmath113 , from which the derivative @xmath114 is easily computed . formation of a superfluid ( of a given particle species ) is marked by the formation of cooper pairs and collapse of the pairs into a bose - einstein condensate . it is the unpaired particles that we are concerned with , since they are the ones that can participate in the bulk - viscosity generating reactions . the free - particle states within the pair - binding energy @xmath115 of the fermi surface are depleted . as a result all phase - space factors ( and effectively the reaction rates ) are decreased by roughly a factor @xmath116 . the effect of superfluidity can be included in our ordinary - fluid bulk viscosity calculation then simply by making the substitution @xmath117 thus when superfluidity is taken into account the equation for the relaxation time , eq . ( [ tau ] ) becomes : @xmath118 since the @xmath15 and @xmath16 hyperons form at the lowest threshold densities , we are most interested in the nonleptonic reactions forming them from neutrons . following jones earlier work @xcite we calculate rates for two reactions , @xmath119 as tree - level feynman diagrams involving the exchange of a @xmath120 boson . in his most recent paper @xcite , jones treats the reaction @xmath121 which is generally the dominant channel for @xmath16 production in laboratory experiments on hypernuclei . this process has no simple @xmath120-exchange contribution . several other processes contribute , but at present the rate can not be well predicted from theory . some of these processes surely operate in reactions ( [ reacs ] ) and ( [ reacl ] ) and modify the rates , perhaps significantly . however , our simple calculations should provide a reasonable lower limit on the rates and thus an upper limit on the bulk viscosity . we calculate reaction rates using the standard techniques of time - dependent perturbation theory in relativistic quantum mechanics . we use the conventions of griffiths @xcite : spinors are normalized to @xmath122 , the fifth dirac matrix is @xmath123 , the metric has negative trace , and @xmath124 . for a single reaction ( [ reacs ] ) or ( [ reacl ] ) between particles with 4-momenta @xmath125 , the differential reaction rate ( number per unit volume per unit time ) is @xmath126 where @xmath127 is the spinor matrix element ( squared and summed over spin states ) , boldface @xmath128 are 3-momenta , and @xmath129 are particle energies . the statistical factor @xmath130 , which compensates for overcounting momentum states of indistinguishable particles , is @xmath131 for reaction ( [ reacs ] ) and @xmath132 for reaction ( [ reacl ] ) . first , consider the @xmath16 reaction ( [ reacl ] ) , which we represent by a single tree - level diagram . labeling the particles 1 ( neutron ) , 2 ( ingoing proton ) , 3 ( outgoing proton ) , 4 ( @xmath16 hyperon ) , we obtain the matrix element ( for a single set of spin states ) @xmath133 .\end{aligned}\ ] ] here @xmath134 is the fermi coupling constant and @xmath135 is the cabibbo weak mixing angle . the quantities @xmath136 , @xmath137 , and @xmath138 are axial - vector couplings ( normalized to the vector coupling ) of the weak interaction , whose deviation from @xmath139 represents the partial nonconservation of the axial current . ( these quantities are often written @xmath140 or @xmath141 in the particle and nuclear physics literature . we add a label to keep track of which nucleon - hyperon line is which . ) we use the values @xmath142mev@xmath143 and @xmath144 from the particle data group @xcite . the axial - vector couplings change with varying momentum transfer and density of the medium in a way that reflects the internal ( strong - interaction ) structure of the baryons and is therefore difficult to calculate . we use the values @xmath145 , @xmath146 , and @xmath147 measured in @xmath11-decay of baryons at rest @xcite . there are theoretical reasons to believe that all axial - vector couplings tend to their asymptotically free values of @xmath139 in a dense medium @xcite . we provide dissipation numbers in later sections both for laboratory values and for asymptotic values of the couplings to give an estimate of the uncertainty in this calculation . to obtain the net reaction rate @xmath148 we sum @xmath127 over all possible initial and final spin states . this is done in the standard way by tracing over outer products of spinors : @xmath149 \nonumber \\ & & \left . + p_1 \cdot p_4\ , p_2 \cdot p_3 \left [ \left(1+{g_{np}}^2\right ) \left(1+{g_{p\lambda}}^2\right ) - 4{g_{np}}{g_{p\lambda}}\right ] \right\},\end{aligned}\ ] ] which in the low - momentum limit used by jones @xcite reduces to @xmath150 note that this limit corresponds to @xmath151 in the notation of jones @xcite , who obtains @xmath152 . the factor of three discrepancy is within the uncertainties of modern nuclear - matter physics , as we see by e.g. taking the asymptotic values of the axial - vector couplings . reaction ( [ reacs ] ) is treated similarly , with particle labels 1 and 2 ( neutrons ) , 3 ( proton ) , and 4 ( @xmath15 hyperon ) . antisymmetrizing with respect to the two indistinguishable neutrons , the matrix element is @xmath153.\end{aligned}\ ] ] the squared sum over spins is given by @xmath154 + 4p_1 \cdot p_2\ , p_3 \cdot p_4 \left [ \left(1+{g_{np}}^2\right ) \left(1+{g_{n\sigma}}^2\right ) + 4{g_{np}}{g_{n\sigma}}\right ] \nonumber \\ & & + p_1 \cdot p_3\ , p_2 \cdot p_4 \left [ \left(1+{g_{np}}^2\right ) \left(1+{g_{n\sigma}}^2\right ) - 4{g_{np}}{g_{n\sigma}}\right ] + p_1 \cdot p_4\ , p_2 \cdot p_3 \left [ \left(1+{g_{np}}^2\right ) \left(1+{g_{n\sigma}}^2\right ) - 4{g_{np}}{g_{n\sigma}}\right ] \left . \right\ } , \end{aligned}\ ] ] which in the low - momentum limit reduces to @xmath155 this expression agrees with eq . ( 7 ) of jones old paper @xcite , allowing for our different placement of the statistical overcounting factor and different normalization of the spinors ( he uses @xmath156 ) . however , we find that for this reaction the low - momentum limit is a very poor approximation to the full result : the collision integral computed in the next subsection can be more than an order of magnitude higher than eq . ( [ low ] ) would suggest . this effectively brings down the coefficient of bulk viscosity @xmath51 by a factor of about 5 at densities 56@xmath157 g/@xmath20 , where there are no @xmath16 hyperons @xcite . reactions ( [ reacs ] ) and ( [ reacl ] ) can be regarded as scattering processes wherein the scattered particles change identity . therefore we can use existing results on collision integrals in the literature on superfluidity ( e.g. @xcite ) with some slight modifications . we now integrate the differential reaction rate ( [ dgam ] ) over momentum space to obtain the total rate @xmath158 ( from here on we use italic @xmath125 to denote the absolute value of the 3-momenta , @xmath159 . ) the pauli blocking factor , @xmath160 where @xmath161\ } , \ ] ] accounts for the degeneracy of the reactant particles and restricts the available phase space to those particles within roughly @xmath162 of their fermi energies . in the case where all particles are degenerate ( which is true except for a very small region just above the threshold density for each hyperon species ) , the collision integral separates into angular and energetic parts . the energetic part can be written in the limit @xmath163 as @xmath164 where the latter integral has the value @xmath165 . and @xmath166 can be rotated out of the page around the long - dashed common axis by an angle @xmath167 for a given @xmath168 and @xmath169.[f : angles],title="fig:",width=240 ] 0.3 cm we address the angular integral with the aid of fig . [ f : angles ] . with no preferred direction ( e.g. , strong magnetic field ) , the @xmath168 volume element is @xmath170 and the @xmath169 volume element is @xmath171 , where @xmath172 and @xmath173 are the polar and azimuthal angles of @xmath169 with respect to @xmath168 . conservation of momentum demands that @xmath174 , determining a common axis . let @xmath175 and @xmath176 be the angles of @xmath168 and @xmath177 with that axis . the volume element of @xmath177 is then @xmath178 , where @xmath167 is the angle between @xmath179 and @xmath180 , multiplied by the area element in the plane containing @xmath177 and @xmath166 . to separate the energetic and angular integrals , it is convenient to write this area element as @xmath181 , where @xmath182 is the angle between @xmath177 and @xmath166 . before leaving fig . [ f : angles ] , note the following useful identities : @xmath183 we use the differential of the first identity ( with @xmath125 held constant ) to write our final result for the volume element as @xmath184 although @xmath173 and @xmath167 are free to range from 0 to @xmath185 , the limits of integration over @xmath182 depend on the relations between the momenta @xmath125 , which are constrained to be close to the fermi momenta . ( therefore from now on we use @xmath125 to refer to the fermi momenta . ) we integrate @xmath182 over the full range @xmath186 to @xmath187 , which is allowed by momentum conservation for @xmath188 in our equation of state these fermi momentum criteria are satisfied for @xmath189g/@xmath20 ( see fig . [ f : p_fermi ] ) . it turns out that almost all of the dissipation takes place below this density , and much above this density the baryons are too closely packed to maintain their separate identities anyway . therefore for our purposes it is sufficient to treat only the case ( [ triangles ] ) . now consider the case that @xmath127 does not depend on any of the angles , such as in the limit @xmath190 . we can integrate trivially over all the angles , with the exception of @xmath191 where @xmath192 is the smallest ( i.e. hyperon ) fermi momentum involved . this motivates us to use eq . ( [ energy ] ) to write @xmath193 where @xmath194 is the angle - averaged value of @xmath195 . we write @xmath127 in terms of the integrals @xmath196 @xmath197.\end{aligned}\ ] ] ( note that some dot products , e.g. @xmath198 , depend on @xmath167 . the average over @xmath167 is then nontrivial but is easily taken by symmetry about the common axis . ) the results for the angle averages are @xmath199 \right.\nonumber\\ & & \left.\times\left [ 5{\epsilon_p}{\epsilon_\lambda}\left ( 6{\epsilon_n}{\epsilon_p}+ 3p_n^2 - { p_\lambda}^2 \right ) + { p_\lambda}^2 \left ( 10{\epsilon_n}{\epsilon_p}+ 5p_n^2 + 10p_p^2 - { p_\lambda}^2 \right ) \right ] + \left [ { \left(1+{g_{np}}^2\right)}{\left(1+{g_{p\lambda}}^2\right)}- 4{g_{np}}{g_{p\lambda}}\right]\right . \nonumber\\ & & \left.\times\left [ 10{\epsilon_n}{\epsilon_\lambda}\left ( 6m_p^2 + 3p_n^2 + { p_\lambda}^2 \right ) + { p_\lambda}^2 \left ( -20{\epsilon_p}^2 + 15p_p^2 - 3{p_\lambda}^2 + 5{(p_n^2 - p_p^2)^2 / ( p_p^2 - { p_\lambda}^2 ) } \right ) \right ] \right\}.\label{m2lambda}\end{aligned}\ ] ] [ note that the denominator in the last term does not diverge while criterion ( [ triangles ] ) holds . ] similarly , for reaction ( [ reacs ] ) , @xmath200 \left [ 10{\epsilon_n}^2 \left ( 3{\epsilon_p}{\epsilon_\sigma}+ { p_\sigma}^2 \right ) + 5{\epsilon_p}{\epsilon_\sigma}\left ( 6p_n^2 - 3p_p^2 - { p_\sigma}^2 \right ) + { p_\sigma}^2 \left ( 10p_n^2 + 5p_p^2 - { p_\sigma}^2 \right ) \right ] \nonumber \\ & & \left . + 10\left [ { \left(1+{g_{np}}^2\right)}{\left(1+{g_{n\sigma}}^2\right)}- 4{g_{np}}{g_{n\sigma}}\right ] \left [ -m_n^2 \left ( 3{\epsilon_p}{\epsilon_\sigma}+ { p_\sigma}^2 \right ) + { \epsilon_n}\left ( 6{\epsilon_n}{\epsilon_p}{\epsilon_\sigma}- 2{\epsilon_p}{p_\sigma}^2 - 3{\epsilon_\sigma}p_p^2 + { \epsilon_\sigma}{p_\sigma}^2 \right ) \right ] \right\}.\label{m2sigma}\end{aligned}\ ] ] these angle averages are inserted into eq . ( [ rate ] ) to yield the net reaction rate per unit volume as a function of fermi momenta . the relaxation timescales for the non - leptonic weak interactions , eqs . ( [ reacs ] ) and ( [ reacl ] ) , can now be determined by combining the microscopic collision rates determined in sec . [ s : rates ] with the thermodynamic quantities evaluated in secs . [ s : eos ] and [ s : bulk ] . at densities above the threshold for the production of @xmath15 , but below the @xmath16 threshold , the final expression for the relaxation timescale @xmath62 is , @xmath201 here the collision cross section @xmath202 is evaluated from eq . ( [ m2sigma ] ) . the fermi momenta that appear in this expression are obtained from the complete description of the equilibrium thermodynamic state as described in sec . [ s : eos ] . this description includes the values of the particle densities ( and hence fermi momenta ) of each species as a function of the total baryon density . the thermodynamic quantity @xmath203 is given by eq . ( [ delta2 ] ) in terms of chemical potentials and their derivatives . once the density increases to the point that both @xmath15 and @xmath16 hyperons are present in the equilibrium state , then the expression for the relaxation timescale becomes , @xmath204 { \delta\mu\over n_b\delta x_n}. \label{tausl}\ ] ] here the collision cross sections are given by eq . ( [ m2lambda ] ) and ( [ m2sigma ] ) , and @xmath203 is given by eq . ( [ delta1 ] ) in this case . we have evaluated this relaxation timescale for neutron - star matter using the equation of state described in sec . [ s : eos ] . figure [ f : tau ] illustrates the density dependence of this timescale for a range of temperatures . in the case of fig . [ f : tau ] we have assumed that the @xmath15 superfluid gap function is given by @xmath205 , while in fig . [ f : tau10 ] we assume @xmath206 . the only significant difference between these two cases comes about in the density range where there exist @xmath15 but not @xmath16 . in that range the timescale is significantly increased in the @xmath206 case by the stronger superfluid effects . all of the curves in these two figures were evaluated using the `` standard '' @xmath11-decay values of the axial - vector coupling constants : @xmath207 , @xmath208 , and @xmath209 @xcite and the effective masses of all of the baryons when evaluating the scattering cross sections . there are theoretical arguments that suggest the coupling constants should approach the values , @xmath210 in a dense medium @xcite . we illustrate the impact this might have on these timescales in fig . [ f : tau_alt ] ( dotted curve ) . we also illustrate in this figure ( dashed curve ) the effect of using the bare masses of the various baryons when computing the scattering cross sections . we see that the overall effect of these changes is to make the timescales shorter ( by up to an order of magnitude ) . this tends to decrease the bulk viscosity by a similar factor , until the temperature drops below the superfluid critical values . g/@xmath46 ) of the relaxation timescale @xmath62 ( in units of s ) for a range of temperatures . these curves were constructed using the assumption @xmath211 for the @xmath15 superfluid gap . [ f : tau],width=288 ] g/@xmath46 ) of the relaxation timescale @xmath62 ( in units of s ) for a range of temperatures . these curves were constructed using the assumption @xmath47 for the @xmath15 superfluid gap . [ f : tau10],width=288 ] g/@xmath46 ) of the relaxation timescale @xmath62 ( in units of s ) for @xmath212k . solid curve uses effective masses and the @xmath11-equilibrium values of the axial - vector couplings . other cuves explore various alternate microphysics assumptions : dashed curve uses bare masses , dotted curve uses @xmath213.[f : tau_alt],width=288 ] finally , we are in a position now to evaluate the bulk viscosity itself . the real part of the bulk viscosity , the part that is responsible for damping the modes of neutron stars , is given by the expression @xmath214 figure [ f : zeta ] illustrates the density and temparature dependence of @xmath51 . ( here we assume that the frequency corresponds to the @xmath215 @xmath0-mode frequency of a maximally rotating neutron star : @xmath216 . ) these figures illustrate the complicated temperature dependence of the visocity due to superfluid effects . for temperatures slightly below the superfluid critical temperature the values of the bulk viscosity are increased over their normal values . this is due to an increase in the timescale @xmath62 which moves it closer to being in resonance with the pulsation period of the mode . once the temperature falls well below the superfluid critical temparature however , we see that the timescale @xmath62 becomes even longer than the pulsation period and so the viscosity becomes smaller again in this case . we note that even for very low temperatures there exists a small range of densities , just above the hyperon threshold densities , where the bulk viscosity remains rather large . this is due to the momentum dependence of the superfluid gap , eq . ( [ lambda_gap ] ) . the gap @xmath115 goes to zero as the fermi momentum of the particle goes to zero . thus just above the threshold density the superfluid gap vanishes ( for any finite temperatue ) so the material in this region will retain the normal - fluid value of the bulk viscosity . our value of @xmath51 is generally much larger than that obtained recently by jones @xcite : at a total density @xmath217g/@xmath20 and temperature @xmath212k , our @xmath51 is larger than jones by a factor of 400 . roughly a factor of 8 is due to the relatively weak coupling we calculate for reaction in eq . ( [ reacl ] ) . ( at this density the @xmath15 hyperons account for only about @xmath218 of the bulk viscosity and can be neglected . ) we note that using the asymptotic values of the weak axial - vector couplings would erase much of this factor of 8 difference , and thus it is indicative of the size of the uncertainties in @xmath51 due to our poor understanding of nuclear - matter physics . the remaining factor of 50 is thermodynamic in origin . jones evaluates various partial derivatives of pressure and chemical potentials , e.g. appearing in eq . ( [ delta1 ] ) of our paper and eq . ( 42 ) of ref . @xcite , using the values for a gas of noninteracting fermions . we include all the mesonic interaction terms , whose effect is to increase significantly these thermodynamic derivatives . since the details of the neutron - star equation of state are uncertain , the precise values of the derivatives are correspondingly uncertain . however , we think it unlikely that the true physical mesonic terms will cancel precisely enough to bring the derivatives down to their noninteracting values . in summary , we think that the true value of @xmath51 is within an order of magnitude of the value we compute here . g/@xmath46 ) of the hyperon bulk viscosity ( in units of g / cm s ) for a range of temperatures . these curves were constructed using the assumption @xmath211 for the @xmath15 superfluid gap . [ f : zeta],width=288 ] g/@xmath46 ) of the hyperon bulk viscosity ( in units of g / cm s ) for a range of temperatures . these curves were constructed using the assumption @xmath47 for the @xmath15 superfluid gap . [ f : zeta10],width=288 ] the basic formalism for evaluating the effects of bulk viscosity on the stability of the @xmath0-modes is well known @xcite . the time derivative of the co - rotating frame energy @xmath219 due to the effects of bulk viscosity is @xmath220 this rotating frame energy @xmath219 is ( to lowest order in the angular velocity of the star ) given by the integral @xmath221 the timescale @xmath222 on which hyperon bulk viscosity damps the mode then is @xmath223 here we have normalized @xmath222 so that @xmath224 is the hyperon bulk viscosity contribution to the imaginary part of the frequency of the mode . for the case of the @xmath0-modes ( in slowly rotating stars ) the integrals that determine @xmath219 and @xmath225 in eqs . ( [ dedt ] ) and ( [ tildee ] ) can be reduced to simple one - dimensional integrals . for the case of @xmath219 this reduction is well known @xcite : @xmath226 here @xmath175 represents the dimensionless amplitude of the @xmath0-mode , and @xmath227 and @xmath228 are the angular velocity and radius of the star respectively . the reduction of @xmath225 to a one - dimensional integral is not so straightforward . in general the expansion of the mode @xmath229 is a complicated function of radius and angle . to lowest order in slowly rotating stars the bulk viscosity @xmath51 will depend only on radius . thus we may always convert eq . ( [ dedt ] ) to a one - dimensional integral by defining the angle averaged expansion squared @xmath230 : @xmath231 while the angle - averaged expansion is in general a complicated function , for the case of the @xmath0-modes it is rather simple . this function has only been determined numerically @xcite , however the simple analytical expression , @xmath232 , \label{expansion}\ ] ] is an excellent fit to those numerical solutions . once the structure of the density function @xmath233 in a stellar model is determined , it is straightforward to evaluate the integrals in eqs . ( [ tildee_rint ] ) and ( [ dedt_rint ] ) using eq . ( [ expansion ] ) to determine the bulk viscosity damping time @xmath222 . the bulk viscosity of interest to us here is very sensitive to the density of hyperons in the stellar core . thus we use the relativistic stellar structure equations to evaluate @xmath233 . this ensures that the size and stucture of the hyperon containing core are sufficiently accurate for our purposes . these functions @xmath233 are illustrated for a range of stellar masses in fig . [ f : density ] based on the equation of state discussed in sec . [ s : eos ] . given these ( numerical ) expressions for @xmath233 it is straightforward then to use the expressions for the hyperon bulk viscosity @xmath51 derived in sec . [ s : times ] to obtain @xmath234 for any given neutron star temperature . together @xmath233 and @xmath234 then determine @xmath222 through eqs . ( [ tildee_rint ] ) , ( [ dedt_rint ] ) and ( [ taub ] ) . while it is straightforward to evaluate these timescales , the result is a rather complicated function of the temperature , angular velocity , and mass of the stellar model and so we do not attempt to illustrate it directly . the most important application of the hyperon bulk viscosity timescale @xmath222 is the analysis of the role this type of dissipation plays in the gravitational radiation driven instability in the @xmath0-modes . gravitational radiation contributes a term to the evolution of the energy @xmath225 that is positive . as is well known by now , gravitational radiation tends to drive the @xmath0-modes unstable in all rotating stars @xcite . as has been discussed in detail elsewhere @xcite the imaginary part of the frequency of the @xmath0-mode may be written as : @xmath235 here @xmath236 represents the timescale for gravitational radiation to effect the @xmath0-mode , @xmath222 is the hyperon bulk viscosity timescale discussed here , and @xmath237 is the modified urca bulk viscosity . detailed expressions for evaluating these other terms are discussed elsewhere and will not be repeated here . suffice it to say that each is a function of the temperature , angular velocity and mass of the neutron star . since @xmath238 is the imaginary part of the frequency of the @xmath0-mode , the mode is stable when @xmath239 and unstable when @xmath240 . for a star of given temperature and mass , the critical angular velocity @xmath241 is defined to be the angular velocity where @xmath242 . stars rotating more rapidly than @xmath241 are unstable while those rotating more slowly are stable . we have evaluated the critical angular velocities @xmath241 numerically using the new hyperon bulk viscosities derived in sec . [ s : times ] . figure [ f : om_crit_m ] illustrates the temperature dependence of the critical angular velocities for a range of neutron - star masses . the more massive neutron stars have larger hyperon cores which suppress the @xmath0-mode instability more effectively . the curves in fig . [ f : om_crit_m ] assume that the @xmath15 superfluid gap function is given by @xmath205 , and that the axial vector coupling coefficients have their @xmath11-decay values . in fig . [ f : om_crit_m1.4_gall ] we compare the critical angular velocity curves for 1.4@xmath243 stellar models using either the @xmath205 ( solid curve ) or the @xmath206 ( dash - dot curve ) assumption about the @xmath15 superfluid gap . the larger value of @xmath244 allows superfluidity to make the bulk viscosity larger over a wider range of temperatures , and hence the @xmath0-mode instability is less effective . also illustrated in fig . [ f : om_crit_m1.4_gall ] are the effects of changing various microphysics assumptions . the dotted curve shows the effect of changing the values of the axial vector coupling constants from their @xmath11-decay values to the asymptotic value -1 . and the dashed curve shows the effect of using bare masses rather than the effective masses in the scattering cross sections . these alternative assumptions make the bulk viscosity less effective and the @xmath0-mode instability operates over a wider range of angular velocities in these stars . however , neither of these effects is as large as that resulting from a change in the superfluid gap . to the maximum mass for this equation of state , 1.79 @xmath243 . these curves assume @xmath205 and use the @xmath11-decay values of the weak axial vector coupling coefficients.[f : om_crit_m],width=288 ] neutron stars as a function of hyperon core temperature . the solid curve assumes that @xmath245 while the dot - dash curve assumes @xmath246 . both curves use the @xmath11-decay values of the weak axial vector coupling coefficients . dotted curve uses @xmath213 , while dashed curve uses bare masses . [ f : om_crit_m1.4_gall],width=288 ] the hyperons primary contribution to the bulk viscosity of neutron - star matter is through the mechanism discussed above . however , as pointed out by jones , the presence of hyperons in the core of a neutron star also make it possible for alternate forms of the direct urca interaction to take place and these too contribute to the bulk viscosity of the material . jones showed that the contributions to the bulk viscosity from this process are given by @xmath247 in cgs units for typical values of neutron star matter , where @xmath248 is the temperature measured in units of @xmath3k . using this expression in the @xmath16 containing core of the neutron star , we have evaluated the effects of this hyperon channel direct urca bulk viscosity on the stability of the @xmath0-modes . these results are illustrated in fig . [ f : om_crit_m1.4_gn_direct ] . the solid curve in fig . [ f : om_crit_m1.4_gn_direct ] includes the effects of the hyperon bulk viscosity discussed above , the hyperon channel direct urca bulk viscosity of eq . ( [ e : direct_urca ] ) , and the ordinary modified urca bulk viscosity . for comparison the dashed curve leaves out the effects of the hyperon channel direct urca bulk viscosity . we see that this direct urca bulk viscosity has only a small effect on the stability of the @xmath0-modes for temperatures around @xmath3k . neutron stars as a function of hyperon core temperature . solid curve includes the effects of hyperon bulk viscosity , hyperon channel direct urca bulk viscosity , and modified urca bulk viscosity . dashed curve leaves out the effects of the direct urca hyperon bulk viscosity.[f : om_crit_m1.4_gn_direct],width=288 ] we have analyzed here the effects of the bulk viscosity due to hyperons on the stability of the @xmath0-modes in rotating neutron stars . hyperons exist only in the high density core of a neutron star where the influence of the @xmath0-mode is quite small . thus to evaluate accurately and reliably the importance of this effect , it was necessary to compute detailed and accurate models of the composition and structure of the neutron star core , and to have an accurate model of the structure of the @xmath0-mode in this region . we use glendenning s @xcite relativistic mean - field equation of state to evaluate the composition of the nuclear matter in the stellar core , and solve the relativistic oppenheimer - volkoff equations to determine the stellar structure . our evaluation of the hyperon bulk viscosity improves on previous work in several ways . first we generalize in eqs . ( [ zeta ] ) and ( [ zeta2 ] ) the standard expression for the bulk viscosity coefficient so that it applies to relativistic fluids such as neutron star matter . second we generalize the expressions in eqs . ( [ delta1 ] ) and ( [ delta2 ] ) for the thermodynamic quantities that relate the microscopic reaction rates to the relaxation time that appears in the expression for the bulk viscosity by including fully interacting nuclear matter . and third we obtain in eqs . ( [ m2lambda ] ) and ( [ m2sigma ] ) the fully relativistic expressions for the relevant hyperon scattering cross sections needed to evaluate the microscopic reaction rates . while our expressions for these cross sections reduce to the published low - momentum results , we find that the difference of eq . ( [ m2sigma ] ) from the low - momentum limit can be an order of magnitude or more and reduces the coefficient of bulk viscosity somewhat at low densities . finally we evaluate the effects of this hyperon bulk viscosity on the @xmath0-modes using an accurate model for the structure of the @xmath0-mode in the core of a neutron star . our results show that the hyperon bulk viscosity does not substantially suppress the gravitational radiation instability of the @xmath0-modes until the temperature of the core of the neutron star drops below a few times @xmath1k . ( this is in spite of the fact that our coefficient of bulk viscosity is actually higher than that of jones @xcite . the expansion of the fluid in the core of the star as given in eq . [ expansion ] is smaller than he estimated . ) below @xmath1k the @xmath0-mode instability is strongly suppressed in all of our models over the range of @xmath15 superfluid gap functions and the range of axial vector coupling constants that we studied . if the core of the neutron star cools according to the standard modified urca process @xcite , then it would remain hot enough for the @xmath0-mode instability to act for about a day . this is enough time for the @xmath0-mode to grow and radiate away through gravitational waves a substantial fraction of the rotational kinetic energy of a rapidly rotating neutron star @xcite . however if the core of the neutron star cools substantially faster than this , then it may not be possible for the @xmath0-mode to grow rapidly enough to effect the star in a substantial way before the hyperon bulk viscosity stabilizes it . cooling by the direct urca process is significantly faster that the modified urca process : cooling the core of a neutron star to a few times @xmath1k within about a second @xcite . cooling by the direct urca process will occur in neutron - star matter whenever the proton / baryon ratio is larger than about 0.15 . since proton fractions in excess of this are now generally expected in neutron star matter , cooling by the direct urca process seems likely at least until the temperature of the core falls below the superfluid transition temperature for neutrons or protons at about @xmath1k . thus it appears likely that the @xmath0-mode instability is effectively suppressed by rapid cooling of the neutron star core and the non - leptonic hyperon bulk viscosity . once a neutron star cools below the transition temperature for the formation of neutron and proton superfluids , the relaxation timescale for the hyperon interactions will increase exponentially compared to the expressions derived here in eqs . ( [ taus ] ) and ( [ tausl ] ) . this sharply reduces via eq . ( [ realzeta ] ) the bulk viscosity from this process at sufficiently low temperatures . further detailed calculations would be needed to determine whether the hyperon bulk viscosity has a significant influence on the @xmath0-mode instability at temperatures of a few times @xmath249k , which are expected to exist in the cores of neutron stars in low - mass x - ray binaries . we did not carry out these calculations in part because solid crust - related shear @xcite and magnetic field @xcite effects are quite effective in suppressing the instability at these low temperatures . how robust is the conclusion that the @xmath0-mode instability is effectively suppressed ? clearly the details of the nuclear physics involving hyperons in neutron star matter are not well understood at this time . however , our conclusion applies to the entire expected range of the most poorly known properties of this material : the superfluid @xmath15 gap function and the axial vector coupling coefficients . in order to escape this conclusion , it would be necessary for neutron star matter to have very few hyperons present at the densities which exist in the cores of real neutron stars . this would require the equation of state to be substantially different from the one studied here , or the masses of neutron stars to be significantly smaller than @xmath250 . rapid rotation also lowers the central density and consequently the size of the hyperon core in a neutron star . the central density of a maximally rotating @xmath250 neutron star is about 73% of its non - rotating value ( for the equation of state studied here ) @xcite . this reduction almost eliminates the hyperon core for this extreme angular velocity , but over almost all of the range of angular velocities , 1.4@xmath243 stars have substantial hyperon cores . finally if the dissipation in the core were sufficiently large it might be possible for the @xmath0-mode eigenfunction to be clamped to zero in the core by the dissipative processes while remaining finite and unstable in the outer parts of the star . the discussion of this possibility in the appendix shows that the hyperon bulk viscosity is not strong enough to clamp the @xmath0-mode in this way . we thank p. b. jones for raising the issue of hyperon bulk viscosity , for communicating his unpublished work to us , and for giving us helpful comments on our work . we also thank s. balberg , d. chernoff , j. friedman , n. glendinning , p. goldreich , j. lattimer , g. mendell , j. miller , e. s. phinney , m. prakash , r. sawyer , and k. thorne for helpful discussions . this work was supported by nsf grants phy-9796079 , phy-0071028 , phy-0079683 , and phy-0099568 , and nasa grants nag5 - 4093 and nag5 - 10707 . bulk viscosity damps a mode by dissipating energy according to the expression @xmath251 for the case of the @xmath0-modes in slowly rotating stars , we may express the energy , and its time derivative as simple radial integrals : @xmath252 @xmath253 where @xmath254 and @xmath255 are the angle averaged energy and energy dissipation rate densities respectively as given in eqs . ( [ tildee_rint ] ) and ( [ dedt_rint ] ) . the mode will be completely suppressed ( clamped ) locally if the amount of energy removed from the mode locally in one oscillation period is comparable to the local energy density of the mode . thus we define the local quality factor of the mode : @xmath256 if @xmath257 the mode will be clamped . for the @xmath0-modes we find that @xmath258 for the @xmath250 neutron star model considered here @xmath259 rad / s , @xmath260 g/@xmath46 and @xmath261 km in the region where hyperons occur , and @xmath262 km . thus @xmath263 for this case . figures [ f : zeta ] and [ f : zeta10 ] show that the bulk viscosity never exceeds about @xmath264 in the temperature range of interest to us . thus , we conclude that the hyperon bulk viscosity represents a small perturbation on the basic hydrodynamic forces of the @xmath0-modes . the condition @xmath265 is violated only for relatively slowly rotating stars . in the domain where the gravitational radiation instability is most likely to be important , the dissipation by hyperons represents a small perturbation on the basic hydrodynamic forces , thus the @xmath0-modes will not be clamped . p. haensel , k. p. levenfish , and d. g. yakovlev , astro - ph/0110575 appeared on the los alamos archive shortly after we submitted this paper . _ perform a hyperon bulk viscosity calculation analogous to ours for the case of the @xmath15 reaction ( [ reacs ] ) only , but they obtain a coefficient of bulk viscosity @xmath51 ten times greater than ours . most of this factor ( 5 + ) is due to their taking the limiting value ( [ low ] ) for @xmath202 , which is anomalously small . the critical angular velocity for the @xmath0-mode instability scales as @xmath266 , and thus haensel _ et al . _ would obtain critical angular velocities about @xmath267 times the values we obtain in sec . [ s : rmodes ] .

the experimental investigation of superfluid turbulence inhe - b is generally not possible with the techniques which have been developed forhe - ii . we describe a new method by which a transient burst of turbulent vortex expansion can be generated inhe - b . it is based on the injection of a few vortex loops into rotating vortex - free flow . the time - dependent evolution of the quantized vorticity is then monitored with nmr spectroscopy . using these techniques the transition between regular ( _ i.e. _ vortex number conserving ) and turbulent vortex dynamics can be recorded at and a number of other characteristics of turbulence can be followed down to a temperature of . pacs numbers : 47.37 , 67.40 , 67.57 .

cond-mat0406202

quantum turbulence , the seemingly chaotic motion of quantized vortex lines in a disordered network , is a prominent characteristic of superfluid @xmath1he - ii , where it has been known to exist for almost half a century.@xcite up to recently , @xmath1he - ii was the only superfluid system in which this phenomenon has been investigated . @xmath1he - ii is believed to display turbulence in the entire superfluid temperature range which so far has been probed with measurements . in superconductors friction in vortex motion is always high and turbulence is not observed : only in the extreme clean limit one might hope to see any signs of other behavior . at present the closest superconducting analogues are dynamically driven vortex avalanches.@xcite superfluid @xmath0he is an intermediate case between these two extremes . the dynamics of vortex motion in the @xmath0he superfluids has primarily been studied in rotating cryostats . in the past the generally accepted view was that the high mutual friction leads to exponentially damped motion and forces vortices to evolve rapidly along well defined trajectories , so that the number of vortex lines remains conserved in dynamic processes . a large number of measurements over the years has proven this to be the case in both @xmath0he - a and @xmath0he - b at @xmath4.@xcite even at low temperatures ( @xmath5 ) a duplication of the classic vinen - vibrating - wire measurement@xcite verified that in @xmath0he - b stable vortex motion persists for hours in the absence of applied flow . in this experiment superfluid circulation was trapped around a thin wire suspended along the symmetry axis of a cylindrical container while the cryostat was rotating . when rotation was stopped , the trapped circulation started to unwind while a vortex filament , stretched between the wire and the cylinder wall , precessed around the wire and spiraled down along the whole length of the 15 mm long cylinder . the precessing spiral motion lasted for more than 35h and the precession frequency remained constant with a precision of 0.5% . the first indications of rapid non - linear vortex proliferation in @xmath0he - b came from measurements of the critical velocity of vortex formation at temperatures @xmath6 ( ref . , see p. 140141 ) . instead of the usual single - vortex formation at a reproducible critical velocity , sudden avalanches of vortices were observed in accelerating rotation . these bursts were interpreted to mean that at lower temperatures mutual - friction damping had decreased sufficiently so that turbulence became possible . the first evidence for tangled vorticity was reported from vibrating wire measurements in a non - rotating cryostat at temperatures below @xmath7.@xcite a vortex network was found to be produced with a vibrating wire resonator when it was driven at high level above a critical velocity of vortex formation . this was inferred by studying the damping of a second wire , vibrating at low level and probing the density of quasiparticle excitations which are andreev retro - reflected by the superflow fields in the vortex tangle . similar later measurements have allowed a determination of the vortex density in the wire - generated network . at zero externally applied pressure and at a temperature of @xmath8 the tangle localized around the vibrating wire turned out to have extremely low density , corresponding to an average inter - vortex distance of 0.2 mm , and to decay away rapidly in about 5s , when the generator wire was switched off.@xcite there exist no immediately obvious techniques by which turbulence can be easily generated and detected in @xmath0he - b . the above two examples suggest two ways which are supplementary , since they work in different temperature regimes . as outlined below in sec . [ experiment ] , nmr on a rotating sample can be used at temperatures down to about @xmath9 . in the zero temperature limit , where the density @xmath10 of the normal component approaches zero , moving objects can be used to generate vortices and , when driven in a high - quality resonance mode , to detect them . in this report we outline the techniques by which turbulence is generated and detected in the intermediate temperature regime 0.4 @xmath11 , where mutual friction damping and @xmath10 are still finite . recently a practical technique was discovered for injecting vortex loops into vortex - free flow.@xcite it then became possible to monitor the nature of vortex dynamics as a function of temperature and flow velocity in a controlled fashion.@xcite the transition , which separates regular vortex motion from turbulence , could now be mapped . the foundation , on which the interpretation of these results is built , was laid down by an earlier set of important hydrodynamic measurements@xcite on mutual friction which provided both the dissipative and reactive mutual friction coefficients @xmath12 and @xmath13 as a function of temperature @xmath14 and pressure @xmath15 . it turns out that in @xmath0he - b the variation of mutual friction with temperature happens to be such that the transition falls in the middle of the experimental temperature range , to 0.5 @xmath11 . these techniques have been employed in three recent studies investigating turbulence in @xmath0he - b.@xcite the first@xcite reports on the flight time of the vorticity when it expands along the rotating column . the second@xcite describes how turbulence starts in rotating vortex - free flow in the presence of neutron radiation . the third@xcite discusses the mutual - friction dependence of the transition from regular to turbulent vortex dynamics , _ i.e. _ the transition line as a function of @xmath16 and @xmath17 at high flow velocities , where the transition is velocity independent . these transition properties will also be illuminated in sec . [ turbtransition ] of this report , but from a different point of view . nmr measurements from the transient period , when the turbulent vorticity decays into the equilibrium rotating state of rectilinear vortex lines , will be analyzed in a forthcoming report.@xcite according to current view no similar kind of turbulence as in @xmath1he - ii or @xmath0he - b exists in the anisotropic superfluid @xmath0he - a . here vortex motion is highly damped at all experimentally accessible temperatures and fast vortex motion is supported by the appearance of a different structure of vorticity , the vortex sheet.@xcite : @xmath18 . _ ( middle ) _ the equilibrium vortex state is the lowest energy state . here an array of rectilinear vortex lines fills essentially all of the container and the superfluid component is on an average in solid body rotation , _ i.e. _ stationary in the rotating frame . the number of vortex lines is @xmath19 if @xmath20rad / s . this is the final state after injection of seed vortex loops , if the vortex expansion has been turbulent . _ ( right ) _ a meta - stable vortex cluster may include any number of rectilinear vortex lines , @xmath21 . within the central coaxial vortex cluster the areal density of the lines @xmath22 corresponds to solid body rotation . the counterflow velocity outside the cluster is @xmath23 , where we define @xmath24 . this is the final state after injection , if loop expansion has been regular and the number of individual vortices has been conserved . in this case @xmath25 after a kelvin - helmholtz injection event ( fig . [ stepdistribution ] ) . ] let us consider a hydrodynamic equation which describes the evolution of quantized vorticity in a limit , which becomes useful in the interpretation of our measurements . we apply dimensional analysis to this equation in the same manner as one characterizes different flow regimes with the navier - stokes equation in viscous hydrodynamics . this immediately leads to the notion that a change in the nature of vortex motion is expected as a function of mutual friction damping . the argument is simple - minded , but makes the suppression of turbulence by mutual friction damping understandable in superfluid hydrodynamics . the equation of motion for a viscous incompressible fluid is the navier - stokes equation , @xmath26 where @xmath27 is an external force per unit volume , @xmath17 pressure , and @xmath28 kinematic viscosity . in order to compare flows with different viscosities it is useful to write the navier - stokes equation in dimensionless units @xmath29 , @xmath30 and @xmath31 , where @xmath32 and @xmath33 are the characteristic length and velocity scales . the relative importance of the inertial term @xmath34 and the viscous term @xmath35 is used to characterize the flow . the ratio of these two terms is called the reynolds number @xmath36 . for flows with small @xmath37 the dissipative forces are dominant and the resulting flow is typically laminar . if @xmath37 is large , the inertial forces are dominant and the flow is usually turbulent . the same type of classification in superfluid hydrodynamics can be based on an equation for the `` coarse - grained '' superfluid velocity @xmath38 . it is obtained by averaging over volumes containing many vortex lines . the equation@xcite is constructed from the euler equation : @xmath39 which can also be written as @xmath40 where @xmath41 plays the role of the chemical potential and @xmath42 is the vorticity . for such coarse - graining to be meaningful the volume , over which we average , should contain many roughly similarly oriented vortex lines . as each vortex carries a fixed quantized circulation @xmath43 the number of vortex lines in the sample can be estimated from below as @xmath44 l^2 \sim v l / \kappa$ ] . we call this combination the `` superfluid reynolds number '' @xmath45 and require that @xmath46 for the coarse - grained equation to be applicable . the euler equation basically expresses the conservation of energy and eq . thus accounts for the kinetic energy of the global flow . the vortex line tension , which is associated with the @xmath47 local velocity profile around the vortex cores , is neglected . the tension can be included by introducing a correction@xcite to the superfluid velocity @xmath48 , where @xmath49 is roughly the intervortex spacing , @xmath50 is the vortex core size , and @xmath51 is the unit vector in the direction of @xmath52 . dimensional arguments give the estimate @xmath53 . thus for values @xmath54 , the tension can be neglected . in the classical equation the term @xmath27 represents some force applied externally to the fluid . in the case of superfluids there exists an intrinsic contribution to @xmath27 , the mutual friction force @xmath55 , which arises from the interaction of vortex lines with normal excitations . this force , acting on a single vortex line element , is@xcite @xmath56- \alpha'\rho_s\bkappa \times(\mathbf{v}_{\rm n}-\mathbf{v}_{\rm sl}),\ ] ] where @xmath12 and @xmath57 are the dissipative and reactive mutual friction coefficients , @xmath58 is the local superfluid velocity at the position of the vortex line , @xmath59 is directed along the vortex line and @xmath60 is the corresponding unit vector . in general the averaging of the non - linear force in eq . over volumes containing many vortex lines can not be done without knowledge of the vortex configuration . here for simplicity we assume a locally polarized vortex tangle,@xcite where all vortex lines in the averaged volume have the same direction @xmath59 . in this case the transformation from @xmath61 to the coarse - grained force in equation is done by replacing @xmath59 with @xmath52 and @xmath58 with @xmath38 . . thus becomes @xmath62\ , . \label{eq : dyn}\ ] ] this equation has traditionally been used to examine small deviations of vortex lines from equilibrium configurations . we use it to highlight the role of mutual friction in superfluid turbulence . the viscosity of the normal component in superfluid @xmath0he is large : @xmath63 . as a result , the normal component is in practice always in well - defined externally imposed laminar motion . the velocity , which is induced in the normal component by the mutual friction force , can be ignored in flows with large @xmath64 : from eq . , by substituting @xmath27 with the mutual friction force , we find that @xmath65 ( in the intermediate temperature range where @xmath66 and @xmath67 ) . thus the influence of the mutual friction force on the normal component can be ignored up to relatively fast motion of the superfluid component with @xmath68 . even beyond that velocity ( which has not been experimentally reached yet ) the motion of the normal component remains laminar as @xmath69 . ( note that these conclusions do not apply to superfluid @xmath1he where @xmath70 . ) in rotating experiments @xmath71 and we call @xmath72 the _ counterflow velocity _ ( fig . [ rotsupfluidstates ] ) . to simplify eq . we change to the rotating reference frame ( where @xmath73 and which we assume to be inertial ) : @xmath74\,.\ ] ] by applying dimensional analysis to this equation , we can construct a quantity , which has similar physical meaning as the reynolds number in the case of the navier - stokes equation : the ratio of the inertial @xmath75 and dissipative @xmath76 \sim \alpha v^2/l^2 $ ] terms is here @xmath77 . similar to the role of the reynolds number in classical hydrodynamics we might expect that when @xmath78 is large the flow is turbulent and when @xmath78 is small the flow is laminar.@xcite unlike in the classical case , here only the intrinsic properties of the fluid , @xmath12 and @xmath57 , determine the nature of the flow , since @xmath78 does not depend on the externally imposed geometry or the velocity of the flow . none of the established methods to generate turbulence in @xmath1he - ii are applicable in @xmath0he - b in the intermediate temperature regime . a new approach is therefore essential . fortunately compared to @xmath1he - ii , other possibilities are available . one major difference is that vortex formation can be experimentally controlled in @xmath0he - b to much better degree . if remanent vortices and other extrinsic sources of vortex formation can be eliminated , then a high energy barrier prevents vortex formation via intrinsic mechanisms up to relatively high counterflow velocities , typically of order 1cm / s . technically the simplest means of generating high vortex - free flow velocities is to rotate the sample container . vortex loops can then be injected in the vortex - free flow by means of a few different techniques and their dynamic evolution can be monitored with nmr . [ rotsupfluidstates ] classifies schematically the different rotating states , from the initial vortex - free state to the equilibrium vortex state . the most versatile injection method at present time is the kelvin - helmholtz instability of the ab phase boundary.@xcite here the injection can be performed at variable values of rotation @xmath79 and temperature @xmath16 . this requires a two - phase sample , like that shown in fig . [ expsetup ] on the left , where the center section of the long sample cylinder is maintained with a magnetic barrier field in the a phase . to study the transition from regular to turbulent vortex dynamics as a function of temperature and flow velocity , we need a three - step process : 1 ) to set up the vortex - free rotating state , 2 ) to trigger the kelvin - helmholtz instability , and 3 ) to distinguish the type of final state . for controlling the experiment with the nmr measurement it needs to be able 1 ) to verify that we are correctly in the vortex - free initial state , 2 ) to indicate that the injection is properly executed , and 3 ) to determine the number of rectilinear vortex lines in the final state . surprisingly it turns out that the transition to turbulence is so sharp that the final state contains either a few rectilinear vortex lines ( fig . [ rotsupfluidstates ] right ) , if the temperature is above the transition , or else the equilibrium number of vortex lines ( fig . [ rotsupfluidstates ] middle ) . as seen from the nmr absorption spectra in fig . [ expsetup ] on the right , the line shapes of these two final states differ so profoundly that the classification in regular and turbulent events can be performed by inspection only . this is the case if the rotation velocity @xmath80rad / s and the so - called `` counterflow absorption peak '' ( cf peak ) , generated by the vortex - free flow , is well developed . in this report we use the kh instability as injection method . it is triggered with a small step increase in rotation @xmath81 which is typically a few percent of the total rotation . this means that in our measurement the externally controlled conditions are maintained constant , since practically even the rotation velocity can be regarded as constant throughout the measurement . this amounts to an important simplification of the analysis . the time - dependent evolution of the injected vorticity lasts from tens of seconds to minutes , depending on temperature and rotation velocity . during this transient period nmr signals can be measured which yield information on the dynamics . one such measurement is the determination of the flight time @xmath82 of the vorticity from the ab interface to a detector coil,@xcite as illustrated for the lower half of the sample container in fig . [ expsetup ] . another measurement involves the configuration in which the vorticity propagates along the rotating column.@xcite the injection of vortex loops can be accomplished also by other means . in ref . neutron irradiation was discussed for injection . in refs . and a sudden burst of vortex loops through the orifice on the bottom of the sample space was observed to start the turbulence . he sample and nmr measurement . _ ( left ) _ the sample container is a quartz tube with diameter 6 mm and length 110 mm . it is separated from the rest of the liquid @xmath0he volume with a partition disc at the bottom of the sample space . in the disc an orifice of 0.75 mm diameter provides the thermal contact to a liquid column which connects to the sintered heat exchanger on the nuclear cooling stage . three superconducting coil systems produce independently axially oriented magnetic fields along different sections of the tube : at both ends are the polarization fields for nmr detection and in the middle the barrier field for stabilizing @xmath0he - a . the dashed curves , approximately symmetric about the middle of the sample tube , show the locations of the ab interfaces in the gradient of the barrier solenoid when its current is @xmath83a ( short section of a phase with parabolic contour of ab interfaces ) and @xmath84a ( longer section of a phase with almost flat interfaces).@xcite _ ( right ) _ two nmr absorption spectra in rotation . in the equilibrium vortex state the line shape borders steeply on the right to the larmor field at 21.0mt , while at low values of field sweep it extends to the temperature dependent cut off of the flare - out texture , here at 19.7mt . in the vortex - free state the line shape displays a large `` counterflow peak '' . its maximum is located at the cut - off value of the `` flare - out texture '' . both line shapes have been measured at the same temperature and have equal integrated absorptions . ] the liquid @xmath0he sample and the nmr measurement have been summarized in fig . [ expsetup ] . additional details have been explained in refs . and . the long sample tube is surrounded by three superconducting end - compensated coil systems . these supply the two constant field regions at both ends of the tube for nmr polarization and the barrier field to maintain the a phase in the center . if the barrier field @xmath85 exceeds the thermodynamic b@xmath86a transition field @xmath87 , two ab interfaces are formed symmetrically around the center of the sample tube . their shapes and locations depend on the current @xmath88 in the barrier magnet and on the externally controlled variables @xmath16 , @xmath17 , and @xmath79.@xcite as an example , two locations of the ab interfaces are shown in fig . [ expsetup ] . for the measurement of the turbulent transition the locations of the ab interfaces are not of importance . however , to determine the flight time @xmath82 the distance @xmath89 has to be accurately known . it is obtained from the calculated field profile of the solenoidal magnet system . the continuous wave nmr measurement is performed at constant frequency @xmath90 , using a linear sweep of the axially oriented polarization field @xmath91 . the two absorption spectra in fig . [ expsetup ] display the extremes of different possible line shapes . in fact , to monitor the transition to turbulence , they turn out to be the only two line shapes needed . the large peak in the vortex - free state is produced by the orienting effect of the counterflow on the order parameter texture and is called the _ counterflow peak_. its shift from the larmor field ( at 21.02mt ) is used for temperature measurement and its height is a sensitive function of the number of vortex lines . when a cluster of rectilinear vortex lines is formed , the cf peak height is reduced . the reduction is linear with vortex number @xmath92 in the limit @xmath93 , if all other variables remain constant . at large values of @xmath92 the reduction becomes nonlinear and finally , if @xmath92 is continuously increased until it reaches the maximum possible value @xmath94 , the spectrum looks entirely different : here much of the absorption is shifted to high fields and borders prominently to the larmor edge at 21.0mt . with increasing @xmath92 the intensity in the cf peak is thus shifted closer to the larmor edge . therefore one may monitor the increase in @xmath92 by recording either the cf peak height or the absorption at the larmor edge . in the latter region the absorption is less affected by temperature changes and often it is more practical to monitor the evolution of the vortex expansion process at fixed polarization field value close to the larmor edge . first kelvin - helmholtz instability events is traversed . _ ( top ) _ the measurement is performed at high temperatures where vortex motion is regular . here the number of vortex loops injected in the b phase equals the number of new rectilinear vortex lines which are added to the b phase sections per instability event . the measurement monitors the increase in rectilinear vortex lines per event . the increase is proportional to the height of the discontinuous steps in the linear staircase patterns and , as seen from the step heights , varies randomly per event . the nmr absorption is measured at constant polarization field at a value just below the larmor edge ( _ e.g. _ for the bottom spectrometer @xmath95mt , see fig . [ expsetup ] ) . _ ( bottom ) _ the rate of rotational acceleration is @xmath96rad / s@xmath97 . ] an example of such a measurement is illustrated in fig . [ kh - steps ] at constant temperature above the transition to turbulence . here the rotation @xmath79 is increased at constant slow rate across the region where the kelvin - helmholtz instability of the ab interface develops . the @xmath79 increase is performed sufficiently slowly so that no dynamic effects are visible . the first critical event we denote with @xmath98 . during further increase of the rotation drive the instability occurs recurrently every time when the counterflow at the outer sample perimeter reaches the critical value @xmath99 , where @xmath100 is the radial value at which the injected vorticity breaks through the ab interface ( sec . [ injection ] ) . simultaneously a central cluster of rectilinear vortex lines starts to develop ( fig . [ rotsupfluidstates ] right ) . the parameters controlling the instability are roughly identical in the upper and lower halves of the container . the first instability event ( with vortex - free flow in the two b - phase sections ) is registered at almost exactly the same value of @xmath79 in both detector coils , as seen in fig . [ kh - steps ] . nevertheless , the instabilities of the two ab interfaces in fig . [ expsetup ] are not coupled , but occur independently . thus the subsequent instabilities in fig . [ kh - steps ] are not synchronous , since the number of vortices injected in the b - phase sections at each event varies randomly . also in the temperature regime where the transition from regular to turbulent vortex expansion takes place , it is quite possible to have a regular event in one half and a turbulent event in the second half . as the two @xmath0he - b samples are independent , one can have , for instance , vortex - free flow in the upper b - phase section @xmath101 and roughly the equilibrium number of vortices in both the a - phase and the lower b - phase sections @xmath102 . as @xmath98 is almost equal for both halves of the container , the first critical event can be triggered with a small step increase in @xmath79 at both ab interfaces simultaneously . this triggers the injection of vortex loops into both @xmath0he - b samples and their dynamic evolution towards the respective final states is then recorded independently . this is our normal measuring procedure when searching for the transition between regular and turbulent vortex expansion . the top and bottom halves of the container give thus similar information , one half would be sufficient to identify the transition , but two halves gives better statistics . for the initial diagnostics , to understand the critical characteristics of the different sources of vortex formation in a new sample tube , the two independent nmr detection systems are most useful . four different quartz tubes were examined in this setup . their properties turned out to vary substantially with respect to vortex formation in the absence of the barrier field and the ab interfaces . this means that at present the cleanliness and the properties of the fused quartz surfaces are not in reliable control . the fourth and last sample tube was by far the best , with the highest overall critical velocity . most of the results of this report were measured with this tube . the classification of the vortex expansion events is based on a measurement of the number of rectilinear vortex lines in the final state . the vortex line number @xmath92 can be accurately determined , if needed , from the linear reduction in the cf peak height as a function of @xmath92 . it turns out that precise measurements of @xmath92 are not needed . because of the narrow width of the transition regime between regular and turbulent vortex dynamics at high rotation velocities , we need to distinguish only between two states , one with a small vortex cluster @xmath103 at temperatures above the transition and one with the equilibrium vortex state @xmath104 at temperatures below the transition . in fig . [ expsetup ] the two line shapes with very different appearance exemplify these two cases . he - b during continuous cool down at constant rotation . _ ( main panel ) _ with decreasing temperature the cf peak is shifted further from the larmor value ( at 21.0mt ) and its height is reduced , since the total absorption is proportional to the static susceptibility @xmath105 . below @xmath9 the textural orienting interaction from the cf vanishes rapidly with decreasing temperature , as shown by this series of cf line shapes recorded with the bottom spectrometer at constant @xmath79 and @xmath106vortex lines . all spectra extend with finite ( but small ) absorption up to the larmor value , but the line shapes above the three bottom examples have been truncated . _ ( inserts ) _ spectra from the top and bottom spectrometers measured simultaneously at the lowest temperature . in both inserts three different situations have been compared : a ) sample not rotating @xmath107 , b ) sample filled with the equilibrium number of vortex lines , and c ) continuous cool - down in rotation at constant @xmath108 with a small stable vortex cluster . ] the large difference between the two line shapes in fig . [ expsetup ] is caused by the textural orienting influence of the macroscopic counterflow . in rotating experiments the textural interaction can be conveniently manipulated with the rotation velocity @xmath79 . the velocity of the counterflow @xmath109 is given by @xmath110 within the vortex - free annulus , which surrounds the central vortex cluster in fig . [ rotsupfluidstates ] ( right ) , _ i.e. _ in the region @xmath111 . here the rotation velocity @xmath112 is an experimentally useful quantity which denotes the rotation at which the cluster of @xmath92 lines fills the whole cross section of the sample , _ ie . _ the rotation velocity at which a given number of lines @xmath92 is in the equilibrium state . no changes in the vortex - line number occur if one sweeps @xmath79 between the limits @xmath113 , where @xmath114 is the relevant critical velocity , at which more vortices start to form . by changing @xmath79 , one scans the magnitude of the textural interaction term@xcite @xmath115 where the superfluid density anisotropy @xmath116 . the applied magnetic field induces an orbital momentum in the direction @xmath117 where the unit vector @xmath118 specifies the direction of the applied field @xmath119 , and @xmath120 is a rotation matrix in the b - phase order parameter ( which minimizes the dipolar spin - orbit interaction in the equilibrium state ) . since @xmath121 , to minimize the free energy @xmath122 in the equilibrium configuration of the texture , the azimuthal cf velocity @xmath123 attempts to orient the orbital anisotropy axis @xmath124 parallel to itself . since the magnetic field induced distortion is of axial symmetry , in first order the anisotropy @xmath125 . in addition @xmath126 ( assuming @xmath127 ) and thus with increasing rotation the orienting effect from the cf becomes comparable sequentially with other textural interactions . in the axially symmetric case ( when the magnetic field is oriented along the symmetry axis of the sample cylinder , @xmath128 ) this leads to a sequence of different axially symmetric textures , which are generally known to be of `` flare - out '' configuration.@xcite the density anisotropy @xmath129 is only weakly temperature dependent , when @xmath130 . this is the regime where all nmr measurements on vortices in rotating @xmath0he - b have so far been performed and where the measuring techniques are well established . below @xmath131 however , @xmath132 drops rapidly with decreasing temperature.@xcite the evolution of the nmr absorption line shape during cooling ( in otherwise constant conditions ) is illustrated in the main panel of fig . [ nmr - cf - spectra ] : with decreasing temperature the peak heights rapidly drop , the line shapes undergo changes owing to textural transitions , standing spin - wave resonances grow in amplitude and decorate the line shape , and finally the orienting influence of the cf on the texture vanishes , the spin wave amplitudes die away , and the nmr shifts saturate below @xmath133 ( bottom spectrum ) . the textural transitions as a function of the magnitude of the cf interaction ( [ cf - interaction ] ) have been studied at temperatures above @xmath131 in ref . . broadly speaking the transitions in fig . [ nmr - cf - spectra ] resemble those which occur at higher temperatures while @xmath79 is reduced at constant temperature . an example is the second cf peak at maximum possible nmr shift which starts to grow in amplitude at @xmath134 . at still lower temperatures this second peak breaks into a twin peak structure while the main cf peak at smaller nmr shift is already declining in amplitude . the doubling of the peak at maximum possible nmr shift has not been observed before and no explanation exists for it . finally below @xmath135 the effect of the counterflow is not visible in the line shape and the final example on the bottom of fig . [ nmr - cf - spectra ] is similar in appearance to the equilibrium vortex state spectrum in fig . [ expsetup ] . the two inserts to fig . [ nmr - cf - spectra ] demonstrate that at the lowest temperature the nmr spectrum is completely insensitive to rotation : the line shapes are identical irrespective whether the sample ( a ) is not rotating at all ( @xmath136 ) , ( b ) rotating in the equilibrium vortex state @xmath137 ) , ( c ) or with vortex - free flow around a small vortex cluster ( @xmath138rad / s ) . the nmr polarization field of the top spectrometer is 40% higher than that of the bottom spectrometer . in the upper insert of fig . [ nmr - cf - spectra ] the textural magnetic healing length@xcite is of order @xmath139 and the line shape is that of a regular flare - out texture . in the lower insert at lower nmr field @xmath140 is larger and the line shape is more distorted . the spectra from the top and bottom spectrometers in these two inserts have been measured simultaneously in zero barrier field ( _ i.e. _ in the absence of an a - phase layer ) at the lowest temperature to which the sample cools . this temperature is limited by the residual heat leak into the @xmath0he - b sample and its thermal contact to the refrigerator . owing to the saturation of the nmr shifts , and to a lesser degree the uncertainty in the value of the larmor edge ( _ i.e. _ the line width from the residual magnetic field inhomogeneity of the nmr polarization magnets , fwhh @xmath141khz ) , with the presently available techniques the lowest temperature can only be assigned an upper limit @xmath142 . an interesting further feature in the line shapes of fig . [ nmr - cf - spectra ] are spin - wave resonances in the two signals just above that on the very bottom . in contrast to the local oscillator model , which yields the overall line shape of the flare - out textures , spin - wave absorption modes represent coherent spin precession over global regions with a smoothly varying texture.@xcite the magnetic healing length of the texture @xmath143 grows with decreasing temperature and becomes finally comparable to the sample radius @xmath144 . this causes gradients in textural orientations to become more and more gradual . thus the spin - wave resonances grow in prominence with decreasing temperature . as seen in fig . [ nmr - cf - spectra ] , both the intensity and the width of successive spin - wave resonances are approximately constant across the entire absorption regime of the signal . in fact , theory predicts that their absorption intensity grows towards low temperatures and finally saturates , similar to nmr shifts . however , in the signal corresponding to the lowest temperature in fig . [ nmr - cf - spectra ] spin - wave resonances are not visible . it is believed that here , owing to the absence of dissipation and rigidity of the order parameter texture , spin - wave resonances are not displayed by our measurement . the most convenient thermometer in these measurements is the horizontal shift of the cf peak from the larmor value . the shift is calibrated in a slow continuous warmup , which is repeated at varying warming rates , against two external thermometers . these are rigidly fixed on the nuclear cooling stage : a capacitive @xmath0he melting pressure gauge and a pulsed nmr thermometer with a platinum wire brush as sample . our temperature calibration of the nmr shift is generally in agreement with those reported in refs . and . however , below 0.7mk the external thermometers lose contact with the @xmath0he - b sample and a smooth extrapolation of the nmr shift has to be used as temperature scale . finally , as seen in fig . [ nmr - cf - spectra ] below @xmath145 , the nmr shifts saturate and become useless for thermometry . to summarize , in the present type of nmr measurements the experimentally vital information is obtained via the image which the nmr provides about the order parameter texture , primarily through the change in the magnitude of the counterflow interaction in eq . ( [ cf - interaction ] ) as a function of the number of rectilinear vortex lines @xmath92 at constant @xmath79 . these techniques work well at temperatures above @xmath146 , so that single - vortex resolution is achieved above @xmath147 . however below @xmath148 the texture - based nmr techniques break down . this is most unfortunate for further efforts to study the properties of turbulence in the zero temperature limit . clearly @xmath149 is not yet the regime of ballistic quasiparticle motion and vanishingly small mutual friction . at present it is not clear if other nmr techniques can be developed for @xmath0he - b to measure the number of rectilinear vortex lines or the vortex density at the lowest temperatures . . the absorption amplitude close to the larmor value has been recorded with the bottom spectrometer . the sloping dashed line is a fit through the instability points and is proportional to @xmath150 . _ ( bottom ) _ the discontinuous cf velocity as a function of @xmath79 at @xmath100 : @xmath151 . the horizontal dashed line is equivalent to the sloping dashed line in the upper panel and defines the mean critical velocity @xmath152 . ] the classic kelvin - helmholtz ( kh ) instability is a generic phenomenon of fluid dynamics which occurs at the interface between two inviscid fluid layers , which move with respect to each other in a state of relative shear flow . if the difference in tangential velocities is small , the interface remains calm , but at some critical value interfacial waves are formed . the ab interface instability is its first superfluid example.@xcite @xmath0he - a and b behave very differently in rotation : the a phase has a low critical velocity and remains close to the equilibrium vortex state at all times , while the b phase is vortex - free before any instabilities have occurred . to maintain this situation , a new structure of vorticity is formed on the a - phase side of the ab interface , a surface vortex layer . it is confined to the interface and gives rise to the tangential velocity difference across the phase boundary.@xcite the instability manifests itself as a sudden event in which some vortices from the surface layer break through the interface and start to evolve in the vortex - free b - phase counterflow . the critical cf velocities across the interface follow the general formula@xcite @xmath153 where the magnetic restoring force on the ab interface is @xmath154 at @xmath155 . here @xmath156 and @xmath105 are the susceptibilities of the a and b phases , @xmath157 the surface tension of the ab interface , and @xmath158 the critical magnetic field of the ab transition . in our experiments @xmath159 and @xmath160 , where @xmath98 is the first critical rotation velocity of the kh instability and @xmath161 the effective radial value for the instability . the wave vector of the mode of interface waves , which causes the instability , is @xmath162 . measurements in widely different conditions agree with eq . ( [ kh - critvelocity ] ) without adjusted or fitted parameters.@xcite formed per event . the smooth curve is a guide for the eye of a differential probability distribution . ] a repetition of fig . [ kh - steps ] at higher signal - to - noise resolution is shown in fig . [ kh - hiprecisionsteps ] for the bottom section of the sample . it illustrates the experimental consequences from eq . ( [ kh - critvelocity ] ) , the signatures of the kh instability . a sequence of instability events has been recorded here in slowly increasing rotation at high temperatures where the number of vortex loops injected in the b phase equals the number of rectilinear vortex lines which are formed after each instability event . the signal monitored in the top panel is the absorption close to the larmor edge where the step - like increase measures the number of new rectilinear vortex lines which are added to the lower b phase section , _ i.e. _ the vertical discontinuous step height is proportional to the increase in vortex number . the simplest technique to assign the correspondence between step height and vortex number @xmath163 is to perform a least squares fit with the constraint that the values for @xmath163 have to be integral numbers . the order of magnitude for the step height signal is @xmath164v / vortex line . this figure refers to the voltage measured at the output of our cryogenic mesfet preamplifier ( with a gain @xmath165 ) which follows the superconducting tank circuit . with this fit we get the vortex line numbers which are denoted in fig . [ kh - hiprecisionsteps ] next to each step . the location of the instability events on the @xmath79 axis has to fulfill the requirement that the critical kh velocity is constant , _ i.e. _ the corner points in the staircase pattern fall on one line . this gives the dashed linear fit with the slope @xmath166v/(rad / s ) . combining the two numbers so far , we get @xmath167lines/(rad / s ) . this means that when @xmath163 new rectilinear lines are added to the central vortex cluster , the counterflow velocity along the effective perimeter decreases by @xmath168 . from this requirement we obtain the effective radial value @xmath169 mm for the instability , which should be compared with the sample radius @xmath170 mm . a lot of similar examples confirm the above value for the effective radius @xmath171 . also more elaborate measurements , which measure the number of vortex lines formed in each instability event with a calibrating procedure , give the same value of @xmath161 in the above regime of @xmath172 . a result from this latter type of precision measurements was reported in ref . . what then is the explanation for @xmath161 ? the vortices which create the staircase pattern in fig . [ kh - hiprecisionsteps ] are detected by the pick - up coil in the bottom of the sample tube 29 mm below the ab interface . the flight time for the vortices from the interface to the top edge of the coil is 5s . at high temperatures the expansion of the vortices is deterministic . the fixed flight time introduces a 5s delay before the information about the events at the interface starts to build up in the pick - up coil . in fig . [ kh - hiprecisionsteps ] the rotation drive is a linear ramp with the acceleration rate @xmath173rad / s@xmath97 . this means that every 6s the counterflow velocity increases at the effective perimeter by the equivalent of one circulation quantum . comparing these two numbers , we interpret the staircase pattern to arise from the events at the interface and not to be blurred by the motion of the vortices along the rotating column . in this case @xmath161 is the radial distance at which the kh instability occurs and the vorticity is injected across the ab interface from the a phase into the b phase . the fact that we measure a definite value for @xmath161 allows us to make a guess of the initial configuration of the injected vortex loops immediately after injection , in the situation from where their expansion in the vortex - free b - phase counterflow starts . this would be a loop of length @xmath174 mm whose one end sticks out of the ab interface at @xmath175 and the second ends on the cylindrical side wall at @xmath176 in the b phase section . the second important question about the injection is the number of such loops . in fig . [ kh - hiprecisionsteps ] we have a small sample of eight instability events , with the number of vortex lines @xmath163 formed per event marked next to each event . it is seen that @xmath163 is a small random number and on an average @xmath177 . in fig . [ stepdistribution ] a histogram of @xmath163 for a large sample of 214 events is shown . such a probability distribution can only be measured at high temperatures with regular vortex motion . it has not been carefully measured as a function of the external variables @xmath178 , but there are no indications that it would be rapidly changing in the conditions used for vortex loop injection in this work . therefore we assume that the distribution behaves in a continuous manner across the transition from regular to turbulent dynamics and is of similar shape in the turbulent regime . the average of the probability distribution in fig . [ stepdistribution ] is 11 vortex lines per event . this agrees with what one would expect for the number of a - phase vortex quanta in the surface vortex layer which cover one trough , or half a wave length , in the wave pattern of the interface ripplons according to eq . ( [ kh - critvelocity ] ) . in the a - phase section , assuming solid - body rotation with zero critical velocity , there are @xmath179 single - quantum vortices ( @xmath180 ) covering the ab interface which all flare out radially perpendicular to the cylindrical wall . measured along the effective perimeter there are thus @xmath181 circulation quanta per unit length and thus in one corrugation @xmath182 , where @xmath183 . from eq . ( [ kh - critvelocity ] ) this is seen to be @xmath184 , which gives 9 vortex lines in fair agreement with the measured average in fig . [ stepdistribution ] ( @xmath185cm / s at @xmath186 , 29.0bar , and @xmath187a ) . from 1.31 to 1.36rad / s in 2s . with vortex - free flow in both b - phase sections , the instability at both ab interfaces falls between these two limits . _ ( top ) _ the trigger step increase @xmath81 is instantaneously registered as an increase in the cf peak height . a rapid decay of the peak height starts when the front of propagating vorticity reaches the closest end of the detector coil and approaches zero when the front leaves the far end of the coil . this signal decay corresponds to the removal of the large - scale azimuthal counterflow just behind the propagating front . the nmr polarization fields are maintained here at constant value at the location of the maximum of the cf peak of the initial state ( _ e.g. _ at 19.37mt for the bottom spectrometer in fig . [ expsetup ] ) . ] as noted in the context of fig . [ kh - steps ] , the first critical event , before any vortices have been injected , takes place almost simultaneously in both b - phase sections . therefore , when the critical rotation velocity @xmath98 is known , the two events can be triggered together with a small step increase @xmath188rad / s in the manner shown in fig . [ injectionturbulence ] . this procedure allows to measure the flight time for the vorticity to propagate from the ab interface to the detector coil . moreover , it allows to set up the nmr measurement so that transient signals can be captured efficiently when the vorticity front travels through the detector coils . in contrast to fig . [ kh - steps ] , in fig . [ injectionturbulence ] the evolution of the cf peak height after injection has been monitored at low temperatures where the expansion of the injected vortex loops is turbulent . what happens at injection ? what is the configuration in which the vorticity expands along a rotating superfluid column ? the answers to these questions are still discussed ; we are not going to delve into details here . clearly the first turbulent proliferation of vorticity takes place immediately at or after injection . if the rotating vortex - free column is infinitely long , it is unrealistic to expect that turbulence would fill at some point of the process the entire column . rather it could be expected to expand into the vortex - free section in a form which after some time travels along the column in a time - invariant average configuration . since the flight time of the vorticity grows as @xmath189 with decreasing friction @xmath190 , the propagating front moves slower with decreasing temperature and has more time to settle down in such a configuration . thus at @xmath191 the transient signals do not appear to depend on the distance @xmath89 . this conclusion is reached from monitoring the transient signals , which the propagating front induces in the two detector coils when turbulence is started in random locations along the sample with a neutron capture process ( in zero barrier field and no a - phase).@xcite with this picture in mind , we take a second look at the timing in fig . [ injectionturbulence ] . the split - half detector coils , whose axes are oriented transverse to the sample column in the setup of fig . [ expsetup ] , have a sharp cutoff in sensitivity at the edge of the coil . thus the flight time from the ab interface to the detector coil is bracketed between the instantaneous increase in cf peak height from the trigger and the arrival of the vorticity front at the closer end of the detector coil ( from where the decay of the cf signal starts in fig . [ injectionturbulence ] ) . in fig . [ expsetup ] this flight time of 41s corresponds to a distance @xmath192 mm ( bottom section of sample ) . the front is expected to travel through the 10 mm long detector coil during the next @xmath193s . from fig . [ injectionturbulence ] we note that the cf peak height of the bottom spectrometer vanishes in @xmath194s . comparing these two numbers , this means that the macroscopic azimuthally flowing counterflow is removed immediately behind the traveling front of vorticity , _ i.e. _ the vorticity behind the first front has the equilibrium polarization of solid body rotation in the direction @xmath195 of the column . other conclusions about the structure of the front in the turbulent temperature regime can be drawn from the transient signal measured at the larmor edge , _ i.e. _ by monitoring the build up of the vorticity instead of the cf decay.@xcite figs . [ kh - hiprecisionsteps ] and [ injectionturbulence ] thus characterize different aspects of how the vorticity expands in the rotating column at high and low temperatures . the most fundamental difference is in the final states : in fig . [ kh - hiprecisionsteps ] only a small number of rectilinear vortex lines is created in each sequential kh event , while in fig . [ injectionturbulence ] the first event fills the vortex - free b - phase section with the equilibrium number of vortex lines . we have measured extensively the first critical velocity @xmath98 of the kh instability as a function of @xmath196 . these measurements give in the @xmath197 plane continuous trajectories,@xcite in which the transition to turbulent vortex dynamics is not visible . there is no indication that the injection process itself would influence which final state is formed , when the temperature is well above or below the transition . rather , all measurements suggest that the kh instability is a predictable injection mechanism with continuous properties and with a reasonably narrow distribution in the number of injected vortex loops . the measurement of flight time @xmath82 in fig . [ injectionturbulence ] gives the propagation velocity of the vorticity along the rotating column : @xmath198 . it turns out to amount to @xmath199 , since this expression gives @xmath190 values in good agreement with the mutual friction measurements in ref . . this is the value expected for the end point of a single vortex spiralling down along the vertical container wall in the vortex - free state . the measurement can thus be used to determine the dissipative mutual friction coefficient @xmath12 . surprisingly the single - vortex value for @xmath200 applies for the propagation velocity both in the regular and turbulent regimes of propagation , _ i.e. _ the velocity of the front of expanding vorticity with the retreating vortex - free flow displays no anomaly at the transition.@xcite when measured with kh injection , the transition from regular to turbulent dynamics is found to be velocity independent and centered at 0.52 0.59@xmath201 , depending on pressure.@xcite it coincides with the condition @xmath202 . however , the transition is liquid pressure dependent , moving with increasing pressure to higher temperatures on the normalized @xmath203 scale and also to higher @xmath204 . thus the transition does not take place at exactly the same @xmath204 value at all pressures and the situation is more complex than the simple theory in sec . [ analysis ] portrays . with kh injection the transition has a narrow width of @xmath205 at the three measured pressures of 10.2 , 29.0 , and 34bar . in this section we describe results from the transition region which shed more light on the kh injection mechanism . , the number of vortices present compared to the equilibrium number . this ratio is obtained using a calibration procedure similar to that explained in ref . . in most cases the result pertains to the first injection event with vortex - free flow in the b - phase section , but some cases have been included where the first or even the second sequential event gives only a small number ( as in fig . [ stepdistribution ] ) , until a later injection event produces a large number of vortices . the measurement demonstrates that the temperature width of the transition regime is narrow . also the number of rectilinear vortex lines produced per injection event is either very small and the process is of the form shown in fig . [ kh - steps ] , or the final vortex number is close to equilibrium and the process is as in fig . [ injectionturbulence ] . in contrast , intermediate vortex clusters are rare . ] a close up of the transition regime between regular and turbulent vortex motion as a function of temperature is shown in fig . [ transitionregime ] . here the number of rectilinear vortex lines has been measured and is expressed as a fraction of the equilibrium state , @xmath206 , where @xmath79 is the critical rotation velocity @xmath98 of kh injection . the conspicuous feature is that intermediate values @xmath207 0.7 are prominently absent : the final state after injection includes either very few vortices or is close to equilibrium . in fact , the only intermediate values which have been measured are a few in the transition regime . this means that turbulence predominantly either switches on and fills the sample with close to the equilibrium number of vortices or it does not and the final state contains only the vortices injected by the kh instability . therefore the transition regime is not characterized during cooling by a gradual increase in the value of @xmath206 , but by the fact that a growing fraction of the expansion processes after injection become turbulent and send the sample in the equilibrium vortex state . the absence of intermediate values for the fraction @xmath206 has an interesting explanation . it suggests that turbulence either switches on or does not depending on initial conditions , as was observed to be the case in neutron capture induced turbulence in ref . . the initial conditions are dictated by the details of the injection process . although the kh instability appears to be a reproducible injection technique , it contains a stochastic component , namely the number of injected vortex loops @xmath163 in fig . [ stepdistribution ] and their exact configuration . this would mean that it is not the expansion process , but the injection situation , which is most influential in starting turbulence : in the same conditions the expansion process might proceed towards only a few vortices or to the equilibrium number and it is the stochastic variability in kh injection which produces the finite width of the transition . these considerations are corroborated by ongoing numerical simulations of the appropriate experimental situation.@xcite a further example of the transition features is illustrated in fig . [ twoturbtransitions ] . it shows the nmr absorption response , measured at constant polarization field close to the larmor edge , during a continuous slow increase of @xmath79 . the beginning of the trace @xmath208rad / s ) shows how the absorption is shifted from the larmor region into the growing cf peak ( compare to the spectra in fig . [ expsetup ] ) . here it can be seen that the cf peak starts to develop only when @xmath209rad / s . above 1rad / s the textural interaction from the cf has become the dominant contribution and from then on changes in the larmor region are minor until the kh instability occurs at 1.45rad / s . the sharp peak here is a transient signal@xcite from a turbulent expansion process which fills the sample with almost the equilibrium number of vortices . during further increase of @xmath79 a central cluster of vortices with @xmath210rad / s is formed , but otherwise a similar cycle of absorption behavior is repeated . finally a second turbulent event takes place at exactly twice higher rotation velocity , as appropriate for the critical velocity of the kh instability according to eq . ( [ cf - velocity ] ) . this latter turbulent event occurs in the annular vortex - free cf region which surrounds the central cluster in fig . [ rotsupfluidstates ] ( right ) . this example shows that ( i ) the central cluster formed from rectilinear vortex lines is stable in increasing rotation and ( ii ) , since the second turbulent event occurs at the expected @xmath79 value of the kh instability , the turbulent event evolves outside the cluster at large radii and cf velocities , so that the cluster is not involved in the process . ) without adjustable parameters.@xcite the insert in the bottom left panel is a cartoon of the interfacial wave on the ab phase boundary . the black half circles represent the a - phase vortex quanta covering the interface and flaring radially outwards ( _ i.e. _ out of the plane of the paper ) . ] in fig . [ transition ] results are displayed on the final state after kh injection , as a function of cf velocity at constant temperature . the two plots examine the velocity dependence of the transition and here exactly in the temperature regime where the transition from regular to turbulent dynamics takes place . the data have been classified as ( a ) a regular expansion event if the final number of rectilinear vortex lines is in the range of the distribution in fig . [ stepdistribution ] , as ( b ) turbulent if the final state contains close to the equilibrium number of rectilinear vortex lines , or as ( c ) mixed if one or a few regular events precede a final turbulent event . the data have been collected by moving as a function of the current @xmath88 in the barrier magnet along an isotherm in the @xmath197 plane . since @xmath16 and @xmath17 are constant in this measurement , also the magnetic field at the interface is constant : @xmath211 . on the left at low currents the curves start when the maximum field in the center of the barrier solenoid at @xmath212 reaches @xmath87 . from here with increasing @xmath88 the ab interfaces move apart along the solenoidal characteristic @xmath213 and the gradient @xmath214 increases . as a result , the kh critical velocity also moves to higher values , which is seen both from eq . ( [ kh - critvelocity ] ) and in fig . [ transition ] . among the many different measuring runs which have been performed , the isotherms in the two panels of fig . [ transition ] happen to be almost exactly in the center of the transition regime at 10.2 and 29.0bar pressures . they were selected to illustrate the behavior in the transition regime . the results show no clear velocity dependence : the mix of different symbols in both panels appears random . this is also the general conclusion reached in ref . when all data from the kh measurements is reviewed as a whole in the @xmath79 range 0.8 1.6rad / s . the solid curves in the panels of fig . [ transition ] illustrate eq . ( [ kh - critvelocity ] ) and its predictions . the panels in the middle show the behavior of the half length of the interface wave at the instability : @xmath215 . inserting the critical kh velocity @xmath216 , we obtain @xmath217 and therefore the trough with the a - phase vortices , which break through the interface , gets rapidly smaller with increasing rotation . the inter - vortex distance @xmath50 in the surface vortex layer on the ab interface is determined by the tangential velocity difference : @xmath218 ( see insert in panel on bottom left ) . the number of vortices in the trough can then be expressed as @xmath219 and thus it also decreases with increasing rotation , as shown in the outermost panels on the right . so far the probability distribution in fig . [ stepdistribution ] has not been measured as a function of @xmath79 , to check the dependence @xmath220 . however , it is instructive to consider the consequences from eq . ( [ deltan ] ) . our measurements at 29.0bar and @xmath221a as a function of temperature are a good example . first @xmath222 was measured using a slowly accelerating rotation drive , as in figs . [ kh - steps ] and [ kh - hiprecisionsteps ] . @xmath222 turned out to increase with decreasing temperature , curving smoothly towards saturation at @xmath223rad / s below @xmath224 . this agrees with eq . and reflects the fact that at 8a the barrier field in the center of the solenoid exceeds @xmath225 . from eq . ( [ deltan ] ) we see that if anything , then @xmath163 decreases towards low temperatures . the measurements were next repeated using the triggering method of fig . [ injectionturbulence ] with @xmath188rad / s . such a measurement proceeded as seen in fig . [ injectionturbulence ] and resulted below the transition regime @xmath226 always in a turbulent event . in view of eq . ( [ deltan ] ) this seems to mean that even with fewer injected vortex loops the first kh instability leads invariably to turbulence , or that even a small number of injected loops , such as @xmath227 ( fig . [ stepdistribution ] ) , has to suffice to start turbulence . this we take as evidence for the importance of the kelvin - wave instability@xcite of a single vortex as the initial source for the proliferation of vorticity and as the mechanism which here starts the turbulence . finally we note in passing two features which are related to any future extension of these measurements : ( i ) with the present barrier magnet the trajectories of the kh instability do not extend much below or above the velocity interval shown in fig . [ transition ] . for lower velocities a barrier magnet with a weaker gradient is needed , for higher velocities it should be steeper ( eq . ) . ( ii ) we have performed measurements with kh injection also at zero pressure , to span the entire range in particle densities . however , at low temperatures below @xmath228 a new interfering feature appears , the continuous proliferation of vorticity when curved vortex lines are present and undergo the kelvin - wave instability.@xcite it then becomes difficult to achieve vortex - free rotation to sufficiently high velocities , as required for the kh instability . this difficulty grows in prominence at low pressures , especially at zero pressure . the kelvin - wave instability starts to fill the sample at a seemingly constant rate with rectilinear vortex lines already at low rotation and proceeds until close to the equilibrium vortex state . the source of the problem are existing curved vortex lines , which at @xmath229 are remanent filaments , left over from earlier rotations . these are difficult to avoid at low temperatures , since the last one or two remanent vortices take exceedingly long times to annihilate in our long sample tube in standstill @xmath230 . rotation in the reverse direction is not helpful since the required velocities are of order @xmath231rad / s and even then more often than not lead to a reorientation of the filament ( a reversal in the orientation of @xmath59 ) and thereby via the kelvin - wave instability to more vorticity . the long life time of vortex filaments at the lowest temperatures in zero applied flow we noted when the example of a single precessing vortex from ref . was described in sec . [ introduction ] we have described the techniques for investigating turbulence in @xmath0he - b in the temperature range @xmath232 . both the methods to generate turbulence and to record it non - invasively are different from what can be done in @xmath1he - ii . in the temperature region of these measurements the vortex mutual friction of @xmath0he - b changes rapidly and proves to control the character of the vortex dynamics . a sharp transition from regular vortex number conserving dynamics at high temperatures to turbulent dynamics at low temperatures is observed at @xmath233 . the best injection mechanism of vortex loops into vortex - free counterflow of @xmath0he - b has been found to be the kelvin - helmholtz instability of the magnetically stabilized interface between @xmath0he - a and @xmath0he - b . in this process only the number of injected vortex loops is not in good control , which is believed to cause the observed finite transition width . it is to be expected that further work on turbulence in @xmath0he - b will allow to place turbulence in @xmath1he - ii , the only previously known example of superfluid turbulence , into a wider context . in contrast to @xmath1he - ii , where in most cases at higher flow velocities one has to deal with the mutual - friction coupled turbulence in both the superfluid and normal components , in @xmath0he - b experiments the normal component is practically always immobile . thus a simpler example of superfluid turbulence is available for studies here . the present nmr technique , which relies on the influence of the vortex - free counterflow on the order parameter texture , stops working below @xmath234 . at present it is not known if other nmr techniques can be developed to investigate turbulence in the zero temperature limit . we thank carlo barenghi , rob blaauwgeers , girgl eska , demosthenes kivotides , nikolai kopnin , makoto tsubota , and grigory volovik for valuable discussions and especially ladislav skrbek and joe vinen for a critical reading of this manuscript . this work has benefited from the eu - ihp ulti-3 visitor program ( hpri - ct-1999 - 00050 ) and the esf conference programs coslab and vortex . the locations and shapes of the ab interfaces@xcite in fig . [ expsetup ] have been calculated at @xmath235bar , @xmath136 , and @xmath236mk , where the b@xmath86a transition field is @xmath237mt . at the barrier current @xmath221a , the interface is almost flat and at a distance @xmath238 mm from the mid plane of the magnet . at the same @xmath221a at @xmath191 , the interface is more curved and at @xmath239 mm . we assume the following values at 10.2bar [ 29.0bar ] : @xmath240erg / cm@xmath97 [ @xmath242erg / cm@xmath97 ] , @xmath244g / cm@xmath0 [ @xmath246g / cm@xmath0 ] , and @xmath248mt [ @xmath249mt ] . for better agreement we take @xmath250 [ @xmath251 .

abstract we present a closed framework of ads / cft with finite-charge chemical potential . we show how the gauge - invariant identification of the chemical potential with the bulk gauge field emerges from the standard ads / cft dictionary . physical importance and necessity of the minkowski embeddings within the present framework is also shown numerically in the d3-d7 systems . we point out that the d3-d7 model with only the black - hole embeddings does not have the low - temperature and low - chemical - potential region in the grand - canonical ensemble , hence it is incomplete . a physical interpretation that explains these numerical results is also proposed . 0.5 cm

0708.2818

ads / cft correspondence @xcite is a useful framework to analyze strongly coupled yang - mills ( ym ) theories , and its application to quark - hadron physics is one of the important subjects from the phenomenological point of view . since macroscopic properties of quark - hadron systems , based on thermodynamics and hydrodynamics , are as important as their microscopic nature , it is quite significant to establish holographic descriptions of thermodynamics and hydrodynamics of the ym theories . finite - temperature ads / cft has been initiated in ref . hydrodynamic quantities of static ym - theory plasma have been computed in ads / cft ( see , for example , reviews @xcite and the references therein ) , holographic description of hydrodynamics of time - dependent ym - theory fluid has been investigated in refs . @xcite . for a complete description of the thermodynamic and hydrodynamic properties , we need to introduce chemical potentials of the conserved charges to the framework . since the lattice gauge theory has a technical difficulty in introduction of finite baryon chemical potential , it is quite significant to establish a holographic description of baryon chemical potential . a holographic description of r - charge chemical potential has been proposed in refs . @xcite and isospin chemical potential in ads / cft has also been studied in refs . however attempts to introduce a baryon chemical potential to ads / cft have been just started recently @xcite , and there still exists a point which is under debate @xcite . the issue under the question is existence of so - called minkowski - embedding phase in the d3-d7 systems @xcite . for recent studies on finite baryon density systems in holographic frameworks , @xcite . in this paper , we present a closed framework of ads / cft with finite @xmath0-charge chemical potential .- charge chemical potential and baryon chemical potential . see for details , section [ discussion ] . ] although our approach has wide overlap with what have been discussed in refs . @xcite , the following points will be clarified in this paper : * a standard dictionary of ads / cft implies that the chemical potential is given by the boundary value of the zeroth component of the bulk @xmath1-gauge field @xcite . however , this identification is not manifestly gauge invariant . a manifestly gauge - invariant identification of the chemical potential and the @xmath1-gauge field has been proposed in refs . we clarify how the gauge - invariant formulation emerges starting from the standard ads / cft dictionary . * the minkowski embeddings at finite baryon density is claimed to be unphysical in ref . however , we will show their necessity and physical significance within the context of our framework . it will be shown that the model with only the black - hole embeddings proposed in ref . @xcite has a serious problem : the model lacks the low - temperature and the low - chemical - potential region of the parameter space in the grand - canonical ensemble . we call this `` incomplete - ness problem '' in this paper . * the incomplete - ness of the model with only the black - hole embeddings can also be seen in terms of thermodynamic instability in the canonical ensemble . there is a parameter region where thermodynamically stable black - hole embeddings do not exist in the canonical ensemble . the minkowski embeddings provide a stable final state in that case . * we will present a possible physical picture which clarify the difference between the model of ref . @xcite and that of this paper . we will also propose an idea which may remedy the incomplete - ness problem in the model of ref . @xcite . although we will work out along the d3-d7 systems , the formalism of the @xmath0-charge chemical potential given in this paper is applicable to general setups of ads / cft with flavor branes . the organization of the paper is as follows . in section [ basic ] , we present a basic setup and notations in our framework . a closed formulation of ads / cft with finite @xmath0-charge chemical potential is given in section [ formulation ] . the numerical results that support consistency of our framework is presented in section [ numerical ] . the incomplete - ness of the model proposed in ref . @xcite is also pointed out there . in section [ evidences ] , we re - examine the importance of the minkowski embeddings from the viewpoint of thermodynamic stability in the canonical ensemble . in the discussion section , we propose a possible physical interpretation of our framework . we also discuss how the discrepancy between ref . @xcite and ref . @xcite can be interpreted . a possible improvement to remedy the incomplete - ness problem of the model of ref . @xcite is also discussed . the system we deal with is so - called d3-d7 system which corresponds to a @xmath2 sym theory with flavor quarks @xcite . we assume that the number of the flavors @xmath3 is small enough comparing to the number of the colors @xmath4 , and we employ the probe approximation where the back reaction from the d7-brane to the bulk geometry is neglected . in order to clarify the notations , we briefly review the basics of the d3-d7 system . our bulk metric in the euclidean signature is given by @xmath5 where @xmath6 and @xmath7 . here , @xmath8 is the t hooft coupling of the ym theory . there is a horizon at @xmath9 , and the hawking temperature is given by @xmath10 introducing a dimensionless coordinate @xmath11 defined by @xmath12 , the bulk geometry becomes @xmath13 where @xmath14 and @xmath15 is the radius of the 3-sphere . @xmath16 and @xmath11 are related by @xmath17 and @xmath18 now , the induced metric on the d7-brane is given by @xmath19 where @xmath20 . the dbi action of the d7-branes is @xmath21 where @xmath22^{-1}$ ] is the d7-brane tension and we have assumed that the non - zero field strength is only @xmath23 . notice that the chern - simons term vanishes within such a setup . introducing @xmath24 , the dbi action is written compactly as @xmath25 here , @xmath26 with @xmath27 and @xmath28 where @xmath29 is the volume of the space where the ym theory lives . we set @xmath30 in this paper and all the extensive quantities should be understood as those per unit volume . there are two big categories of the d7-brane solutions ( embeddings ) in the present setup : minkowski embeddings and black - hole embeddings . the minkowski embeddings are the d7-brane configurations which do not touch the black - hole horizon , and the black - hole embeddings are those connected to the black - hole horizon . let us define the two foregoing embeddings by using the bulk coordinate . @xmath31 on the d7-brane worldvolume starts at a particular value at the boundary ( which we define @xmath32 ) and one finds that it decreases monotonically as @xmath15 decreases . the minimum value of @xmath33 ( which we define @xmath34 ) is realized at the minimum @xmath15 ( which is denoted by @xmath35 ) . since the location of the horizon is where @xmath36 , the brane touches the horizon at @xmath37 if @xmath38 . this defines the black - hole embeddings . the minkowski embeddings are those with @xmath39 and @xmath40 . a further detailed classification of the brane embeddings is given in appendix [ structure - can ] . in fact , @xmath41 , which is the asymptotic value of @xmath33 is a parameter of the theory which describes the current quark mass @xmath42 . the relationship between the mass and @xmath41 is given by @xmath43 since we usually fix the current quark mass and the t hooft coupling in the analysis , @xmath44 is equivalent to the temperature up to a proportional constant . let us consider a quark - current operator @xmath45 in the ym theory . sym theory as well , we do not write the super - partner part explicitly in this paper for notational simplicity . ] here , @xmath46 is the quark field , and we define the quark - number density @xmath47 as @xmath48 is proportional to @xmath49 in this convention . ] . adding a term @xmath50 to the ym - theory lagrangian corresponds to switching on a one - form bulk field @xmath51 that has the following asymptotic behaviour : @xmath52 where @xmath53 for the one - form field and @xmath54 is a constant which will be determined later . @xmath55 is the conformal dimension of the operator @xmath56 that is 3 in this case . in the ym - theory side , @xmath0-charge chemical potential @xmath57 in the grand - canonical ensemble is introduced by adding an operator @xmath58 to the euclidean ym lagrangian . here , @xmath59 is the quark - charge chemical potential and we may use @xmath60 rather than @xmath57 when it is convenient . what we need to do first in the gravity dual is to find an appropriate one - form field @xmath51 . a proposal given in refs . @xcite is that @xmath51 is the @xmath1 gauge field @xmath61 on the flavor d - branes . @xmath0 symmetry is the @xmath1-diagonal part of the global flavor symmetry in the ym side . an important fact is that the global flavor symmetry is promoted to a _ gauge _ symmetry on the flavor branes in the gravity dual . for example , the global @xmath62 symmetry in the @xmath2 sym theory is realized as @xmath62 gauge symmetry on the flavor d7-branes . hence the @xmath0 symmetry corresponds to the @xmath1 gauge symmetry on the flavor brane , and the @xmath0 charge is identified as the @xmath1 `` electric '' charge on the brane in terms of the gravity dual . then it is quite natural to infer that the chemical potential conjugate to the @xmath0 charge is given by the boundary value of @xmath63 that is the conjugate field to the `` electric '' charge : @xmath64 here , we have normalized the unit electric charge so that it has the unit quark - number charge ( that is @xmath65 times the unit baryon - number charge ) rather than the unit baryon - number charge . this normalization comes from the fact that a quark is represented by a single fundamental string , and the end point of the fundamental string carries unit electric charge with respect to @xmath63 . now , a problem comes up at ( [ dictio-2 ] ) ; @xmath63 is not a gauge invariant quantity in the bulk theory while @xmath60 and @xmath48 in the right - hand side are physical quantities . we will provide a gauge - invariant definition of the quark / baryon - charge chemical potential in the next subsection . in this section , we construct a gauge - invariant formulation of the @xmath0-charge chemical potential in ads / cft . a gauge - invariant definition of the chemical potential has already been introduced in refs . however , the aim of this section is to present a mathematically consistent framework by showing how the manifestly gauge - invariant formulation is obtained starting with the standard ads / cft dictionary ( [ dictio-2 ] ) . in the grand - canonical ensemble , the grand potential can be evaluated from an on - shell euclidean bulk lagrangian obtained by fixing the chemical potential at a given value . let us begin with a bulk lagrangian given by @xmath66 where we have explicitly introduced a lagrange multiplier , denoted by @xmath47 , that defines the chemical potential . we omit the lagrangian of the bulk gravity , since it is independent of the @xmath0-charge under the probe approximation . ( [ bulk - action ] ) can be re - written in a way that makes its ads / cft antecedents explicit : @xmath67 the last term is the usual bulk - boundary coupling in ads / cft which directly allows us to interpret @xmath47 as the conserved @xmath68 charge of the field theory . the first two terms then may be viewed as the lagrangian of the d7-branes with a charged source at @xmath35 . we note that in the present background ( with the only gauge field present being @xmath69 ) , the d7-brane lagrangian does not have a coupling of the form @xmath70 ; thus latter term must be due to some other dynamical object . we will discuss the nature of the source term later . the third term is understood in the following way . in field theory , the grand - canonical partition function is defined as @xmath71=\int d\phi \ : dq \exp[-\beta ( l-\mu q)],\ ] ] where @xmath47 is the charge and @xmath72 denotes the fields whose lagrangian is @xmath41 . the third term in the bulk action then corresponds to the field theoretic term @xmath73 which is present in the grand - canonical ensemble . in the grand - canonical ensemble , the amount of the charge fluctuates and its expectation value may be determined ( approximately ) by the saddle point value of the partition function . in bulk terms , this is equivalent to determining the value of @xmath47 by minimizing the bulk action ( [ bulk - action ] ) at the given chemical potential . thus , the saddle - point value ( the on - shell value ) of the bulk action is identified with the grand - potential of the dual field theory . note that , in the usual ads / cft correspondence , field theory quantities are defined by boundary values of bulk fields . our definition of the chemical potential in ( [ bulk - action ] ) is however @xmath74 to understand how this can be the case , it is worthwhile recalling how the grand potential in the r - charged system has been computed in terms of gravity dual . in ref . @xcite , the grand potential is identified with the on - shell value of the einstein - maxwell - anti - desitter effective lagrangian in the bulk computed by solving the equations of motion with the boundary value of the time component of the vector potential kept fixed . this matches the picture that the control parameter of the grand - canonical ensemble , the r - charge chemical potential in this case , is given by the boundary value of the potential . let us start with a slightly different bulk action by following this spirit : @xmath75 in this action , we have a source of strength @xmath47 at @xmath35 , while the chemical potential has been defined to be the boundary value of the bulk gauge field using a lagrange multiplier @xmath76 ; our ads / cft dictionary is ( [ dictio-2 ] ) at this stage . however , demanding gauge invariance forces @xmath77 . then , extremizing over @xmath76 leads us to the definition of chemical potential as in ( [ chem - def ] ) . on the other hand , if we first extremize over @xmath76 , then it is easily seen that the chemical potential is obtained as the asymptotic value of @xmath78 , which in turn is obtained by solving the equations of motion of the gauge field in the presence of a charged source @xmath47 at @xmath35 . @xmath79 is then obtained as a consequence of flux conservation . the difference , of course , is that it is only in the first case that we have off - shell gauge invariance . in the grand - canonical ensemble , since the charge q fluctuates , the off - shell gauge invariance is required . thus we must first impose gauss law and then extremize over the lagrange multiplier . this explains how , starting from the standard ads / cft prescription , we arrive at the definition ( [ chem - def ] ) . indeed , the definition of the chemical potential ( [ chem - def ] ) is consistent with that in ref . @xcite since @xmath80 is set to be zero there . one interesting observation is that the right - hand side of ( [ chem - def ] ) is nothing but a work necessary to bring a unit charge from the boundary to @xmath35 against the electric field @xmath81 along the worldvolume of the d7-brane . it would be interesting if we can connect this picture to some physical process in the ym theory . now the dictionary ( [ dictio-2 ] ) is modified to be @xmath82 this new recipe also matches the standard computation of the grand potential in thermal field theories . let us rewrite @xmath83 to be @xmath84 where @xmath85 + q\left\{a_{0}(\infty)-a_{0}(\rho_{min})\right\}.\end{aligned}\ ] ] the grand potential is the on - shell value of @xmath86 obtained by solving the equations of motion of the fields and @xmath47 . this is nothing but the standard computational recipe of the grand potential if we regard @xmath87 corresponds to the effective lagrangian of the field theory . indeed , @xmath87 is more consistent than @xmath88 as a d7-brane lagrangian since the conservation of the electric flux is manifest . in summary , the grand potential is determined by the following steps in the gravity dual : 1 . we start with @xmath83 . the chemical potential @xmath60 is our input parameter . 2 . we solve the equations of motion . the boundary condition for @xmath63 is given by our input @xmath60 through @xmath89 that comes from equation of motion of @xmath47 . the on - shell value of @xmath83 ( see appendix [ regularization ] for the renormalization ) gives the grand potential ( density ) @xmath90 of the flavor part . the on - shell value of @xmath47 is the thermal expectation value of the quark - charge density . we employ this gauge - invariant formulation of the chemical potential in the present paper , and we choose @xmath91 gauge in the remaining part . a connection between the equations of motion and a thermodynamic relation can also be seen in the following way . the equations of motion of @xmath63 from ( [ bulk - action ] ) are @xmath92 solving the differential equation ( [ eom1 ] ) with boundary conditions ( [ bound - cond-1 ] ) and ( [ bound - cond-2 ] ) , we obtain @xmath93 which suggests @xmath94 . since @xmath47 does not depend on @xmath15 , we have @xmath95 which is equivalent to the thermodynamic relation @xmath96 here we have used the definition of the chemical potential ( [ chem - def ] ) and the fact that the on - shell value of @xmath83 is given by that of @xmath97 . it is interesting that the gauss - law constraint ( [ gauss ] ) in the `` electro - magnetism '' on the flavor brane corresponds to the thermodynamic relation ( [ thermo - rel ] ) in the ym theory . the coefficient @xmath54 in ( [ dictio ] ) can also be determined by the boundary condition ( [ bound - cond-1 ] ) . substituting the asymptotic expansions ( [ dictio - inv ] ) and @xmath98 , @xmath99 to make the foregoing expression to be consistent with our identification @xmath100 , @xmath54 is determined to be @xmath101 the helmholtz free energy @xmath102 per unit volume is obtained by the legendre transformation @xmath103 . this is nothing but a procedure to obtain an _ on - shell _ `` hamiltonian '' ( if we regard @xmath15 as a `` time '' formally ) in terms of the d7-brane dynamics : @xmath104_{on - shell}. \label{legendre}\end{aligned}\ ] ] here in the right - hand side , @xmath105 is interpreted as the `` conjugate momentum '' of @xmath63 by ( [ gauss ] ) , and all the quantities in ( [ legendre ] ) are the on - shell values . then , we can make the following recipe in the _ canonical ensemble_. 1 . we define the hamiltonian @xmath106 where @xmath107 and @xmath108 are not necessarily at on - shell , but the gauss - law constraint ( [ gauss ] ) is imposed to eliminate the electric field in favor of @xmath47 . now our input parameter is @xmath47 but not @xmath60 . 2 . we solve the equations of motion to find the on - shell value of @xmath107 which gives @xmath102 . the thermal expectation value of @xmath60 is given by @xmath109 , where @xmath81 is obtained from the gauss - law constraint ( [ gauss ] ) . the hamiltonian for our system is explicitly given by @xmath110 the equation of motion after eliminating the gauge field is written explicitly by @xmath111 and the gauss - law constraint in the explicit form is @xmath112 notice that @xmath113 can be regarded as an effective tension of the d7-branes which is affected by the presence of the gauge field . in particular , the effective tension becomes larger if the charge increase , and the brane is attracted to the black - hole more strongly . therefore if we keep the current quark mass fixed , the minimum `` height '' @xmath34 of the d7-brane decreases as we increase the charge density . it may be useful to define @xmath114 since both @xmath113 and ( [ f - tilde ] ) are written in terms of @xmath115 . we may use @xmath115 as a parameter of the system in the canonical ensemble in the later analysis if it is convenient . before closing this section , we would like to make a few comments . the gauss - law constraint in the form of ( [ f - tilde ] ) gives a relationship between @xmath60 and @xmath47 : an equation of state . equation ( [ f - tilde ] ) itself can be used both in the grand - canonical ensemble and the canonical ensemble . we should be careful only in the interpretation of the variables : in the ( grand- ) canonical ensemble , @xmath60 ( @xmath47 ) obtained from ( [ f - tilde ] ) is the thermal expectation value while @xmath47 ( @xmath60 ) is the control parameter of the theory . another comment is about the framework presented in ref . there is a `` quick '' computational method based on our formalism . since the on - shell value of @xmath83 is the same as that of @xmath88 , the correct grand potential is also obtained by starting with @xmath88 alone , and solving the equations of motion by imposing the gauss - law constraint @xmath116 by hand . this is nothing but the computational method employed in ref . this means that ref . @xcite produces the results of the present framework . the last comment is about the consistency of the present formalism . we would like to stress that the charged source whose necessity is pointed out in ref . @xcite has already been introduced at the stage of ( [ d7+source ] ) . thus , there is no need for introducing any further sources , and all equations are correctly satisfied . now , we would like to bring our attention to the claim presented in ref . @xcite . the authors of ref . @xcite pointed out the necessity of the charged source at @xmath117 , and they provided it by adding @xmath0-charge carrying objects to the system . their idea is to put fundamental strings ( f1 s ) between the d7-branes and the black - hole horizon , that are interpreted as quarks . then , they found that the minkowski embeddings in ref . @xcite are unstable due to the tension of the f1 s and there is no way to keep the d7-branes off the horizon . this is why the minkowski embeddings at finite baryon - number density are concluded to be unphysical in ref . @xcite . however , one should notice that this is not what we are doing in the present paper . the necessary charged source term has been introduced as an _ external _ source to the d7-brane dbi theory at ( [ d7+source ] ) and we have not added any corresponding nambu - goto action of the f1 there . this means the system we are dealing with is something different from that in ref . since we have not introduced the additional f1 s , we expect that in the present framework , the minkowski embeddings are physical as well . in this section , we will show that it is indeed the case . we will present numerical results , @xmath118 so that @xmath119 and @xmath120 in the numerical analysis . the numerical values of the free - energy densities @xmath102 and @xmath90 are in the unit of @xmath121 , those of the chemical potential and @xmath47 are in the unit of @xmath122 and @xmath49 , respectively . the unit of the entropy density and the quark condensate is also @xmath49 in the numerical analysis . ] which indicate consistency of the present formalism and the necessity of the minkowski embeddings . a physical interpretation of our setup will be proposed in section [ discussion ] . in order to show a consistency of our framework , we will employ @xmath123-@xmath47 diagrams where the relationship between @xmath60 and @xmath47 obtained from ( [ f - tilde ] ) is drawn . ( we may use @xmath123 as the meaning of @xmath60 . ) let us present basic explanations on the @xmath123-@xmath47 diagram in this subsection as preparation . an example of the @xmath123-@xmath47 diagram is given at fig . [ fig : ptq01 ] ( or at fig . [ fig : ptm01 ] ) @xcite . the d7-brane solutions that belong to the minkowski embeddings start from @xmath124 ( which is the origin of the plane ) and go through @xmath125 and @xmath126 until @xmath127 where the line meets the vertical axes . the black - hole embeddings exactly start from @xmath127 where the minkowski embeddings terminate , and go through @xmath128 , @xmath102 , @xmath129 , @xmath130 , @xmath131 and extend to the large-@xmath47 and large-@xmath123 region . it is worthwhile checking a consistency of the @xmath123-@xmath47 diagram . [ fig : fq01 ] and fig . [ fig : am01 ] show the relationship between the helmholtz free energy density and @xmath47 , and that between the grand potential density and @xmath123 , respectively . although the thermodynamic potentials obtained from all the possible solutions are indicated , what we should take is the line which has the minimum value . then , we find first - order phase transitions . let us see fig . [ fig : am01 ] , for example . the phase transition is a jump between a minkowski embedding and a black - hole embedding at the critical chemical potential @xmath132 . since the grand potentials in both embeddings have the same values at @xmath132 , the integral @xmath133 below @xmath132 and that above @xmath132 have to be same because the integrals compute @xmath134 by virtue of @xmath135 . this is the maxwell construction . one can see that the maxwell construction works in a non - trivial way in good accuracy in fig . [ fig : ptm01 ] . we have checked numerically that the area of the shaded regions marked `` @xmath136 '' agrees with the area of the regions with `` @xmath137 '' , due to a non - trivial collaboration of the minkowski and the black - hole embeddings . we can also see that the maxwell construction works in the canonical ensemble in fig . [ fig : ptq01 ] , again by virtue of the collaboration of the two types of the embeddings . now , we are ready to make some important comments based on the numerical results . fig . [ fig : muq ] shows that how the @xmath123-@xmath47 diagram is deformed if we make the temperature lower . the temperature goes down as we move from fig . [ fig : muq01 ] to fig . [ fig : muq04 ] . an important feature is that the cusp @xmath130 located at the origin on fig . [ fig : muq01 ] `` goes up '' ( fig . [ fig : muq02 ] ) and disappears ( fig . [ fig : muq03 ] and fig . [ fig : muq04 ] ) along the cooling process . the cusp on fig . [ fig : muq02 ] gives the minimum value of the chemical potential within the black - hole embeddings . this means that the low - chemical - potential region disappears from the parameter space of the theory at the sufficiently low temperature if we abandon the minkowski embeddings . we can also interpret the behaviour of the diagram in a different way . suppose that we are in the grand - canonical ensemble and we examine a process in which we vary the temperature with maintaining the chemical potential . then , we encounter a problem that there is no low - temperature region in our parameter space at sufficiently small chemical potential . the lack of the low - temperature region can be seen more clearly in fig . [ mubranch ] , where the d7-brane solutions on the @xmath41-@xmath138 plane are given . the minkowski embeddings ( @xmath39 ) and the black - hole embeddings ( @xmath139 ) are connected at @xmath140 there . notice that all the temperature region is covered by using the two types of embeddings . however , if we abandon the minkowski embeddings , the theory can not cover the low - temperature region . ( recall that @xmath141 . ) we can conclude from the above that if we abandon the minwkowski embeddings , the low - temperature and the low - chemical - potential region in the grand - canonical ensemble does not exist in the formalism ; we have no way to introduce the flavor degree of freedom in that region . this is an incomplete - ness of the formalism . on the other hand , if we include the minkowski embeddings together with the black - hole embeddings , all the parameter region of the theory is covered in harmony of the two types of embeddings . we can also see that the two embeddings are connected into a single family of the solutions on fig . [ fig : ptc ] , fig . [ fig : muq ] and fig . [ mubranch ] . we present some numerical results in the grand - canonical ensemble in appendix [ grand - more ] to make the above points vivid . incidentally , we can understand why we did not encounter the above `` incomplete - ness '' problem manifestly in ref . let us look at fig . [ fig : muq ] again . we notice that the entire region of @xmath47 can always be covered even if we use only the black - hole branch . this means that if we examine the process in which we vary the temperature with fixing @xmath47 , we have always at least one solution inside the black - hole branch at any temperature ; all the region on the @xmath47-@xmath142 plane can be covered even if we use only the black - hole embeddings . namely , we do not have a missing parameter region in the _ canonical _ ensemble . the reason why we do not see the incomplete - ness in ref . @xcite in a manifest way is that the analysis is given in the canonical ensemble there . however , our claim is that the canonical ensemble with only the black - hole embeddings is transformed to the incomplete formalism of the grand - canonical ensemble through the legendre transformation : the formalism lacks something necessary even in the canonical ensemble . in the next section , we will see on this point and we will find the minkowski embeddings play an important role in the canonical ensemble as well . we will further look into the importance of the minkowski embeddings by investigating the thermodynamic properties of the system in this section . a key issue here is thermodynamic stability of the system . we will examine the thermodynamic stability based on the following two conditions in the canonical ensemble : 1 . @xmath143 : + if this condition does not hold , the system can lower its free energy by separating into two phases with densities @xmath144 @xcite ; the baryon - charge density tends to be inhomogeneous . we call this instability `` number - density instability '' in this paper . existence of this instability in the d3-d7 systems has been originally discovered in ref . @xmath145 : + @xmath146 denotes the entropy density of the flavor part . this condition is equivalent to the positivity of the specific heat ( in the canonical ensemble ) , hence it has to hold . the foregoing conditions have also been examined with the black - hole embeddings in ref . @xcite based on the canonical ensemble . the number - density instability has been found , while the instability based on the condition 2 has not been found there . a detailed study on thermodynamic instability at zero charge density is found in ref . @xcite . in the present setup , we will also find the number - density instability . however , we will see that the minkowski embeddings provide a stable final state . we will also comment on the instability based on the condition 2 . let us go back to the @xmath123-@xmath47 diagram , fig . [ fig : ptq01 ] , and examine the number - density stability in the canonical ensemble . there is a region where @xmath147 between points @xmath127 and @xmath129 @xcite . ( we call this region `` black - hole b branch . '' see for the details , appendix [ structure - can ] . ) the region between @xmath128 and @xmath102 has the lowest free energy if we assume the charge density is uniformly distributed . however , the free energy may be lowered by allowing the system to be a mixture of two phases of different densities , as it is discussed in ref . @xcite . let us remind us of the necessary conditions to form a stable final state as a mixture of two types of domains of different phases . since the charge can move from one domain to the other , the condition for the stability of each domain is what we may usually use in the context of the grand - canonical ensemble : the chemical potentials of the two domains have to be equal to each other in order to achieve thermal equilibrium , and the condition @xmath143 has to hold in _ each _ domain to make the thermal equilibrium to be stable . in the present setup , we have two first - order phase transitions at @xmath148 ( from @xmath128 to @xmath125 ) and at @xmath149 ( from @xmath102 to @xmath131 ) where the system jumps into the stable phases . since the chemical potential at @xmath125 is larger than that of at @xmath131 , the final value of the chemical potential should be somewhere between the two , namely the points indicated by @xmath150 and @xmath151 in fig . [ fig : ptm01 ] . therefore , the final state can be realized as a _ stable _ mixed phase of the minkowski phase at @xmath150 and the black - hole c phase ( see for the details , appendix [ structure - can ] ) at @xmath151 . however , if we do not have the minkowski embeddings as the model in ref . @xcite , the low - density region is not bounded by any stable phase , and one of the domains which may form the mixed phase is still in the unstable black - hole b branch . this is a problem : there is no way to achieve the thermodynamic stability in this case . indeed , the problem is seen as absence of phase transition in the grand - canonical ensemble . points @xmath150 and @xmath151 in fig . [ fig : ptm01 ] which forms the stable mixed phase in the canonical ensemble are nothing but the points we have a phase transition from the minkowski to the black - hole embeddings in the grand - canonical ensemble . if we remove the minkowski embeddings , we have no phase transition in the grand - canonical ensemble . it would be useful to present the phase structure of the system both in the canonical and the grand - canonical ensemble to overview the above mentioned properties . the phase diagrams in the canonical ensemble are shown in fig . [ fig : tq ] . ( see also ref . @xcite . ) as it has been reported in ref . @xcite , there are two phase transitions in the small region of @xmath152 the phase transitions are of the first order except for the endpoint of the upper line at @xmath153 where a second - order phase transition takes place . a part of the phase diagram is zoomed in fig . [ fig : ptinset ] . the detailed analysis shows that the three different phase boundaries meet at @xmath154 with different slopes . here , the low - temperature phase is the minkowski phase , the phase between the two transitions is the black - hole b phase and the high - temperature region is the black - hole c phase . ( see appendix [ structure - can ] , for the detailed classification of the phase . ) the region where the number - density instability exists is indicated as a shaded region in fig . [ fig : tq ] . we can see that the unstable region is surrounded by the stable regions . this means that the system can reach a final stable mixed phase in any case . if we remove the minkowski embeddings , the minkowski phase disappears . the lower phase boundary disappears and we have only one phase transition ( this is essentially what is reported in ref . @xcite ) . in this case the unstable region reaches the @xmath155 axes and it is not bounded by the stable region @xcite . the phase structure in the grand - canonical ensemble is presented in fig . [ fig : ptgc ] . phase diagram in the grand - canonical ensemble , title="fig:",scaledwidth=45.0% ] a significant difference from the canonical - ensemble case is that we have only one phase transition , as one may expect from the discussion based on fig . [ fig : ptm01 ] and fig . [ fig : am01 ] . the low - temperature phase is the minkowski phase while the high - temperature phase is the black - hole phase . the phase transition is of the first order and the value of @xmath47 jumps at the transition point ; the unstable region we found in the canonical ensemble is `` skipped '' by the jump at the transition ( see fig . [ fig : ptm01 ] again ) . we did not find any number - density instability within our analysis in the grand - canonical ensemble . if we remove the minkowski embeddings , the minkowski phase disappears and the system has no phase transition . furthermore , the low - temperature low - chemical - potential region will not be covered by any embeddings . we can see the cooperation of the minkowski and the black - hole embeddings in the grand - canonical ensemble vividly in appendix [ grand - more ] . here , we would like to comment on the temperature dependence of the entropy density @xmath146 . we may use @xmath156 rather than @xmath157 , where @xmath102 is the helmholtz free energy density computed only from the dbi theory of the flavor d7-branes and @xmath158 is the quark condensate . ( [ st - eom ] ) is derived by using the fact that the free energy is a function of @xmath159 and @xmath160 in the form of @xmath161 . ( [ st - eom ] ) may be technically more useful than @xmath157 in the numerical analysis since we need not to differentiate the numerical data . notice that the above entropy density is not that of the total system . since we are not taking account of the bulk ( adjoint fields ) free energy , the entropy we discuss here is that of the flavor part . a typical , but preliminary result on the temperature dependence of the entropy density in the large-@xmath47 region is plotted in fig . [ fig : qentropy ] . we find that the entropy still remains finite at @xmath155 in the black - hole phase at this parameter region . this means that the third law of thermodynamics will be broken if we abandon the minkowski embeddings . indeed , @xmath162 that counts the degree of freedom per unit volume of the system diverges at @xmath155 in the black - hole phase . however , this is not physically realized in our setup since the system at @xmath155 is in the minkowski phase where the third law of thermodynamics seems to hold . this is another example of the role of the minkowski embeddings . one may notice that we still have a region of @xmath163 in the black - hole phase even if we take the minkowski embeddings into account . the specific heat of the flavor brane in this region is negative and the flavor part is not at a stable thermal equilibrium . however , the entropy density we analyzed is only that in the flavor part ; a stable phase may be realized after interaction of the flavor part and the bulk part . this implies that we may need to go beyond the probe approximation . ( this also matches the fact that this instability is observed only in the large-@xmath47 region where the d7-brane s effective tension @xmath113 becomes large . ) because of this reason , the statement in this subsection should be understood as a preliminary comment . we leave detailed study on this negative - slope behaviour for future work . we have presented a gauge - invariant formulation of @xmath0-charge chemical potential in ads / cft . the framework was constructed in a mathematically closed way . we have also numerically examined the properties of the system . the results indicate that it is consistent to employ all the possible types of the d7-brane embeddings in our framework , and the necessity of the minkowski embeddings has been shown . then , how can we reconcile the claim in ref . @xcite with our formalism ? the purpose of this section is to present one possible physical interpretation of our framework , and to clarify the difference between our system and that of ref . @xcite . to this end , let us consider the lagrangian @xmath164 the second term which is the source term does not arise from any known coupling of the d7-brane ( in this background ) . the remaining terms however , can be argued for , from the standard ads / cft correspondence ( as was done earlier ) . thus , one is forced to regard the second term as coming from a charged object which intersects the d7-brane at @xmath35 . the authors of ref . @xcite have asked what is the physical object carrying the @xmath0-charge in the gravity - dual side . their proposal is that this object is a fundamental string connecting the d7-branes and the black - hole horizon which corresponds to a single quark in the ym theory . hence they added the fundamental strings ( f1 s ) further to the system . the bulk action for the system in ref . @xcite in our language is then , @xmath165 where @xmath166 is the nambu - goto lagrangian of the f1 , and @xmath167 represents the contribution of the bundle of the f1 s that are homogeneously distributed in the three - dimensional space with number density @xmath47 @xcite . in this picture , the second term in ( [ action - prelim-2 ] ) is induced by the end points of the f1 s on the d7-branes . now the equations of motion of the system will be modified because of @xmath167 . in particular , the effect of the tension of the bundle of the f1 s has to be taken into account ( this force balance condition may be regarded as arising from the surface terms in the equations of motion ) . the conclusion of ref . @xcite is that the bundle s tension is always strong enough to pull down the d7-branes to the horizon , and we have only black - hole embeddings if we start from ( [ kmmmt - action ] ) . however , this is clearly different from what we are doing in this paper . since the nambu - goto lagrangian is absent from our action ( [ action - prelim-2 ] ) , there is no force - balance problem between the d7-branes and the f1 s , and there is no reason to abandon the minkowski embeddings _ a priori _ in our framework . then what is the physical interpretation of our setup ? let us look into ( [ kmmmt - action ] ) in more details to clarify the physical interpretation . if we start from ( [ kmmmt - action ] ) and carry out constructing the formalism ( by regarding @xmath47 as a legendre multiplier ) , the definition of the chemical potential will be modified to @xmath168 the mass of the f1 ( which is defined to be @xmath166 here ) is now included in the definition of the chemical potential . here we should notice that we could have added baryons instead of quarks to the system a la ref . a baryon is represented by a d5-brane wrapped on the @xmath169 with @xmath4 fundamental strings attached @xcite , which is often called a baryon vertex . ( see also refs . @xcite . ) if we add ( a bundle of ) the baryon vertices instead of the f1 s , the second term in the right - hand side of ( [ action - prelim-2 ] ) is understood as a term induced by the chern - simons term of the d5-brane action . we should also add the `` baryon - vertex lagrangian '' ( d5-brane dbi lagrangian and nambu - goto lagrangians of the fundamental strings that constitute the baryon vertex ) to ( [ action - prelim-2 ] ) instead of the f1 s nambu - goto lagrangian : @xmath170 where @xmath171 is the lagrangian of the baryon vertex . now the definition of the chemical potential will be modified to be @xmath172 . however , this is not what we are doing , either . the point is that the definition of our chemical potential @xmath60 is blind to what the baryon - charge carrying object is . therefore , one natural interpretation of our formalism is that we are dealing with a system of mesons with an _ external _ @xmath0-charged source in the canonical ensemble ( and its conjugate system in the grand - canonical ensemble ) . this is why we have used an expression `` @xmath0-charge chemical potential '' rather than `` baryon chemical potential '' in this paper . this interpretation comes form the fact that the d7-brane s dbi action provides an effective action of mesons , and we have inserted only the charged source term to it in the gravity - dual side . one may wonder how to imagine the system of mesons with an external @xmath0-charged source . an intuition may be obtained by considering so - called walecka model ( or the @xmath173-@xmath174 model ) @xcite . the walecka model is a model of nucleon - meson systems . the baryon - number current couples to the @xmath174-meson through the yukawa coupling in that model . therefore , the baryon - number density ( which is charged under the @xmath0 symmetry ) can act as a source for the @xmath175 mesons there . we will not try to connect the present setup with any particular phenomenological model of the sym theory in this paper . however , our baryon - number current can couple to our mesons in the same way as it does in the walecka model , in principle .- meson field by a constant and there is no effect on the scalar mesons ( @xmath173 mesons ) because of the absence of direct @xmath173-@xmath174 coupling . however , in the present d3-d7 setup , we do have couplings between the scalar mesons and the vector mesons due to the non - linearity of the d7-brane dbi action . this is a reason why the presence of @xmath0 charge can change the shape of the d7-branes . ] let us re - interpret what we have observed based on the above interpretation . first of all , the existence of the physical minkowski embeddings is explained . since we are dealing with a system of mesons with an external @xmath0-charged source inserted , we _ need not _ to add any @xmath0-charge carrying dynamical object to the system by hand . therefore , the fundamental strings that make the minkowski embeddings to be unstable can never come into the present setup by definition . the instability of the minkowski embeddings discussed in ref . @xcite may also be interpreted in the following way . the color non - singlet quarks have been added to the phase ( minkowski branch ) where the physical mesons exists . the instability of the minkowski embeddings with f1 s may correspond to the instability of a confinement phase with color non - singlet quarks . the reason why the framework in ref . @xcite is incomplete may also be understood . if we wish to analyze the system which contains physical baryon - number charged particles , we also have to examine what happens if we add the baryon vertices instead of the f1 s to the minkowski embeddings . this analysis has not yet been done , and what we should do at this stage is to postpone the conclusion about the fate of the minkowski embeddings rather than to abandon them , until the analysis with baryons is accomplished . an argument that the charged source can be d5-branes is the following . let us imagine that we introduce charged sources keeping the d7-branes fixed . we can then compare the energies of various candidate sources . that is to say , we compare the energy of @xmath176 f1 s stretching between @xmath177 and the horizon , and the energy of a single d5-brane ( wrapping the @xmath178 ) placed at @xmath179 . we can immediately see , that if @xmath138 is sufficiently large , d5-branes are lower in energy@xcite . thus , baryons are more appropriate as charged sources for low enough temperature ( i.e. , large enough @xmath138 ) . ( a related argument that d5-branes with attached fundamental strings can sit outside the horizon in the black - hole geometry for low enough temperatures has been demonstrated in @xcite ) . another candidate source is d1-branes that are transverse to the d7-branes ( i.e. , wrap the @xmath180 direction of the metric ( [ metric ] ) ) . it can be argued that the open strings from the d1-branes to the d7-branes are fermionic @xcite . further , the d1-branes can form a fuzzy @xmath178 by a version of the myers effect @xcite . indeed we have a preliminary observation of what happens if we add baryon vertices to the minkowski embeddings . in fig . [ fig : mod - ft ] , the free energy densities with / without adding the d5-brane mass are shown there . [ fig : mod - phase ] shows the phase diagram in the canonical ensemble with / without adding the d5-brane mass . the d5-brane s mass is computed based on the assumption that the configuration of the d5 is spherical and it is attached to the minkowski embedding at @xmath181 without taking account of interactions between the d7-brane ( see , for the details , appendix [ baryon - mass ] ) . the correction to the black - hole embeddings is zero under this assumption since the d5-brane is on top of the horizon . this is why the line of the phase transition between the black - hole embeddings ( between the b - branch and the c - branch ) is not modified in fig . [ fig : mod - phase ] . we understand that the assumption employed here may be invalidated if we analyze the dynamics of the full system of the d5 s and the d7-branes . however , what we would like to comment from fig . [ fig : baryon - add ] is that the naive estimation of the d5-brane s energy is not too large to make the phase transition impossible ; the modified free energy of the minkowski embeddings still intersects that of the black - hole embeddings . this suggests that we have a chance to obtain a complete setup for the finite density systems of dynamical baryons , by including the baryon - vertex d5-branes in an appropriate way . notice that our estimation of the d5-brane mass qualitatively matches what is discussed in ref . @xcite ; the d5-brane mass is a decreasing function of @xmath47 since @xmath34 is . improvement of the framework in this direction is certainly an important subject . finally , we would like to comment on the location of the charge on the d7-branes . we have assumed that the location of the charge on the d7-branes is @xmath117 at ( [ d7+source ] ) . however , there is a freedom in the choice of the location in principle . for example , the charge can be delocalized in the @xmath15-direction , or the position of the charge can be different from @xmath117 . if we take the picture where we have dynamical charged objects , the distribution of the charge in the @xmath15-direction may be determined by the dynamics of the d7-branes and the charge - carrying objects . however , since the charge in the this paper with the above interpretation is non - dynamical , the distribution should be given by hand in the present setup . one may wonder how to determine the location of the charge in the @xmath15-direction and what is its physical interpretation . in this paper , we have taken the most simplest choice for the charge distribution , and we leave these problem open . all the points discussed in this section have to be clarified more in further studies . + note added : while the manuscript of the present paper is prepared , we received related papers @xcite , @xcite , @xcite and @xcite . * acknowledgments * the authors thank to kazuo ghoroku , deog ki hong , masafumi ishihara , youngman kim , eiji nakano , kazuaki ohnishi , mannque rho , shigeki sugimoto and ho - ung yee for useful discussions . this work was supported by kosef grant r01 - 2004 - 000 - 10520 - 0 and the src program of the kosef through the center for quantum space - time of sogang university with grant number r11 - 2005 - 021 . the brane embeddings in our system have rich variety . we will present the detailed classification of the brane embeddings and the phases based on the canonical ensemble process in the canonical ensemble.]in this appendix . + * * + the key feature of the minkowski embeddings is captured by fig . [ fig : y0min ] , where we plot @xmath138 , the value of @xmath33 at @xmath182 , against the asymptotic height @xmath183 . vs. @xmath34 for the minkowski embeddings for various @xmath184 . [ fig : y0min ] , title="fig:",scaledwidth=45.0% ] in the case of zero temperature where we do not have black hole in the bulk , the equations of motion become very simple . one can easily show that the obtained brane configuration is just horizontal , namely @xmath185 , regardless of the value of @xmath47 . let us consider more general setup for finite temperature cases . for large values of @xmath138 , the branes are hardly affected by the black hole and the brane profile is pretty much similar to the horizontal configuration of the @xmath155 case . this means that @xmath41 decreases linearly ( with the slope @xmath186 ) regardless of the value of @xmath115 as we decrease @xmath138 at large @xmath138 ( see the large-@xmath138 region in fig . [ fig : y0min ] ) . however , there is a critical value @xmath187 at which @xmath41 reaches a minimum . for smaller values of @xmath138 , the branes seem strongly `` repelled '' by the black hole and @xmath41 increases again steeply . we will refer to those minkowski embeddings in the region of @xmath188 as _ minkowski a branch _ and those in the region of @xmath189 as _ minkowski b branch_. these two branches exist even when @xmath190 . the a - branch always has lower free energy than the b - branch , hence the b - branch will not be physically realized . as we increase @xmath115 , @xmath187 moves to a larger value and the corresponding minimum value of @xmath41 also increases . this means we do not have any minkowski embeddings in the region of sufficiently large @xmath115 and sufficiently high temperature , in the canonical ensemble . + * * + the relationship between @xmath138 and @xmath41 for the black - hole embeddings is shown in fig . [ fig : y0bh1 ] . vs. @xmath34 for the black - hole embeddings for various @xmath115.[fig : y0bh1 ] , title="fig:",scaledwidth=45.0% ] as we increase @xmath34 from zero , @xmath41 initially increases linearly with the common slope regardless of the value of @xmath191 . for very small densities @xmath192 ( that corresponds to @xmath193 ) , @xmath41 attains a local maximum ( at @xmath194 ) , and then decreases to a local minimum ( at @xmath195 ) . this cubic behaviour is reminiscent of the van der waals @xmath196 diagram . the local maximum of @xmath41 is present even for @xmath190 . as we increase @xmath115 , @xmath194 becomes larger while @xmath195 goes smaller , and they merge at the critical value of @xmath197 ( that corresponds to @xmath153 ) . this is again analogous to the behaviour of the van der waals @xmath196 curves with @xmath115 playing the role of temperature . when @xmath34 becomes nearly @xmath198 , @xmath41 increases sharply . in the language of configuration , the worldvolume starts from the horizon and grows rapidly in the @xmath33 direction within a short range in @xmath15 like the minkowski b - type branes . we will classify the black - hole embeddings into three branches when we have the local maximum / minimum . those in the region of @xmath199 are called as black - hole c branch . those in the region @xmath200 will be referred to black - hole d branch , and those with @xmath201 is defined as black - hole b branch . if we do not have the local maximum / minimum , there is no further classification of the black hole embeddings . ( we may call them as black - hole a branch . ) the d - branch always has higher free energy than that of the black - hole b or c branch , hence it will not be realized as a physical phase . the black - hole b branch can have the lowest free energy among them in some case , however it has the number - density instability discussed in section [ numb - dens - instability ] . we present some more numerical results in order to make the collaboration of the minkowski and the black - hole embeddings vivid . temperature dependences of various quantities are shown in fig . [ fig : grand ] . here , the entropy density ( of the flavor part ) is computed by using @xmath202 where @xmath203 is the quark condensate . each diagram is represented as a connected line within the resolution of the numerical data that consists of the minkowski branch ( low - temperature region ) and the black - hole branch ( high - temperature region ) . because of the cusp - shaped structure ( see fig . [ fig : entro2 ] , for example ) , the physical value of the entropy density , the number density and the quark condensate jump at the critical temperature of the ( first - order ) phase transition . we will not get into the details of the physical interpretation of the results in this paper , but one can see how the minkowski and the black - hole embeddings cover the full temperature region from these examples . the on - shell value of @xmath204 is in fact divergent due to the integral over @xmath15 from @xmath35 to infinity . this ir divergence in the bulk is understood as a uv divergence of the boundary theory and therefore we need to renormalize it . we introduce a cut - off @xmath205 . one can check that the integrand @xmath206 behaves @xmath207 the renormalization scheme we adopt is the minimal subtraction , namely we add the counter term @xmath208 to @xmath204 . indeed , our renormalization method is the same as that of the holographic renormalization given in ref . @xcite . the mass of single d5-brane spherically wrapped on the @xmath169 is given by @xmath209 where @xmath210^{-1}$ ] is the tension of the d5-brane and @xmath211 is the volume of the @xmath169 . substituting the metric ( [ adsm ] ) and the constants into ( [ mb-1 ] ) , we obtain @xcite @xmath212 where @xmath213 is the location of the baryon vertex in the @xmath16 direction . it is convenient to express @xmath214 by using the @xmath215-coordinate . notice that we have the following relations : @xmath216 where we have assumed that the location of the d5-brane on the @xmath182 axes is @xmath34 . plugging @xmath217 into the formulae , we obtain @xmath218 the correction we added to the helmholtz free energy of the minkowski embeddings in section [ discussion ] is @xmath219 . j. m. maldacena , adv . * 2 * 231 ( 1998 ) ; arxiv : hep - th/9711200 . s. s. gubser , i. r. klebanov and a. m. polyakov , phys . b428 * 105 ( 1998 ) , arxiv : hep - th/9802109 ; e. witten , adv . * 2 * 253 ( 1998 ) , arxiv : hep - th/9802150 ; o. aharony , s. s. gubser , j. m. maldacena , h. ooguri , and y. oz , phys . rept . * 323 * 183 ( 2000 ) . r. a. janik and r. peschanski , phys . d * 73 * 045013 ( 2006 ) , arxiv : hep - th/0512162 ; s. nakamura and s. j. sin , jhep * 0609 * 020 ( 2006 ) , arxiv : hep - th/0607123 ; s. j. sin , s. nakamura and s. p. kim , arxiv : hep - th/0610113 ; r. a. janik and r. peschanski , phys . rev . * d74 * 046007 ( 2006 ) , arxiv : hep - th/0606149 ; r. a. janik , phys . lett . * 98 * 022302 ( 2007 ) , arxiv : hep - th/0610144 ; d. bak and r. a. janik , phys . lett . * b645 * 303 ( 2007 ) , arxiv : hep - th/0611304 ; m. p. heller and r. a. janik , arxiv : hep - th/0703243 ; y. v. kovchegov , a. taliotis , arxiv:0705.1234[hep - ph ] ; k. kajantie , jorma louko and t. tahkokallio , arxiv:0705.1791[hep - th ] . r. apreda , j. erdmenger , n. evans and z. guralnik , phys . rev . * d71 * 126002 ( 2005 ) , arxiv : hep - th/0504151 ; j. erdmenger , m. kaminski and f. rust , arxiv:0704.1290[hep - th ] . k. y. kim , s. j. sin and i. zahed , arxiv : hep - th/0608046 . n. horigome and y. tanii , jhep * 0701 * 072 ( 2007 ) , arxiv : hep - th/0608198 . s. nakamura , y. seo , s. j. sin and k. p. yogendran , `` a new phase at finite quark density from ads / cft , '' arxiv : hep - th/0611021 . s. kobayashi , d. mateos , s. matsuura , r. c. myers and r. m. thomson , jhep * 0702 * 016 ( 2007 ) , arxiv : hep - th/0611099 . s. k. domokos and j. a. harvey , arxiv:0704.1604[hep - ph ] ; a. karch and a. obannon , arxiv:0705.3870[hep - th ] ; d. yamada , arxiv:0707.0101[hep - th ] ; y. kim , b - h . lee , s. nam , c. park and s - j sin , arxiv:0706.2525[hep - ph ] ; s - j sin , arxiv:0707.2719[hep - th ] . j. babington , j. erdmenger , n. evans , z. guralnik and i. kirsch , phys . rev . * * 066007 ( 2004 ) , arxiv : hep - th/0306018 ; i. kirsch , fortsch . phys . * 52 * 727 ( 2004 ) , arxiv : hep - th/0406274 ; r. apreda , j. erdmenger , n. evans and z. guralnik , phys . rev . * d71 * 126002 ( 2005 ) , arxiv : hep - th/0504151 ; d. mateos , r. c. myers and r. m. thomson , phys . lett . * 97 * , 091601 ( 2006 ) , arxiv : hep - th/0605046 ; t. albash , v. filev , c. v. johnson and a. kundu , arxiv : hep - th/0605088 ; t. albash , v. filev , c. v. johnson and a. kundu , arxiv : hep - th/0605175 ; a. karch and a. obannon , phys . rev . * d74 * 085033 ( 2006 ) , arxiv : hep - th/0605120 ; v. g. filev , c. v. johnson , r. c. rashkov and k. s. viswanathan , arxiv : hep - th/0701001 . y. imamura , prog . phys . * 100 * 1263 ( 1998 ) , arxiv : hep - th/9806162 ; y. imamura , nucl . b * 537 * 184 ( 1999 ) , arxiv : hep - th/9807179 ; c. g. callan , a. guijosa and k. g. savvidy , nucl . b * 547 * 127 ( 1999 ) , arxiv : hep - th/9810092 .

we present thermal sunyaev - zeldovich effect ( sze ) measurements for 42 galaxy clusters observed at 150 ghz with the apex - sz experiment . for each cluster , we model the pressure profile and calculate the integrated comptonization to estimate the total thermal energy of the intracluster medium ( icm ) . we compare the measured values to x - ray observables of the icm from the literature ( cluster gas mass , temperature , and ) that relate to total cluster mass . we measure power law scaling relations , including an intrinsic scatter , between the sze and x - ray observables for three subsamples within the set of 42 clusters that have uniform x - ray analysis in the literature . we observe that differences between these x - ray analyses introduce significant variability into the measured scaling relations , particularly affecting the normalization . for all three subsamples , we find results consistent with a self - similar model of cluster evolution dominated by gravitational effects . comparing to predictions from numerical simulations , these scaling relations prefer models that include cooling and feedback in the icm . lastly , we measure an intrinsic scatter of per cent in the scaling relation for all three subsamples . 500 @jnl#1#1 galaxies : clusters : general ( cosmology : ) cosmic background radiation cosmology : observations .

1404.7103

as the largest gravitationally collapsed objects in the universe , clusters of galaxies provide a unique opportunity to study the evolution of large - scale structure . the distribution and abundance of clusters is sensitive to both the geometry of the universe and the growth of density perturbations ( e.g. , * ? ? ? * ; * ? ? ? currently , cluster - based constraints on cosmology are limited by systematic uncertainties in relating observables to cluster masses . most of the cluster mass is in the form of dark matter and therefore is not directly observable . instead , cluster masses are inferred through scaling relations with observable signals such as x - ray luminosity , galaxy velocity distribution , weak - lensing shear and sunyaev - zeldovich effect ( sze ) brightness . under the model of self - similarity , where cluster evolution is dominated by gravitational processes , cluster mass scales with observable signals through simple power law relations @xcite . in this model , the intracluster medium ( icm ) is in hydrostatic equilibrium , for which the scaling to cluster mass can be predicted for a given observable . however , self - similarity does not take into account the role of non - thermal mechanisms such as turbulent gas motions in estimating cluster mass . this simple model also neglects the effects of physical processes internal to the cluster such as feedback from active galactic nuclei and star formation . numerical simulations predict that while the power law exponent of sze scaling relations will be consistent with self - similarity , the normalization does depend on the internal cluster astrophysics @xcite . an empirical measurement of the scaling relations therefore informs models of cluster astrophysics , progressing towards the needed calibration for cosmology . currently , cluster surveys are operating in the optical ( e.g. , * ? ? ? * ) , x - ray ( e.g. , * ? ? ? * ; * ? ? ? * ) , and the millimeter ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) wavelength regimes . however , optical and x - ray measurements of clusters suffer from cosmological dimming , and only the brightest and most massive clusters are detected at high redshifts ( @xmath7 ) . in contrast , the millimeter - wavelength thermal sze @xcite , where cosmic microwave background ( cmb ) photons inverse - compton scatter off hot intracluster electrons , is redshift - independent . sze surveys with sufficient resolution to resolve clusters , such as those performed with the south pole telescope @xcite and the atacama cosmology telescope @xcite detect clusters with a mass selection nearly independent of redshift . precision cosmology requires that mass - observable scaling relations be characterized with a high level of accuracy , including both the measurement uncertainty and the intrinsic scatter from cluster - to - cluster differences . numerical simulations suggest that sze observations are relatively insensitive to the details of cluster astrophysics ( e.g. , * ? ? ? * ; * ? ? ? * ) , resulting in low intrinsic scatter scaling relations and a tighter constraint on cluster mass . previous measurements of sze scaling relations include interferometric observations from the ovro / bima @xcite and sza @xcite arrays as well as imaging studies with the south pole telescope @xcite , bolocam @xcite , atacama cosmology telescope @xcite , and the planck mission @xcite . in general , those studies find that observational sze scaling relations agree with expectations based on self - similarity and that simulations including additional non - adiabatic physics are preferred . while cluster samples selected from large surveys ( both sze and x - ray ) have well - known selection functions , studies such as the one presented in this paper that target known clusters often select them in an _ ad hoc _ manner . the influence of this sample selection on sze scaling relations is unknown . in this paper , we study how the sze signal scales with x - ray observables related to cluster mass using observations from the apex - sz imaging bolometer array . we measure sze scaling relations using x - ray observables ( @xmath8 , @xmath1 , and @xmath2 ) as proxies for the total cluster mass . apex - sz observed a small set of 11 clusters selected from the reflex x - ray survey ( referred to as the reflex - dxl sample @xcite ) as well as an additional 31 clusters selected in an _ ad hoc _ manner . within this full sample are clusters drawn from the ( * ? ? ? * hereafter z08 ) and ( * ? ? ? * hereafter m10 ) samples . we measure power law scaling relations for each of these three subsamples ( reflex - dxl , z08 , m10 ) and compare the results to the expectations of the self - similar model as well as numerical simulations that incorporate different physical processes in the icm . we compare the results between the three cluster samples as well as to the full apex - sz sample to investigate the effects of sample selection and uniform analysis on the measured power law and intrinsic scatter of the sze scaling relations . in this paper , we assume the _ _ wmap7__+bao+@xmath9 @xmath10cdm best - fit cosmology in which @xmath11 km@xmath12@xmath13 , @xmath14 , and @xmath15 @xcite . the structure of the paper is as follows . section [ sec : theory ] reviews the sze and the associated scaling relations . section [ sec : observations ] introduces the clusters in this study and observations thereof . section [ sec : szdatareduction ] briefly describes the data reduction process for the apex - sz instrument . we present the sze - x - ray scaling relations and discuss these results in section [ sec : results ] . a summary and an outlook on future work are provided in section [ sec : conclusions ] . the sze occurs when photons from the cmb inverse compton scatter off hot electrons in the intracluster medium to higher energies @xcite . this process distorts the cmb blackbody spectrum with an amplitude proportional to the gas pressure integrated along the line - of - sight . in units of temperature the change is @xmath16 where @xmath17 and @xmath18 are the electron density and temperature of the icm , @xmath19 is the temperature of the cmb , @xmath20 is the thomson cross - section , @xmath21 is the boltzmann constant , and @xmath22 is the electron rest mass energy . the integrated pressure is often parametrized in terms of the comptonization , @xmath23 . the frequency dependence @xmath24 is given by @xmath25 where @xmath26 and @xmath27 is a correction for relativistic electrons ( see , e.g. , * ? ? ? * ) . for frequencies below 217 ghz , the sze manifests itself as a decrement in the cmb spectrum , while at frequencies above 217 ghz the sze results in an increment . equation [ eqn : szeffect ] demonstrates two important characteristics of the sze . first , the sze is redshift independent since its surface brightness is expressed as a fractional change in the cmb temperature . second , the dependence on integrated pressure directly measures the thermal energy of the icm . the comptonization integrated over the solid angle of the cluster on the sky , @xmath28 measures the total thermal energy of a cluster , and is therefore expected to be a robust proxy for total mass . in this work , we integrate a parametric model of @xmath0 to a radius of @xmath29 , the radius at which the mean matter density of the cluster is 500 times the critical density of the universe , @xmath30 . assuming that clusters are in hydrostatic equilibrium , the integrated comptonization scales as a function of the icm properties according to @xmath31 where @xmath32 is the expansion rate of the universe normalized to its present value , @xmath33 , @xmath34 is the gas mass fraction , and @xmath35 is the angular diameter distance of the cluster @xcite . a third proxy for cluster mass can be constructed from the combination of temperature and gas mass , @xmath36 @xcite . @xmath0 scales with @xmath8 as @xmath37 and is expected to have lower intrinsic scatter than @xmath38 or @xmath39 due to the anti - correlation of systematic deviations between @xmath40 and @xmath1 . apex - sz @xcite was a transition - edge - sensor ( tes ) bolometer array located on the 12-meter atacama pathfinder experiment ( apex ) telescope in northern chile @xcite . the focal plane comprised a total of 280 bolometers distributed on six wafers which were cooled with closed - cycle refrigerators to an average operating temperature of @xmath41300 mk and read out using frequency domain multiplexing @xcite . the apex - sz experiment imaged sze decrements at 150 ghz with arcmin resolution over a 23 arcmin field of view ( fov ) . additional details of the apex - sz instrument are presented in @xcite and @xcite . apex - sz completed two commissioning and seven observing runs from 2005 to 2010 . during this time , apex - sz observed the 42 clusters presented here for a total of approximately 760 hours . apex - sz executed a circular drift scan pattern that concentrated the integration time in a region around the known cluster coordinates while limiting overhead time due to telescope turn - arounds . the telescope repeated the circular scan while centered on a constant azimuth and elevation , allowing the target to drift through the fov . after completing between 11 and 20 circles ( 47 seconds per circle ) , the telescope slewed to track the target and then repeated the circular pattern . the radius of each circle ( @xmath42@xmath43 arcmin ) was chosen from the expected radial extent of each cluster to ensure that the detectors observed the sky background during each scan . these scans resulted in a sky coverage area of approximately @xmath44 . clusters were targeted based on the availability of published x - ray measurements . in general , massive clusters ( @xmath45 kev ) were selected and the sample included both dynamically relaxed and disturbed clusters . additionally , apex - sz targeted clusters from the reflex - dxl x - ray sample @xcite , which is discussed in further detail in [ sec : sampleintro ] . coordinates and map depth for each cluster are given in table [ tbl : clustercoord ] . lcccc + cluster name & right ascension & declination & redshift & map depth@xmath46 + & ( _ h m s _ ) & ( ) & & ( @xmath47 cmb ) + + abell 2744 & 00 14 18.6 & -30 23 15.4 & 0.307 & 12 + abell 2813 & 00 43 24.5 & -20 37 31.2 & 0.292 & 21 + abell 209 & 01 31 52.6 & -13 36 35.5 & 0.209 & 16 + xlss j022145.2 - 034617 & 02 21 45.2 & -03 46 17.4 & 0.430 & 7 + rxcj0232.2 - 4420 & 02 32 18.8 & -44 20 51.9 & 0.284 & 17 + abell 383 & 02 48 03.3 & -03 31 43.6 & 0.187 & 14 + rxcj0437.1 + 0043 & 04 37 09.5 & + 00 43 54.5 & 0.284 & 14 + ms0451.6 - 0305 & 04 54 11.3 & -03 00 52.6 & 0.550 & 21 + abell 520 & 04 54 09.0 & + 02 55 18.0 & 0.203 & 14 + rxcj0516.6 - 5430 & 05 16 35.2 & -54 30 36.8 & 0.294 & 10 + rxcj0528.9 - 3927 & 05 28 52.5 & -39 28 16.7 & 0.284 & 11 + rxcj0532.9 - 3701 & 05 32 55.9 & -37 01 34.5 & 0.275 & 16 + abell 3404 & 06 45 30.0 & -54 13 42.1 & 0.164 & 13 + 1es 0657 - 56 & 06 58 30.2 & -55 56 33.7 & 0.296 & 19 + abell 907 & 09 58 21.9 & -11 03 48.2 & 0.160 & 11 + xmmxcsj095940.8 + 023111.3 & 09 59 40.8 & + 02 31 11.3 & 0.720 & 14 + rxcj1023.6 + 0411 & 10 23 39.6 & + 04 11 12.0 & 0.291 & 12 + ms1054.4 - 0321 & 10 56 59.0 & -03 37 37.0 & 0.830 & 13 + macsj1115.8 + 0129 & 11 15 51.9 & + 01 29 55.0 & 0.355 & 16 + abell 1300 & 11 31 54.7 & -19 55 40.5 & 0.308 & 18 + rxcj1206.2 - 0848 & 12 06 12.3 & -08 48 06.0 & 0.439 & 18 + xmmuj1230.3 + 1339 & 12 30 16.9 & + 13 39 04.3 & 0.975 & 11 + rdcsj1252 - 2927 & 12 52 54.4 & -29 27 17.0 & 1.240 & 8 + macsj1311.0 - 0311 & 13 11 01.7 & -03 10 37.6 & 0.494 & 12 + abell 1689 & 13 11 29.3 & -01 20 26.7 & 0.184 & 19 + rxcj1347.5 - 1144 & 13 47 30.8 & -11 45 09.0 & 0.451 & 28 + macsj1359.1 - 1929 & 13 59 10.3 & -19 29 24.0 & 0.447 & 27 + abell 1835 & 14 01 01.9 & + 02 52 35.5 & 0.253 & 36 + rxcj1504.1 - 0248 & 15 04 07.6 & -02 48 16.0 & 0.215 & 21 + abell 2163 & 16 15 46.0 & -06 08 54.0 & 0.203 & 30 + abell 2204 & 16 32 47.1 & + 05 34 32.3 & 0.152 & 10 + macsj1931.8 - 2635 & 19 31 49.6 & -26 34 34.0 & 0.352 & 30 + rxcj2011.3 - 5725 & 20 11 27.1 & -57 25 09.8 & 0.279 & 11 + rxcj2014.8 - 2430 & 20 14 49.7 & -24 30 30.0 & 0.161 & 15 + macsj2046.0 - 3430 & 20 46 00.5 & -34 30 17.0 & 0.423 & 15 + rxcj2214.9 - 1359 & 22 14 57.4 & -14 00 10.8 & 0.503 & 23 + xmmxcsj2215.9 - 1738 & 22 15 58.0 & -17 38 02.5 & 1.450 & 11 + xmmuj2235.3 - 2557 & 22 35 20.6 & -25 57 42.0 & 1.393 & 9 + rxcj2243.3 - 0935 & 22 43 21.4 & -09 35 43.0 & 0.447 & 20 + abell s1077 & 22 58 48.1 & -34 47 59.4 & 0.313 & 17 + abell 2537 & 23 08 22.0 & -02 11 30.0 & 0.297 & 11 + rxcj2337.6 + 0016 & 23 37 37.8 & + 00 16 15.5 & 0.278 & 20 + @xmath46map depth is measured within the central arcminute from the standard deviation of 500 jackknife noise map realizations . + lccccc + cluster name & @xmath2 & ref . & @xmath1 & ref . & @xmath8 + & kev & & @xmath48 & & @xmath49kev@xmath50 + + abell 2744 & @xmath51 & 2 & @xmath52 & 2 & @xmath53 + & @xmath54 & 3&@xmath55 & 3&@xmath56 + abell 2813 & @xmath57 & 2 & @xmath58 & 2 & @xmath59 + & @xmath60 & 1 & @xmath61&1 & @xmath62 + & @xmath63 & 3 & @xmath64 & 3&@xmath65 + abell 209 & @xmath66 & 1 & @xmath67 & 1 & @xmath68 + & @xmath69 & 3 & @xmath70 & 3&@xmath71 + xlss j022145.2 - 034617 & @xmath72 & 7 & @xmath73 & @xmath73 & @xmath73 + rxcj0232.2 - 4420 & @xmath74 & 2 & @xmath75 & 2 & @xmath76 + & @xmath77 & 1&@xmath78&1&@xmath79 + & @xmath80 & 3 & @xmath81 & 3 & @xmath82 + abell 383 & @xmath83 & 1 & @xmath84 & 1 & @xmath85 + rxcj0437.1 + 0043 & @xmath86 & 2 & @xmath87 & 2 & @xmath88 + ms0451.6 - 0305 & @xmath89 & 11 & @xmath90 & 11 & @xmath91 + abell 520 & @xmath92 & 3 & @xmath93 & 3 & @xmath94 + rxcj0516.6 - 5430 & @xmath95 & 2 & @xmath96 & 2 & @xmath97 + & @xmath98 & 1 & @xmath99&1&@xmath100 + rxcj0528.9 - 3927 & @xmath101 & 2 & @xmath102 & 2 & @xmath103 + & @xmath104 & 1 & @xmath105&1 & @xmath106 + & @xmath107&3&@xmath108&3&@xmath109 + rxcj0532.9 - 3701 & @xmath110 & 2 & @xmath58 & 2 & @xmath111 + & @xmath112 & 1&@xmath113&1 & @xmath114 + abell 3404 & @xmath115 & 1 & @xmath116 & 1 & @xmath117 + 1es 0657 - 56 & @xmath118 & 2 & @xmath119 & 2 & @xmath120 + & @xmath121 & 1 & @xmath122 & 1 & @xmath123 + & @xmath124&3&@xmath125&3&@xmath126 + abell 907 & @xmath127 & 1 & @xmath128 & 1 & @xmath129 + xmmxcsj095940.8 + 023111.3 & @xmath130 & 9 & @xmath73 & @xmath73 & @xmath73 + rxcj1023.6 + 0411 & @xmath131 & 3 & @xmath132 & 3 & @xmath133 + ms1054.4 - 0321 & @xmath134 & 14 & @xmath73 & @xmath73 & @xmath73 + macsj1115.8 + 0129 & @xmath135 & 3 & @xmath136 & 3 & @xmath137 + abell 1300 & @xmath138 & 2 & @xmath96 & 2 & @xmath139 + & @xmath140 & 3&@xmath141 & 3&@xmath142 + rxcj1206.2 - 0848 & @xmath143 & 3 & @xmath144 & 3 & @xmath145 + xmmuj1230.3 + 1339 & @xmath146 & 4 & @xmath147 & 4 & @xmath148 + rdcsj1252 - 2927 & @xmath149 & 6 & @xmath150 & 6 & @xmath151 + macsj1311.0 - 0311 & @xmath152 & 11 & @xmath153 & 11 & @xmath154 + abell 1689 & @xmath155 & 1 & @xmath156 & 1 & @xmath157 + rxcj1347.5 - 1144 & @xmath158 & 3 & @xmath159 & 3 & @xmath160 + macsj1359.1 - 1929 & @xmath161 & 13 & @xmath73 & @xmath73 & @xmath73 + abell 1835 & @xmath162 & 1 & @xmath163 & 1 & @xmath164 + & @xmath165&3&@xmath166&3&@xmath167 + rxcj1504.1 - 0248 & @xmath168 & 3 & @xmath169 & 3 & @xmath170 + abell 2163 & @xmath171 & 3 & @xmath172 & 3 & @xmath173 + abell 2204 & @xmath174 & 1 & @xmath175 & 1 & @xmath176 + macsj1931.8 - 2635 & @xmath177 & 3 & @xmath178 & 3 & @xmath179 + rxcj2011.3 - 5725 & @xmath180 & 3 & @xmath181 & 3 & @xmath182 + rxcj2014.8 - 2430 & @xmath183 & 12 & @xmath184 & 12 & @xmath185 + macsj2046.0 - 3430 & @xmath186 & 13 & @xmath73 & @xmath73 & @xmath73 + rxcj2214.9 - 1359 & @xmath187 & 8 & @xmath73 & @xmath73 & @xmath73 + xmmxcsj2215.9 - 1738 & @xmath188 & 6 & @xmath189 & 6 & @xmath190 + xmmuj2235.3 - 2557 & @xmath191 & 5 & @xmath192 & 6 & @xmath193 + cluster name & @xmath2 & ref . & @xmath1 & ref . & @xmath8 + & kev & & @xmath48 & & @xmath49kev@xmath50 + + rxcj2243.3 - 0935 & @xmath194 & 3 & @xmath195 & 3 & @xmath196 + abell s1077 & @xmath197 & 10 & @xmath73 & @xmath73 & @xmath73 + abell 2537 & @xmath198 & 2 & @xmath199 & 2 & @xmath200 + & @xmath201 & 1 & @xmath202&1 & @xmath203 + & @xmath204&3&@xmath205&3&@xmath206 + rxcj2337.6 + 0016 & @xmath207 & 2 & @xmath208 & 2 & @xmath209 + & @xmath210 & 1&@xmath211&1&@xmath212 + 1 . @xcite @xmath213 2 . @xcite @xmath214 3 . @xcite @xmath215 4 . @xcite @xmath216 5 . @xcite @xmath217 6 . @xcite @xmath218 7 . @xcite @xmath219 8 . @xcite @xmath220 9 . @xcite 10 . @xcite @xmath221 11 . @xcite @xmath222 12 . @xcite @xmath222 13 . @xcite @xmath223 14 . raw apex - sz timestream data contain the faint sze cluster signal , scan synchronous signals due to ground pick - up and instrumental thermal fluctuations , atmospheric fluctuations , and instrument noise . in order to measure the sze surface brightness accurately , the array properties must be well characterized and contamination cleaned from the data . the beam position and shape for each bolometer in the focal plane are measured from a daily raster scan of a calibration target ( mars , uranus , or saturn ) . in addition to the 58 arcsec fwhm gaussian main beam , apex - sz beams exhibit significant sidelobes @xcite . we characterize the sidelobes by mapping the individual detector beams and combining them into a composite beam . the total beam area , including the gaussian main beam and the sidelobes , is then taken into account during further calibration . the beam size is also corrected to account for the angular extent of the source and detector saturation effects . absolute flux calibration is performed based on the overall amplitude of response for each detector from a raster scan on a known celestial source . apex - sz observed two primary flux calibrators : mars and uranus . the planetary disk size and brightness temperature are taken from the rudy model for mars @xcite and from the jcmt fluxes model for uranus . as described in @xcite , we refine the absolute brightness temperatures for mars and uranus using _ wmap _ results @xcite to improve calibration accuracy . we find that the rudy temperatures are systematically higher than _ wmap _ by a factor of @xmath225 . the calculation of this factor includes the extrapolation of the rudy and _ wmap _ 94 ghz brightness temperatures to 150 ghz , 1.016@xmath226 0.009 , which is included in the total calibration uncertainty . similarly , we use the data of @xcite to extrapolate the _ wmap7 _ 94 ghz measurement of uranus to @xmath227 k. several times during observations primary planetary calibrators were unavailable due to angular proximity to the sun . the stable galactic h ii regions rcw38 and iras12073 - 6233 were used as secondary calibrators during these periods , with brightness temperatures bootstrapped from back - to - back scans of the primary and secondary calibrators . the measured calibration is corrected for differences in atmospheric opacity between the data and calibration scans using the measured value of precipitable water vapour ( pwv ) from the apex radiometer . following @xcite , we extrapolate the atmospheric optical depth based on the atm model @xcite and correct accordingly . in addition to the calibration scan , routine pointing scans of bright quasars are performed throughout the night . we find an average pointing jitter of approximately 5.7 arcsec in both azimuth and elevation . this is significantly less than the size of the apex - sz beam and has a negligible effect on the measurements presented here . combining the corrections discussed in this section ( beam solid angle , calibrator temperature , atmospheric opacity ) with the uncertainty in the frequency band center ( see * ? ? ? * ) we estimate the gaussian uncertainty in the flux calibration to be @xmath2268 per cent . ideally , the responsivity of each bolometer is constant despite changing amounts of incident optical power @xcite . however , when a bolometer is biased near the upper edge of the superconducting transition , the responsivity changes as a result of varying optical power from three sources : the elevation dependence of atmospheric airmass , the overall atmospheric opacity , and the power from an astronomical source . the last effect is sub - dominant , but it can not be neglected for the case of a bright calibrator . for most detectors in the apex - sz array , suppression of responsivity is approximately a linear function of incident optical power . however , some channels exhibit more extreme changes . we measure the responsivity of each bolometer using the elevation(@xmath228 ) dependent signal @xmath229 introduced by the circular scan pattern . this is repeated for every two minutes of data , normalizing by the changing zenith emissivity to measure @xmath230 . channels that exhibit strong nonlinear response are discarded from use in further analysis . for the remaining channels , we model the response as a linear function and use it to correct the absolute flux calibration , discarding data that deviate more than @xmath231 from the best - fit model . data with extremely large overall correction factors ( greater than 50 per cent ) are discarded . we also correct for two second order effects . first , as a bright calibration source is scanned , variable responsivity will slightly distort the measured beam shape . second , the fractional solid angle of the sidelobes relative to the main beam will be affected . based on the rms scatter around the best - fit gain models , we estimate the uncertainty in the responsivity corrections to be @xmath2265 per cent . combining this uncertainty with the calibration uncertainty from the previous section , we find that the total uncertainty in the apex - sz temperature calibration is @xmath22610 per cent . the apex - sz data set was measured in widely varying atmospheric conditions , a significant challenge when trying to measure the relatively faint sze signal . in general , the atmospheric signal is highly correlated between bolometers as spatial fluctuations in the atmosphere are much larger than the apex - sz fov . additionally , the bolometers exhibit a correlated signal due to heating of the detector array from telescope vibrations . we employ several different filtering techniques to clean the data , similar to @xcite . in order to optimize the final maps , filtering is tailored to the specific needs of each target with the goal of sufficiently removing contamination and minimizing noise while preserving cluster signal . the first step to condition apex - sz timestreams is to parse the data into separate circles made on the sky according to the scan pattern . the circles are grouped together based on a common central azimuth and elevation and data not in a circle set are discarded . initial channel and data cuts are performed on the indexed data based on several criteria . channels that are optically unresponsive ( optical efficiency less than 6 per cent ) are rejected . apex - sz timestreams occasionally exhibit large spikes or glitches in one or many bolometers due to either cosmic rays or electrical interference . glitches are located and removed from the timestreams using both the timestream derivative and a simple s / n threshold cut . when a glitch is simultaneously detected in several channels ( more than 7 per cent ) , that particular section of data is discarded for all channels . as a final step , channels with excess noise between 12 and 18 hz are discarded . after the initial data cuts we deconvolve an optical time constant for each channel from the timestreams . the median measured optical time constant for apex - sz bolometers is @xmath4112 ms , however , it is as high as @xmath4160 ms for some channels . to prevent the amplification of high frequency noise during deconvolution , an additional low - pass filter ( @xmath232 $ ] ) is applied . we define @xmath233 where @xmath234 is the median scan velocity of the telescope and @xmath235 is the full width at half - maximum of the apex - sz beam . the final step in processing the timestream data is to remove the dominant atmospheric and thermal signals . the majority of the timestream cleaning is performed through a combination of two methods : high - pass time - domain filters and the subtraction of spatial polynomial functions across the array , taking advantage of the correlation between bolometers . an initial high - pass filter removes the static offset for each channel as well as low - frequency noise in the timestreams . removing the timestream signal due to stage heating is especially important because the subsequent spatial template removal steps use relative gain coefficients that are based solely on the atmospheric response . the high - pass filtering is implemented in two forms : an n - order polynomial removal and a butterworth frequency domain filter . the polynomial filter , described in detail in @xcite , simultaneously removes a scan synchronous cosecant signal . after the first high - pass filter , we remove a 2-dimensional spatial polynomial function across the array @xcite at each sample in the timestream . during this process , referred to as _ spatial template removal _ , we fit the polynomial either across the entire array , or individually for each of the six detector wafers to improve removal of correlated signals on scales smaller than the @xmath236 field - of - view . often , this filter is applied multiple times , discarding channels with excess variance in between . a final polynomial removal is performed to remove any remaining contamination . figure [ fig : tspowspect ] shows the power spectrum of an apex - sz timestream at three different steps in the reduction process . once the timestreams are filtered , we create sky maps by binning the timestream data into @xmath237 arcsec pixels . sky coordinates for each bolometer are determined from the absolute telescope boresight pointing and individual bolometer positional offsets . as in @xcite , bolometers and scans are combined with a minimum variance weighting for each cluster . the resulting set of 42 sky maps is shown in figure [ fig : skymaps ] . each map has been convolved with a one arcmin fwhm gaussian for visual presentation . however , all subsequent analysis is performed on the unsmoothed version . in conjunction with the sky map , we quantify the effects of the apex - sz instrumental beam and analysis filtering on the sky signal . a simulated cluster map , created using the profile discussed in section [ sec : szmmodel ] , is convolved with the composite beam profile of the array and mapped into timestreams using the telescope pointing information . each data cut and filtering process performed on the real cluster data is repeated on these timestreams . the filtered simulation is mapped and coadded using the same minimum variance weights as the cluster map . an individualized transfer function map , @xmath238 , is created for each cluster by normalizing the fourier transform of the filtered map to that of the original simulation . we create difference noise maps for each cluster to characterize the map noise . in this process , random halves of the individual scan maps are multiplied by -1 before coadding . this removes any astrophysical signal that is consistent between all maps , including primary cmb anisotropy . to account for the noise due to the cmb , we convolve a realization of the cmb created from the planck + wp best - fit power spectrum @xcite with a cmb transfer function ( created in the same way as the cluster transfer function ) and add it to the difference noise map . in addition to the cmb , a contribution from a background of point sources is expected . we find that the power from poisson distributed point sources @xcite , including an extra 50 per cent variance due to lensing @xcite , is negligible in comparison to the instrument noise . in order to calculate the integrated cluster signal , we need to extrapolate the measured cluster signal to spatial scales that have been filtered out due to timestream processing and the angular resolution of the instrument . we adopt the universal pressure profile of @xcite , @xmath239^{(\beta-\gamma)/\alpha}}.\ ] ] here @xmath240 is the normalization of the pressure profile , @xmath241 is the scale radius , and ( @xmath242 ) describe the slope of the profile for @xmath243 , @xmath244 and @xmath245 , respectively . the scale radius is often defined in terms of a concentration parameter @xmath246 , @xmath247 . proposed by @xcite , this generalized navarro - frenk - white model ( gnfw ) , is a generalization of the numerical results from @xcite for the distribution of mass in a dark matter halo . @xcite combined simulations and x - ray observations to measure a best - fit model of ( @xmath246 , @xmath242 ) = ( 1.177 , 1.0510 , 5.4905 , 0.3081 ) , which we use for the measurements presented here . applying the ideal gas law , @xmath248 , to equation [ eqn : gnfwmodel ] and substituting into equation [ eqn : szeffect ] the sze profile becomes @xmath249 we reparametrize equation [ eqn : szgnfwmodel ] in terms of a normalization @xmath250 , @xmath251^{(\beta-\gamma)/\alpha}}dl.\ ] ] due to the compact scan pattern and strong atmospheric filtering required , the apex - sz data poorly constrain power on scales larger than 10 arcmin in the maps . therefore , we infer @xmath29 from the x - ray based @xmath252 scaling relation of @xcite for each cluster and fix the model centroid at the x - ray centroid . the @xcite profile for each cluster serves as the simulated input for creating the individualized transfer functions described in @xmath253[sec : transferfunc ] and is used to fit for the normalization , @xmath250 . we calculate the likelihood for each model as a function of @xmath250 , @xmath254 . first , we create a 2-dimensional model map @xmath255 and convolve it with the transfer function map @xmath238 , @xmath256 . the model @xmath257 and sky maps @xmath258 are binned into 1-arcmin radial bins around the x - ray centroid . using the central 10 arcmin of these profiles , we construct the @xmath259 statistic , @xmath260 where @xmath261 is the noise covariance matrix estimated by averaging 500 radially - binned difference noise realizations . the best - fit parameters along with the x - ray based @xmath262 and respective probability to exceed ( pte ) are given in table [ tbl : mcmc ] . for each cluster , we calculate the spherical integral of the assumed model profile within the x - ray defined @xmath29 and use the measured @xmath250 to estimate the integrated comptonization @xmath6 . we use the work of @xcite and the x - ray temperatures from table [ tbl : physparam ] to calculate the relativistic correction @xmath263 to the frequency dependent term in equation [ eqn : szefreq ] . on average , this correction is of order @xmath264 per cent . the measured values of @xmath6 are given in table [ tbl : mcmc ] . we use the method described in section [ sec : clustermodel ] to measure the spherically - integrated comptonization for each cluster . we assume the cluster follows an @xcite pressure profile ( see equation [ eqn : gnfwmodel ] ) with radius @xmath29 , where @xmath29 is inferred from the @xcite @xmath265 relation . we use three different proxies for total cluster mass : spectroscopic x - ray temperature , gas mass , and @xmath8 , the values for which are taken from the literature . the redshifts and x - ray parameters for the 42 clusters included in this sample are given in table [ tbl : physparam ] along with the references . for clusters found in more than one of the reflex - dxl , z08 , or m10 subsamples , the additional parameter values are also given . all measurements were made with either the _ xmm - newton _ or _ chandra _ x - ray observatories . x - ray temperatures are used that were derived from temperature maps with the central region excluded when available . inclusion of the core region can lead to an underestimated global temperature for the cool - core clusters present within our sample . @xmath1 values , and by extension @xmath8 , were available for only 35 clusters . lccccc + cluster name & @xmath250 & @xmath266 & ref . & pte & @xmath6 + & ( @xmath267k cmb ) & ( @xmath268 ) & & & ( @xmath269 ) + + abell 2744 & @xmath270 & @xmath271&2 & 0.11 & @xmath272 + & @xmath273 & @xmath274&3 & 0.07 & @xmath275 + abell 2813 & @xmath276 & @xmath277&2 & 0.14 & @xmath278 + & @xmath279 & @xmath280&1 & 0.24 & @xmath281 + & @xmath282 & @xmath283&3 & 0.14 & @xmath284 + abell 209 & @xmath285 & @xmath286&1 & 0.24 & @xmath287 + & @xmath288 & @xmath289&3 & 0.32 & @xmath290 + xlss j022145.2 - 034617 & @xmath291 & @xmath292&8 & 0.21 & @xmath293 + rxcj0232.2 - 4420 & @xmath294 & @xmath295&2 & @xmath296 & @xmath297 + & @xmath298 & @xmath299&1 & @xmath300 & @xmath301 + & @xmath302 & @xmath303&3 & @xmath296 & @xmath304 + abell 383 & @xmath305 & @xmath306&1 & 0.79 & @xmath307 + rxcj0437.1 + 0043 & @xmath308 & @xmath309&2 & @xmath310 & @xmath311 + ms0451.6 - 0305 & @xmath312 & @xmath313&12 & 0.21 & @xmath314 + abell 520 & @xmath315 & @xmath316&4 & @xmath317 & @xmath318 + rxcj0516.6 - 5430 & @xmath319 & @xmath320&2 & @xmath321 & @xmath322 + & @xmath323 & @xmath324&1 & @xmath325 & @xmath326 + rxcj0528.9 - 3927 & @xmath327 & @xmath328&2 & @xmath329 & @xmath330 + & @xmath331 & @xmath299&1 & @xmath332 & @xmath333 + & @xmath334 & @xmath335&3 & @xmath336 & @xmath337 + rxcj0532.9 - 3701 & @xmath338 & @xmath339&2 & @xmath340 & @xmath341 + & @xmath342 & @xmath343&1 & @xmath344 & @xmath345 + abell 3404 & @xmath346 & @xmath347&1 & @xmath348 & @xmath349 + 1es 0657 - 56 & @xmath350 & @xmath351&2 & @xmath352 & @xmath353 + & @xmath354 & @xmath355&1 & @xmath356 & @xmath357 + & @xmath358 & @xmath359&3 & @xmath360 & @xmath361 + abell 907 & @xmath362 & @xmath363&1 & 0.71 & @xmath364 + xmmxcsj095940.8 + 023111.3 & @xmath365 & @xmath366&10 & @xmath367 & @xmath368 + rxcj1023.6 + 0411 & @xmath369 & @xmath370&4 & 0.16 & @xmath371 + ms1054.4 - 0321 & @xmath372 & @xmath373&15 & 0.98 & @xmath374 + macsj1115.8 + 0129 & @xmath375 & @xmath376&4 & 0.12 & @xmath377 + abell 1300 & @xmath378 & @xmath379&2 & 0.73 & @xmath380 + & @xmath381 & @xmath382&3 & 0.83 & @xmath383 + rxcj1206.2 - 0848 & @xmath384 & @xmath385&4 & 0.18 & @xmath386 + xmmuj1230.3 + 1339 & @xmath387 & @xmath388&5 & 0.31 & @xmath389 + rdcsj1252 - 2927@xmath390 & @xmath391 & @xmath392&7 & 0.28 & @xmath393 + macsj1311.0 - 0311@xmath390 & @xmath394 & @xmath395&12 & 0.88 & @xmath396 + abell 1689 & @xmath397 & @xmath398&1 & 0.23 & @xmath399 + rxcj1347.5 - 1144 & @xmath400 & @xmath401&4 & 0.24 & @xmath402 + macsj1359.1 - 1929@xmath390 & @xmath403 & @xmath404&14 & 0.02 & @xmath405 + abell 1835 & @xmath406 & @xmath407&1 & 0.42 & @xmath408 + & @xmath409 & @xmath410&3 & 0.36 & @xmath411 + rxcj1504.1 - 0248 & @xmath412 & @xmath413&4 & 0.52 & @xmath414 + abell 2163 & @xmath415 & @xmath416&4 & 0.90 & @xmath417 + abell 2204 & @xmath418 & @xmath419&4 & 0.01 & @xmath420 + macsj1931.8 - 2635@xmath390 & @xmath421 & @xmath422&4 & 0.20 & @xmath423 + rxcj2011.3 - 5725 & @xmath424 & @xmath425&3 & 0.16 & @xmath426 + rxcj2014.8 - 2430@xmath390 & @xmath427 & @xmath428&13 & 0.58 & @xmath429 + macsj2046.0 - 3430 & @xmath430 & @xmath431&14 & 0.72 & @xmath432 + rxcj2214.9 - 1359 & @xmath433 & @xmath434&9 & @xmath435 & @xmath436 + xmmxcsj2215.9 - 1738@xmath390 & @xmath437 & @xmath438&7 & 0.24 & @xmath439 + xmmuj2235.3 - 2557@xmath390 & @xmath440 & @xmath441&6 & 0.51 & @xmath442 + rxcj2243.3 - 0935 & @xmath443 & @xmath444&4 & 0.15 & @xmath445 + abell s1077 & @xmath446 & @xmath447&11 & 0.08 & @xmath448 + abell 2537 & @xmath449 & @xmath277&2 & 0.02 & @xmath450 + & @xmath451 & @xmath452&1 & 0.03 & @xmath450 + & @xmath453 & @xmath454&3 & 0.02 & @xmath455 + rxcj2337.6 + 0016 & @xmath456 & @xmath457&2 & 0.12 & @xmath458 + & @xmath459 & @xmath460&1 & 0.18 & @xmath461 + + @xmath6 is the spherical integration of best - fit pressure profile within @xmath29 . non - detections are denoted with @xmath462 . references correspond to those in table [ tbl : physparam ] and denote the x - ray measurement of @xmath463 used to determine @xmath29 . values are given for each subsample where appropriate : [ 1 ] z08 , [ 2 ] reflex - dxl , [ 3 ] m10 . to derive best - fit scaling relations parameters we perform a regression analysis of the data shown in figure [ fig : scalefullbeta ] . we assume that the data follow the power law relationship , @xmath464 where @xmath465 is the normalization , @xmath466 is the power law exponent , and @xmath467 denotes an x - ray observable . a pivot point , @xmath468 , is chosen for each observable ( @xmath469 = 7.5 kev , @xmath470 , @xmath471 ) to reduce the correlations between @xmath465 and @xmath466 . we use the self - similar values of @xmath472 , respectively ( see equation [ eqn : scale ] ) . the intrinsic astrophysical scatter in the scaling relation is represented through a log - normal distribution with mean @xmath473 and variance @xmath474 in the base 10 logarithm of the distribution . this regression is often performed in the log basis , where equation [ eqn : powlaw ] transforms to a simple linear relationship with gaussian intrinsic scatter . in agreement with ( * hereafter k07 ) , we find from simulated data sets that common methods for linear regression with uncertainty in both the x- and y - variables , such as those in @xcite and @xcite , introduce systematic bias into the measured parameters . the regression parameters are further biased by the symmetrization and transformation of the @xmath6 likelihood distribution from the linear basis to the logarithmic basis . lastly , the measured @xmath6 non - detections in the apex - sz data can not be fully represented in the log basis . for example , when we place a prior that @xmath475 and perform a linear regression using the bayesian method of sampling the posterior probability from k07 , we find that the resulting maximum likelihood parameters are biased at the @xmath476 level . to address the difficulties introduced by the log basis , we instead perform the regression analysis in the linear basis . we construct the likelihood of the scaling relations parameters from the probability densities of 500 measured with apex - sz and the x - ray observable ( e.g. , @xmath8 ) . we approximate the measured probabilities in @xmath6 as asymmetric gaussian distributions . following the methodology of k07 , we model the distribution of @xmath467 as a weighted sum of gaussian functions . we sample the posterior likelihood distribution using an mcmc method with a metropolis - hastings algorithm . using simulated data sets with x - ray properties based on those of the apex - sz cluster sample , we find that we are able to recover unbiased estimates of the scaling relation parameters . [ cols="<,^ , > " , ] we divide our investigation of possible systematic contamination of these results into two categories : instrumental and analysis systematics and astrophysical effects . in the following two sections we discuss these systematics including the role of sample selection , correlations between the x - ray and sze measurements , variations in x - ray measurements , known clusters that apex - sz did not detect , the assumptions inherent in the cluster profile modelling , contamination due to point sources , and a non - universal gas mass fraction . a portion of the measured intrinsic scatter in @xmath477 and @xmath478 can be attributed to correlation introduced by the fact that we use the x - ray temperature @xmath2 to estimate @xmath29 and therefore 500 . we measure this correlation in the scaling relations by simulating the apex - sz dataset . for each cluster , a new temperature is drawn from its measured probability distribution and @xmath29 is determined from the @xmath252 relation . the @xmath6 distributions are recalculated and the regression analysis is performed on each realization of the scaling relation . we find that this correlation accounts for an intrinsic scatter of @xmath416 per cent in the scaling relations , which is less than half the total measured scatter for the subsample with the lowest value ( m10 ) . the x - ray observables used in the apex - sz scaling relations are drawn from several different pre - existing studies in the literature . we select the clusters in each of our scaling relations subsamples to ensure that a uniform x - ray analysis is used for each . however , differences between x - ray analyses translate into a systematic bias in the @xmath6 scaling relations for each sample . @xcite used x - ray observables from several different studies and found that the measured , and by extension , relations changed significantly depending on the choice of study . @xcite compared hydrostatic mass estimates of common clusters in three different x - ray samples . these authors found total mass differences as large at 45 per cent at a redshift of 0.2 , resulting in differences in @xmath29 and the aperture used to measure @xmath1 and @xmath2 . these aperture differences are in addition to those regarding the specific fraction of @xmath29 ( e.g. , @xmath479 ) used by individual authors for @xmath2 measurements ( see the note in table [ tbl : physparam ] ) . for cool - core clusters in particular the core region is known to bias estimates of the cluster temperature and is often excluded , but the exact boundary chosen varies between authors . additionally , the outer radius is often determined by data quality limitations . @xcite measured @xmath8 from _ chandra _ observations for a subset of the clusters in @xcite and compared these data with the _ planck _ sze measurements to constrain the relation . these authors found a significantly lower level of intrinsic scatter in comparison with the original _ planck _ + _ xmm - newton _ analysis of the full sample ( @xmath480 per cent and @xmath481 per cent , respectively ) . when considering the same subset of clusters as the _ chandra _ analysis , @xcite measure a @xmath482 per cent intrinsic scatter . @xcite extends this analysis by comparing _ data @xcite with @xmath8 from @xcite and suggest an upper limit on the intrinsic scatter of 15 per cent . these authors conclude that the observed variations are the result of systematic differences in the x - ray analysis . we investigate the importance of uniform x - ray analysis by fitting scaling relations to the full apex - sz sample of 42 clusters and comparing the results to the subsets in table [ tbl : linefit ] . for all three scaling relations , we find the measured intrinsic scatter for a uniformly analyzed subset is less than when considering the full sample . the most dramatic improvement in scatter is seen for the @xcite sample , decreasing for the relation from 41 per cent to 14 per cent . by drawing random sets of clusters from the full sample and performing the regression analysis , we conclude that this level of scatter has a 1 per cent chance of occurring based on the parent distribution and that the observed decrease in intrinsic scatter is significant . as discussed in section [ sec : reflexdxl ] , we find that there are significant variations in both the normalization and exponent for the relation between the three independent subsamples . the relation has less variation in the exponent , while the normalization changes by a factor of two between the @xcite and @xcite subsamples . the relation shows a similar factor of two variation in the normalization . overall , we find that using x - ray parameters from multiple studies introduces significant variability and uncertainty in the power law parameters due to systematic analysis differences . additionally , the level of intrinsic scatter is dominated by systematics when using multiple studies . this increase is expected as the intrinsic scatter attempts to account for the difference in normalizations . there are seven clusters in the full sample that apex - sz did not detect ( defined by @xmath483 ) . these clusters are specified in table [ tbl : mcmc ] . included in the non - detections are the three clusters with the highest redshifts : rdcsj1252 - 2927 ( @xmath484 ) , xmmxcsj2215.9 - 1738 ( @xmath485 ) , and xmmuj2235.3 - 2557 ( @xmath486 ) . these clusters were also observed at 31 ghz by @xcite using the sunyaev - zeldovich array ( sza ) . of the three , only xmmuj2235.3 - 2557 was detected by sza . the apex - sz @xmath6 likelihood distributions for rdcsj1252 - 2927 and xmmxcsj2215.9 - 1738 are consistent with the sza upper limits . @xcite conclude that rdcsj1252 - 2927 and xmmxcsj2215.9 - 1738 are lower mass systems than expected from x - ray measurements . the seven non - detections are not part of the reflex - dxl or z08 samples , and therefore will not impact the results presented in table [ tbl : linefit ] . the m10 sample contains a single non - detection . we investigate the influence of this cluster by comparing the measured scaling relation regression parameters with and without it included and find no signficant difference . a final analysis systematic to consider is the assumption that the @xcite model with our inferred values of @xmath29 accurately represents the true cluster pressure profile . @xcite find that 500 varies by @xmath487 per cent under a variety of model assumptions , including the @xcite profile . the limited angular extent and strong timestream filtering of the apex - sz data prevent us from exploring different model parametrizations . however , we list the pte for the @xcite model fit to each cluster in table [ tbl : mcmc ] . if we exclude clusters which have a poor fit to the chosen @xcite model ( @xmath488 per cent of the sample with pte @xmath489 per cent or pte @xmath490 per cent ) , the power law regression parameters do not change significantly . at the apex - sz observing frequency , emission from point sources can mask the sze signal from a cluster . we estimate the effect of radio sources by extrapolating the flux densities of sources in the nrao vla sky survey ( 1.4 ghz , * ? ? ? * ) , vla first ( 1.4 ghz , * ? ? ? * ) , and parkes - mit - nrao ( 4.85 ghz , * ? ? ? * ) surveys to 150 ghz . assuming a power law spectral energy distribution , @xmath491 , with @xmath492 , there are six clusters with known bright sources with @xmath493k within two arcmin of the x - ray centroid . of these six , two ( rxcj2014.8 - 2430 and macsj1931.8 - 2635 ) are not detected by apex - sz . simulations from @xcite suggest that only 1 per cent of clusters with masses similar to those observed in this paper will have radio sources at 150 ghz that contaminate the integrated cluster signal by 20 per cent or more . for the apex - sz clusters , this implies that a single cluster would suffer from radio contamination and that our estimate of six contaminated clusters is extremely conservative . we have also observed a subset of the apex - sz clusters with the carma array , searching for point source contamination . nine clusters were observed at 100 ghz and three at 230 ghz with an average map rms of approximately 0.9 and 4.4 mjy@xmath494 , respectively . we detect a 7.4 mjy source at 100 ghz in the map of rxcj1504.1 - 0248 . of the three non - detection clusters observed with carma ( xmmuj2235.3 - 2557 , xmmxcsj2215.9 - 1738 , xmmuj1230.3 + 1339 ) no point sources are detected . we test for the effects of these bright point sources by excluding the associated clusters from the appropriate subsamples ( five clusters for the m10 sample , and one each for the reflex - dxl and z08 samples ) . there is no significant difference in the measured scaling relations parameters when removing these potentially contaminated clusters . the final systematic we investigate is a gas mass fraction that changes as a function of total mass . there is evidence in both observations @xcite and simulations @xcite that the gas mass fraction changes due to the thermodynamics and feedback processes in the intracluster medium . as seen in equation [ eqn : scale ] , both the and relation depend on the gas mass fraction . all of the results presented in table [ tbl : linefit ] assume a constant gas mass fraction embedded in the normalization term . we test the effect of a non - universal gas mass fraction by first estimating cluster mass using a scaling relation with @xmath2 from @xcite . we adopt the functional form for gas mass fraction given by the same authors , @xmath495 , where @xmath496 is the total cluster mass @xmath497 in units of @xmath498 . we then look for the effect of changing @xmath34 on the power law exponent parameter in the measured scaling relations . when accounting for the non - universal @xmath34 we find a power law exponent of @xmath499 , @xmath500 , and @xmath501 for the reflex - dxl , z08 , and m10 samples , respectively . these values are consistent with the results where we assume a constant @xmath34 , and are consistent with self - similarity . in addition , the intrinsic scatter increases from @xmath502 to @xmath503 for the reflex - dxl and m10 samples . however , the stated uncertainty in the @xmath34 relation is @xmath504 per cent @xcite and our use of @xmath2 combined with the scatter already present in the relation most likely explains the increase . in summary , we examine the role of millimeter - wavelength point sources and a non - universal gas mass fraction as astrophysical systematics in the apex - sz scaling relations . we find no evidence for bias or additional intrinsic scatter due to point sources . accounting for a non - universal gas mass fraction results in a slight change to the measured power law and increases the intrinsic scatter . we present sunyaev - zeldovich effect ( sze ) observations of galaxy clusters measured with the apex - sz experiment and use them to investigate the scaling of the sze with cluster mass . we model the thermal pressure of the intracluster medium using the @xcite universal pressure profile and use the results to calculate the spherical integrated comptonization 500 for each cluster . we compare these measurements of 500 to x - ray estimates of cluster mass taken from the literature . using these two sets of observables , we measure the , , and scaling relations , finding the best - fit power law and intrinsic scatter for each . these scaling relations are measured for three subsamples of the 42 apex - sz clusters that have uniform x - ray analysis : the reflex - dxl sample , 15 clusters from z08 , and 19 clusters from m10 . we find when all 42 clusters with varying x - ray analyses are included , significant systematics are introduced into the power law regression parameters and the intrinsic scatter increases . for all three subsamples , we find that the best - fit power laws for the , , and relations have exponents consistent with those predicted by the self - similar model . we compare the measured normalizations for each scaling relation to numerical simulations to probe the underlying astrophysics of the intracluster medium ( icm ) . the measured normalization of the relation for two of the subsamples implies a higher @xmath505 than seen in previous studies . however , there is a large variation in the normalization between the three samples , despite the uniformly analyzed x - ray data . therefore , we associate a large systematic uncertainty with the high @xmath505 and do not draw further conclusions from it . we compare the normalization for the and relations to the numerical simulations of @xcite and find a weak preference for models that included radiative cooling and feedback in the icm as well as standard gas dynamics . finally , we find that the levels of intrinsic scatter for the and relations are consistent with previous measurements . the uncertainties in the intrinsic scatter are large due to the small number of clusters in the three subsamples . a larger sample is key to improving this measurement of intrinsic scatter in 500 and its application as an estimator for total cluster mass in cluster - based cosmological constraints . a uniform x - ray analysis for the full sample of apex - sz clusters is ongoing . this analysis will resolve the main systematic limitation of the full cluster sample and will allow us to use its large number of clusters to improve on the constraints presented here for the three subsamples . additionally , a follow - up program of optical observations has been concluded to estimate the total cluster mass independently using weak - lensing measurements . future papers will use the additional x - ray and weak - lensing information to improve further our understanding of how 500 scales with total cluster mass and facilitate its use in exploring the physics of the icm as well as in constraining cosmological parameters using galaxy clusters . we thank the staff at the apex telescope site , at the time of observations led by david rabanus and previously by lars - ke nyman , for their dedicated and exceptional support . we also thank bradford benson for useful comments on a draft version of the manuscript as well as lbnl engineers john joseph and chinh vu for their work on the readout electronics . apex - sz is funded by the national science foundation under grant nos . ast-0138348 & ast-0709497 . work at lbnl is supported by the director , office of science , office of high energy and nuclear physics , of the u.s . department of energy under contract no . de - ac02 - 05ch11231 . work at mcgill is supported by the natural sciences and engineering research council of canada , the canada research chairs program , and the canadian institute for advanced research . this work has been partially supported by the dfg transregio program tr33 `` the dark universe '' . fp acknowledges support from the bmbf / dlr grant nr . 50 or 1117 . ch acknowledges support from the barbro osher pro suecia foundation and from the swedish research council under grants 2006 - 3356 and 2009 - 4027 . m.k . acknowledges support by the german bmwi through the verbundforschung under grant 50or1107 . nwh acknowledges support from the alfred p. sloan foundation .

sub - subgiant stars ( ssgs ) lie to the red of the main - sequence and fainter than the red giant branch in cluster color - magnitude diagrams ( cmds ) , a region not easily populated by standard stellar evolution pathways . while there has been speculation on what mechanisms may create these unusual stars , no well - developed theory exists to explain their origins . here we discuss three hypotheses of ssg formation : ( 1 ) mass transfer in a binary system , ( 2 ) stripping of a subgiant s envelope , perhaps during a dynamical encounter , and ( 3 ) reduced luminosity due to magnetic fields that lower convective efficiency and produce large star spots . using the stellar evolution code mesa , we develop evolutionary tracks for each of these hypotheses , and compare the expected stellar and orbital properties of these models with six known ssgs in the two open clusters m67 and ngc 6791 . all three of these mechanisms can create stars or binary systems in the ssg cmd domain . we also calculate the frequency with which each of these mechanisms may create ssg systems , and find that the magnetic field hypothesis is expected to create ssgs with the highest frequency in open clusters . mass transfer and envelope stripping have lower expected formation frequencies , but may nevertheless create occasional ssgs in open clusters . they may also be important mechanisms to create ssgs in higher mass globular clusters .

1703.10181

optical color - magnitude diagrams reveal that @xmath0 of the evolved stars in older open clusters do not fall along standard single - star evolutionary tracks . these stars include the well - known blue stragglers , but also the yellow giants and sub - subgiants . sub - subgiant stars ( ssgs ) were first identified in the color - magnitude diagram ( cmd ) of the open cluster m67 @xcite . these two ssgs fall to the red of both the main - sequence and main - sequence binary track and well below the subgiant branch . both ssgs have high membership probabilities based on both proper - motion and radial - velocity ( rv ) data , leaving a negligible probability that both are field interlopers . broadly speaking , the populations of ssgs in globular and open clusters share similar characteristics . they fall to the red of the main sequence and below the subgiant and giant branch on optical cmds , a region that can not be easily populated by either single - star evolutionary theory or by any combination of two normal cluster stars . they are also typically x - ray sources with @xmath1 erg s@xmath2 and photometric variables with periods between 1 and 20 days . where binary status is known , they are often found to be close binary systems with orbital periods on the order of 1 - 10 days . similar x - ray sources and photometric variables are also found to red of the rgb . we call these stars red stragglers " rather than sub - subgiants , though the two types may be related and have similar formation mechanisms . geller et al . 2017a give a census of the open cluster and globular cluster red stragglers and ssgs known from the literature . no well - developed theory has yet been presented for the origin and evolutionary status of these non - standard stars . @xcite suggest they may be products of close stellar encounters involving binaries , or stars with enhanced extinction ( i.e. due to the presence of circumstellar material ) . other authors invoke mass transfer and stellar collision events to form ssgs @xcite . while many ssgs are kinematic cluster members , most do not have binary orbital information ( geller et al . 2017a ) . the sample of ssgs with both high quality 3d kinematic memberships and known orbital solutions for the binaries is small , consisting of 6 stars in two open clusters in the wiyn open cluster study ( wocs ; @xcite ) : 4 ssgs ( 3 binaries and one single star ) in ngc 6791 @xcite and 2 binaries in m67 @xcite . we use this sample to guide the formation of an origin theory that matches the observed properties of this sample . we focus on three hypotheses for ssg formation : mass transfer in a binary system , stripping of a subgiant s envelope , or a reduced luminosity due to the presence of a strong magnetic field . our sample of six ssgs is drawn from two wocs open clusters : m67 and ngc 6791 . cmds for both clusters are shown in figure [ cmds ] with the locations of the ssgs highlighted . located at @xmath3 , @xmath4 ( j2000 ) , m67 is an old , solar - metallicity open cluster ( e.g. @xcite ) . distance measurements for the cluster range from 800 - 900 pc , with reddening measurements ranging from e@xmath5 to 0.056 @xcite . for this study , we adopt e@xmath6 @xcite and @xmath7 @xcite . age determinations put the cluster at around 4 gyr ( e.g. @xcite ) with a main - sequence - turnoff mass of @xmath8 m@xmath9 . located at @xmath10 , @xmath11 ( j2000 ) , ngc 6791 is an old ( 8 gyr ; @xcite ) and metal - rich ( [ fe / h]=+0.40 , @xcite ) open cluster . distance measurements put the cluster at around 4 kpc ( e.g. @xcite ) . the turn - off mass of the cluster is @xmath12 m@xmath9 @xcite . for this study we use the distance modulus and reddening values found by @xcite : e@xmath13 , @xmath14 . in table 1 we list the wocs i d , coordinates , proper - motion membership probabilities ( p@xmath15 ) , and the radial - velocity membership probabilities ( p@xmath16 ) for the six ssgs in our sample . we also include the @xmath17 photometry from @xcite for ngc 6791 , and @xcite for m67 . in the comments section we include other identifiers for these targets from previous studies . five of the six ssgs in our sample are binary systems , and for these we also list periods and eccentricities in table 1 . one ssg is a double - lined spectroscopic binary ( sb2 ; wocs 15028 ) , and the other four are single - lined ( sb1s ) . all but one of these binaries are circular , and they all have short periods ranging from 2.8 to 18.4 days . while it is possible that any one ssg could be a field contaminant , given the kinematic memberships the probability that all of these systems are field stars is quite low . @xcite provide a membership analysis for the m67 ssgs , calculating a @xmath18 probability that one of the 246 3d kinematic members in their sample is a nonmember . the probability of finding 2 nonmembers is just 0.4@xmath19 . this is within their entire sample of kinematic members , so the likelihood that the 2 ssgs specifically are field stars is smaller still . @xcite provide a similar analysis of the ngc 6791 ssgs . their analysis , based on the kinematic membership probabilities and the cmd location of the stars , indicates that it is highly unlikely for all four stars to be field contaminants . specifically , they calculate a 17% probability that one of the four ssgs is a field star , dropping to just 1.8% chance that two ssgs are field stars , 0.13% for 3 , and 0.007 % for all 4 . we are thus confident that our sample of 6 ssgs can not be explained simply by field contamination . cccccccccccrrc[htbp ] m67 & 15028 & 08 51 25.30 & + 12 02 56.3 & 97 & 99 & 13.77 & 1.01 & & 2.823094 & 0 & s1113 + & & & & & & & & & @xmath20 & @xmath21 & + + m67 & 13008 & 08 51 13.36 & + 11 51 40.1 & 98 & 98 & 13.79 & 1.05 & & 18.396 & 0.26 & s1063 + & & & & & & & & & @xmath22 & @xmath23 & + + ngc 6791 & 130013 & 19 21 25.22 & + 37 45 49.82 & 99 & 84 & 17.65 & & 1.53 & 7.7812 & 0.015 & 15561 + & & & & & & & & & @xmath24 & @xmath25 & + + ngc 6791 & 131020 & 19 20 10.61 & + 37 51 11.20 & 96 & 85 & 18.30 & & 1.50 & & & 83 + & & & & & & & & & & & + + ngc 6791 & 147014 & 19 20 21.48 & + 37 48 21.60 & 99 & 95 & 17.96 & 1.35 & 1.39 & 11.415 & 0.05 & 746 + & & & & & & & & & @xmath26 & @xmath27 & + + ngc 6791 & 170008 & 19 20 38.88 & + 37 49 04.29 & 99 & 63 & 17.96 & 1.15 & 1.29 & 5.8248 & 0.013 & 3626 + & & & & & & & & & @xmath28 & @xmath29 & + in order to measure the physical characteristics of the open cluster ssgs , we pieced together spectral energy distributions ( seds ) from existing optical observations @xcite and photometry from the two - micron all - sky survey ( 2mass ; @xcite ) , wide field infrared explorer ( wise ; @xcite ) , and spitzer space telescope infrared array camera @xcite . we used these seds to fit a temperature and radius to each star and determine the bolometric luminosities of the systems . l r r r r r r r r r r r [ htbp ] 15028 & 15.3 & 14.78 & 13.77 & 13.09 & & 11.671 & 11.123 & 10.971 & 10.84 & 10.822 & 10.681 + & & & & & & @xmath30.021&@xmath30.023&@xmath30.022&@xmath30.023&@xmath30.021&@xmath30.086 + + 13008 & 15.56 & 14.84 & 13.79 & & 12.59 & 11.657 & 11.058 & 10.958 & 10.810 & 10.855 & 10.643 + & & & & & & @xmath30.022&@xmath30.019&@xmath30.018&@xmath30.022&@xmath30.019&@xmath30.101 + + 130013 & & & 17.654 & & 16.127 & 15.197 & 14.495 & 14.485 & 14.16 & 14.337 + & & & @xmath30.0052 & & @xmath30.0117 & @xmath30.047&@xmath30.050&@xmath30.086&@xmath30.028&@xmath30.043 & + + 131020 & & & 18.3 & & 16.8 & 15.791 & 15.040 & 14.780 & & & + & & & & & & @xmath30.069&@xmath30.071&@xmath30.101 & & & + + 147014 & & 19.305 & 17.957 & & 16.563 & 15.532 & 14.810 & 14.707 & 14.643 & 14.77 & + & & @xmath30.0076 & @xmath30.0028 & & @xmath30.01 & @xmath30.059&@xmath30.057&@xmath30.101&@xmath30.033&@xmath30.051 & + + 170008 & & 19.116 & 17.962 & & 16.670 & 15.795 & 15.248 & 15.055 & & & + & & @xmath30.0069&@xmath30.0012&&@xmath30.0011&@xmath30.061&@xmath30.08&@xmath30.1029 & & & + we performed a @xmath31-minimization between the observed photometry and a grid of castelli - kurucz models @xcite convolved with filter transmission functions . we fit only t@xmath32 ( k ) and r ( r@xmath9 ) while fixing the distance and reddening to cluster values . altering the distance does significantly affect the values of radius and bolometric luminosity we determine , and so we ran our code using a range of distance values ( 3900 - 4100 pc for ngc 6791 ; 800 - 900 pc for m67 ) found in the literature to better determine the range in radius and luminosity . photometry used in these seds is listed in table [ tab : photometry ] . because 5 of the 6 systems are known photometric variables , the larger source of error in some bandpasses is the intrinsic variability of the system and not the photon statistics . the amplitude of this variability ranges from a few percent in the v band , up to 0.26 mags for wocs 130013 , the most variable ssg @xcite . in order to fit the sed we therefore use the amplitude of the variability of the star rather than the photometric errors for the optical photometry . this variability is known for each star in at least the v band . where the amplitude of variability is known , we use that as our error . if we do not have a measurement of variability in an optical band , we use the v - band variability . spot modeling predicts that ir observations are much less effected by spot modulation , and therefore for 2mass , wise , and spitzer observations we use the photometric errors for the sed fits . in most cases we expect the amplitude of the variability in these bands to be less than the photometric errors . fit parameters for each of the ssgs are given in table 3 . for comparison , we also used our code to fit subgiants near the base of the rgb in ngc 6791 and m67 : wocs 10006 in m67 and wocs 12270 in ngc 6791 . these fits suggest that ssgs are slightly cooler and larger than a typical cluster subgiant . the ssg radii would place them on the lower rgb or near the end of the subgiant branch , but with cooler temperatures and lower luminosities than expected for a typical cluster giant . we note that a few of the stars show tentative evidence of an ir excess . however , given the large uncertainties on some of the wise photometry and the variable nature of the stars , its not clear that this excess is significant . l l c c c l [ fitparams ] m67 & 15028 & 4500 & 2.5 - 2.9 & 2.32 - 3.13 & ssg + m67 & 13008 & 4500 & 2.8 - 3.1 & 2.92 - 3.57&ssg + ngc 6791 & 130013 & 4250 & 2.9 - 3.2 & 2.50 - 3.03&ssg + ngc 6791 & 147014 & 4500 & 2.3 - 2.5 & 1.97 - 2.32&ssg + ngc 6791 & 170008 & 4750 & 1.9 - 2.1 & 1.67 - 2.03&ssg + ngc 6791 & 131020 & 4250 & 2.3 - 2.5 & 1.56 - 1.85 & ssg + & & & & + + m67 & 10006 & 5250 & 2.4 - 2.7 & 3.97 - 5.02 & subgiant + ngc 6791 & 12270 & 5000 & 1.9 - 2.1 & 2.04 - 2.50 & subgiant + one hypothesis for ssg formation is that roche lobe overflow reduces the mass of a subgiant star , lowering its luminosity and moving it into the ssg cmd region . in order to investigate this idea we employ two different stellar evolution codes : binary star evolution ( bse ; @xcite ) and modules for experiments in stellar astrophysics ( mesa ; @xcite ) . we use bse as an efficient tool to search the large progenitor - binary parameter space . we use mesa to produce more detailed models of the evolution of systems bse indicates may produce ssgs . we first used bse to simulate binaries in clusters with parameters matching those of ngc 6791 and m67 ( see section 2.1).the genetic algorithm creates 100 generations of 5000 binaries each , and for each cluster we perform 20 simulations . to begin , we define a sample of 5000 binaries for the first generation with : * primary masses chosen randomly from a uniform distribution between 0.7 m@xmath9 ( well below the msto in both clusters ) and twice the turnoff mass of the cluster , * secondary masses chosen randomly from a uniform distribution between 0.1 m@xmath9 and twice the turnoff mass , * periods chosen randomly from a uniform distribution between 3 days and 5000 days , * and eccentricities chosen randomly from a uniform distribution between 0 and 1 . these distributions cover the relevant initial parameter space , but are not meant to reproduce the true shapes of these binary distributions , for example as observed in open clusters . bse then evolves these 5000 systems up to the age of the cluster ( 4 gyr for m67 , 8 gyr for ngc 6791 ) . we use the default parameters from bse , but we make two changes to the code : 1 ) we increase the strength of the convective tidal damping coefficient by a factor of 100 to correspond with the findings of @xcite , and 2 ) we fix a bug in the implementation of equation 32 in @xcite once the systems have been evolved to the age of the cluster , we evaluate the fitness of each model . we evaluate fitness based on two criteria : the observed location of a system in a bv cmd , and the period of the final binary system . specifically , we define the fitness ( f ) as : @xmath33 where @xmath34 and @xmath35where i = b - v , v , or p , respectively , o refers to the observed color , magnitude or period of the ssgs , and s the color , magnitude , or period of the bse model . for m67 , we took @xmath36 and @xmath37 . for ngc 6791 we took @xmath38 and @xmath39 . for both clusters , we select for short period systems by taking @xmath40 days , and we adopt @xmath41 days , @xmath42 , and @xmath43 . for a second round of models , we sought to produce systems matching the orbital periods of the ssgs in ngc 6791 and m67 . for this we re - ran the genetic algorithm for each cluster , this time taking @xmath44 to be the specific orbital period of each ssg . for 15028 , the sb2 ssg , we also included a fitness term for the mass ratio of the system ( @xmath45 ) with @xmath46=0.7 . after evaluating the fitness of each of the 5000 first - generation models , we take any models with non - zero fitness and these models become parents " for the next generation of models . technically we limit the number of parents per generation to 1000 , but we rarely find more than a few hundred . parents are then allowed to mate " with each other to produce two `` children '' per parent - parent pair in the next generation . to define the children , we begin with the parent that has the highest fitness value , allow it to mate with all other parents , repeat this process for the parent with the second highest fitness value , and so on until we obtain all parent - parent combinations or we produce 3000 children . to produce a child we take a random combination of the initial parameters from each parent . specifically , to determine each initial binary parameter for a child , we draw a random number from a uniform distribution between 0 and 1 ; if the number is @xmath47 , we choose the initial parameter of the first parent , and otherwise we choose the initial parameter from the second parent for the child . if a child duplicates a binary already in the subsequent generation , we impose a mutation where at least one of the binary initial parameters is chosen randomly from the same respective distribution defining the initial parameters if the first generation , and the number of mutated parameters is chosen randomly . we take the @xmath48 children produced in this manner , and fill the remaining @xmath49 spots with binaries whose parameters are chosen from the same distributions as the initial generation . we then evolve this new generation of 5000 binaries with bse up to the cluster age , and the process is repeated for 100 generations . through this procedure , we ensure that subsequent generations climb to higher and higher fitness values , retain the best fitting binaries throughout the generations , and introduce a fresh sample of random binaries in each generation to fill out the parameter space . note that the genetic algorithm does not uniformly sample parameter space , and therefore can potentially miss a peak in the fitness surface . this is alleviated somewhat by the introduction of binaries with randomly chosen initial parameters into each generation . we ran 20 simulations of the genetic algorithm ( with different initial random seeds ) and combine the results to further alleviate this issue . in both sets of runs , those selecting for an orbital period of 10 days and those selecting for specific ssg orbital periods , bse was able to produce systems in the ssg region of a cmd . inspection of these results show that there is a family of solutions that create these ssg systems . in m67 , the progenitor binaries are generally high - eccentricity systems with periods between 3 and 1000 days , with the longest period binaries requiring eccentricities approaching 1 . the secondary star can possess a range of masses between 0.3 and 0.8 m@xmath9 , with the majority being drawn from the lower end of this range . the primary star is a @xmath8 m@xmath9 star that begins roche lobe overflow somewhere on the subgiant branch . this requires that tidal forces circularize and shrink the initial orbit of these systems so that when they evolve through the subgiant branch they have periods of @xmath50 day . we note that our choice of a large tidal strength factor may artificially allow the wide , high eccentricity systems to circularize , but this does not change the conclusion that we require @xmath51 day binaries on the subgiant branch to create ssgs via mass transfer . the final ssg systems are short period ( p @xmath52 days ) circular binaries . the highest fitness solutions have reduced the primary mass to @xmath530.2 m@xmath9 , and increased the secondary mass to @xmath531.0 m@xmath9 , though a wide range of final primary and secondary masses are produced by the algorithm . in ngc 6791 , results are similar . the progenitor binaries are again high - eccentricity systems with periods from 3 days up to 1000 days , with longer period systems requiring higher eccentricities . initial primary masses are @xmath54 m@xmath9 , with secondaries ranging from 0.3 - 0.5 m@xmath9 , with the majority drawn from the lower mass end . the final systems are circular binaries periods of just a few days . the longest period system created was a 3.26 day binary . bse was unable to produce ssgs with periods above 3.5 days in ngc 6791 , or 7 days in m67 . these periods are shorter than 5 out of the 6 ssgs in our sample.the only individual ssg that we had some success at reproducing using the genetic algorithm was 15028 . this 2.8 day binary falls solidly in the period domain that can be created by mass transfer . however , creating an ssg with the observed mass ratio of this system ( @xmath55 ) proved problematic . the genetic algorithm strongly favored smaller @xmath56 values for longer period ssgs , with a @xmath55 possible only for shorter period binaries with @xmath57 day . we note that the algorithm produces longer period ssgs in m67 than in ngc 6791 , which is as expected . the roche lobe radius is given by @xcite as : @xmath58 where @xmath59 is the roche lobe radius , a is the orbital separation , and @xmath56 is the mass ratio of the binary system . the roche lobe radius depends only on the mass ratio and the orbital separation of a binary . the stellar radius depends on a star s mass and evolutionary state . since subgiants in ngc 6791 are lower mass than in m67 , they begin with smaller radii . therefore , for a given mass ratio and orbital separation , ngc 6791 subgiants must reach a more evolved state to exceed their roche lobes . this means they will have evolved further up the giant branch than their m67 counterparts , and thus not evolve through the ssg domain as they lose mass . in order for an ngc 6791 subgiants to evolve through the ssg region as they lose mass , they must then have smaller orbital separations than are required for m67 subgiants . the results indicate that , while we do expect mass transfer to create ssg systems , we would expect the observed periods to be in the range of just a few days in both clusters , shorter than most of the observed ssgs . bse therefore suggests that none of the ngc 6791 ssgs have short enough periods to be mass transfer systems . similarly , while mass transfer may create a 2.8 day binary like the one observed in m67 , the expected mass ratio would be much smaller than the observed @xmath55 , and bse is unable to create the other 18.4 day binary . the bse - based genetic algorithm is an excellent tool for exploring a wide range of parameter space , but the stellar evolution and mass transfer calculations are highly parameterized . we therefore also use the more detailed evolution code modules for experiments in stellar astrophysics ( mesa ; @xcite ) to create ssgs via stable mass transfer , and compare the results of these mesa models to our observations . for m67 we initially ran a coarse grid of models with input based on the bse results . these models all had a primary mass of @xmath60 m@xmath9 and companion masses in the range @xmath61 m@xmath9@xmath62 m@xmath9 . we evolved each component of the binary up to core hydrogen exhaustion before placing it in a binary with a period between 1 and 10 days and allowing the evolution to proceed . we start from the test suite case ` binary_both_stars ` , use non - rotating models , and do not include magnetic braking . we used three different mass transfer efficiencies : @xmath63 , and @xmath64 . @xmath65 is defined such that fully conservative mass transfer has @xmath66 , and if no mass is transfered @xmath67 . mass and angular momentum are presumed to be lost from the vicinity of the primary in these models . models with periods above @xmath68 days began roche lobe overflow after beginning their ascent up the red giant branch , and mass transfer products did not evolve through the ssg region . in many cases mass transfer was dynamically unstable and we terminated their evolution , as mesa can not handle these cases . it is generally assumed that binaries undergoing unstable case b mass transfer enter a common envelope phase that ends with the spiraling in of the binary and an ejection of the common envelope material @xcite . the end product of this phase is either a shorter period white dwarf - main sequence binary or a merger to create a single star . if this is correct , we conclude these longer period systems do not create ssgs . models with periods less than @xmath68 days began roche lobe overflow while still on the subgiant branch and did proceed through the ssg region as mass transfer proceeded . for these shorter period models we ran a more detailed grid of models with periods between 0.6 and 2.0 days , @xmath60 m@xmath9 , varying @xmath69 from @xmath61 m@xmath9@xmath62 m@xmath9 , and using three mass transfer efficiencies @xmath66 , @xmath70 , and @xmath64 . our models indicate that systems with initial periods @xmath71 days begin rlo prior to evolution onto the subgiant branch and do not pass through the ssg region . models with @xmath72 days@xmath73 days will often evolve through the ssg region depending on @xmath69 and mass transfer efficiency . models with @xmath74 days may have primaries that evolve through the ssg region , but mass transfer is only stable in these systems if they have a near equal - mass secondary . due to the required mass of the secondary , the combined light of the binary does not pass into the ssg domain . therefore , it appears that only a very narrow range of systems with periods right around @xmath75 day begin mass transfer at the right time in their evolution , with faint enough companions to move through the ssg domain . as an example , we show a grid of mesa models with a 1.0 day period , 1.3 m@xmath9 primary and a range of secondary masses from 0.3 m@xmath9 to 1.25 m@xmath9 in figure [ fig : mesagrid ] with the secondary mass indicated by the color of the track . in this grid , for a large range of secondary masses and mass transfer efficiencies , the primary will evolve near the domain of the m67 ssgs ( see figure 3 , middle row . however , only for models beginning with a fairly low - mass secondary and a moderate degree of non - conservativeness will the combined light of the binary evolve through this region ( figure 3 , top row ) . for example , see the top middle plot . only the model binaries with progenitor secondaries of 0.5 and 0.7 m@xmath9 are observed in the ssg region during their evolution . the evolution of the accreting stars in these models is also interesting . we assume the accreting star in these binaries is a main sequence star with initial mass ranging from 0.3 to 1.25 m@xmath9 . for models where mass transfer can proceed stably and we assume a significant fraction of the mass lost from the primary is accreted by the secondary , the secondaries may gain several tenths of a solar mass of material . this accretion causes them to move up the main sequence to a position corresponding to their new larger mass . if the accreting star starts close enough to the main sequence turn off , or we assume a large mass - transfer efficiency , the accreting star may move above the main - sequence turn - off and in to the blue straggler region ( for example , see figure 3 , lower left plot ) . as these accretors begin to evolve off the main sequence , their higher mass causes them to follow an evolutionary track brighter and bluer than the cluster isochrone . such ` yellow stragglers ' have been observed in m67 and other clusters , and are believed to be evolved blue stragglers @xcite . in this mass transfer scenario , then , ssgs could be part of the same evolutionary pathway that leads to the formation of other non - standard stars like the blue and yellow stragglers . we ran another grid for ngc 6791 , using a primary of @xmath76 m@xmath9 , varying @xmath69 from @xmath61 m@xmath9 @xmath77 m@xmath9 , using three mass transfer efficiencies @xmath66 , @xmath70 , and @xmath64 , and periods between 0.6 and 10.0 days . results were similar to those of m67 . the models indicate that systems with initial periods @xmath78 days moved through the ssg region during rlo . as in m67 , longer period systems began mass transfer on the lower giant branch , and if mass loss proceeded stably the model evolved to the red of the rgb , not down into the ssg region . we compare this finding to the results of the bse models . shown in figure [ period_histograms ] is a histogram of orbital periods of the ssg binaries produced in bse models of m67 and ngc 6791 ( section [ section : bse ] ) . we show both the orbital period of the system at onset of rlo , and the final period of the binary at 4 gyr when it is observed as an ssg . we also show the shortest and longest period ssg created with mesa with vertical dashed blue lines . note that our grid of mesa models has a resolution of 0.1 days in period . we conclude that the mesa results are consistent with our findings from bse that only binaries in @xmath51 day orbits at onset of rlo will move through the ssg region . given this period range for ssg formation , we can use the period distribution found by @xcite to estimate the number of predicted ssg systems in a cluster . specifically , @xcite fit a gaussian function to the distribution of orbital periods found in a large sample of binary systems . they find @xmath79=5.03 and @xmath80 . integrating this function using the upper and lower period bounds found for m67 ( @xmath81 ) and ngc 6791 ( @xmath78 ) we would expect only @xmath82 of binaries to go through this evolution . assuming a @xmath83 binary fraction and given a population of @xmath84 subgiant stars in m67 , and @xmath85 subgiant stars in ngc 6791 ( see section 6.4 ) , this results in a poisson probability of observing one or more mass transfer ssgs in m67 of 4@xmath19 and in ngc 6791 of @xmath86 . this estimate should be regarded as an upper limit , as period is not the only factor which determines whether a star moves through the ssg region . for example , if the companion mass is very small , mass transfer may be unstable and the system would not evolve as an ssg . conversely , if the companion is too close to the main sequence turn off , the secondary may overwhelm the lower - luminosity primary and move the system into a more standard region of the cmd or into the blue straggler domain . given this small number , it is unlikely , though not impossible that we would observe a mass - transfer ssg in ngc 6791 or m67 . however , this mechanism is unlikely to explain all the ssgs observed in m67 or ngc 6791 . the poisson probability of producing 2 ssgs in m67 or 4 in ngc 6791 from mass transfer is negligible . however , in a larger cluster it may be quite likely to observe at least one mass transfer ssg . in a companion paper , geller et al . 2017b ( in preparation ) , we investigate in more detail the expected formation frequency of this and other formation mechanisms across a wide range of cluster properties . a serious mismatch between mass transfer models and observations is that model mass transfer ssgs have shorter periods than most ssgs observed in m67 and ngc 6791 . one method to produce longer period binaries undergoing mass loss is to adopt a model in which the primary can lose substantial mass via a wind while still well within its roche lobe . the tidally enhanced wind model proposed by @xcite proposes that tidal interactions and magnetic activity drive a stronger stellar wind in close binary systems than in a typical single star . they assume that the wind can be described by the standard reimer s wind for rgb stars , multiplied by a factor that has the same dependence on stellar radius ( @xmath87 ) and roche lobe radius ( @xmath88 ) as a tidal torque . specifically , their expression is : @xmath89)\\ \mbox{where } \dot{m}_{reimers}=-4 \times 10^{-13 } \bigg ( \frac{rl}{m}\bigg ) \end{aligned}\ ] ] where r , l , and m are in solar units and time is in years . here they assume that wind mass loss saturates when @xmath90 . this wind prescription includes a constant multiplicative factor ( @xmath91 ) that may be varied to achieve greater or lesser mass loss rates . @xcite calibrate this constant to match the properties of the system z her , a detached rs cvn binary with a mass ratio inversion in which the more evolved star is near the end of the subgiant branch and is less massive than its near - turnoff companion . its orbital period is @xmath92 days . they find @xmath93 well matches the observed mass loss from the primary . using a tidally enhanced wind model can reproduce an ssg similar to 15028 ( figure 5 ) . this model has an initial primary mass of 1.3 m@xmath9 , a period of 2.8 days , and a coefficient of @xmath94 , twice as large as that proposed in @xcite . the mass loss rates on the subgiant branch required are on the order of @xmath95 m@xmath9 yr@xmath2 . the star has a mass of just 0.95 m@xmath9 when it reaches the cmd location of 15028 . while we have no direct mass measurement of 15028 , we do have a mass ratio of @xmath55 from the orbital solution . given this mass ratio , the 0.95 m@xmath9 mass from the tidally - enhanced wind model would imply a secondary of @xmath96 m@xmath9 , in which case the observed secondary is substantially hotter and more luminous than expected for a @xmath96 m@xmath9 star @xcite . alternatively , assuming the rotational and orbital axes are aligned , a secondary mass of @xmath97 m@xmath9 well matches the photometry , spectroscopic temperature , and mean density of the secondary star @xcite . this would imply a mass of 1.3 m@xmath9 for the primary , indicating a subgiant that has not lost substantial mass . however , the luminosity ratio of the system is not consistent with the alignment of the axes , and @xcite were not able to find a fully self - consistent model for the system . the ssgs in ngc 6791 are in longer - period orbits than 15028 . they also presumably start with smaller radii if they are normal cluster subgiants undergoing mass loss . therefore , the tidal wind enhancement does not produce a noticeable effect in the models until the stars have evolved substantially up the giant branch . even increasing the b parameter by a factor of 10 is unable to create observed systems near the location of the ngc 6791 ssgs . similarly , a tidally enhanced wind model for the 18.4 day binary in m67 also does not produce significant mass loss until the primary is substantially more evolved . these stellar models never move through the ssg region . overall the wind prescription of @xcite is unable to reproduce the cmd location of any of the ngc 6791 ssgs or the 18.4 day ssg in m67 using a value of b close to what is typically assumed . these stars are just not close enough to their roche radii to have large mass loss rates using this model . this wind prescription can create the 2.8 day ssg in m67 by losing @xmath98 m@xmath9 from a subgiant primary , but it is not clear from the observational evidence that this star has lost substantial mass . we conclude that wind mass loss rates are likely not large enough to be the sole reason for the ssgs under - luminosity . another possibility is that ssgs could be created as a result of removing the envelope of a subgiant star . rapid envelope mass loss yields a rapid decrease in luminosity , and subsequent evolution below the current subgiant branch to a lower - mass red giant branch . such stripping could occur in a number of ways . one suggestion is that such mass loss may occur if a subgiant star has a close encounter with a passing star . this may occur , for example , during a resonant binary encounter ( e.g. @xcite ) . if the impact parameter of the passage is sufficient to disrupt and remove a large fraction of the stellar envelope , but not close enough to lead to a merger , an ssg - like star might result . in order to explore the effect of envelope mass loss on a subgiant , we remove mass from a subgiant star at a high , constant rate using mesa , stopping the mass loss after the star has lost a few tenths of a solar mass of material from its envelope and allowing the star to continue to evolve without further mass loss . in principle , any mass loss rate in which the subgiant can lose a few tenths of a solar mass of material within its subgiant lifetime can move the subgiant into the ssg domain . this requires mass loss rates @xmath99 m@xmath9 @xmath100 . in practice , mesa does not handle dynamical mass loss , and thus the highest mass loss rate for which we achieve numerical stability is @xmath101 m@xmath9 @xmath2 . as an example , we show an m67 model in which we strip mass from a 1.3 m@xmath9 subgiant star at a rate of @xmath101 m@xmath9 yr@xmath2 , stopping the mass loss when the subgiant reaches 0.8 , 0.9 , 1.0 , or 1.1 m@xmath9 . for an ngc 6791 model , we strip mass from a 1.1 m@xmath9 star , stopping mass loss when the subgiant reaches 0.7 , 0.8 , or 0.9 m@xmath9 . plots of the resulting mesa evolutionary tracks are shown in figure [ ngc6791_dynamical ] . the high mass loss rates produce models that are out of thermal equilibrium , causing a rapid drop in luminosity . when mass loss is terminated , the models quickly return to equilibrium and resume evolution along a subgiant / giant track . due to their newly reduced masses , these tracks are at lower temperature and luminosity than a normal cluster giant . a mass loss rate this large or larger would be required to strip enough envelope material to produce an ssg during a short - duration event like a dynamical encounter however , lower mass loss rates between @xmath102 and @xmath103 m@xmath9 yr@xmath2 could also produce stars in the ssg region if the duration of the stripping event is longer . lower rates produce a more gradual decline in luminosity and do not drive the star out of thermal equilibrium ( similar to the mass transfer models in fig . 3 ) , but the models still move through the ssg region if they begin mass loss early enough on the subgiant branch to lose several tenths of a solar mass before beginning their ascent up the giant branch . these models indicate that mass loss of 0.3 - 0.4 m@xmath9 on the subgiant branch can produce stars in the ssg domain for a wide range of mass loss rates . these mesa models indicate that if a subgiant loses significant envelope mass it will move in to the ssg cmd region . we conjecture that one possible mechanism to remove this envelope mass would be a grazing dynamical encounter . in this encounter scenario , another star would have to pass close enough to a subgiant to tidally strip envelope material , but not close enough to lead to a merger . a similar mechanism has been proposed to explain the depletion of red giants near the milky way galactic center @xcite . in this scenario , encounters between rgb stars and black holes can eject the red giant core from the envelope . the core retains only a fraction of the envelope material , creating a giant with a significantly reduced envelope mass . they also find rgb - ms encounters capable of ejecting envelope material if the impact parameter is small enough . similar models of encounters at the galactic center have found that encounters between rgb stars and ms stars , white dwarfs , or neutron stars can cause significant stripping of a giant envelope , though the amount of mass loss varies substantially between these studies . depending on the specifics of the encounter and assumptions of the models , some conclude less than 10 % of the envelope mass will be ejected @xcite , and others find nearly the entire envelope may be lost @xcitef . at this stage of evolution , losing even a large fraction of the envelope does not prevent the giant from continuing its evolution up the giant branch , and it does not move into the ssg cmd region @xcite here we suggest similar encounters with subgiants or early giants may create ssgs . however , no existing studies focus specifically on subgiant encounters . new hydrodynamic simulations for subgiants would be necessary to advance this hypothesis , specifically determining the range of impact parameters that yield substantial mass loss while avoiding common envelope mergers as well as determining the possible orbital parameters for a post - encounter binary . to explore the frequency of such dynamical encounters , we consider the case of the m67 ssgs . using the encounter rates presented in @xcite , we find the time between single - binary encounters to be @xmath104 yrs . to get the encounter rate for subgiants , we can scale this rate by the fraction of stars in the cluster that are subgiants or early giants . in m67 , we observe @xmath105 subgiants or early giants . adopting the total number of stars in m67 to be @xmath106 @xcite results in a subgiant fraction of @xmath107 . scaling our single - binary encounter rate by @xmath107 , we find a time between single - binary encounters involving a subgiant to be @xmath108 years . if we assume that all of these encounters lead to an ssg , and that the ssg is observable for its entire subgiant lifetime of @xmath109 myr in m67 , we find the poisson probability of observing an ssg formed via a single - binary encounter in m67 to be @xmath110 . using the same assumptions , binary - binary encounters also result in a @xmath111 chance of observing an ssg . while encounters with triples may also play a role , the smaller number of triples makes binary - binary or single - binary encounters the dominant encounter types . more likely only a small fraction of encounters involving subgiants would strip the subgiant s envelope rather than leading to a merger , resulting in a very low probability of observing a dynamically formed ssg in m67 . dynamically formed ssgs may be more likely to be observed in larger clusters . we investigate this channel in more detail in the companion paper geller et al . 2017b , including in globular cluster environments where the larger core densities and higher encounter rates may make this a more likely mechanism . we have proposed two mechanisms that create ssgs via mass loss from a subgiant : 1 ) mass transfer in a binary system and 2 ) tidal stripping of a subgiant s envelope during a dynamical encounter . whether from mass transfer or envelope stripping , the essential finding in sections 3 and 4 is that mass loss from a subgiant of several tenths of a solar mass successfully creates stars in the ssg cmd region . our explorations suggest that neither roche lobe overflow nor tidal stripping via dynamical encounters produce ssgs with high enough frequency to explain the observations . however , we have not fully explored all mechanisms of stellar mass loss , and there may be other ways for a subgiant to lose substantial envelope mass . for example , if a binary system with a subgiant primary went through a period of common envelope evolution , ejecting some but not all of a giant s envelope material before the mass loss stabilized , the remaining system might resemble an ssg . @xcite create an ssg in an n - body simulation of m67 using a similar mechanism . in their model , two stars merge in a common envelope event , ejecting 0.29 m@xmath9 masses of material in the process . the resulting star lies below the subgiant branch . as a second example , an ssg with a millisecond pulsar companion has been found in the globular cluster ngc 6397 @xcite . the ssg is a giant that is extremely under - massive ( 0.22 m@xmath9- 0.32 m@xmath9 ) because it is being evaporated by the wind from its pulsar companion @xcite . while the n - body ssg is not in a binary and open clusters are unlikely to have millisecond pulsars , we encourage further exploration of more mass - loss hypotheses both in open cluster and other environments . it is also worth noting that dynamical encounters between main sequence stars are common , and can lead to the collision of two main sequence stars to form a single object . such collision products are out of thermal equilibrium immediately after collision , and become much brighter than the main sequence due to the energy deposition in their envelopes during the encounter @xcite . as these products settle back into equilibrium , they contract and move back towards the main sequence . this occurs over a thermal timescale of a few myr , but during this time such a collision product may be found in the ssg cmd domain @xcite . due to the short duration of this phase compared to the single - binary and binary - binary encounter rate for m67 derived in section [ sec : encounterrate ] , the poisson probability of observing such a collision product is just a few percent . these encounter products are expected to be rapidly rotating single stars , not close binaries . scattering experiments find that it is difficult for collision products to retain close binary companions ( e.g. @xcite ) , and thus we do not consider this a likely explanation for the systems in our sample . while we do determine one of the ssgs to be a single star , it is not observed to be rotating rapidly , and thus is unlikely to be a recent collision product . again , collisions are more common in denser globular clusters , and we explore the frequency of this formation mechanism in geller et al . 2017b . five of the six stars show evidence that they possess strong surface magnetic fields : x - rays from a hot corona , h@xmath65 emission from chromospheric activity , and starspots from areas of concentrated magnetic flux that inhibit convection and lower surface temperatures . @xcite determine 0.1 - 2.4 kev x - ray luminosities for the two ssgs in m67 of @xmath112 ergs s@xmath2 using rosat observations . @xcite find a 0.3 - 7 kev luminosity of @xmath113 ergs s@xmath2for wocs 13008 using chandra - acis , while wocs 15028 is outside their field of view . @xcite obtain chandra observations of ngc 6791 , detecting 3 of the 4 ssgs as 0.3 - 7 kev x - ray sources with luminosities of @xmath114 ergs s@xmath2(15561 ) , @xmath115 ergs s@xmath2(746 ) , and @xmath116 ergs s@xmath2(3626 ) . these x - ray luminosities are consistent with coronal emission due to magnetic activity . all five of these ssg x - ray sources are also observed to be photometric variables with periods on the order of a few days . the lowest amplitude variable has @xmath117v=0.02 - 0.09 depending on the variability survey ( wocs 170008 ; @xcite ) . the largest amplitude variable has @xmath117v=0.26 ( wocs 130013 ; @xcite ) . in all cases , the variability is attributed primarily to spot activity on the primary star , with perhaps ellipsoidal variations contributing in some cases . @xcite present an overview of all known measurements of variability in the ssgs in ngc 6791 . @xcite present variability information for the 2 ssgs in m67 . for 4 of these 5 stars , the variability is found to have periods close to ( but not exactly at ) the orbital period of the system . for the 5th system , 13008 in m67 , they find variability , but do not have a time baseline long enough to determine the periodicity of the variability . finally , these 5 ssgs are also observed to have h@xmath65 emission @xcite , indicative of chromospheric activity . these 5 ssgs are the 5 binaries in our sample , all with orbital periods between 2.8 and 18.4 days . taken together , these observations indicate these 5 ssgs are magnetically active binaries similar to rs cvn systems . the 6th ssg ( wocs 131020 ) shows no evidence of x - ray emission , h@xmath65 emission , or photometric variability . it is also the only one of the 6 stars that is not observed to be a binary . because of this lack of evidence for magnetic activity in wocs 131020 , we suggest that the magnetic field hypothesis is not well suited to explaining its existence . @xcite conclude that while it is unlikely all 4 ssgs in ngc 6791 are field contaminants , it is possible that at least one of them is . perhaps 131020 is this field contaminant . it is also possible that wocs 131020 is formed via a different formation channel that does not yield binarity , rapid rotation , and magnetic activity . @xcite determine spectroscopic temperatures a few hundred k warmer than our best - fit sed temperatures in the m67 ssgs . one possible explanation is that the ssgs are known photometric variables , and spot activity on the primary could skew the sed fits towards lower @xmath118 . here we revisit our analysis of the ssg seds from section 2 to analyze the impact of spots on the determination of radius , temperature , and luminosity . for example , we refit the 15028 sed to include a 3500 k spot while varying the spot covering fraction . we perform this fit using the same approach detailed in section 2.3 , but add the spot covering fraction ( @xmath119 ) as an additional free parameter . the 3500 k temperature is motivated by the known relation between photospheric temperature and spot temperature @xcite . a 4500 - 5000 k would have a temperature contrast of @xmath531000 k according to this relation . here we model the sed as a combination of two seds weighted by the covering fraction ( @xmath119 ) , one representing the temperature of the unspotted photosphere , and the other a 3500 k sed representing the spotted photosphere . the new best - fit parameters for 15028 including a spot are t@xmath32=4750 k , r=3.1 r@xmath9 , with a 3500 k spot ( or spots ) accounting for 40@xmath19 of the surface area . adding a spot to our model provides a slightly better fit to the photometry and is consistent with the spectroscopically determined @xmath118 of 4800 @xmath30 150 k @xcite . because the spectroscopic temperature was measured in an optical spectral window , the flux from the star would be dominated by the 4750 k surface rather than the 3500 k spot . similarly , adding a spot to 13008 changes the best - fit parameters to 4750 k and 3.4 r@xmath9 , closer to the 5000 k spectroscopic temperature @xcite . @xcite also find a discrepancy between the radius of 15028 inferred from optical photometry ( r=2.0 r@xmath9 ) and the radius inferred from geometry assuming a tidally synchronized rotation rate ( r=4.0 r@xmath9 ) . the best - fit radius assuming a spot is 3.1 r@xmath9 , a substantially larger photometric radius than determined in @xcite . while this does not fully explain the discrepancy between the photometric and geometric determinations of the radius , it does bring the two radius measurements closer together . @xcite also observe a flux ratio between 15028 s secondary and primary of 0.35 , much higher than the expected ratio of 0.11 given the spectroscopic temperatures and geometric radii of the primary and secondary . our sed temperature and radius imply an expected v - band luminosity ratio of @xmath120 , closer to the observations but still lower than observed . spot covering fractions for rs cvn have been measured using various methods ( e.g. tio band strength ; @xcite , doppler imaging ; @xcite ) ) these measurements find covering fractions around 30 - 40@xmath19 and sometimes up to 50@xmath19 @xcite , in line with the sed best - fits to the m67 ssgs . we conclude that if ssgs have a substantial spot covering fraction , the best - fit temperatures from our sed fits in section [ sec : sedfits ] may be too cool by a few hundred k and our best - fit radii may be too small by a few tenths of a solar radius . we encourage future observational efforts to determine the spot sizes and temperatures in order to better correct for this effect . while the interaction between magnetic fields and convection remains unclear , theory suggests that a magnetic field may act to reduce the efficiency of convective energy transport in stars with sufficiently large field strengths @xcite . observational evidence suggests the presence of magnetic fields in m - dwarfs and solar - type binaries can create stars with temperatures and radii that deviate from model predictions . radius determinations of eclipsing low - mass main - sequence stars ( m @xmath121 m@xmath9 ) are inflated by @xmath122 from model predictions @xcite and appear redder and cooler than typical low mass stars by a few percent @xcite . similar to our sample of ssgs , these stars display evidence of strong magnetic fields including x - ray and h@xmath65 emission . many have modeled these observations as an effect of inhibited convection due to the presence of magnetic fields@xcite . we suggest a similar effect may be at work in the ssgs . fully modeling the effect of magnetic fields on stellar evolution requires a 3d magnetic stellar evolution code and is not currently technically feasible . instead , two main approaches have been used to produce 1d models of magnetically active stars . one approach is to introduce magnetic perturbations to the stellar structure equations , equation of state , and mixing - length theory of convection ( e.g. @xcite ) . the other has been to reduce the mixing length coefficient ( @xmath123 ; @xcite ) . the argument for using a reduced mixing length has been laid out in several papers @xcite . in brief , the argument assumes that the star possesses a turbulent dynamo that sources energy directly from convective motions . the moving plasma in a stellar convective region induces a lorentz force that preferentially opposes the movement of fluid across magnetic field lines . thus the motion of a convective bubble will be slowed as some of its kinetic energy is diverted into the local magnetic field . for convective motions to be significantly slowed the local alven velocity must approach the convective velocity . given typical solar values , this requires internal magnetic field strengths of the order @xmath124 g , comparable to the equipartition field expected for rapid rotators . the result is a reduced heat flux transported by convection , which can be expressed in the framework of mixing length theory as a smaller characteristic convective length scale , or lowered mixing length coefficient @xmath123 . @xcite simply alter @xmath123 to fit the observed properties of magnetically active stars . they do not attempt to tie the value of @xmath123 to a specific magnetic field strength or topology . in complementary work , @xcite compare stellar models with reduced mixing length to models using a slightly different implementation of a turbulent dynamo , as well as to models using a magnetically modified schwarzschild criterion . they also develop an expression relating the reduction in mixing length to a magnetic field strength . they find that the reduced mixing length models produce stellar structure nearly identical to their turbulent dynamo models , and that these models can reproduce the observed radius inflation among low mass main sequence stars using plausible internal magnetic field strengths and surface magnetic fields comparable to the observations . given the exploratory nature of this work , we elect to use the simple and easily implemented approach of @xcite of altering mixing length to match the observed temperatures and radii of the sub - subgiants . however , the question of how magnetic fields impact stellar structure is far from settled . we consider this a useful first test , and expect future comparisons with sub - subgiant models using other approaches to implementing magnetic fields in stellar evolution codes will be necessary . following the approach of @xcite , we run a 1.3 m@xmath9 model and a 1.1 m@xmath9 model in mesa using different mixing length coefficients in order to explore the impact this may have on the star s global properties . mesa s standard mixing length coefficient is @xmath125 . groups that model this reduced convective efficiency in m - dwarf models found they required @xmath65 to be around 0.5 to reproduce the observed mass - radius relationship derived from m - dwarf eclipsing binaries when assuming a uniform photospheric temperature . a larger mixing length of @xmath126 was sufficient to reproduce observations when using a two temperature model for the photosphere to account for the effects of starspots to lower surface flux @xcite . in figure [ mltmodels ] we show the evolution of an ssg using various mixing length parameters in a hertzprung - russell ( hr ) diagram and a r - t@xmath32 diagram . also plotted with diamonds are the ssg best - fit radii , temperatures , and luminosity from the sed fitting assuming an unspotted surface ( see section [ sec : sedfits ] ) . we compare to the unspotted sed temperature because the mesa models use a single temperature photosphere , and because the spot filling factors for the ssgs are still uncertain . we also plot for comparison the location of a normal cluster star located at the base of the rgb ( open square ) . the models indicate that lowering mixing lengths creates cooler , larger stars at all points during the evolution of the star , but the luminosity remains unchanged at most stages of evolution regardless of mixing length . however , the altered mixing length does create lower luminosity stars near the end of the subgiant branch and the beginning of the rgb . at this stage in evolution , the expanding shell absorbs enough energy to lower the luminosity for a time . lowering the mixing length parameter leads to a greater dip in the luminosity here , and this dip occurs at lower @xmath118 . the ssgs fall near this dip closest to the @xmath127 track . we also compare the cmd locations of ssgs to our models with lowered mixing length in figure [ mltcmd ] . the color transformation is done using the mesa colors module . this transformation assumes a uniform , unspotted surface temperature for the star . a more accurate treatment of the star would be to include both a lowered mixing length coefficient , and to assume a two ( or more ) temperature model for the surface flux that includes contributions from an unspotted photosphere and a spotted region . in fact , the temperature structure may be even more complex , with spots of different temperatures or hot plages surrounding the spots contributing to the emission . we therefore show the cmd to demonstrate the approximate region in which these lowered mixing length models would appear , but caution that the colors may not be accurate for highly spotted stars . while these models do move through the ssg region in a b - v or v - i cmd , we have less success producing the specific locations of the ssgs in m67 than in the hr diagram . these models predict stars that are redder , but brighter than the observed ssgs . we suggest this discrepancy may be due to our assumption of a single temperature photosphere in the mesa models . a better measurement of the spot temperatures and covering fraction of the ssgs and a stellar evolution code capable of modeling a spotted photosphere would provide more reliable color transformations . in ngc 6791 , the models do better at reproducing the observed locations of the magnetically active ssgs . these stars fall between @xmath127 and 1.5 , similar to the results in the hr diagram . the cmd makes clear that changing mixing length becomes most noticeable near the end of the subgiant branch and through the rgb . while the tracks show cooler , more expanded stars on the main sequence and early subgiant branch , the spread in the tracks falls within the scatter of stars that fall on a normal isochrone , and therefore would not be noticed based on cmd position alone . we also expect that the such short - period magnetically active binaries are not observed all the way up the rgb , as they will evolve off the giant branch once they begin roche lobe overflow . this model therefore predicts we should only observe ssgs in a small region just below or to the red of the lower rgb . \1 ) we start with the assumption that the binary fraction for systems with @xmath128 days in both clusters is @xmath0 . this is the average of binary fractions determined in other wocs clusters of various ages : m35 ( @xmath129 , @xcite ) , ngc 6819 ( @xmath130 , @xcite ) and ngc 188 ( @xmath131 , @xcite ) . \2 ) we assume that all binaries with periods less than 18 days ( the longest period ssg in our sample ) will produce magnetic fields on the subgiant branch that cause them to move through the ssg region . the tidal - circularization period for m67 is observed to be @xmath132 days @xcite , so this cutoff seems reasonable . \3 ) we adopt the log - normal period distribution observed by @xcite . integrating this distribution , we find that @xmath133 of binaries with @xmath128 days have @xmath134 days . equivalently , we could say that @xmath135 of all objects are binaries with @xmath134 days . \4 ) we count the number of objects observed in each cluster on the later half of the subgiant branch or the lower rgb . these are the areas predicted by mesa models to appear underluminous due to magnetic fields . in m67 , where we have 3d kinematic memberships , we just count the number of stars in this region and find 20 . in ngc 6791 , we only have proper motion memberships . here we take all members with p@xmath136 from @xcite . we correct for field contamination using the result of @xcite that @xmath137 of the @xmath138 stars are confirmed as rv members . we find 100 members in our region of interest . \5 ) multiplying our number of stars by @xmath139 , we expect to find 0.55 ssgs in m67 and 2.75 ssgs in ngc 6791 formed through this mechanism . \6 ) we calculate the cumulative poisson probability to determine our odds of observing ssgs in m67 and ngc 6791 . we find that the chance of observing 1 or more magnetic ssgs in m67 is @xmath140 . the chance of observing 1 or more magnetic ssgs ngc 6791 is @xmath141 . the chance of observing 2 magnetic ssgs in m67 is @xmath18 , while the chance of observing 4 magnetic ssgs in ngc 6791 is @xmath142 . if we assume that only 3 of the 4 ssgs are magnetic since the 4th shows no signs of binarity or magnetic activity , we find a @xmath130 chance of observing 3 magnetic ssgs in ngc 6791 . we conclude that if this magnetic field mechanism can indeed create stars in the ssg cmd region , it is likely that several of the stars in our sample are created in this way . here we put forth three hypotheses of ssg formation : 1 ) mass transfer in a binary system , 2 ) stripping of a subgiant s envelope , and 3 ) reduced luminosity due to magnetic fields that inhibit convection and produce large stellar spots . we demonstrate that stellar models for each of these methods evolve through the ssg domain . models of mass transfer in binaries containing subgiant stars can produce binary systems with combined light in the ssg cmd region . this requires binary systems with orbital periods around 1 day as they evolve along the subgiant branch , as longer period binaries begin rlo once they have started ascending the rgb and do not move through the ssg region . additionally , the binary must have a small enough secondary that the secondary light does not push the combined light into a more populated cmd region ( i.e. the blue straggler domain ) . due to these restrictions on period and secondary mass , ssgs formed through mass transfer are expected to be rare , and we would not expect to see many , if any , in open clusters . however , with a larger sample of subgiant stars , e.g. in a massive globular cluster , we may observe ssgs formed in such a way . furthermore , mass transfer models produce binaries with shorter periods than the observed orbital periods of the ssgs in m67 and ngc 6791 . we test the tidally enhanced wind model of @xcite , and find that even with this elevated wind mass loss we can not achieve the necessary mass loss rates to produce ssgs with the observed periods . mesa models in which several tenths of a solar mass of material is rapidly stripped from a subgiant s envelope can also produce stars in the ssg domain . we conjecture that this may happen during grazing dynamical encounters in which a star passes close enough to tidally strip material from a subgiant s envelope but avoids merging . additional scattering experiments and hydrodynamic simulations are necessary to determine if this mechanism is viable , and if binaries with @xmath143 day orbital periods are an expected end product of such an interaction . as an upper limit on formation rate , we assume that all single - binary or binary - binary encounters with subgiants lead to the formation of ssgs . even with this very optimistic assumption , the expected rates of formation in open clusters are low enough that we would not expect to observe such stars in m67 or ngc 6791 . ssgs may also form as the result of main - sequence collisions during dynamical encounters , but this too should be rare in open clusters . as with mass transfer , this may be a more relevant formation mechanism in massive globular clusters where encounter rates are higher and there are larger populations of subgiant stars , a possibility we investigate in another paper ( geller et al . 2017b , in preparation ) . while envelope stripping during dynamical encounters and roche lobe overflow perhaps produce ssgs infrequently , clearly substantial mass loss from a subgiant star can create stars in the ssg cmd region . there may be other mass loss mechanisms that we have not explored that create ssgs more frequently , such as common envelope evolution . stellar magnetic fields in - and - of themselves may be sufficient to explain the anomalous luminosities and colors of sub - subgiants . five of our six ssgs show spot variability , x - rays , and h@xmath65 emission indicative of surface magnetic activity . research in the field of low - mass eclipsing binaries suggests magnetic fields can cause inflated radii and lower effective temperatures . the seds of the ssgs suggest that the stars do have lower temperatures than normal subgiants in the clusters . simple 1d models of stars with reduced mixing lengths can fairly well reproduce the observed temperatures , radii , and luminosities of the systems . we have less success matching the models to the ssg cmd locations , which may be a consequence of assuming a single temperature photosphere when performing the color transformation . a stellar evolution code that allows for a multi - temperature photosphere is necessary to test these results and produce more accurate model colors , as are better measurements of spot temperatures and covering fractions ( e.g. using tio bands ) . a calculation of the frequency with which magnetic fields should produce these stars yields an expectation of at least one such star in ngc 6791 , and possibly one in m67 . the formation frequencies indicate that several of the ssgs in our sample could likely have formed in this way , with a probability of a few percent that all five ssgs showing signs of magnetic activity could have been produced by this mechanism . the sixth ssg shows no signs of magnetic activity or binarity , and we conclude this mechanism is not likely to explain the origin of this system . of course , these lowered - mixing length models are simplistic and do not include a physical treatment of the interaction between magnetic fields and convection . without full 3d magnetic stellar evolution codes , fully implementing all the required physics is impossible , but several other stellar evolution codes use other approaches to model the effects of magnetic fields on evolution @xcite . comparing the results of these codes to our models would be a useful test of our approach . finally , the discovery of more ssg sb2 binaries or eclipsing binaries in order to infer masses would be an excellent test . while the magnetic field hypothesis requires ssgs to be similar in mass to normal cluster subgiants , mass - loss mechanisms such as mass transfer in a binary or envelope stripping require significant amounts of mass loss to produce ssgs . a sample of ssg systems in which the mass could be well determined would be a strong test of which hypothesis is best . this research was supported by a grant - in - aid of research from sigma xi , the scientific research society and by nasa through grants hst - ar-13910 and hst- go-13354.001-a from the space telescope science institute , which is operated by aura , inc . , under nasa contract nas 5 - 26555 . this work was also funded by the national science foundation grant ast-0908082 to the university of wisconsin - madison .

background : large - scale biological jobs on high - performance computing systems require manual intervention if one or more computing cores on which they execute fail . this places not only a cost on the maintenance of the job , but also a cost on the time taken for reinstating the job and the risk of losing data and execution accomplished by the job before it failed . approaches which can proactively detect computing core failures and take action to relocate the computing core s job onto reliable cores can make a significant step towards automating fault tolerance . + method : this paper describes an experimental investigation into the use of multi - agent approaches for fault tolerance . two approaches are studied , the first at the job level and the second at the core level . the approaches are investigated for single core failure scenarios that can occur in the execution of parallel reduction algorithms on computer clusters . a third approach is proposed that incorporates multi - agent technology both at the job and core level . experiments are pursued in the context of genome searching , a popular computational biology application . + result : the key conclusion is that the approaches proposed are feasible for automating fault tolerance in high - performance computing systems with minimal human intervention . in a typical experiment in which the fault tolerance is studied , centralised and decentralised checkpointing approaches on an average add 90% to the actual time for executing the job . on the other hand , in the same experiment the multi - agent approaches add only 10% to the overall execution time .

1403.0500

he scale of resources and computations required for executing large - scale biological jobs are significantly increasing @xcite . with this increase the resultant number of failures while running these jobs will also increase and the time between failures will decrease @xcite . it is not desirable to have to restart a job from the beginning if it has been executing for hours or days or months @xcite . a key challenge in maintaining the seamless ( or near seamless ) execution of such jobs in the event of failures is addressed under research in fault tolerance @xcite . many jobs rely on fault tolerant approaches that are implemented in the middleware supporting the job ( for example @xcite ) . the conventional fault tolerant mechanism supported by the middleware is checkpointing @xcite , which involves the periodic recording of intermediate states of execution of a job to which execution can be returned if a fault occurs . such traditional fault tolerant mechanisms , however , are challenged by drawbacks such as single point failures @xcite , lack of scalability @xcite and communication overheads @xcite , which pose constraints in achieving efficient fault tolerance when applied to high - performance computing systems . moreover , many of the traditional fault tolerant mechanisms are manual methods and require human administrator intervention for isolating recurring faults . this will place a cost on the time required for maintenance . self - managing or automated fault tolerant approaches are therefore desirable , and the objective of the research reported in this paper is the development of such approaches . if a failure is likely to occur on a computing core on which a job is being executed , then it is necessary to be able to move ( migrate ) the job onto a reliable core @xcite . such mechanisms are not readily available . at the heart of this concept is mobility , and a technique that can be employed to achieve this is using multi - agent technologies @xcite . two approaches are proposed and implemented as the means of achieving both the computation in the job and self - managing fault tolerance ; firstly , an approach incorporating agent intelligence , and secondly , an approach incorporating core intelligence . in the first approach , automated fault tolerance is achieved by a collection of agents which can freely traverse on a network of computing cores . each agent carries a portion of the job ( or sub - job ) to be executed on a computing core in the form of a payload . fault tolerance in this context can be achieved since an agent can move on the network of cores , effectively moving a sub - job from one computing core which may fail onto another reliable core . in the second approach , automated fault tolerance is achieved by considering the computing cores to be an intelligent network of cores . sub - jobs are scheduled onto the cores , and the cores can move processes executed on them across the network of cores . fault tolerance in this context can be achieved since a core can migrate a process executing on it onto another core . a third approach is proposed which combines both agent and core intelligence under a single umbrella . in this approach , a collection of agents freely traverse on a network of virtual cores which are an abstraction of the actual hardware cores . the agents carry the sub - jobs as a payload and situate themselves on the virtual cores . fault tolerance is achieved either by an agent moving off one core onto another core or the core moving an agent onto another core when a fault is predicted . rules are considered to decide whether an agent or a core should initiate the move . automated fault tolerance can be beneficial in areas such as molecular dynamics @xcite . typical molecular dynamics simulations explore the properties of molecules in gaseous , liquid and solid states . for example , the motion of molecules over a time period can be computed by employing newton s equations if the molecules are treated as point masses . these simulations require large numbers of computing cores that run sub - jobs of the simulation which communicate with each other for hours , days and even months . it is not desirable to restart an entire simulation or to loose any data from previous numerical computations when a failure occurs . conventional methods like periodic checkpointing keep track of the state of the sub - jobs executed on the cores , and helps in restarting a job from the last checkpoint . however , overzealous periodic checkpointing over a prolonged period of time has large overheads and contributes to the slowdown of the entire simulation @xcite . additionally , mechanisms will be required to store and handle large data produced by the checkpointing strategy . further , how wide the failure can impact the simulation is not considered in checkpointing . for example , the entire simulation is taken back to a previous state irrespective of whether the sub - jobs running on a core depend or do not depend on other sub - jobs . one potential solution to mitigate the drawbacks of checkpointing is to proactively probe the core for failures . if a core is likely to fail , then the sub - job executing on the core is migrated automatically onto another core that is less likely to fail . this paper proposes and experimentally evaluates multi - agent approaches to realising this automation . genome searching is considered as an example for implementing the multi - agent approaches . the results indicate the feasibility of the multi - agent approaches ; they require only one - fifth the time compared to that required by manual approaches . the remainder of this paper is organised as follows . the methods section presents the three approaches proposed for automated fault tolerance . the results section highlights the experimental study and the results obtained from it . the discussion section presents a discussion on the three approaches for automating fault tolerance . the conclusions section summarises the key results from this study . three approaches to automate fault tolerance are presented in this section . the first approach incorporates agent intelligence , the second approach incorporates core intelligence , and in the third a hybrid of both agent and core intelligence is incorporated . a job , @xmath0 , which needs to be executed on a large - scale system is decomposed into a set of sub - jobs @xmath1 . each sub - job @xmath1 is mapped onto agents @xmath2 that carry the sub - jobs as payloads onto the cores , @xmath3 as shown in figure 1 . the agents and the sub - job are independent of each other ; in other words , an agent acts as a wrapper around a sub - job to situate the sub - job on a core . there are three computational requirements of the agent to achieve successful execution of the job : ( a ) the agent needs to know the overall job , @xmath0 , that needs to be achieved , ( b ) the agent needs to access data required by the sub - job it is carrying and ( c ) the agent needs to know the operation that the sub - job needs to perform on the data . the agents then displace across the cores to compute the sub - jobs . intelligence of an agent can be useful in at least four important ways for achieving fault tolerance while a sub - job is executed . firstly , an agent knows the landscape in which it is located . knowledge of the landscape is threefold which includes ( a ) the knowledge of the computing core on which the agent is located , ( b ) knowledge of other computing cores in the vicinity of the agent and ( c ) knowledge of agents located in the vicinity . secondly , an agent identifies a location to situate within the landscape . this is possible by gathering information from the vicinity using probing processes and is required when the computing core on which the agent is located is anticipated to fail . thirdly , an agent predicts failures that are likely to impair its functioning . the prediction of failures ( for example , due to the failure of the computing core ) is along similar lines to proactive fault tolerance . fourthly , an agent is mobile within the landscape . if the agent predicts a failure then the agent can relocate onto another computing core thereby moving off the job from the core anticipated to fail ( refer figure 2 ) . and @xmath4 are situated on cores @xmath5 and @xmath6 respectively . a failure is predicted on core @xmath7 . the agent @xmath8 moves onto core @xmath9.,scaledwidth=40.0% ] the intelligence of agents is incorporated within the following sequence of steps that describes an approach for fault tolerance : ' '' '' _ agent intelligence based fault tolerance _ + ' '' '' 1 . decompose a job , @xmath0 , to be executed on the landscape into sub - jobs , @xmath1 2 . each sub - job provided as a payload to agents , @xmath2 3 . agents carry jobs onto computing cores , @xmath3 4 . for each agent , @xmath10 located on computing core @xmath11 , where @xmath12 to @xmath13 1 . periodically probe the computing core @xmath11 2 . if @xmath11 predicted to fail , then 1 . agent , @xmath10 moves onto an adjacent computing core , @xmath9 2 . notify dependent agents 3 . agent @xmath10 establishes dependencies 5 . collate execution results from sub - jobs ' '' '' a failure scenario is considered for the agent intelligence based fault tolerance concept . in this scenario , while a job is executed on a computing core that is anticipated to fail any adjacent core onto which the job needs to be reallocated can also fail . the communication sequence shown in figure 3 is as follows . the hardware probing process on the core anticipating failure , @xmath14 notifies the failure prediction to the agent process , @xmath15 , situated on it . since the failure of a core adjacent to the core predicted to fail is possible it is necessary that the predictions of the hardware probing processes on the adjacent cores be requested . once the predictions are gathered , the agent process , @xmath15 , creates a new process on an adjacent core and transfers data it was using onto the newly created process . then the input dependent ( @xmath16 ) and output dependent ( @xmath17 ) processes are notified . the agent process on @xmath14 is terminated thereafter . the new agent process on the adjacent core establishes dependencies with the input and output dependent processes . a job , @xmath0 , which needs to be executed on a large - scale system is decomposed into a set of sub - jobs @xmath1 . each sub - job @xmath1 is mapped onto the virtual cores , @xmath18 , an abstraction over @xmath3 respectively as shown in figure 4 . the cores referred to in this approach are virtual cores which are an abstraction over the hardware computing cores . the virtual cores are a logical representation and may incorporate rules to achieve intelligent behaviour . and @xmath19 are situated on virtual cores @xmath20 and @xmath21 respectively . a failure is predicted on core @xmath7 and @xmath22 moves the job @xmath23 onto virtual core @xmath24.,scaledwidth=40.0% ] intelligence of a core is useful in a number of ways for achieving fault tolerance . firstly , a core updates knowledge of its surrounding by monitoring adjacent neighbours . independent of what the cores are executing , the cores can monitor each other . each core can ask the question ` are you alive ? ' to its neighbours and gain information . secondly , a core periodically updates information of its surrounding . this is useful for the core to know which neighbouring cores can execute a job if it fails . thirdly , a core periodically monitors itself using a hardware probing process and predicts if a failure is likely to occur on it . fourthly , a core can move a job executing on it onto an adjacent core if a failure is expected and adjust to failure as shown in figure 4 . once a job has relocated all data dependencies will need to be re - established . the following sequence of steps describe an approach for fault tolerance incorporating core intelligence : ' '' '' _ core intelligence based fault tolerance _ + ' '' '' 1 . decompose a job , @xmath0 , to be executed on the landscape into sub - jobs , @xmath1 2 . each sub - job allocated to cores , @xmath18 3 . for each core , @xmath25 , where @xmath12 to @xmath13 until sub - job @xmath26 completes execution 1 . periodically probe the computing core @xmath11 2 . if @xmath11 predicted to fail , then 1 . migrate sub - job @xmath26 on @xmath25 onto an adjacent computing core , @xmath24 4 . collate execution results from sub - jobs ' '' '' figure 5 shows the communication sequence of the core failure scenario considered for the core intelligence based fault tolerance concept . the hardware probing process on the core predicted to fail , @xmath14 notifies a predicted failure to the core . the job executed on @xmath27 is then migrated onto an adjacent core @xmath28 once a decision based on failure predictions are received from the hardware probing processes of adjacent cores . the hybrid approach acts as an umbrella bringing together the concepts of agent intelligence and core intelligence . the key concept of the hybrid approach lies in the mobility of the agents on the cores and the cores collectively executing a job . decision - making is required in this approach for choosing between the agent intelligence and core intelligence approaches when a failure is expected . a job , @xmath0 , which needs to be executed on a large - scale system is decomposed into a set of sub - jobs @xmath1 . each sub - job @xmath1 is mapped onto agents @xmath2 that carry the sub - jobs as payloads onto the virtual cores , @xmath18 which are an abstraction over @xmath3 respectively as shown in figure 1 . the following sequence of steps describe the hybrid approach for fault tolerance incorporating both agent and core intelligence : ' '' '' _ hybrid intelligence based fault tolerance _ + ' '' '' 1 . decompose a job , @xmath0 , to be executed on the landscape into sub - jobs , @xmath1 2 . each sub - job provided as a payload to agents , @xmath2 3 . agents carry jobs onto virtual cores , @xmath18 4 . for each agent , @xmath10 located on virtual core @xmath25 , where @xmath12 to @xmath13 1 . periodically probe the computing core @xmath11 2 . if @xmath11 predicted to fail , then 1 . if ` agent intelligence ' is a suitable mechanism , then 1 . agent , @xmath10 , moves onto an adjacent computing core , @xmath24 2 . notify dependent agents 3 . agent @xmath10 establishes dependencies 1 . else if ` core intelligence ' is a suitable mechanism , then 1 . core @xmath25 migrates agent , @xmath10 onto an adjacent computing core , @xmath24 5 . collate execution results from sub - jobs ' '' '' when a core failure is anticipated both an agent and a core can make decisions which can lead to a conflict . for example , an agent can attempt to move onto an adjacent core while the core on which it is executing would like to migrate it to an alternative adjacent core . therefore , an agent and the core on which it is located need to negotiate before either of them initiate a response to move ( see figure 6 ) . the rules for the negotiation between the agent and the core in this case are proposed from the experimental results presented in this paper ( presented in the decision making rules sub - section ) . and @xmath4 are situated on virtual cores @xmath20 and @xmath21 which are mapped onto computing cores @xmath5 and @xmath6 respectively . a failure is predicted on core @xmath7 . the agent @xmath8 and @xmath22 negotiate to decide who moves the sub - job onto core @xmath24.,scaledwidth=40.0% ] in this section , the experimental platform is considered followed by the experimental studies and the results obtained from experiments . four computer clusters were used for the experiments reported in this paper . the first was a cluster available at the centre for advanced computing and emerging technologies ( acet ) , university of reading , uk . thirty three compute nodes connected through gigabit ethernet were available , each with pentium iv processors and 512 mb-2 gb ram . the remaining three clusters are compute resources , namely brasdor , glooscap and placentia , all provided by the atlantic computational excellence network ( acenet ) @xcite , canada . brasdor comprises 306 compute nodes connected through gigabit ethernet , with 932 cores and 1 - 2 gb ram . glosscap comprises 97 nodes connected through infiniband , with 852 cores and 1 - 8 gb ram . placentia comprises 338 compute nodes connected through infiniband , with 3740 cores and 2 - 16 gb ram . the cluster implementations in this paper are based on the message passing interface ( mpi ) . the first approach , incorporating agent intelligence , is implemented using open mpi @xcite , an open source implementation of mpi 2.0 . the dynamic process model which supports dynamic process creation and management facilitates control over an executing process . this feature is useful for implementing the first approach . the mpi functions useful in the implementation are ( i ) mpi_comm_spawn which creates a new mpi process and establishes communication with an existing mpi application and ( ii ) mpi_comm_accept and mpi_comm_connect which establishes communication between two independent processes . the second approach , incorporating core intelligence , is implemented using adaptive mpi ( ampi ) @xcite , developed over charm++ @xcite , a c++ based parallel programming language . the aim of ampi is to achieve dynamic load balancing by migrating objects over virtual cores and thereby facilitating control over cores . core intelligence harnesses this potential of ampi to migrate a job from a core onto another core . a strategy to migrate a job using the concepts of processor virtualisation and dynamic job migration in ampi and charm++ is reported in @xcite . parallel reduction algorithms @xcite which implement the bottom - up approach ( i.e. , data flows from the leaves to the root ) are employed for the experiments . these algorithms are of interest for three reasons . firstly , the algorithm is used in a large number of scientific applications including computational biological applications in which optimizations are performed ( for example , bootstrapping ) . incorporating self - managing fault tolerant approaches can make these algorithms more robust and reliable @xcite . secondly , the algorithm lends itself to be easily decomposed into a set of sub - jobs . each sub - job can then be mapped onto a computing core either by providing the sub - job as a payload to an agent in the first approach or by providing the job onto a virtual core incorporating intelligent rules . thirdly , the execution of a parallel reduction algorithm stalls and produces incorrect solutions if a core fails . therefore , parallel reduction algorithms can benefit from local fault - tolerant techniques . figure 7 is an exemplar of a parallel reduction algorithm . in the experiments reported in this paper , a generic parallel summation algorithm with three sets of input is employed . firstly , @xmath29 , @xmath30 @xmath31 @xmath32 , secondly , @xmath33 , @xmath34 @xmath31 @xmath35 , and thirdly , @xmath36 @xmath31 @xmath37 . the first level nodes which receive the three sets of input comprise three set of nodes . firstly , @xmath38 , @xmath39 @xmath31 @xmath40 , secondly , @xmath41 , @xmath42 @xmath31 @xmath43 , and thirdly , @xmath44 , @xmath45 @xmath31 @xmath46 . the next level of nodes , @xmath47 , @xmath48 and @xmath49 receive inputs from the first level nodes . the resultant from the second level nodes is fed in to the third level node @xmath50 . the nodes reduce the input through the output using the parallel summation operator ( @xmath51 ) . and the three levels of nodes are denoted by @xmath52 . the inputs are passed to the nodes @xmath53 which are then reduced and passed to nodes @xmath54 and finally onto @xmath55 for the output.,scaledwidth=45.0% ] the parallel summation algorithm can benefit from the inclusion of fault tolerant strategies . the job , @xmath0 , in this case is summation , and the sub - jobs , @xmath1 are also summations . in the first fault tolerant approach , incorporating mobile agent intelligence , the data to be summed along with the summation operator is provided to the agent . the agents locate on the computing cores and continuously probe the core for anticipating failures . if an agent is notified of a failure , then it moves off onto another computing core in the vicinity , thereby not stalling the execution towards achieving the summation job . in the second fault tolerant approach , incorporating core intelligence , the sub - job comprising the data to be summed along with the summation operator is located on the virtual core . when the core anticipates a failure , it migrates the sub - job onto another core . a failure scenario is considered for experimentally evaluating the fault tolerance strategies . in the scenario , when a core failure is anticipated the sub - job executing on it is relocated onto an adjacent core . of course this adjacent core may also fail . therefore , information is also gathered from adjacent cores as to whether they are likely to fail or not . this information is gathered by the agent in the agent - based approach and the virtual core in the core - based approach and used to determine which adjacent core the sub - job needs to be moved to . this failure scenario is adapted to the two strategies giving respectively the agent intelligence failure scenario and the core intelligence failure scenario ( described in the methods section ) . figures 8 through 13 are a collection of graphs plotted using the parallel reduction algorithm as a case study for both the first ( agent intelligence - figure 8 , figure 10 and figure 12 ) and second ( core intelligence - figure 9 , figure 11 and figure 13 ) fault tolerant approaches . each graph comprises four plots , the first representing the acet cluster and the other three representing the three acenet clusters . the graphs are also distinguished based on the following three factors that can affect the performance of the two approaches : * the number of dependencies of the sub - job being executed denoted as @xmath56 . if the total number of input dependencies is @xmath57 and the total number of output dependencies is @xmath58 , then @xmath59 . for example , in a parallel summation algorithm incorporating binary trees , each node has two input dependencies and one output dependency , and therefore @xmath60 . in the experiments , the number of dependencies is varied between 3 and 63 , by changing the number of input dependencies of an agent or a core . the results are presented in figure 8 and figure 9 . * the size of the data communicated across the cores denoted as @xmath61 . in the experiments , the input data is a matrix for parallel summation and its size is varied between @xmath62 to @xmath63 kb . the results are presented in figure 10 and figure 11 . * the process size of the distributed components of the job denoted as @xmath64 . in the experiments , the process size is varied between @xmath62 to @xmath63 kb which is proportional to the input data . the results are presented in figure 12 and figure 13 . figure 8 is a graph of the time taken in seconds for reinstating execution versus the number of dependencies in the agent intelligence failure scenario . the mean time taken to reinstate execution for 30 trials , @xmath65 , is computed for varying numbers of dependencies , @xmath56 ranging from 3 to 63 . the size of the data on the agent is @xmath66 kilo bytes . the approach is slowest on the acet cluster and fastest on the placentia cluster . in all cases the communication overheads result in a steep rise in the time taken for execution until @xmath67 . the time taken on the acet cluster rises once again after @xmath68 . figure 9 is a graph of the time taken in seconds for reinstating execution versus the number of dependencies in the core intelligence failure scenario . the mean time taken to reinstate execution for 30 trials , @xmath69 , is computed for varying number of dependencies , @xmath56 ranging from 3 to 63 . the size of the data on the core is @xmath66 kilo bytes . the approach requires almost the same time on the four clusters for reinstating execution until @xmath67 , after which there is divergence in the plots . the approach lends itself well on placentia and glooscap . figure 10 is a graph showing the time taken in seconds for reinstating execution versus the size of data in kilobytes ( kb ) , @xmath70 , where @xmath71 , carried by an agent in the agent intelligence failure scenario . the mean time taken to reinstate execution for 30 trials , @xmath65 , is computed for varying sizes of data ranging from @xmath62 to @xmath63 kb . the number of dependencies @xmath56 is 10 for the graph plotted . placentia and glooscap outperforms acet and brasdor in the agent approach for varying size of data . figure 11 is a graph showing the time taken in seconds for reinstating execution versus the size of data in kilobytes ( kb ) , @xmath70 , where @xmath71 , on a core in the core intelligence failure scenario . the mean time taken to reinstate execution for 30 trials , @xmath69 , is computed for varying sizes of data ranging from @xmath62 to @xmath63 kb . the number of dependencies @xmath56 is 10 for the graph plotted . in this graph , nearly similar time is taken by the approach on the four clusters with the acet cluster requiring more time than the other clusters for @xmath72 . figure 12 is a graph showing the time taken in seconds for reinstating execution versus process size in kilobytes ( kb ) , @xmath73 , where @xmath71 , in the agent intelligence failure scenario . the mean time taken to reinstate execution for 30 trials , @xmath65 , is computed for varying process sizes ranging from @xmath62 to @xmath63 kb . the number of dependencies @xmath56 is 10 for the graph plotted . the second scenario performs similar to the first scenario . the approach takes almost similar times to reinstate execution after a failure on the four clusters , but there is a diverging behaviour after @xmath74 . figure 13 is a graph showing the time taken in seconds for reinstating execution versus process size in kilobytes ( kb ) , @xmath73 , where @xmath71 , in the core intelligence failure scenario . the mean time taken to reinstate execution for 30 trials , @xmath69 , is computed for varying process sizes ranging from @xmath62 to @xmath63 kb . the number of dependencies @xmath56 is 10 for the graph plotted . the approach has similar performance on the four clusters , though placentia performs better than the other three clusters for a process size of more than @xmath75 kb . parallel simulations in molecular dynamics model atoms or molecules in gaseous , liquid or solid states as point masses which are in motion . such simulations are useful for studying the physical and chemical properties of the atoms or molecules . typically the simulations are compute intensive and can be performed in at least three different ways @xcite . firstly , by assigning a group of atoms to each processor , referred to as atom decomposition . the processor computes the forces related to the group of atoms to update their position and velocities . the communication between atoms is high and effects the performance on large number of processors . secondly , by assigning a block of forces from the force matrix to be computed to each processor , referred to as force decomposition . this technique scales better than atom decomposition but is not a best solution for large simulations . thirdly , by assigning a three dimensional space of the simulation to each processor , referred to as spatial decomposition . the processor needs to know the positions of atoms in the adjacent space to compute the forces of atoms in the space assigned to it . the interactions between the atoms are therefore local to the adjacent spaces . in the first and second decomposition techniques interactions are global and thereby dependencies are higher . agent and core based approaches to fault tolerance can be incorporated within parallel simulations in the area of molecular dynamics . however , which of the two approaches , agent or core intelligence , is most appropriate ? the decomposition techniques considered above establish dependencies between blocks of atoms and between atoms . therefore the degree of dependency affects the relocation of a sub - job in the event of a core failure and reinstating it . the dependencies of an atom in the simulation can be based on the input received from neighbouring atoms and the output propagated to neighbouring atoms . based on the number of atoms allocated to a core and the time step of the simulation the intensity of numerical computations and the data managed by a core vary . large simulations that extend over long periods of time generate and need to manage large amounts of data ; consequently the process size on a core will also be large . therefore , ( i ) the dependency of the job , ( ii ) the data size and ( iii ) the process size are factors that need to be taken into consideration for deciding whether an agent - based approach or a core - based approach needs to come into play . along with the observations from parallel simulations in molecular dynamics , the experimental results provide an insight into the rules for decision - making for large - scale applications . from the experimental results graphed in figure 8 and figure 9 , where dependencies are varied , core intelligence is superior to agent intelligence if the total dependencies @xmath56 is less than or equal to 10 therefore , 1 . if the algorithm needs to incorporate fault tolerance based on the number of dependencies , then if @xmath76 use core intelligence , else use agent or core intelligence . from the experimental results graphed in figure 10 and figure 11 , where the size of data is varied , agent intelligence is more beneficial than core intelligence if the size of data @xmath61 is less than or equal to @xmath77 kb . therefore , 1 . if the algorithm needs to incorporate fault tolerance based on the size of data , then if @xmath78 kb , then use agent intelligence , else use agent or core intelligence . from the experimental results graphed in figure 12 and figure 13 , where the size of the process is varied , agent intelligence is more beneficial than core intelligence if the size of the process @xmath64 is less than or equal to @xmath77 kb . therefore , 1 . if the algorithm needs to incorporate fault tolerance based on process size , then if @xmath79 kb , then use agent intelligence , else use agent or core intelligence . the number of dependencies , size of data , and process size are the three factors taken into account in the experimental results . the results indicate that the approach incorporating core intelligence takes lesser time than the approach incorporating agent intelligence . there are two reasons for this . firstly , in the agent approach , the agent needs to establish the dependency with each agent individually , where as in the core approach as a job is migrated from a core onto another its dependencies are automatically established . secondly , agent intelligence is a software abstraction of the sub - job , thereby adding a virtualised layer in the communication stack . this increases the time for communication . the virtual core is also an abstraction of the computing core but is closer to the computing core in the communication stack . the above rules can be incorporated to exploit both agent - based and core - based intelligence in a third , hybrid approach . the key concept of the hybrid approach combines the mobility of the agents on the cores and the cores collectively executing a job . the approach can select whether the agent - based approach or the core - based approach needs to come to play based on the rules for decision - making . the key observation from the experimental results is that the cost of incorporating intelligence at the job and core levels for automating fault tolerance is less than a second , which is smaller than the time taken by manual methods which would be in the order of minutes . for example , in the first approach , the time for reinstating execution with over 50 dependencies is less than 0.55 seconds and in the second approach , is less than 0.5 seconds . similar results are obtained when the size of data and the process are large . the proposed multi - agent approaches and the decision making rules considered in the above sections are validated using a computational biology job . a job that fits the criteria of reduction algorithms is considered . in reduction algorithms , a job is decomposed to sub - jobs and executed on multiple nodes and the results are further passed onto other node for completing the job . one popular computational biology job that fits this criteria is searching for a genome pattern . this has been widely studied and fast and efficient algorithms have been developed for searching genome patterns ( for example , @xcite , @xcite and @xcite ) . in the genome searching experiment performed in this research multiple nodes of a cluster execute the search operation and the output produced by the search nodes are then combined by an additional node . the focus of this experimental study is not parallel efficiency or scalability of the job but to validate the multi - agent approaches and the decision making rules in the context of computational biology . hence , a number of assumptions are made for the genome searching job . first , redundant copies of the genome data are made on the same node to obtain a sizeable input . secondly , the search operation is run multiple times to span long periods of time . thirdly , the jobs are executed such that they can be stopped intentionally by the user at any time and gather the results of the preceding computations until the execution was stopped . the placentia cluster is chosen for this validation study since it was the best performing cluster in the empirical study presented in the previous sections . the job is implemented using r programming which uses mpi for exploiting computation on multiple nodes of the placentia cluster . bioconductor packages are required for supporting the job . the job makes use of bsgenome.celegans.ucsc.ce2 , bsgenome.celegans.ucsc.ce6 and bsgenome.celegans.ucsc.ce10 as input data which are the ce2 , ce6 and ce10 genome for chromosome i of caenorhabditis elegans @xcite . a list of 5000 genome patterns each of which is a short nucleotide sequence of 15 to 25 bases is provided to be searched against the input data . the forward and reverse strands of seven caenorhabditis elegans chromosomes named as chri , chrii , chriii , chriv , chrv , chrx , chrm are the targets of the search operation . when there is a target hit the search nodes provide to the node that gathers the results the name of the chromosome where the hit occurs , two integers giving the starting and ending positions of the hit , an indication of the hit either in the forward or reverse strand , and unique identification for every pattern in the dictionary . the results are tabulated in an output file in the combining node . a sample of the output is shown in figure 14 . redundant copies of the input data are made to obtain 512 mb ( which is @xmath62 kb ) and the job is executed for one hour . in a typical experiment the number of dependencies , @xmath56 was set to 4 ; three nodes of the cluster performed the search operation while the fourth node combined the results passed on to it from the three search nodes . in the agent intelligence based approach the time for predicting the fault is 38 seconds , the time for reinstating execution is 0.47 seconds , the overhead time is over 5 minutes and the total time when one failure occurs per hour is 1 hour , 6 minutes and 17 seconds . in the core intelligence based approach the time for predicting the single node failure is similar to the agent intelligence approach ; the time for reinstating execution is 0.38 seconds , the overhead time is over 4 minutes and the total time when one failure occurs per hour is 1 hour , 5 minutes and 8 seconds . in another experiment for 512 mb size of input data the number of dependencies was set to 12 ; eleven nodes for searching and one node for combining the results provided by the eleven search nodes . in the agent intelligence based approach the time for reinstating execution is 0.54 seconds , the overhead time is over 6 minutes and the total time when one failure occurs per hour is 1 hour , 7 minutes and 34 seconds . in the core intelligence based approach the time for reinstating execution is close to 0.54 seconds , the overhead time is over 6 minutes and the total time when one failure occurs per hour is 1 hour , 7 minutes and 48 seconds . the core intelligence approach requires less time than the agent intelligence approach when @xmath80 , but the times are comparable when @xmath81 . so , the above two experiments validate rule 1 for decision making considered in the previous section . experiments were performed for different input data sizes ; in one case @xmath82 kb and in the other @xmath83 kb . the agent intelligence approach required less time in the former case than the core intelligence approach . the time was comparable for the latter case . hence , the genome searching job in the context of the experiments validated rule 2 for decision making . similarly , when process size was varied rule 3 was found to be validated . the genome searching job is used as an example to validate the use of the multi - agent approaches for computational biology jobs . the decision making rules empirically obtained were satisfied in the case of this job . the results obtained from the experiments for the genome searching job along with comparisons against traditional fault tolerance approaches , namely centralised and decentralised checkpointing are considered in the next section . all fault tolerance approaches initiate a response to address a failure . based on when a response is initiated with respect to the occurrence of the failure , approaches can be classified as proactive and reactive . proactive approaches predict failures of computing resources before they occur and then relocate a job executing on resources anticipated to fail onto resource that are not predicted to fail ( for example @xcite ) . reactive approaches on the other hand minimise the impact of failures after they have occurred ( for example checkpointing @xcite , rollback recovery @xcite and message logging @xcite ) . a hybrid of proactive and reactive , referred to as adaptive approaches , is implemented so that failures that can not be predicted by proactive approaches are handled by the reactive approaches @xcite . the control of a fault tolerant approach can be either centralised or distributed . in approaches where the control is centralised , one or more servers are used for backup and a single process responsible for monitoring jobs that are executed on a network of nodes . the traditional message logging and checkpointing approach involves the periodic recording of intermediate states of execution of a job to which execution can be returned if faults occur . such approaches are susceptible to single point failure , lack scalability over a large network of nodes , have large overheads , and require large disk storage . these drawbacks can be minimised or avoided when the control of the approaches is distributed ( for example , distributed diagnosis @xcite , distributed checkpointing @xcite and diskless checkpointing @xcite ) . in this paper two distributed proactive approaches towards achieving fault tolerance are proposed and implemented . in both approaches a job to be computed is decomposed into sub - jobs which are then mapped onto the computing cores . the two approaches operate at the middle levels ( between the sub - jobs and the computing cores ) incorporating agent intelligence . in the first approach , the sub - jobs are mapped onto agents which are released onto the cores . if an agent is notified of a potential core failure during execution of the sub - job mapped onto it , then the agent moves onto another core thereby automating fault tolerance . in the second approach the sub - jobs are scheduled on virtual cores , which are an abstraction of the computing cores . if a virtual core anticipates a core failure then it moves the sub - job on it to another virtual core , in effect onto another computing core . the two approaches achieve automation in fault tolerance using intelligence in agents and using intelligence in cores respectively . a third approach is proposed which brings together the concepts of both agent intelligence and core intelligence from the first two approaches . the conventional approaches to fault tolerance such as checkpointing have large communication overheads based on the periodicity of checkpointing . high frequencies of checkpointing can lead to heavy network traffic since the available communication bandwidth will be saturated with data transferred from all computing nodes to the a stable storage system that maintains the checkpoint . this traffic is on top of the actual data flow of the job being executed on the network of cores . while global approaches are useful for jobs which are less memory and data intensive and can be executed over short periods of time , they may constrain the efficiency for jobs using big data in limited bandwidth platforms . hence , local approaches can prove useful . in the case of the agent based approaches there is high periodicity for probing the cores in the background but very little data is transferred while probing unlike in checkpointing . hence , communication overhead times will be significantly lesser . lack of scalability is another issue that affects efficient fault tolerance . many checkpointing strategies are centralised ( with few exceptions , such as @xcite ) thereby limiting the scale of adopting the strategy . this can be mitigated by using multiple centralised checkpointing servers but the distance between the nodes and the server discounts the scalability issue . in the agent based approaches , all communications are short distance since the cores only need to communicate with the adjacent cores . local communication therefore increases the scale to which the agent based approaches can be applied . checkpointing is susceptible to single point failures due to the failure of the checkpoint servers . the job executed will have to be restarted . the agent - based approaches are also susceptible to single point failures . while they incorporate intelligence to anticipate hardware failure the processor core may fail before the sub - job it supports can be relocated to an adjacent processor core , before the transfer is complete , or indeed the core onto which it is being transferred may also fail . however , the incorporation of intelligence on the processor core , specifically the ability to anticipate hardware failure locally , means that the numbers of these hardware failures that lead to job failure can be reduced when compared to traditional checkpointing . but since there is the possibility of agent failure the retention of some level of human intervention is still required . therefore , we propose combining checkpointing with the agent - based approaches , the latter acting as a first line of anticipatory response to hardware failure backed up by traditional checkpointing as a second line of reactive response . figure 15 shows the execution of a job between two checkpoints , @xmath84 and @xmath85 , where @xmath86 is the predicted failure and @xmath87 is the actual failure of the node on which a sub - job is executing . figure 15(a ) shows when there are no predicted failures or actual failures that occur on the node . figure 15(b ) shows when a failure occurs but could not be predicted . in this case , the system fails if the multi - agent approaches are solely employed . one way to mitigate this problem is by employing the multi - agent approaches in conjunction with checkpointing as shown in the next section . figure 15(c ) shows when the approaches predict a failure which does not happen . if a large number of such predictions occur then the sub - job needs to be shifted often from one node to the other which adds to the overhead time for executing the job . this is not an ideal case and makes the job unstable . figure 15(d ) shows the ideal case in which a fault is predicted before it occurs . and @xmath85 . ( a ) ideal state of the job when no faults occur . ( b ) failure state of the job when a fault occurs but is not predicted . ( c ) unstable state of the job when a fault is predicted but does not occur . ( d ) ideal prediction state of the job when a fault is predicted and occurs thereafter.,scaledwidth=40.0% ] failure prediction is based on a machine learning approach that is incorporated within multi - agents . this prediction is based on a log that is maintained on the health of the node and its adjacent nodes . each agent sends out are you alive signals to adjacent nodes to gather the state of the adjacent node . the machine learning approach is constantly evaluating the state of the system against the log it maintains , which is different across the nodes . the log can contain the state of the node from past failures , work load of the nodes when it failed previously and even data related to patterns of periodic failures . however , this prediction method can not predict a range of faults due to deadlocks , hardware and power failures and instantaneously occurring faults . hence , the multi - agent approaches are most useful when used along with checkpointing . it was observed that nearly 29% of all faults occurring in the cluster could be predicted . although this number is seemingly small it is helpful to not have to rollback to a previous checkpoint when a large job under time constraints is executed . accuracy of the predictions were found to be 64% ; the system was found to be stable in 64 out of the 100 times a prediction was made . to increase the impact of the multi - agent approaches more faults will need to be captured . for this extensive logging and faster methods for prediction will need to be considered . these approaches will have to be used in conjunction with checkpointing for maximum effectiveness . the instability due to the approaches shifting jobs between nodes when there is a false prediction will need to be reduced to improve the overall efficiency of the approaches . for this , the state of the node can be compared with other nodes so that a more informed choice is made by the approaches . table 1 shows a comparison between a number of fault tolerant strategies , namely centralised and decentralised checkpointing and the multi - agent approaches . an experiment was run for a genome searching job that was executed multiple times on the placentia cluster . data in the table was obtained to study the execution of the genome searching job between two checkpoints ( @xmath84 and @xmath85 ) which are one hour apart . the execution is interrupted by failure @xmath87 as shown in figure 16 . two types of single node failure are simulated in the execution . the first is a periodic node failure which occurs at 15 minutes after @xmath84 and 45 minutes before @xmath85 ( refer figure 16(a ) ) , and the second is a random node failure which occurs @xmath88 minutes after @xmath84 and @xmath89 minutes before @xmath85 ( refer figure 16(b ) ) . the average time when a random failure occurs is found to be 31 minutes and 14 seconds for 5000 trials . the size of data , @xmath82 kb and the number of dependencies , @xmath90 . in table 1 , the average time taken for reinstating execution , for the overheads and for executing the job between the checkpoints is considered . the time taken for reinstating execution is for bringing execution back to normal after a failure has occurred . the reinstating time is obtained for one periodic single node failure and one random single node failure . the overhead time is for creating the checkpoints and transferring data for the checkpoint to the server . the overhead time is obtained for one periodic single node failure and one random single node failure . the execution time without failures , when one periodic failure occurs per hour and when five random failures occur per hour is obtained . centralised checkpointing using single and multiple servers is considered when the frequency of checkpointing is once every hour . in the case of both single and multiple server checkpointing the time taken for reinstating execution regardless of whether it was a periodic or random failure is 14 minutes and 8 seconds . on a single server the overhead is 8 minutes and 5 seconds where as the overhead to create the checkpoint is 9 minutes and 14 seconds which is higher than overheads on a single server and is expected . the average time taken for executing the job when one failure occurs includes the elapsed execution time ( 15 minutes for periodic failure and 31 minutes and 14 seconds for random failure ) until the failure occurred and the combination of the time for reinstating execution after the failures and the overhead time . for one periodic failure that occurs in one hour the penalty of execution when single server checkpointing is 62% more than executing without a failure ; in the case of a random failure that occurs in one hour the penalty is 89% more than executing without a failure . if five random failure occur then the penalty is 445% , requiring more than five times the time for executing the job without failures . centralised checkpointing with multiple servers requires more time than with a single server . this is due to the increase in the overhead time for creating checkpoints on multiple servers . hence , checkpointing with multiple servers requires 64% and 91% more time than the time for executing the job without any failures for one periodic and one random failure per hour respectively . on the other hand executing jobs when decentralised checkpointing on multiple servers is employed requires similar time to that taken by centralised checkpointing on a single server . the time for reinstating execution is higher than centralised checkpointing methods due to the time required for determining the server closest to the node that failed . however , the overhead times are lower than other checkpoint approaches since the server closest to the node that failed is chosen for creating the checkpoint which reduces data transfer times . the multi - agent approaches are proactive and therefore the average time taken for predicting single node failures are taken into account which is nearly 38 seconds . the time taken for reinstating execution after one periodic single node failure for the agent intelligence approach is 0.47 seconds and for the core intelligence approach is 0.38 seconds . since @xmath76 the core intelligence approach is selected . in this case , the core intelligence approach is faster than the agent intelligence approach in the total time it takes for executing the job when there is one periodic or random fault and when there are five faults that occur in the job . the multi - agent approaches only require one - fifth the time taken by the checkpointing methods for completing execution . this is because the time for reinstating and the overhead times are significantly lower than the checkpointing approaches . table 2 shows a comparison between centralised and decentralised checkpointing and the multi - agent approaches for a genome searching job that is executed on the placentia cluster for five hours . the checkpoint periodicity is once every one , two and four hours as shown in figure 17 . similar to table 1 periodic and random failures are simulated . figure 17(a ) shows the start and completion of the job without failures or checkpoints . when the checkpoint periodicity is one hour there are four checkpoints , @xmath7 , @xmath91 , @xmath6 and @xmath92 ( refer figure 17(b ) ) ; a periodic node failure is simulated after 14 minutes from a checkpoint and the average time at which a random node failure occurs is found to be 31 minutes and 14 seconds from a checkpoint for 5000 trials . when checkpoint periodicity is two hours there are two checkpoints , @xmath7 and @xmath91 ( refer figure 17(c ) ) ; a periodic node failure is simulated after 28 minutes from a checkpoint and the average time a random node failure occurs is found to be after 1 hour , 3 minutes and 22 seconds from a checkpoint for 5000 trials . when checkpoint periodicity is four hours there is only one checkpoint @xmath7 ( refer figure 17(d ) ) ; a periodic node failure is simulated after 56 minutes from a checkpoint and the average time at which a random failure occurs is found to be after 2 hours , 8 minutes and 47 seconds from each checkpoint for 5000 trials . similar to table 1 , in table 2 , the average time taken for reinstating execution , for the overheads and for executing the job from the start to finish with and without checkpoints is considered . the time to bring execution back to normal after a failure has occurred is referred to as reinstating time . the time to create checkpoints and transfer checkpoint data to the server is referred to as the overhead time . the execution of the job when one periodic and one random failure occurs per hour and when five random failures occur per hour is considered . without checkpointing the genome searching job is run such that a human administrator monitors the job from its start until completion . in this case , if a node fails then the only option is to restart the execution of the job . each time the job fails and given that the administrator detected it using cluster monitoring tools as soon as the node failed approximately , then at least ten minutes are required for reinstating the execution . if a periodic failure occurred once every hour from the 14th minute from execution then there are five periodic faults . once a failure occurs the execution will always have to come back to its previous state by restarting the job . hence , the five hour job , with just one periodic failure occurring every hour will take over 21 hours . similarly , if a random failure occurred once every hour ( average time of occurrence is 31 minutes and 14 seconds after execution starts ) , then there are five failure points , and over 23 hours are required for completing the job . when five random failures occur each hour of the execution then more than 80 hours are required ; this is nearly 16 times the time for executing the job without a failure . centralised checkpointing on a single server and on multiple servers and decentralised checkpointing on multiple servers for one , two and four hour periodicity in the network are then considered in table 2 . for checkpointing methods when one hour frequency is chosen more than five times the time taken for executing the job without failures is required . when the frequency of checkpointing is every two hours then just under four times the time taken for executing the job without failures is required . in the case when the checkpoint is created every four hours just over 3 times the time taken for executing the job without failures is required . the multi - agent approaches on the other hand take only one - fourth the time taken by traditional approaches for the job with five single node faults that occur each hour . this is significant time saving for running jobs that require many hours for completing execution . the agent and core intelligence approaches are similar in at least four ways . firstly , the objective of both the approaches is to automate fault tolerance . secondly , the job to be executed is broken down into sub - jobs which are executed . thirdly , fault tolerance is achieved in both approaches by predicting faults likely to occur in the computing core . fourthly , technology enabling mobility is required by both the approaches to carry the sub - job or to push the sub - job from one core onto another . these important similarities enable both the agent and core approaches to be brought together to offer the advantages as a hybrid approach . while there are similarities between the agent and core intelligence approaches there are differences that reflect in their implementation . these differences are based on : ( i ) where the job is situated - in the agent intelligence approach , the sub - job becomes the payload of an agent situated on a computing core . in the core intelligence approach , the sub - job is situated on a virtual core , which is an abstraction of the computing core . ( ii ) who predicts the failures - the agent constantly probes the compute core it is situated on and predicts failure in the agent approach , whereas in the core approach the virtual core anticipates the failure . ( iii ) who reacts to the prediction - the agent moves onto another core and re - establishes its dependencies in the agent approach , whereas the virtual core is responsible for moving a sub - job onto another core in the core approach . ( iv ) how dependencies are updated - an agent requires to carry information of its dependencies when it moves off onto another core and establishes its dependencies manually in the agent approach , whereas the dependencies of the sub - job on the core do not require to be manually updated in the core approach . ( v ) what view is obtained - in the agent approach , agents have a global view as they can traverse across the network of virtual cores , which is in contrast to the local view of the virtual cores in the core approach . the agent based approaches described in this paper offer a candidate solution for automated fault tolerance or in combination with checkpointing as proposed above offer a means of reducing current levels of human intervention . the foundational concepts of the agent and core based approaches were validated on four computer clusters using parallel reduction algorithms as a test case in this paper . failure scenarios were considered in the experimental studies for the two approaches . the effect of the number of dependencies of a sub - job being executed , the volume of data communicated across cores , and the process size are three factors considered in the experimental studies for determining the performance of the approaches . the approaches were studied in the context of parallel genome searching , a popular computational biology job , that fits the criteria of a parallel reduction algorithm . the experiments were performed for both periodic and random failures . the approaches were compared against centralised and decentralised checkpointing approaches . in a typical experiment in which the fault tolerant approaches are studied in between two checkpoints one hour apart when one random failure occurs , centralised and decentralised checkpointing on an average add 90% to the actual time for executing the job without any failures . on the other hand , in the same experiment the multi - agent approaches add only 10% to the overall execution time . the multi - agent approaches can not predict all failures that occur in the computing nodes . hence , the most efficient way of incorporating these approaches is to use them on top of checkpointing . the experiments demonstrate the feasibility of such approaches for computational biology jobs . the key result is that a job continues execution after a core has failed and the time required for reinstating execution is lesser than checkpointing methods . future work will explore methods to improve the accuracy of prediction as well as increase the number of faults that can be predicted using the multi - agent approaches . the challenge to achieve this will be to mine log files for predicting a wide range of faults and predict them as quickly as possible before the fault occurs . although the approaches can reduce human administrator intervention they can be used independently only if a wider range of faults can be predicted with greater accuracy . until then the multi - agent approaches can be used in conjunction with checkpointing for improving fault tolerance . the authors would like to thank the administrators of the compute resources at the centre for advanced computing and emerging technologies ( acet ) , university of reading , uk and the atlantic computational excellence network ( acenet ) . cappello f ( 2009 ) fault tolerance in petascale / exascale systems : current knowledge , challenges and research opportunities . international journal of high performance computing supplications , 23(3 ) : 212 - 226 . engelmann c , vallee gr , naughton t and scott sl ( 2009 ) proactive fault tolerance using preemptive migration . proceedings of the 17th euromicro international conference on parallel , distributed and network - based processing . 252 - 257 . vallee g , engelmann c , tikotekar a , naughton t , charoenpornwattana k , leangsuksun c and scott sl ( 2008 ) a framework for proactive fault tolerance . proceedings of the 3rd international conference on availability , reliability and security . 659 - 664 . fagg ge , gabriel e , chen z , angskun t , bosilca g , grbovic jp , dongarra j ( 2005 ) process fault - tolerance : semantics , design and applications for high performance computing . international journal for high performance applications and supercomputing . 19(4 ) : 465 - 477 . yeh ch ( 2003 ) the robust middleware approach for transparent and systematic fault tolerance in parallel and distributed systems . proceedings of the international conference on parallel processing . 61 - 68 . mourino jc , martin mj , gonzalez p and doallo r ( 2007 ) fault - tolerant solutions for a mpi compute intensive application . proceedings of the 15th euromicro international conference on parallel , distributed and network - based processing . 246 - 253 . tsai j , kuo sy and wang ym ( 1998 ) theoretical analysis for communication - induced checkpointing protocols with rollback - dependency trackability . ieee transactions on parallel and distributed systems . 9(10 ) : 963 - 971 . chtepen m , claeys fha , dhoedt b , de turuck f , demeester p and vanrolleghem pa ( 2009 ) adaptive task checkpointing and replication : toward efficient fault - tolerant grids . ieee transactions on parallel and distributed systems . 20(2 ) : 180 - 190 . sankaran s , squyres jm , barrett b , sahay v , lumsdaine a , duell j , hargrove p and roman e ( 2005 ) the lam / mpi checkpoint / restart framework : system - initiated checkpointing . international journal of high performance computing applications . 19(4 ) : 479 - 493 . hursey j , squyres jm , mattox ti , and lumsdaine a ( 2007 ) the design and implementation of checkpoint / restart process fault tolerance for open mpi . proceedings of the 12th ieee workshop on dependable parallel , distributed and network - centric systems . bowers kj , chow e , xu h , dror ro , eastwood mp , gregersen ba , klepeis jl , kolossvary i , moraes ma , sacerdoti fd , salmon jk , shan y and shaw de ( 2006 ) scalable algorithms for molecular dynamics simulations on commodity clusters . proceedings of the acm / ieee conference on supercomputing . je , tobias dj , brooks iii cl and singh uc ( 1991 ) vector and parallel algorithms for the molecular dynamics simulation of macromolecules on shared - memory computers . journal of computational chemistry . 12(10 ) : 1270 - 1277 . oliner aj , sahoo rk , moreira je , gupta m ( 2005 ) perfomance implications of periodic checkpointing on large - scale cluster systems . proceedings of the 19th ieee international parallel and distributed processing symposium , 2005 . gabriel e , fagg ge , bosilca g , angskun t , dongarra j , squyres jm , sahay v , kambadur p , barrett b , lumsdaine a , castain rh , daniel dj , graham rl , woodall ts ( 2004 ) open mpi : goals , concept , and design of a next generation mpi implementation . proceedings of the 11th european pvm / mpi users group meeting . 97 - 104 . chakravorty s , mendes cl and kale lv ( 2006 ) proactive fault tolerance in mpi applications via task migration . proceedings of ieee international conference on high performance computing , springer . lncs 4297 : 485 - 496 . janakiraman g , santos jr and subhraveti d ( 2005 ) cruz : application - transparent distributed checkpoint - restart on standard operating systems . proceedings of the international conference on dependable systems and networks . 260 - 269 . valle g , charoenpornwattana k , engelmann c , tikotekar a , leangsuksun c , naughton t and scott sl ( 2008 ) a framework for proactive fault tolerance . proceedings of the 3rd ieee international conference on availability , reliability and security . 659 - 664 . & predicting one single node failure & reinstating execution after one periodic single node failure & reinstating execution after one random single node failure & overheads related to one periodic single node failure & overheads related to one random single node failure & + & & & & & & without failures and checkpoints & with one periodic failure per hour & with one random failure per hour & with five random failures per hour + + 1 hour periodicity & - & 00:14:08 & 00:14:08 & 00:08:05 & 00:08:05 & 01:00:00 & 01:37:13 & 01:53:27 & 05:27:15 + + 1 hour periodicity & - & 00:14:08 & 00:14:08 & 00:09:14 & 00:09:14 & 01:00:00 & 01:38:22 & 01:54:36 & 05:33:00 + + 1 hour periodicity & - & 00:15:27 & 00:15:27 & 00:06:44 & 00:06:44 & 01:00:00 & 01:37:11 & 01:53:25 & 05:27:05 + + agent intelligence & 00:00:38 & 00:00:0.47 & 00:00:0.47 & 00:05:14 & 00:05:14 & & 01:06:17 & 01:06:17 & 01:32:27 + core intelligence & 00:00:38 & 00:00:0.38 & 00:00:0.38 & 00:04:27 & 00:04:27 & & 01:05:08 & 01:05:08 & 01:25:42 + hybrid intelligence & 00:00:38 & 00:00:0.38 & 00:00:0.38 & 00:04:27 & 00:04:27 & & 01:05:08 & 01:05:08 & 01:25:42 + & predicting one single node failure & reinstating execution after one periodic single node failure & reinstating execution after one random single node failure & all overheads related to one periodic single node failure & all overheads related to one random single node failure & + & & & & & & without failures & with one periodic failure per hour & with one random failure per hour & with five random failures per hour + cold restart with no failure tolerance & - & 00:10:00 & 00:10:00 & - & - & 05:00:00 & 21:15:17 & 23:01:00 & 80:31:04 + + 1 hour periodicity & - & 00:14:08 & 00:14:08 & 00:08:05 & 00:08:05 & & 08:01:05 & 09:27:15 & 27:16:15 + 2 hour periodicity & - & 00:15:40 & 00:15:40 & 00:10:17 & 00:10:17 & & 07:41:51 & 07:58:38 & 19:53:10 + 4 hour periodicity & - & 00:16:27 & 00:16:27 & 00:11:53 & 00:11:53 & & 06:24:20 & 07:37:07 & 18:05:35 + + 1 hour periodicity & - & 00:14:08 & 00:14:08 & 00:09:14 & 00:09:14 & & 08:07:14 & 09:33:23 & 27:45:00 + 2 hour periodicity & - & 00:15:40 & 00:15:40 & 00:12:22 & 00:12:22 & & 07:47:52 & 08:07:18 & 20:01:16 + 4 hour periodicity & - & 00:16:27 & 00:16:27 & 00:13:57 & 00:13:57 & & 07:04:28 & 07:52:27 & 18:45:22 + + 1 hour periodicity & - & 00:15:27 & 00:15:27 & 00:06:44 & 00:06:44 & & 08:00:55 & 09:27:05 & 27:15:25 + 2 hour periodicity & - & 00:17:23 & 00:17:23 & 00:09:46 & 00:09:46 & & 07:40:18 & 07:57:36 & 19:48:00 + 4 hour periodicity & - & 00:18:33 & 00:18:33 & 00:13:03 & 00:13:03 & & 06:27:36 & 07:40:23 & 18:21:55 + + 1 hour periodicity & & & & 00:05:14 & 00:05:14 & & 05:31:14 & 05:31:14 & 07:37:44 + 2 hour periodicity & & & & 00:06:38 & 00:06:38 & & 05:20:34 & 05:20:34 & 06:42:41 + 4 hour periodicity & & & & 00:07:41 & 00:07:41 & & 05:16:27 & 05:16:27 & 05:39:16 + + 1 hour periodicity & & & & 00:04:27 & 00:04:27 & & 05:26:13 & 05:26:13 & 07:11:37 + 2 hour periodicity & & & & 00:05:37 & 00:05:37 & & 05:16:22 & 05:16:22 & 06:22:34 + 4 hour periodicity & & & & 00:06:29 & 00:06:29 & & 05:13:32 & 05:13:32 & 05:31:21 +

we study the noise of the cotunneling current through one or several tunnel - coupled quantum dots in the coulomb blockade regime . the various regimes of weak and strong , elastic and inelastic cotunneling are analyzed for quantum - dot systems ( qds ) with few - level , nearly - degenerate , and continuous electronic spectra . in the case of weak cotunneling we prove a non - equilibrium fluctuation - dissipation theorem which leads to a universal expression for the noise - to - current ratio ( fano factor ) . the noise of strong inelastic cotunneling can be super - poissonian due to switching between qds states carrying currents of different strengths . the transport through a double - dot ( dd ) system shows an aharonov - bohm effect both in noise and current . in the case of cotunneling through a qds with a continuous energy spectrum the fano factor is very close to one .

cond-mat0211024

in recent years , there has been great interest in the shot noise in mesoscopic systems @xcite , because it contains additional information about correlations , which is not contained , e.g. , in the linear response conductance . the shot noise is characterized by the fano factor @xmath0 , the dimensionless ratio of the zero - frequency noise power @xmath1 to the average current @xmath2 . while it assumes the poissonian value @xmath3 in the absence of correlations , it becomes suppressed or enhanced when correlations set in as e.g. imposed by the pauli principle or due to interaction effects . in the present paper we study the shot noise of the cotunneling @xcite current . we consider the transport through a quantum - dot system ( qds ) in the coulomb blockade ( cb ) regime , in which the quantization of charge on the qds leads to a suppression of the sequential tunneling current except under certain resonant conditions . we consider the transport away from these resonances and study the next - order contribution to the current . this suppression can be interpreted @xcite as being a result of the natural correlations imposed by charge conservation . ] ( see fig . [ energies ] ) . we find that in the weak cotunneling regime , i.e. when the cotunneling rate @xmath4 is small compared to the intrinsic relaxation rate @xmath5 of the qds to its equilibrium state due to the coupling to the environment , @xmath6 , the zero - frequency noise takes on its poissonian value , as first obtained for a special case in @xcite . this result is generalized here , and we find a universal relation between noise and current for the qds in the first nonvanishing order in the tunneling perturbation . because of the universal character of this result eq . ( [ db - fdt ] ) we call it the nonequilibrium fluctuation - dissipation theorem ( fdt ) @xcite in analogy with linear response theory . one might expect however that the cotunneling , being a two - particle process , may lead to strong correlations in the shot noise and to the deviation of the fano factor from its poissonian value @xmath3 . we show in sec . [ super - poissonian ] that this is indeed the case for the regime of strong cotunneling , @xmath6 . specifically , for a two - level qds we predict giant ( divergent ) super - poissonian noise @xcite ( see sec . [ degenerate ] ) : the qds goes into an unstable mode where it switches between states 1 and 2 with ( generally ) different currents . in sec . [ dd - system ] we consider the transport through a double - dot ( dd ) system as an example to illustrate this effect ( see eq . ( [ dd - noise ] ) and fig . [ double - d ] ) . the fano factor turns out to be a periodic function of the magnetic flux through the dd leading to an aharonov - bohm effect in the noise @xcite . in the case of weak cotunneling we concentrate on the average current through the dd and find that it shows aharonov - bohm oscillations , which are a two - particle effect sensitive to spin entanglement . finally , in sec . [ continuum ] we discuss the cotunneling through large qds with a continuum spectrum . in this case the correlations in the cotunneling current described above do not play an essential role . in the regime of low bias , elastic cotunneling dominates transport,@xcite and thus the noise is poissonian . in the opposite case of large bias , the transport is governed by inelastic cotunneling , and in sec . [ continuum ] we study heating effects which are relevant in this regime . in general , the qds can contain several dots , which can be coupled by tunnel junctions , the dd being a particular example @xcite . the qds is assumed to be weakly coupled to external metallic leads which are kept at equilibrium with their associated reservoirs at the chemical potentials @xmath7 , @xmath8 , where the currents @xmath9 can be measured and the average current @xmath2 through the qds is defined by eq . ( [ current - noise ] ) . using a standard tunneling hamiltonian approach @xcite , we write @xmath10 where the terms @xmath11 and @xmath12 describe the leads and qds , respectively ( with @xmath13 and @xmath14 from a complete set of quantum numbers),and tunneling between leads and qds is described by the perturbation @xmath15 . the interaction term @xmath16 does not need to be specified for our proof of the universality of noise in sec . [ fdt - double ] . the @xmath17-electron qds is in the cotunneling regime where there is a finite energy cost @xmath18 for the electron tunneling from the fermi level of the lead @xmath19 to the qds ( @xmath20 ) and vice versa ( @xmath21 ) . this energy cost is of the order of the charging energy @xmath22 and much larger than the temperature , @xmath23 , so that only processes of second order in @xmath15 are allowed . via tunneling barriers . the tunneling between the qds and the leads is parametrized by the tunneling amplitudes @xmath24 , where the lead and qds quantum numbers @xmath13 and @xmath14 have been dropped for simplicity , see eq . ( [ tunneling ] ) . the leads are at the chemical potentials @xmath25 , with an applied bias @xmath26 . the eigenstates of the qds with one added electron ( @xmath27 electrons in total ) are indicated by their energies @xmath28 , @xmath29 , etc . , with average level - spacing @xmath30 . in the cotunneling regime there is a finite energy cost @xmath18 for the electron tunneling from the fermi level of the lead @xmath19 to the qds ( @xmath20 ) and vice versa ( @xmath21 ) , so that only processes of second order in @xmath15 ( visualized by two arrows ) are allowed . ] to describe the transport through the qds we apply standard methods @xcite and adiabatically switch on the perturbation @xmath15 in the distant past , @xmath31 . the perturbed state of the system is described by the time - dependent density matrix @xmath32 , with @xmath33 being the grand canonical density matrix of the unperturbed system , @xmath34 , where we set @xmath35 . because of tunneling the total number of electrons in each lead @xmath36 is no longer conserved . for the outgoing currents @xmath37 we have @xmath38 = ei(d^{\dag}_l - d_l)\ , . \label{currents}\ ] ] the observables of interest are the average current @xmath39 through the qds , and the spectral density of the noise @xmath40 , @xmath41 where @xmath42 . below we will use the interaction representation where eq . ( [ current - noise ] ) can be rewritten by replacing @xmath43 and @xmath44 , with @xmath45\ , . \label{u - operator}\ ] ] in this representation , the time dependence of all operators is governed by the unperturbed hamiltonian @xmath46 . in this section we prove the universality of noise of tunnel junctions in the weak cotunneling regime @xmath47 keeping the first nonvanishing order in the tunneling hamiltonian @xmath15 . since our final result ( [ db - fdt ] ) can be applied to quite general systems out - of - equilibrium we call this result the non - equilibrium fluctuation - dissipation theorem ( fdt ) . in particular , the geometry of the qds and the interaction @xmath16 are completely arbitrary for the discussion of the non - equilibrium fdt in this section . we note that the two currents @xmath48 are not independent , because @xmath49\neq 0 $ ] , and thus all correlators @xmath50 are nontrivial . the charge accumulation on the qds for a time of order @xmath51 leads to an additional contribution to the noise at finite frequency @xmath52 . thus , we expect that for @xmath53 the correlators @xmath50 can not be expressed through the steady - state current @xmath2 only and thus @xmath2 has to be complemented by some other dissipative counterparts , such as differential conductances @xmath54 . on the other hand , at low enough frequency , @xmath55 , the charge conservation on the qds requires @xmath56 . below we concentrate on the limit of low frequency and neglect contributions of order of @xmath57 to the noise power . in the appendix we prove that @xmath58 ( see eq . ( [ a09 ] ) ) , and this allows us to redefine the current and the noise power as @xmath59 and @xmath60 . , on a qds are also relevant for device applications such as set @xcite . while we focus on current fluctuations in the present paper , we mention here that in the cotunneling regime the noise power @xmath61 does not vanish at zero frequency , @xmath62 . our formalism is also suitable for studying such charge fluctuations ; this will be addressed elsewhere . ] in addition we require that the qds is in the cotunneling regime , i.e. the temperature is low enough , @xmath63 , although the bias @xmath64 is arbitrary as soon as the sequential tunneling to the dot is forbidden , @xmath65 . in this limit the current through a qds arises due to the direct hopping of an electron from one lead to another ( through a virtual state on the dot ) with an amplitude which depends on the energy cost @xmath66 of a virtual state . although this process can change the state of the qds ( inelastic cotunneling ) , the fast energy relaxation in the weak cotunneling regime , @xmath67 , immediately returns it to the equilibrium state ( for the opposite case , see sec . [ super - poissonian ] ) . this allows us to apply a perturbation expansion with respect to tunneling @xmath15 and to keep only first nonvanishing contributions , which we do next . it is convenient to introduce the notation @xmath68 . we notice that all relevant matrix elements , @xmath69 , @xmath70 , are fast oscillating functions of time . thus , under the above conditions we can write @xmath71 , and even more general , @xmath72 ( note that we have assumed earlier that @xmath73 ) . using these equalities and the cyclic property of the trace we obtain the following results ( for details of the derivation , see appendix a ) , @xmath74\rangle,\qquad a = d_2\bar d^{\dag}_1+d^{\dag}_1\bar d_2\ , , \label{db - current}\\ & & s(\omega)=e^2\int\limits^{\infty}_{-\infty}dt\ , \cos ( \omega t)\langle\{a^{\dag}(t),a(0)\}\rangle\ , , \label{db - noise}\end{aligned}\ ] ] where we have dropped a small contribution of order @xmath57 and used the notation @xmath75 . next we apply the spectral decomposition to the correlators eqs . ( [ db - current ] ) and ( [ db - noise ] ) , a similar procedure to that which also leads to the equilibrium fluctuation - dissipation theorem . the crucial observation is that @xmath76=0 $ ] , @xmath8 . therefore , we are allowed to use for our spectral decomposition the basis @xmath77 of eigenstates of the operator @xmath78 , which also diagonalizes the grand - canonical density matrix @xmath33 , @xmath79 $ ] . we introduce the spectral function , @xmath80 and rewrite eqs . ( [ db - current ] ) and ( [ db - noise ] ) in the matrix form in the basis @xmath81 taking into account that the operator @xmath82 , which plays the role of the effective cotunneling amplitude , creates ( annihilates ) an electron in the lead 2 ( 1 ) ( see eqs . ( [ tunneling ] ) and ( [ db - current ] ) ) . we obtain following expressions @xmath83{\cal a}(\delta\mu)\ , , \label{db - current2 } \\ & & s(\omega,\delta\mu)=\frac{e^2}{2}\sum_{\pm } { \cal a}(\delta\mu\pm\omega)\ , . \label{db - noise2}\end{aligned}\ ] ] we note that because of additional integration over time @xmath84 in the amplitude @xmath82 ( see eq . ( [ db - current ] ) ) , the spectral density @xmath85 depends on @xmath86 and @xmath87 separately . however , away from the resonances , @xmath88 , only @xmath64-dependence is essential , and thus @xmath85 can be regarded as being one - parameter function . -shift of the energy denominators @xmath89 , which is equivalent to neglecting small terms of order @xmath90 in eq . ( [ db - noise2 ] ) . ] comparing eqs . ( [ db - current2 ] ) and ( [ db - noise2 ] ) , we obtain @xmath91 i(\delta\mu\pm\omega ) \label{db - fdt}\ ] ] up to small terms on the order of @xmath90 . this equation represents our nonequilibrium fdt for the transport through a qds in the weak cotunneling regime . a special case with @xmath92 , giving @xmath93 , has been derived earlier @xcite . to conclude this section we would like to list again the conditions used in the derivation . the universality of noise to current relation eq . ( [ db - fdt ] ) proven here is valid in the regime in which it is sufficient to keep the first nonvanishing order in the tunneling @xmath15 which contributes to transport and noise . this means that the qds is in the weak cotunneling regime with @xmath94 , and @xmath47 . in this section we consider the qds in the strong cotunneling regime , @xmath95 . under this assumption the intrinsic relaxation in the qds is very slow and will in fact be neglected . thermal equilibration can only take place via coupling to the leads ( see sec . [ continuum ] ) . due to this slow relaxation in the qds we find that there are non - poissonian correlations @xmath96 in the current through the qds because the qds has a `` memory '' ; the state of the qds after the transmission of one electron influences the transmission of the next electron . the microscopic theory of strong cotunneling has been developed in ref . @xcite based on the density - operator formalism and using the projection operator technique . here we discuss the assumptions and present the results of the theory , equations ( [ mastereq ] ) , ( [ golden - rule ] ) , and ( [ avcurrent]-[correlator - delta - s ] ) , which are the basis for our further analysis in the secs . [ degenerate ] and [ dd - system ] . first , we assume that the system and bath are coupled only weakly and only via the perturbation @xmath15 , eq . ( [ tunneling ] ) . the interaction part @xmath16 of the unperturbed hamiltonian @xmath46 , eq . ( [ hamiltonian ] ) , must therefore be separable into a qds and a lead part , @xmath97 . moreover , @xmath46 conserves the number of electrons in the leads , @xmath76=0 $ ] , where @xmath98 . the assumption of weak coupling allows us to keep only the second - order in @xmath15 contributions to the `` golden rule '' rates ( [ golden - rule ] ) for the master equation ( [ mastereq ] ) . second , we assume that in the distant past , @xmath99 , the system is in an equilibrium state @xmath100 where @xmath101 $ ] , @xmath102 , and @xmath7 is the chemical potential of lead @xmath19 . note that both leads are kept at the same temperature @xmath103 . physically , the product form of @xmath33 in eq . ( [ initial ] ) describes the absence of correlations between the qds and the leads in the initial state at @xmath104 . furthermore , we assume that the initial state @xmath33 is diagonal in the eigenbasis of @xmath46 , i.e. that the initial state is an incoherent mixture of eigenstates of the free hamiltonian . finally , we consider the low - frequency noise , @xmath105 , i.e. we neglect the accumulation of the charge on the qds ( in the same way as in the sec . [ fdt - double ] ) . thus we can write @xmath106 . this restriction will be lifted in the end of the sec . [ dd - weak ] . we note that the above assumptions limit the generality of the results of present section as compared to those of sec . [ fdt - double ] . on the other hand , they allow us to reduce the problem of the noise calculations to the solution of the master equation @xmath107,\ ] ] with the stationary state condition @xmath108 . this `` classical '' master equation describes the dynamics of the qds , i.e. it describes the rates with which the probabilities @xmath109 for the qds being in state @xmath110 change . the rates @xmath111 are the sums of second - order `` golden rule '' rates @xmath112 for all possible cotunneling transitions from lead @xmath19 to lead @xmath113 . in the last expression , @xmath114 denotes the chemical potential drop between lead @xmath19 and lead @xmath113 , and @xmath115 . we have defined the second order hopping operator @xmath116 where @xmath117 is given in eq . ( [ tunneling ] ) , and @xmath118 . note , that @xmath119 is the amplitude of cotunneling from the lead @xmath19 to the lead @xmath113 ( in particular , we can write @xmath120 , see eq . ( [ db - current ] ) ) . the combined index @xmath121 contains both the qds index @xmath122 and the lead index @xmath123 . correspondingly , the basis states used above are @xmath124 with energy @xmath125 , where @xmath126 is an eigenstate of @xmath127 with energy @xmath128 , and @xmath129 is an eigenstate of @xmath130 with energy @xmath131 . for the average current @xmath2 and the noise power @xmath132 we obtain @xcite @xmath133\bar\rho_m + \delta s(\omega ) , \label{ndnoise}\\ \delta s(\omega ) & = & e^2 \!\!\!\!\!\!\sum_{n , m , n',m ' } w^i_{nm } \delta \rho_{mn'}(\omega ) w^i_{n'm'}\bar\rho_{m'},\label{correlator - delta - s}\end{aligned}\ ] ] where @xmath134 , and @xmath135 is the stationary density matrix . here , @xmath136 is the fourier - transformed conditional density matrix , which is obtained from the _ symmetrized _ solution @xmath137 of the master equation eq . ( [ mastereq ] ) with the initial condition @xmath138 . an explicit result for the noise in this case can be obtained by making further assumptions about the qds and the coupling to the leads , see the following sections . for the general case , we only estimate @xmath96 . the current is of the order @xmath139 , with @xmath140 some typical value of the cotunneling rate @xmath141 , and thus @xmath142 . the time between switching from one dot - state to another due to cotunneling is approximately @xmath143 . the correction @xmath96 to the poissonian noise can be estimated as @xmath144 , which is of the same order as the poissonian contribution @xmath145 . thus the correction to the fano factor is of order unity . ( note however , that under certain conditions the fano factor can diverge , see secs . [ degenerate ] and [ dd - system ] . ) in contrast to this , we find that for elastic cotunneling the off - diagonal rates vanish , @xmath146 , and therefore @xmath147 and @xmath148 . moreover , at zero temperature , either @xmath149 or @xmath150 must be zero ( depending on the sign of the bias @xmath64 ) . as a consequence , for elastic cotunneling we find poissonian noise , @xmath151 . suppose the qds has nearly degenerate states with energies @xmath152 , and level spacing @xmath153 , which is much smaller than the average level spacing @xmath30 . in the regime , @xmath154 , the only allowed cotunneling processes are the transitions between nearly degenerate states . the noise power is given by eqs . ( [ ndnoise ] ) and ( [ correlator - delta - s ] ) , and below we calculate the correlation correction to the noise , @xmath96 . to proceed with our calculation we rewrite eq . ( [ mastereq ] ) for @xmath155 as a second - order differential equation in matrix form @xmath156 where @xmath157 is defined as @xmath158 . we solve this equation by fourier transformation , @xmath159 where we have used @xmath160 . we substitute @xmath161 from this equation into eq . ( [ correlator - delta - s ] ) and write the result in a compact matrix form , @xmath162_{nm}. \label{result01}\ ] ] this equation gives the formal solution of the noise problem for nearly degenerate states . as an example we consider a two - level system . using the detailed balance equation , @xmath163 , we obtain for the stationary probabilities @xmath164 , and @xmath165 . from eq . ( [ avcurrent ] ) we get @xmath166 a straightforward calculation with the help of eq . ( [ delta - rho ] ) gives for the correction to the poissonian noise @xmath167}\times\nonumber \\ & \times & \left[w^{i}_{11}w_{12}w_{21}+w^{i}_{12}w^2_{21}- ( 1\leftrightarrow 2)\right ] . \label{two - noise}\end{aligned}\ ] ] in particular , the zero frequency noise @xmath168 diverges if the `` off - diagonal '' rates @xmath141 vanish . this divergence has to be cut at @xmath52 , or at the relaxation rate @xmath5 due to coupling to the bath ( since @xmath169 in this case has to be replaced with @xmath170 ) . the physical origin of the divergence is rather transparent : if the off - diagonal rates @xmath171 are small , the qds goes into an unstable state where it switches between states 1 and 2 with different currents in general @xcite . the longer the qds stays in the state 1 or 2 the larger the zero - frequency noise power is . however , if @xmath172 , then @xmath173 is suppressed to 0 . for instance , for the qds in the spin - degenerate state with an odd number of electrons @xmath174 , since the two states @xmath175 and @xmath176 are physically equivalent . the other example of such a suppression of the correlation correction @xmath96 to noise is given by a multi - level qds , @xmath177 , where the off - diagonal rates are small compared to the diagonal ( elastic ) rates @xcite . indeed , since the main contribution to the elastic rates comes from transitions through many virtual states , which do not participate in inelastic cotunneling , they do not depend on the initial conditions , @xmath178 , and cancel in the numerator of eq . ( [ two - noise ] ) , while they are still present in the current . thus the correction @xmath179 vanishes in this case . further below in this section we consider a few - level qds , @xmath180 , where @xmath181 . to simplify further analysis we consider for a moment the case , where the singularity in the noise is most pronounced , namely , @xmath182 and @xmath183 , so that @xmath184 , and @xmath185 . then , from eqs . ( [ two - current ] ) and ( [ two - noise ] ) we obtain @xmath186 where @xmath187 is the current through the @xmath188-th level of the qds . thus in case @xmath183 the following regimes have to be distinguished : ( 1 ) if @xmath189 , then @xmath190 , @xmath191 , and thus both , the total current @xmath192 , and the total noise @xmath193 are linear in the bias @xmath64 ( here @xmath194 is the conductance of the qds ) . the total shot noise in this regime is super - poissonian with the fano factor @xmath195 . ( 2 ) in the regime @xmath196 the noise correction ( [ two - noise2 ] ) arises because of the thermal switching the qds between two states @xmath197 , where the currents are linear in the bias , @xmath198 . the rate of switching is @xmath199 , and thus @xmath200 . since @xmath201 , the noise correction @xmath96 is the dominant contribution to the noise , and thus the total noise @xmath1 can be interpreted as being a thermal telegraph noise @xcite . ( 3 ) finally , in the regime @xmath202 the first term on the rhs of eq . ( [ ndnoise ] ) is the dominant contribution , and the total noise becomes an equilibrium nyquist noise , @xmath203 . , @xmath87 . the tunneling amplitudes between dots and leads are denoted by @xmath204 . the tunneling ( @xmath205 ) between the dots results in a singlet - triplet splitting @xmath206 with the singlet being a ground state . the tunneling path between dots and leads 1 and 2 forms a closed loop ( shown by arrows ) so that the aharonov - bohm phase @xmath207 will be accumulated by an electron traversing the dd . right : the fano factor @xmath208 , with the noise power @xmath132 given in eqs . ( [ ndnoise ] ) and ( [ dd - noise ] ) , is plotted as a function of the aharonov - bohm phase @xmath207 for the normalized bias @xmath209 and for four different normalized frequencies @xmath210=0.1 $ ] , @xmath211 , @xmath212 , and @xmath213 . inset : the same , but with fixed frequency @xmath214 , where the bias @xmath215 takes the values @xmath216 , @xmath217 , and @xmath218 . ] we notice that for the noise power to be divergent the off - diagonal rates @xmath169 and @xmath219 have to vanish simultaneously . however , the matrix @xmath141 is not symmetric since the off - diagonal rates depend on the bias in a different way . on the other hand , both rates contain the same matrix element of the cotunneling amplitude @xmath220 , see eqs . ( [ golden - rule ] ) and ( [ matrix - element ] ) . although in general this matrix element is not small , it can vanish because of different symmetries of the two states . to illustrate this effect we consider the transport through a double - dot ( dd ) system ( see ref . @xcite for details ) as an example . two leads are equally coupled to two dots in such a way that a closed loop is formed , and the dots are also connected , see fig . [ double - d ] . thus , in a magnetic field the tunneling is described by the hamiltonian eq . ( [ tunneling ] ) with @xmath221 where the last equation expresses the equal coupling of dots and leads and @xmath207 is the aharonov - bohm phase . each dot contains one electron , and weak tunneling @xmath205 between the dots causes the exchange splitting @xcite @xmath206 ( with @xmath222 being the on - site repulsion ) between one spin singlet and three triplets @xmath223|0\rangle\ , , \nonumber\\ & & |t_0\rangle=\frac{1}{\sqrt{2 } } [ d^{\dag}_{1\uparrow}d^{\dag}_{2\downarrow}\!+ d^{\dag}_{1\downarrow}d^{\dag}_{2\uparrow}]|0\rangle\ , , \label{basis}\\ & & |t_+\rangle = d^{\dag}_{1\uparrow}d^{\dag}_{2\uparrow}|0\rangle\ , , \quad \nonumber\end{aligned}\ ] ] in the case of zero magnetic field , @xmath224 , the tunneling hamiltonian @xmath15 is symmetric with respect to the exchange of electrons , @xmath225 . thus the matrix element of the cotunneling transition between the singlet and three triplets @xmath226 , @xmath227 , vanishes because these states have different orbital symmetries . a weak magnetic field breaks the symmetry , contributes to the off - diagonal rates , and thereby reduces noise . next , we consider weak and strong cotunneling regimes . in this regime , @xmath47 , according to the non - equilibrium fdt ( see sec . [ fdt - double ] ) the zero - frequency noise contains the same information as the average current ( the fano factor @xmath3 ) . therefore , we first concentrate on current . we focus on the regime , @xmath228 , where inelastic cotunneling @xcite occurs with singlet and triplet contributions being different , and where we can neglect the dynamics generated by @xmath229 compared to the one generated by the bias ( `` slow spins '' ) . close to the sequential tunneling peak , @xmath230 , we keep only the term @xmath231 in the amplitude ( [ db - current ] ) . after some calculations we obtain @xmath232,\label{factor1}\end{aligned}\ ] ] where @xmath233 is the conductance of a single dot in the cotunneling regime @xcite , and we assumed fermi liquid leads with the tunneling density of states @xmath234 . ( [ current4 ] ) shows that the cotunneling current depends on the properties of the equilibrium state of the dd through the coherence factor @xmath235 given in ( [ factor1 ] ) . the first term in @xmath236 is the contribution from the topologically trivial tunneling path ( phase - incoherent part ) which runs from lead 1 through , say , dot 1 to lead 2 and back . the second term ( phase - coherent part ) in @xmath236 results from an exchange process of electron 1 with electron 2 via the leads 1 and 2 such that a closed loop is formed enclosing an area @xmath82 ( see fig . [ double - d ] ) . note that for singlet and triplets the initial and final spin states are the same after such an exchange process . thus , in the presence of a magnetic field @xmath237 , an aharonov - bohm phase factor @xmath238 is acquired . next , we evaluate @xmath239 explicitly in the singlet - triplet basis ( [ basis ] ) and discuss the applications to the physics of quantum entanglement ( see the ref . note that only the singlet @xmath240 and the triplet @xmath241 are entangled epr pairs while the remaining triplets are not ( they factorize ) . assuming that the dd is in one of these states we obtain the important result @xmath242 thus , we see that the singlet ( upper sign ) and the triplets ( lower sign ) contribute with opposite sign to the phase - coherent part of the current . one has to distinguish , however , carefully the entangled from the non - entangled states . the phase - coherent part of the entangled states is a genuine _ two - particle _ effect , while the one of the product states can not be distinguished from a phase - coherent _ single - particle _ effect . indeed , this follows from the observation that the phase - coherent part in @xmath236 factorizes for the product states @xmath243 while it does not so for @xmath244 . also , for states such as @xmath245 the coherent part of @xmath236 vanishes , showing that two different ( and fixed ) spin states can not lead to a phase - coherent contribution since we _ know _ which electron goes which part of the loop . finally , we present our results @xcite for the high - frequency noise in the quantum range of frequancies , @xmath246 , and in the slow - spin regime @xmath228 . this range of frequancies is beyond the regime of the applicability of the non - equilibrium fdt , and therefore there is no simple relation between the average current and the noise ( see the sec . [ fdt - double ] ) . after lengthy calculations using the perturbation expansion of ( [ u - operator ] ) up to third order in @xmath15 we obtain @xmath247 , \nonumber \\ & & imx_{\omega}={{c(\varphi ) } \over { 2\omega}}\left[\theta ( \mu_1-\omega ) -\theta ( \mu_2-\omega ) \right ] , \\ & & rex_{\omega}={{c(\varphi ) } \over { 2\pi\omega}}sign(\mu_1-\mu_2+\omega ) \ln|{{(\mu_1+\omega ) ( \mu_2-\omega ) } \over { \mu_1\mu_2}}| \nonumber \\ & & -{1\over { 2\pi \omega } } \left[\theta ( \omega -\mu_1)\ln \theta ( \omega -\mu_2)\ln \label{noise3}\end{aligned}\ ] ] where @xmath248 . thus the real part of @xmath132 is even in @xmath52 , while the imaginary part is odd . a remarkable feature here is that the noise acquires an imaginary ( i.e. odd - frequency ) part for finite frequencies , in contrast to single - barrier junctions , where im@xmath249 always vanishes since we have @xmath250 for all times . in double - barrier junctions considered here we find that at small enough bias @xmath251 , the odd part , im@xmath249 , given in ( [ noise2 ] ) exhibits two narrow peaks at @xmath252 , which in real time lead to slowly decaying oscillations , @xmath253 these oscillations again depend on the phase - coherence factor @xmath236 with the same properties as discussed before . these oscillations can be interpreted as a temporary build - up of a charge - imbalance on the dd during an uncertainty time @xmath254 , which results from cotunneling of electrons and an associated time delay between out- and ingoing currents . the fact that in the perturbation @xmath15 all spin indices are traced out helps us to map the four - level system to only two states @xmath240 and @xmath255 classified according to the orbital symmetry ( since all triplets are antisymmetric in orbital space ) . in appendix b we derive the mapping to a two - level system and calculate the transition rates @xmath256 ( @xmath257 for a singlet and @xmath258 for all triplets ) using eqs . ( [ golden - rule ] ) and ( [ matrix - element ] ) with the operators @xmath117 given by eq . ( [ dd - d ] ) . doing this we obtain the following result @xmath259 which holds close to the sequential tunneling peak , @xmath230 ( but still @xmath260 ) , and for @xmath261 . we substitute this equation into the eq . ( [ two - noise ] ) and write the correction @xmath173 to the poissonian noise as a function of normalized bias @xmath262 and normalized frequency @xmath263 $ ] @xmath264 ^ 2(1-\cos\phi)}{(2v-1)^3[\omega^2+(1-\cos\phi ) ^2]}. \label{dd - noise}\ ] ] from this equation it follows that the noise power has singularities as a function of @xmath52 for zero magnetic field , and it has singularities at @xmath265 ( where @xmath122 is integer ) as a function of the magnetic field ( see fig . [ double - d ] ) . we would like to emphasize that the noise is singular even if the exchange between the dots is weak , @xmath266 . in the case @xmath267 the transition from the singlet to the triplet is forbidden by conservation of energy , @xmath268 , and we immediately obtain from eq . ( [ two - noise ] ) that @xmath269 , i.e. the total noise is poissonian ( as it is always the case for elastic cotunneling ) . in the case of large bias , @xmath228 , two dots contribute independently to the current @xmath270 , and from eq . ( [ dd - noise ] ) we obtain the fano factor @xmath271 this fano factor controls the transition to the telegraph noise and then to the equilibrium noise at high temperature , as described above . we notice that if the coupling of the dots to the leads is not equal , then @xmath272 serves as a cut - off of the singularity in @xmath173 . finally , we remark that the fano factor is a periodic function of the phase @xmath207 ( see fig . [ double - d ] ) ; this is nothing but an aharonov - bohm effect in the noise of the cotunneling transport through the dd . however , in contrast to the aharonov - bohm effect in the cotunneling current through the dd which has been discussed earlier in the sec . [ dd - weak ] , the noise effect does not allow us to probe the ground state of the dd , since the dd is already in a mixture of the singlet and three triplet states . we consider now the transport through a multi - level qds with @xmath177 . in the low bias regime , @xmath273 , the elastic cotunneling dominates transport @xcite , and according to the results of sec . [ super - poissonian ] the noise is poissonian . here we consider the opposite regime of inelastic cotunneling , @xmath274 . since a large number @xmath275 of levels participate in transport , we can neglect the correlations which we have studied in secs . [ degenerate ] and [ dd - system ] , since they become a @xmath276-effect . instead , we concentrate on the heating effect , which is not relevant for the 2-level system considered before . the condition for strong cotunneling has to be rewritten in a single - particle form , @xmath277 , where @xmath278 is the single - particle energy relaxation time on the qds due to the coupling to the environment , and @xmath279 is the time of the cotunneling transition , which can be estimated as @xmath280 ( where @xmath281 is the density of qds states ) . since the energy relaxation rate on the qds is small , the multiple cotunneling transitions can cause high energy excitations on the dot , and this leads to a nonvanishing backward tunneling , @xmath282 . in the absence of correlations between cotunneling events , eqs . ( [ avcurrent ] ) and ( [ ndnoise ] ) can be rewritten in terms of forward and backward tunneling currents @xmath283 and @xmath284 , @xmath285 where the transition rates are given by ( [ golden - rule ] ) . it is convenient to rewrite the currents @xmath286 in a single - particle basis . to do so we substitute the rates eq . ( [ golden - rule ] ) into eq . ( [ ipm1 ] ) and neglect the dependence of the tunneling amplitudes eq . ( [ tunneling ] ) on the quantum numbers @xmath13 and @xmath14 , @xmath287 , which is a reasonable assumption for qds with a large number of electrons . then we define the distribution function on the qds as @xmath288 and replace the summation over @xmath14 with an integration over @xmath289 . doing this we obtain the following expressions for @xmath290 @xmath291 , \label{cpm}\end{aligned}\ ] ] where @xmath292 are the tunneling conductances of the barriers 1 and 2 , and where we have introduced the function @xmath293 with @xmath294 being the step - function . in particular , using the property @xmath295 and fixing @xmath296=0 , \label{symmetry}\ ] ] ( since @xmath286 given by eq . ( [ ipm2 ] ) and eq . ( [ cpm ] ) do not depend on the shift @xmath297 ) we arrive at the following general expression for the cotunneling current @xmath298\geq 0 , \label{upsilon}\end{aligned}\ ] ] where the value @xmath299 has the physical meaning of the energy acquired by the qds due to the cotunneling current through it . we have deliberately introduced the functions @xmath300 in the eq . ( [ ipm2 ] ) to emphasize the fact that if the distribution @xmath301 scales with the bias @xmath64 ( i.e. @xmath302 is a function of @xmath303 ) , then @xmath300 become dimensionless universal numbers . thus both , the prefactor @xmath304 ( given by eq . ( [ prefactor ] ) ) in the cotunneling current , and the fano factor , @xmath305 take their universal values , which do not depend on the bias @xmath64 . we consider now such universal regimes . the first example is the case of weak cotunneling , @xmath306 , when the qds is in its ground state , @xmath307 , and the thermal energy of the qds vanishes , @xmath308 . then @xmath309 , and eq . ( [ i - continuum ] ) reproduces the results of ref . @xcite . as we have already mentioned , the backward current vanishes , @xmath310 , and the fano factor acquires its full poissonian value @xmath3 , in agreement with our nonequilibrium fdt proven in sec . [ fdt - double ] . in the limit of strong cotunneling , @xmath277 , the energy relaxation on the qds can be neglected . depending on the electron - electron scattering time @xmath311 two cases have to be distinguished : the regime of cold electrons @xmath312 and regime of hot electrons @xmath313 on the qds . below we discuss both regimes in detail and demonstrate their universality . in this regime the electron - electron scattering on the qds can be neglected and the distribution @xmath301 has to be found from the master equation eq . ( [ mastereq ] ) . we multiply this equation by @xmath314 , sum over @xmath188 and use the tunneling rates from eq . ( [ golden - rule ] ) . doing this we obtain the standard stationary kinetic equation which can be written in the following form @xmath315 \nonumber \\ & & \qquad\qquad\qquad = \int d\varepsilon ' \sigma(\varepsilon -\varepsilon')f(\varepsilon)[1-f(\varepsilon ' ) ] , \label{kineq1}\\ & & \sigma(\varepsilon)=2\lambda\theta(\varepsilon)+ \sum_{\pm}\theta(\varepsilon \pm\delta\mu ) , \label{kernel}\end{aligned}\ ] ] where @xmath316 arises from the equilibration rates @xmath317 . ( we assume that if the limits of the integration over energy @xmath289 are not specified , then the integral goes from @xmath318 to @xmath319 . ) from the form of this equation we immediately conclude that its solution is a function of @xmath303 , and thus the cold electron regime is universal as defined in the previous section . it is easy to check that the detailed balance does not hold , and in addition @xmath320 . thus we face a difficult problem of solving eq . ( [ kineq1 ] ) in its full nonlinear form . fortunately , there is a way to avoid this problem and to reduce the equation to a linear form which we show next . we group all nonlinear terms on the rhs of eq . ( [ kineq1 ] ) : @xmath321 , where @xmath322\right\}$ ] . the trick is to rewrite the function @xmath323 in terms of known functions . for doing this we split the integral in @xmath323 into two integrals over @xmath324 and @xmath325 , and then use eq . ( [ symmetry ] ) and the property of the kernel @xmath326 to regroup terms in such a way that @xmath323 does not contain @xmath301 explicitly . taking into account eq . ( [ upsilon ] ) we arrive at the following linear integral equation @xmath327f(\varepsilon ) , \label{kineq2}\ ] ] where the parameter @xmath328 is the only signature of the nonlinearity of eq . ( [ kineq1 ] ) . since eq . ( [ kineq2 ] ) represents an eigenvalue problem for a linear operator , it can in general have more than one solution . however , there is only one physical solution , which satisfies the conditions @xmath329 indeed , using a standard procedure one can show that two solutions of the integral equation ( [ kineq2 ] ) , @xmath330 and @xmath331 , corresponding to different parameters @xmath332 should be orthogonal , @xmath333 . this contradicts the conditions eq . ( [ conditions ] ) . the solution is also unique for the same @xmath328 , i.e. it is not degenerate ( for a proof , see the ref . @xcite ) . from eq . ( [ kineq1 ] ) and conditions eq . ( [ conditions ] ) it follows that if @xmath301 is a solution then @xmath334 also satisfies eqs . ( [ kineq1 ] ) and ( [ conditions ] ) . since the solution is unique , it has to have the symmetry @xmath335 . in the expression ( [ i - continuum ] ) for the cotunneling current characterizes a universal cotunneling transport in the regime of weak cotunneling , @xmath306 , ( @xmath309 , see ref . @xcite ) , and in the regime of strong cotunneling , @xmath277 ( @xmath336 ) . here @xmath304 is plotted as a function of @xmath337 ( same as a function of @xmath338 ) for the strong cotunneling , for the cold - electron case , @xmath312 ( solid line ) and for the hot - electron case , @xmath313 ( dotted line ) . @xmath339 are the tunneling conductances of a junctions connecting leads 1 and 2 with the qds . ] we solve eqs . ( [ kineq2 ] ) and ( [ conditions ] ) numerically and use eqs . ( [ cpm ] ) and ( [ f - continuum ] ) to find that the fano factor is very close to 1 ( it does not exceed the value @xmath340 ) . next we use eqs . ( [ prefactor ] ) and ( [ upsilon ] ) to calculate the prefactor @xmath304 and plot the result as a function of the ratio of tunneling conductances , @xmath337 , ( fig . [ coldhot ] , solid line ) . for equal coupling to the leads , @xmath341 , the prefactor @xmath304 takes its maximum value @xmath342 , and thus the cotunneling current is approximately twice as large compared to its value for the case of weak cotunneling , @xmath306 . @xmath304 slowly decreases with increasing asymmetry of coupling and tends to its minimum value @xmath343 for the strongly asymmetric coupling case @xmath337 or @xmath344 . in the regime of hot electrons , @xmath313 , the distribution is given by the equilibrium fermi function @xmath345^{-1}$ ] , while the electron temperature @xmath346 has to be found self - consistently from the kinetic equation . ( [ kineq1 ] ) has to be modified to take into account electron - electron interactions . this can be done by adding the electron collision integral @xmath347 to the rhs of ( [ kineq1 ] ) . since the form of the distribution is known we need only the energy balance equation , which can be derived by multiplying the modified equation ( [ kineq1 ] ) by @xmath289 and integrating it over @xmath289 . the contribution from the collision integral @xmath347 vanishes , because the electron - electron scattering conserves the energy of the system . using the symmetry @xmath348 we arrive at the following equation @xmath349 \sigma(\varepsilon'-\varepsilon)\varepsilon = 0 . \label{kineq3}\ ] ] next we regroup the terms in this equation such that it contains only integrals of the form @xmath350 . this allows us to get rid of nonlinear terms , and we arrive at the following equation , @xmath351 + 3\upsilon^2=\frac{(\delta\mu)^4}{8(1+\lambda)}\ , , \label{kineq4}\ ] ] which holds also for the regime of cold electrons . finally , we calculate the integral in eq . ( [ kineq4 ] ) and express the result in terms of the dimensionless parameter @xmath352 , @xmath353^{1/4}. \label{alpha}\ ] ] thus , since the distribution again depends on the ratio @xmath303 , the hot electron regime is also universal . the next step is to substitute the fermi distribution function with the temperature given by eq . ( [ alpha ] ) into eq . ( [ cpm ] ) . we calculate the integrals and arrive at the closed analytical expressions for the values of interest , @xmath354e^{-\alpha n } , \label{f - hot}\end{aligned}\ ] ] where again @xmath355 . it turns out that similar to the case of cold electrons , sec . [ cold ] , the fano factor for hot electrons is very close to @xmath213 ( namely , it does not exceed the value @xmath356 ) . therefore , we do not expect that the super - poissonian noise considered in this section ( i.e. the one which is due to heating of a large qds caused by inelastic cotunneling through it ) will be easy to observe in experiments . on the other hand , the transport - induced heating of a large qds can be observed in the cotunneling current through the prefactor @xmath304 , which according to eq . ( [ i - hot ] ) takes its maximum value @xmath357 for @xmath341 and slowly reaches its minimum value @xmath213 with increasing ( or decreasing ) the ratio @xmath337 ( see fig . [ coldhot ] , dotted line ) . surprisingly , the two curves of @xmath304 vs @xmath337 for the cold- and hot - electron regimes lie very close , which means that the effect of the electron - electron scattering on the cotunneling transport is rather weak . here we give a short summary of our results . in sec . [ fdt - double ] , we have derived the non - equilibrium fdt , i.e. the universal relation ( [ db - fdt ] ) between the current and the noise , for qds in the weak cotunneling regime . taking the limit @xmath358 , we show that the noise is poissonian , i.e. @xmath3 . in sec . [ super - poissonian ] , we present the results of the microscopic theory of strong cotunneling , ref . @xcite : the master equation , eq . ( [ mastereq ] ) , the average current , eq . ( [ avcurrent ] ) , and the current correlators , eqs . ( [ ndnoise ] ) and ( [ correlator - delta - s ] ) , for a qds system coupled to leads in the strong cotunneling regime @xmath95 at small frequencies , @xmath359 . in contrast to sequential tunneling , where shot noise is either poissonian ( @xmath3 ) or suppressed due to charge conservation ( @xmath360 ) , we find that the noise in the inelastic cotunneling regime can be super - poissonian ( @xmath361 ) , with a correction being as large as the poissonian noise itself . in the regime of elastic cotunneling @xmath3 . while the amount of super - poissonian noise is merely estimated at the end of sec . [ super - poissonian ] , the noise of the cotunneling current is calculated for the special case of a qds with nearly degenerate states , i.e. @xmath362 , in sec . [ degenerate ] , where we apply our results from sec . [ super - poissonian ] . the general solution eq . ( [ result01 ] ) is further analyzed for two nearly degenerate levels , with the result eq . ( [ two - noise ] ) . more information is gained in the specific case of a dd coupled to leads considered in sec . [ dd - system ] , where we determine the average current eqs . ( [ current4]-[factor1 ] ) and noise eqs . ( [ noise2]-[noise3 ] ) in the weak cotunneling regime and the correlation correction to noise eq . ( [ dd - noise ] ) in the strong cotunneling regime as a function of frequency , bias , and the aharonov - bohm phase threading the tunneling loop , finding signatures of the aharonov - bohm effect and of the quantum entanglement . finally , in sec . [ continuum ] , another important situation is studied in detail , the cotunneling through a qds with a continuous energy spectrum , @xmath363 . here , the correlation between tunneling events plays a minor role as a source of super - poissonian noise , which is now caused by heating effects opening the possibility for tunneling events in the reverse direction and thus to an enhanced noise power . in eq . ( [ f - continuum ] ) , we express the fano factor @xmath364 in the continuum case in terms of the dimensionless numbers @xmath365 , defined in eq . ( [ cpm ] ) , which depend on the electronic distribution function @xmath301 in the qds ( in this regime , a description on the single - electron level is appropriate ) . the current eq . ( [ i - continuum ] ) is expressed in terms of the prefactor @xmath304 , eq . ( [ prefactor ] ) . both @xmath364 and @xmath304 are then calculated for different regimes . for weak cotunneling , we immediately find @xmath3 , as anticipated earlier , while for strong cotunneling we distinguish the two regimes of cold ( @xmath312 ) and hot ( @xmath313 ) electrons . for both regimes we find that the fano factor is very close to one , while @xmath304 is given in fig . [ coldhot ] . this work has been partially supported by the swiss national science foundation . in this appendix we present the derivation of eqs . [ db - current ] and [ db - noise ] . in order to simplify the intermediate steps , we use the notation @xmath366 for any operator @xmath367 , and @xmath368 . we notice that , if an operator @xmath367 is a linear function of operators @xmath117 and @xmath369 , then @xmath370 ( see the discussion in sec . [ fdt - double ] ) . next , the currents can be represented as the difference and the sum of @xmath371 and @xmath372 , @xmath373 where @xmath374 , and @xmath375 . while for the perturbation we have @xmath376 first we concentrate on the derivation of eq . ( [ db - current ] ) and redefine the average current eq . ( [ current - noise ] ) as @xmath377 ( which gives the same result anyway , because the average number of electrons on the qds does not change @xmath378 ) . to proceed with our derivation , we make use of eq . ( [ u - operator ] ) and expand the current up to fourth order in @xmath379 : @xmath380 next , we use the cyclic property of trace to shift the time dependence to @xmath381 . then we complete the integral over time @xmath84 and use @xmath382 . this procedure allows us to combine first and second term in eq . ( [ a04 ] ) , @xmath383v(t)\bar v(t)\rangle+{\rm c.c . } \label{a05}\,\ ] ] now , using eqs . ( [ a01 ] ) and ( [ a03 ] ) we replace operators in eq . ( [ a05 ] ) with @xmath384 and @xmath385 in two steps : @xmath386v(t)\bar v(t)\rangle+{\rm c.c.}$ ] , where some terms cancel exactly . then we work with @xmath387 and notice that some terms cancel , because they are linear in @xmath388 and @xmath389 . thus we obtain @xmath390 [ x^{\dag}(t)\bar x^{\dag}(t)+x(t)\bar x(t)]\rangle + { \rm c.c.}$ ] . two terms @xmath391 and @xmath392 describe tunneling of two electrons from the same lead , and therefore they do not contribute to the normal current . we then combine all other terms to extend the integral to @xmath319 , @xmath393 finally , we use @xmath394 ( since @xmath395 ) to get eq . ( [ db - current ] ) with @xmath396 . here , again , we drop terms @xmath397 and @xmath398 responsible for tunneling of two electrons from the same lead , and obtain @xmath82 as in eq . ( [ db - current ] ) . next , we derive eq . ( [ db - noise ] ) for the noise power . at small frequencies @xmath359 fluctuations of @xmath399 are suppressed because of charge conservation ( see below ) , and we can replace @xmath372 in the correlator eq . ( [ current - noise ] ) with @xmath381 . we expand @xmath400 up to fourth order in @xmath379 , use @xmath401 , and repeat the steps leading to eq . ( [ a05 ] ) . doing this we obtain , @xmath402[\bar v , \hat i_d]\rangle\ , . \label{a08}\ ] ] then , we replace @xmath15 and @xmath381 with @xmath384 and @xmath385 . we again keep only terms relevant for cotunneling , and in addition we neglect terms of order @xmath57 ( applying same arguments as before , see eq . ( [ a09 ] ) ) . we then arrive at eq . ( [ db - noise ] ) with the operator @xmath82 given by eq . ( [ db - current ] ) . finally , in order to show that fluctuations of @xmath399 are suppressed , we replace @xmath381 in eq . ( [ a08 ] ) with @xmath403 , and then use the operators @xmath404 and @xmath405 instead of @xmath384 and @xmath385 . in contrast to eq . ( [ a07 ] ) terms such as @xmath406 do not contribute , because they contain integrals of the form @xmath407 . the only nonzero contribution can be written as @xmath408[\bar y^{\dag},\bar y]\rangle\ , , \label{a09}\ ] ] where we have used integration by parts and the property @xmath409 . compared to eq . ( [ db - noise ] ) this expression contains an additional integration over @xmath84 , and thereby it is of order @xmath410 . in this appendix we calculate the transition rates eq . ( [ golden - rule ] ) for a dd coupled to leads with the coupling described by eqs . ( [ dd - d ] ) and ( [ equal ] ) and show that the four - level system in the singlet - triplet basis eq . ( [ basis ] ) can be mapped to a two - level system . for the moment we assume that the indices @xmath188 and @xmath122 enumerate the singlet - triplet basis , @xmath411 . close to the sequential tunneling peak , @xmath412 , we keep only terms of the form @xmath413 . calculating the trace over the leads explicitly , we obtain at @xmath290 , @xmath414 with @xmath415 , and @xmath416 , and we have assumed @xmath417 . since the quantum dots are the same we get @xmath418 , and @xmath419 . we calculate these matrix elements in the singlet - triplet basis explicitly , @xmath420 assuming now equal coupling of the form eq . ( [ equal ] ) we find that for @xmath421 the matrix elements of the singlet - triplet transition vanish ( as we have expected , see sec . [ degenerate ] ) . on the other hand the triplets are degenerate , i.e. @xmath422 in the triplet sector . then from eq . ( [ c1 ] ) it follows that @xmath423 . next , we have @xmath424 , since for nearly degenerate states we assume @xmath425 , and thus @xmath426 . finally , for @xmath427 we obtain , @xmath428 next we prove the mapping to a two - level system . first we notice that because the matrix @xmath429 is symmetric , the detailed balance equation for the stationary state gives @xmath430 , @xmath431 . thus we can set @xmath432 , for @xmath433 . the specific form of the transition matrix eqs . ( [ c5 - 1]-[c5 - 4 ] ) helps us to complete the mapping by setting @xmath434 , @xmath435 , and @xmath436 , so that we get the new transition matrix eq . ( [ dd - rates ] ) , while the stationary master equation for the new two - level density matrix does not change its form . if in addition we set @xmath437 , @xmath438 , and @xmath439 , then the master equation eq . ( [ mastereq ] ) for @xmath440 and the initial condition @xmath441 do not change either . finally , one can see that under this mapping eq . ( [ correlator - delta - s ] ) for the correction to the noise power @xmath173 remains unchanged . thus we have accomplished the mapping of our singlet - triplet system to the two - level system with the new transition matrix given by eq . ( [ dd - rates ] ) . for the super - poissonian noise in resonant double - barrier structures , see : g. iannaccone _ et al . _ , phys . lett . * 80 * , 1054 ( 1998 ) ; v. v. kuznetsov _ et al . _ , phys . b * 58 * , r10159 ( 1998 ) ; ya . m. blanter and m. bttiker , phys . b * 59 * , 10217 ( 1999 ) . one could view this as an analog of a whistle effect , where the flow of air ( current ) is strongly modulated by a bistable state in the whistle , and vice versa . the analogy , however , is not complete , since the current through the qds is random due to quantum fluctuations . note that the aharonov - bohm effect is not suppressed by this inelastic cotunneling , since the _ entire _ cotunneling process involving also leads is elastic : the initial and final states of the _ entire _ system have the same energy .

we use new and archival _ chandra _ and _ rosat _ data to study the time variability of the x - ray emission from the pulsar wind nebula ( pwn ) powered by psr b1509 on timescales of one week to twelve years . there is variability in the size , number , and brightness of compact knots appearing within 20 of the pulsar , with at least one knot showing a possible outflow velocity of ( assuming a distance to the source of 5.2 kpc ) . the transient nature of these knots may indicate that they are produced by turbulence in the flows surrounding the pulsar . a previously identified prominent jet extending 12 pc to the southeast of the pulsar increased in brightness by 30% over 9 years ; apparent outflow of material along this jet is observed with a velocity of . however , outflow alone can not account for the changes in the jet on such short timescales . magnetohydrodynamic sausage or kink instabilities are feasible explanations for the jet variability with timescale of 1.3 years . an arc structure , located 30 north of the pulsar , shows transverse structural variations and appears to have moved inward with a velocity of over three years . the overall structure and brightness of the diffuse pwn exterior to this arc and excluding the jet has remained the same over the twelve year span . the photon indices of the diffuse pwn and possibly the jet steepen with increasing radius , likely indicating synchrotron cooling at x - ray energies .

astro-ph0512047

the pulsar b1509@xmath058 and the supernova remnant ( snr ) [email protected] ( msh 15@xmath05__2 _ _ ) represent one of approximately 20 known associations between a pulsar and a snr . this pulsar is one of the most energetic known , with a period ( @xmath7 ) of 150 ms , a period derivative ( @xmath8 ) of 1.2@xmath9 s s@xmath10 , a characteristic age @xmath11 yr , a spin - down luminosity @xmath12 ergs s@xmath10 ( for a moment of inertia @xmath13 g @xmath14 ) , and an inferred dipole surface magnetic field @xmath15 g @xcite . snr [email protected] has been well studied at radio , optical , and x - ray wavelengths . the radio morphology consists of a partial shell to the southeast and a series of bright clumps @xmath16 to the northwest @xcite that coincide with the optical nebula rcw 89 @xcite . the x - ray morphology consists of a bright , elongated pulsar wind nebula ( pwn ) with a collimated jet extending @xmath17 to the southeast ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * hereafter g02 ) . to the northwest are thermal clumps associated with the radio clumps and rcw 89 @xcite . in addition , identified several compact knots close to the pulsar , plus two semicircular arcs at a distance of 17@xmath1 and 30@xmath1 to the north of the pulsar . the toroidal morphology and collimated jet are reminiscent of structures found in the pwne powered by the crab and vela pulsars @xcite . the arcs have been proposed as being due to ion - compression in the particle - dominated equatorial flow from the pulsar and were interpreted by as analogs of the `` wisps '' found in the crab nebula . snr [email protected] has recently been observed in very high energy @xmath18-rays by hess . the emission is elongated along the pwn axis possibly indicating inverse compton scattering of relativistic electrons @xcite . the structures in the crab and vela pwne are known to vary in brightness and position over short timescales ( days to months ) . in the case of the crab nebula , the outward moving x - ray and optical wisps ( with velocity @xmath19 ) are thought to mark the pwn termination shock , while the small - scale x - ray and optical knots are thought to identify unstable , quasi - stationary shocks in the pulsar wind @xcite . radio wisps , which rarely correspond to optical wisps , develop and move outward at slower velocities ( @xmath20 ) and there are even more slowly moving radio features ( @xmath21 km s@xmath10 ) farther away from the pulsar @xcite . recent observations of the wisps and polar knots in the near infrared indicate brightness variations on time scales as short as 20 minutes @xcite . in the case of the vela pwn , the x - ray arcs also move outward and vary in brightness by up to 30% @xcite . the variability of the vela pwn jet observed in x - rays has been attributed to both kink instabilities , to account for the dramatic shape and brightness changes over the course of days , and sausage instabilities , to account for the relativistically moving `` blobs '' ( @xmath19 ; * ? ? ? * ) . the crab nebula jet , on the other hand , shows only weak x - ray morphological variations on year - long time scales @xcite and relativistic outflow identified in the optical with @xmath22 @xcite . the pwn powered by psr b1509@xmath058 represents a unique opportunity to study variability in pwne . the physical size of the pwn is approximately 10 and 100 times larger than the crab and vela pwne , respectively . given the observed variability in the crab and vela pwne and the distance to psr b1509@xmath058 ( 5.2@xmath23 kpc ; * ? ? ? * ) , predicted measurable variability in the pwn of psr b1509@xmath058 on time scales of a few years . to that end , we obtained new _ chandra x - ray observatory _ observations of psr b1509@xmath058 and its pwn . in this paper we compare our new images to existing _ rosat _ pspc and hri images and _ chandra _ acis - i images and report on the variations observed over time scales of 1 week to 12 years . we observed [email protected] with the _ chandra _ acis - i detector on 2003 apr 21 for a total of 9.6 ks , on 2003 apr 28 for a total of 10 ks , and on 2003 oct 18 for a total of 19.4 ks . the data were calibrated using ciao version 3.1 and caldb version 2.28 . after filtering for good time intervals , the total useable exposure times were 9469 s , 9497 s , and 18739 s for the apr 21 , apr 28 , and oct 18 observations , respectively . exposure corrected images were constructed for each epoch between 0.3 and 8 kev as outlined in . the telescope was not dithered during the 2003 apr 28 observation so that the effects of the chip gaps and dead columns could not be removed at that epoch . [email protected] had been previously observed with the _ chandra _ acis - i detector on 2000 aug 14 for a total of 19.3 ks . the archival data products were obtained from the _ chandra _ archive and were recalibrated using ciao version 3.1 and caldb version 2.28 to make use of the newest gain solutions and correct the geometry of the data set . the data were filtered for good time intervals resulting in a final exposure time of 17863 s and an exposure corrected image was made as outlined above . we compare these _ chandra _ data to archival observations with the _ rntgen satellite _ ( _ rosat _ ; * ? ? ? * ; * ? ? ? [email protected] was observed with the _ rosat _ pspc detector on 1991 feb 22 mar 8 and 1992 feb 25 27 for a total of 9.1 ks @xcite . data were also taken with the _ rosat _ hri detector on 1994 feb 10 23 and 1994 sep 15 for a total of 22.5 ks @xcite . the archival data products were obtained from nasa s heasarc . exposure corrected images were made using the _ xselect _ utility and the _ hriexpmap _ and _ pcexpmap _ tasks within version 5.3.1 of the ftools package . the exposure times of the pspc and hri data are approximately matched to produce the same number of counts on the two images . to facilitate comparison to the _ chandra _ images , exposure corrected images were also made from 2 ks each of the 2000 aug 14 and 2003 oct 18 _ chandra _ data sets to match the energy range ( @xmath24 kev ) and number of counts on the _ rosat _ pspc and hri images . the four _ chandra _ images were registered to each other using a brightness - weighted mean of psr b1509@xmath058 , the star muzzio 10 ( located at ( j2000 ) r.a . @xmath25 , decl . @xmath26 ; * ? ? ? * ) , and four other background sources in the field ( located at ( j2000 ) r.a . @xmath27 , decl . @xmath28 ; r.a . @xmath29 , decl . @xmath30 ; r.a . @xmath31 , decl . @xmath32 ; r.a . @xmath33 , decl . @xmath34 ) . the relative registration error is @xmath35 with rotation constrained to @xmath36 . although the pulsar does not have a significant proper motion ( @xmath37 mas yr@xmath10 , @xmath38 mas yr@xmath10 , * ? ? ? * ) , its suitability for registration purposes is in question because the strong pile - up enhances the azimuthal brightness asymmetry that results from a misalignment of the telescope mirrors in the innermost shell @xcite . therefore , we also determined the registration solution excluding the pulsar . again using a brightness - weighted mean , we find a relative registration error @xmath35 for the 2000 aug 14 and 2003 oct 18 images and @xmath39 for the two 2003 apr images . spectra were extracted for selected sources of interest using the ftools _ xselect _ utility for the _ rosat _ pspc data and the ciao _ acisspec _ script for the _ chandra _ data . background spectra were extracted from an annular region surrounding each source of interest . the spectra were rebinned to have a minimum of 20 counts per spectral channel . xspec version 11.3.1 was used for the spectral analysis . unless otherwise noted , we present the 90% confidence limit for errors . all physical size and distance calculations assume that the source is 5.2 kpc away @xcite . shown in figure [ allim ] is a summed and exposure corrected image of 50 ks of the _ chandra _ data _ excluding _ the 2003 apr 28 observation , which was not dithered . the energy range is [email protected] kev and the image has been convolved with a gaussian of fwhm 10@xmath1 . the pulsar position is indicated by the white cross . in addition to the jet and inner and outer arcs ( g02 features `` c '' , `` 5 '' , and `` e '' , respectively ) , we identify two faint arc - like structures at distances of @xmath4 110@xmath1 and 160@xmath1 from the pulsar . these newly identified features are labelled as `` arc a '' and `` arc b '' and are @xmath420@xmath1 in width . arc a can be identified in the individual 20 ks 2000 aug and 2003 oct images and also appears in more recent _ chandra _ images ( p. slane , private communication ) . the newer _ chandra _ data will require more analysis before we can determine if arc b is also a robust structure . to determine if the total flux of the diffuse pwn varied over time , we performed simultaneous absorbed power law fits to the _ rosat _ pspc and the _ chandra _ data sets , requiring the same absorbing column ( @xmath40 ) and photon index ( @xmath41 ) for all data sets , but allowing the normalization to vary between epochs ( to account for both calibration uncertainties and actual , evolutionary brightness variations ) . we define the _ diffuse _ pwn to be the low brightness emission surrounding the pulsar , interior to the rcw 89 region to the northwest , and extending slightly beyond the jet to the southeast . we specifically exclude the jet and inner and outer arc structures from this definition . the exact region used for the analysis is the same as shown in figure 7 of . the best fit values are @xmath42 @xmath43 and @xmath44 . table [ fluxes ] shows the unabsorbed fluxes obtained from the simultaneous spectral fit for the energy range [email protected] kev . note that the loss of dithering for the 2003 apr 28 observation resulted in lower flux values for the diffuse pwn and jet . the total unabsorbed flux of the diffuse pwn has remained steady at @xmath45 erg @xmath43 s@xmath10 from 1991/1992 to 2003 . lcclcc diffuse pwn & s & r & 1991/1992 & [email protected] & 57@xmath463 + & s & c & 2000 aug 14 & & 58@xmath463 + & s & c & 2003 apr 21 & & 58@xmath463 + & s & c & 2003 apr 28 & & 53@xmath463 + & s & c & 2003 oct 18 & & 58@xmath463 + & i & c & 2000 aug 14 & [email protected] & 58@xmath463 + & i & c & 2000 apr 21 & [email protected] & 59@xmath463 + & i & c & 2003 apr 28 & [email protected] & 53@xmath463 + & i & c & 2003 oct 18 & [email protected] & 58@xmath463 + jet & s & r & 1991/1992 & [email protected] & [email protected] + & s & c & 2000 aug 14 & & [email protected] + & s & c & 2003 apr 21 & & [email protected] + & s & c & 2003 apr 28 & & [email protected] + & s & c & 2003 oct 18 & & [email protected] + & i & c & 2000 aug 14 & [email protected] & [email protected] + & i & c & 2003 apr 21 & [email protected] & [email protected] + & i & c & 2003 apr 28 & [email protected] & [email protected] + & i & c & 2003 oct 18 & [email protected] & [email protected] + outer arc & s & r & 1991/1992 & [email protected] & [email protected] + & s & c & 2000 aug 14 & & [email protected] + & s & c & 2003 apr 21 & & [email protected] + & s & c & 2003 apr 28 & & [email protected] + & s & c & 2003 oct 18 & & [email protected] + the value of the absorbing column does not match that of within the errors . we fitted just the recalibrated 2000 data set studied by and found @xmath47 @xmath43 and @xmath48 , which is consistent with the multi - epoch fit . we also fixed @xmath40 at @xmath49 @xmath43 and @xmath41 at 2.05 ( as found by ) to determine what , if any difference there is in total flux . for the energy range [email protected] kev , we obtained an unabsorbed flux of @xmath50 erg @xmath43 s@xmath10 for both sets of values for the absorbing column and the photon index indicating that these parameters are sufficiently degenerate that the slightly different values do not significantly affect the fitted flux . this flux is within the error of that reported by . we conclude that the mismatch in our absorbing column and photon index values compared to the results is most likely due to updated _ chandra _ calibration solutions . unless otherwise indicated , spectral fits in this paper hold @xmath40 fixed at the simultaneous epoch fit value of 8.6@xmath51 @xmath43 . fluxes are reported for an energy range of [email protected] kev and have been corrected for interstellar absorption . to determine if the pulsar flux has varied over time , we compared the background subtracted count rate from a @xmath52 radius region around the pulsar for each _ chandra _ epoch over the energy ranges [email protected] kev and [email protected] kev . note that these count rates do not reflect the true flux from the pulsar because of pile - up . we chose the two different energy ranges to determine if any variation in count rate could be due to pile - up differences between epochs . a reduction in pile - up might result from the extra absorption at low energies caused by the time - dependent acis filter contamination . the count rates are shown in table [ psrcts ] . in 2000 , the pulsar count rates between [email protected] kev and between [email protected] kev were @xmath48% less than in 2003 , representing less than a 3@xmath53 difference . because the difference is approximately the same for the two energy ranges , it is unlikely that we have a significant bias from a reduction of pile - up at low energies . we also determined the flux by performing a power law fit to each of the _ chandra _ data sets correcting for interstellar absorption and pile - up @xcite . we held @xmath40 fixed at @xmath54 @xmath43 and we held @xmath41 fixed at 1.19 @xcite for each epoch but allowed the normalization and pile - up grade morphing parameter to vary . as shown in table [ psrcts ] , these fits indicate that the flux of the pulsar has remained steady at @xmath55 erg @xmath43 s@xmath10 . lccc 2000 aug 14 & [email protected] & [email protected] & [email protected] + 2003 apr 21 & [email protected] & [email protected] & [email protected] + 2003 apr 28 & [email protected] & [email protected] & [email protected] + 2003 oct 18 & [email protected] & [email protected] & [email protected] + figure [ smallscale ] shows changes in the structure immediately surrounding the pulsar between 2000 aug and 2003 oct . in 2000 , the structure consisted of four compact knots @xmath56 across ( corresponding to a physical size of [email protected] pc ) and located along the jet axis between distances of @xmath57 on both sides of the pulsar . three years later in 2003 apr , this structure has completely changed , with approximately nine small , unresolved knots ( @xmath58 , @xmath59pc ) again located along the jet axis between distances of @xmath57 on both sides of the pulsar . no apparent changes were evident between 2003 apr 21 and 2003 apr 28 . approximately six months later in 2003 oct , the structure has again changed , with six knots ranging in size from @xmath60 to 4@xmath1 and again located along the jet axis within 17@xmath1 of the pulsar . in all cases , the knot activity is greater to the northwest than to the southeast and in no case are knots found beyond the inner arc at a radius of 17@xmath1 . there is the possibility that some of the knot variability we observe could be due to artifacts . it is known that artificial structures may appear near sources with strong pile - up . however , we believe that most of the variability is real because the typical count rates are @xmath4 0.005 - 0.05 counts s@xmath10 ( corresponding to 100 - 1000 counts on the 2000 aug and 2003 oct images ) higher than would be expected for artificial structures . another indication that the small - scale knots are genuine comes from our monte carlo simulations discussed in [ sec : arcs ] . in each of our ten poisson noise simulations , prior to smoothing to 10@xmath1 , we recover all of the knot structures . furthermore , in analyzing their knot 1 , g02 specifically ruled out asymmetries in the wings of the point - spread function and pulsar photons assigned to the wrong location on the sky as possible explanations . finally , the small - scale knots are apparent on unsmoothed images and are not due to the 1@xmath1 fwhm gaussian convolution applied to figure [ smallscale ] . the knot structure changes so drastically that identifying motion is practically impossible . however , if the knot to the southeast of the pulsar in 2003 apr and 2003 oct is indeed the same feature ( identified with arrows on figure [ smallscale ] ) , then the 4@xmath1 motion results in a velocity of @xmath2 ( assuming outflow along the jet with a 30@xmath61 inclination jet inclination to the line - of - sight based on doppler boosting and radio polarization arguments , while @xcite derive an inclination angle @xmath62 based on an interaction between the unseen northwest jet and the rcw 89 region . ] to the line of sight and correcting for relativistic doppler boosting ) . we performed simultaneous absorbed power law fits to the _ rosat _ pspc and _ chandra _ data for the jet in the same manner as for the diffuse pwn in [ sec : dpwn ] . we fixed @xmath40 at @xmath63 @xmath43 and required the same photon index for all the data sets but allowed the normalization to vary between epochs . the photon indices and fluxes are indicated in table [ fluxes ] . as noted by , the photon index of the jet is flatter than the diffuse pwn . from the spectral fits , we note that the jet brightened by @xmath430% between the _ rosat _ pspc observation in 1991/1992 and the _ chandra _ observations , however the flux remained steady from 2000 to 2003 . the flux difference between the 1991/1992 _ rosat _ data and the 2000 _ chandra _ data is slightly greater than the 90% confidence limit . this flux difference might be caused by the differing spectral responses of _ rosat _ and _ chandra _ , however this is presumably accounted for by the corresponding auxiliary response files . we can also exclude a systematic uncertainty in the calibration of the two observatories as contributing to the flux difference by examining the flux of the diffuse pwn . as seen in table [ fluxes ] , the flux from the extended emission remains constant as expected . we also compared the 1991/1992 _ rosat _ pspc image , 1994 _ rosat _ hri image and 2000 aug and 2003 oct _ chandra _ images of the jet as shown in figure [ rcjet ] . to facilitate comparison between different observatories that might arise from the different responses of _ rosat _ and _ chandra _ , the _ chandra _ images are produced from events restricted to the 0.1 - 2.5 kev energy band of _ rosat_. the resultant _ chandra _ images , as well as the _ rosat _ hri image , are then convolved with a gaussian of fwhm 25@xmath1 resolution to match the angular resolution of the _ rosat _ pspc . finally , to ensure each data set has approximately the same number of total counts as the _ rosat _ pspc and hri observations , only a fraction of the _ chandra _ observations were used to generate the images as discussed in [ sec : xobs ] . the images clearly show variability in the jet structure and brightness over a twelve - year period , supporting the flux measurements from the spectral fits . indeed , the jet is dim and poorly defined in 1991/1992 . in 1994 , however , part of the jet structure @xmath64 ( corresponding to a distance of 9 pc , correcting for a jet inclination of 30@xmath61 to the line of sight ) to the southeast of the pulsar has brightened over the two years since the previous observation . by 2000 , the jet is well - defined with structure extending 4@xmath65 from the pulsar and it has a curved appearance . the 2003 _ chandra _ observation shows that part of the jet @xmath66 to the southeast of the pulsar has dimmed and the structure at the end of the jet has become less well - defined . the persistence of the inner- and outer - arc structures during the entire time range , taken with variability of the jet structure and relative intensity , indicates that the observed changes are in fact significant and real since it is difficult to imagine an artifact that would only apply to discrete parts of the images . hence , the jet brightening indicates the existence of an underlying mechanism , as discussed in [ sec : disjet ] . despite remaining steady in total flux between 2000 and 2003 , the jet shows variability on @xmath67 size scales , as shown in figure [ chjet ] . in 2000 , we identify four large ( @xmath68 , physical size of 0.5 pc ) clumps in the jet ( labeled as 1@xmath04 on the top panel of figure [ chjet ] ) located between @xmath66 ( a separation from the pulsar of 3 pc at a 30@xmath61 inclination ) and @xmath69 away from the pulsar . in 2003 , there are two @xmath70-sized clumps located @xmath71 and 1@xmath72 away from the pulsar with fainter emission further down the jet . if we interpret the clumps as the same structure having moved along the jet , then the velocity is @xmath3 ( assuming a 30@xmath61 inclination to the line of sight and correcting for relativistic doppler boosting ) . due to the loss of dithering for the 2003 apr 28 observation , we can not determine if the jet structure changed on week - long time scales . however , the 10 ks of data in the 2003 apr 21 observation shows jet clumps with similar sizes and locations as in the 2003 oct observation indicating that the jet structure changes on timescales longer than 6 months . _ rosat _ images indicated the presence of a cross - like structure around the pulsar @xcite . with the improved resolution of _ chandra _ , the `` cross '' was resolved into inner and outer arcs ( features `` 5 '' and `` e '' of ) located approximately 17@xmath1 and 40@xmath1 ( distances of 0.4pc and 1.2 pc at a 60@xmath61 inclination perpendicular to the jet ) away from the pulsar . to determine if the arcs have changed structure , we constructed _ chandra _ and _ rosat _ hri images matched in energy range , counts , and resolution to the _ rosat _ pspc images as described in [ sec : xobs ] and [ sec : jet ] . due to the @xmath73 resolution of the _ rosat _ pspc detector , emission from the pulsar and small - scale knots surrounding the pulsar extends into the inner and outer arc structures . therefore , we could not exclude the pulsar before smoothing without adversely affecting the arc structures . figure [ rcjet ] shows the comparison between the _ rosat _ and _ chandra _ images . there appears to be structural changes to both the inner and outer arcs between 1991/1992 , 1994 , 2000 , and 2003 . in the same manner as for the diffuse pwn in [ sec : dpwn ] , we performed simultaneous absorbed power law spectral fits to the outer arc using the _ rosat _ pspc and _ chandra _ data holding @xmath40 fixed at @xmath63 @xmath43 , requiring @xmath41 to be the same for all the data sets , and allowing the normalization to vary between epochs . for the _ chandra _ data , we reproduce the flux and photon index reported by within the errors . as noted by , the outer arc has the same photon index as the jet and the photon index is flatter than for the diffuse pwn . the outer arc appears to have _ decreased _ in brightness between 1991/1992 and 2000 by @xmath420% as shown in table [ fluxes ] , however this is only a 1@xmath53 result ( 68% confidence limit ) . this slight flux decrease may just be an artifact of the @xmath73 resolution of the _ rosat _ pspc detector and the pile - up of the pulsar in the _ chandra _ data ( the low resolution of _ rosat _ results in pulsar counts contaminating the outer arc region and the pile - up of the pulsar with _ results in less contamination from the pulsar at the same resolution ) . if we just consider the _ data , the outer arc shows time variability between 2000 aug and 2003 oct as shown in figure [ charc ] . in 2000 aug , the outer arc contained two major clumps the smaller clump having a 10@xmath1 diameter , and the larger clump elongated along the arc with a size of @xmath74 ( 0.25 pc @xmath75 0.5 pc ) . in 2003 oct , the clump locations had changed within the outer arc , although the clump sizes were comparable to the 2000 clumps . similar clump variability is also seen in the outer arc between 2003 apr 21 and 2003 apr 28 as shown in figure [ charc2 ] indicating that the timescale for arc variability might be as short as a few days . the white arrows on figures [ charc ] and [ charc2 ] identify locations of transverse structural change between 2000 and 2003 . note that the changes observed in the outer arc between the 2003 apr images are only slightly greater than one would expect from poisson statistics . the greatest variation between the two 2003 apr images , indicated by the rightmost arrow on figure [ charc2 ] , is only 3@xmath53 on unsmoothed images . spectral fits indicate that the total brightness of the outer arc has remained the same from 2000 to 2003 . the inner arc also shows a brightening to the west of the pulsar between 2000 and 2003 , as seen in figures [ chjet ] and [ charc ] , however this is most likely due to the bright , pointlike source near ( j2000 ) r.a . @xmath76 , decl . @xmath77 in the 2003 oct image seen most clearly in the bottom panel of figure [ smallscale ] . a simbad search shows no variable source in this location . while it is possible that this source could be related to the small - scale knots identified in [ sec : knots ] , we believe that it is more likely an artifact of a `` hot pixel '' in the data . the southwestern tip of the inner arc does appear to vary slightly between 2003 apr 21 and 2003 apr 28 ( see figure [ charc2 ] ) , however , this may be due in part to the lower signal - to - noise ratio on the 10 ks exposures . the inner and outer arcs were predicted to have outward motions of a few arcseconds per year . to determine the proper motion of the outer arc , we convolved the 2000 aug and 2003 oct _ chandra _ images with a gaussian of fwhm 10@xmath1 . we then constructed an angle - averaged radial profile over the entire length of the outer arc and centered on the pulsar for each epoch , as shown in figure [ arcmov ] . we measured the motion of the brightness profiles by minimizing @xmath78 of the difference between the profiles at each epoch as a function of radial shift and amplitude scaling factor . we used data between radii of 38@xmath1 and 55@xmath1 ( corresponding to the peak and outside edge of the brightness profiles ) to measure the motion . it was necessary to use the peaks of the brightness profiles to break the degeneracy between motion and scaling differences between epochs . data at radii less than 38@xmath1 were excluded for the motion measurement because of the brightness variations caused by the small - scale knots near the inner arc . although the small - scale knots are unresolved at 10@xmath1 resolution , they do contribute significantly to the flux of the inner arc region . the error was determined at the 68% confidence limit of the @xmath78 distribution . we confirmed the error by repeating the @xmath78 measurement for ten monte carlo simulations of the poisson noise and calculating the rms scatter of the minimum @xmath78 position . the outer arc appears to have moved _ inward _ by @xmath79 over the 3-year timespan from 2000 to 2003 . this corresponds to a velocity of @xmath80 ( assuming outflow perpendicular to the jet with a 60@xmath61 inclination to the line of sight and correcting for relativistic doppler boosting ) . the convolution of the images and the angle - averaging of the outer arc results in a measurement of the motion of the ensemble average of the material comprising the arc which allows us to make a comparison to the model of . thus , we can report that the average arc structure appears to move inward . however , given the degree of structural change in this timespan , we can not definitively determine if this represents a real inward motion or is just a result of clump variation . we can rule out unaliased outward motion of the outer arc at the 5@xmath53 level . for the inner arc , there was too much structural change due to the nearby small - scale , compact knots to determine proper motions for that feature . although the jet structure has changed over the 3-year timespan of the _ chandra _ observations , individual spectral fits to the 2000 aug and 2003 oct data indicate that there was no photon index evolution in the jet and diffuse pwn during that time as shown in table [ fluxes ] . to determine if the photon index of the diffuse pwn and jet varied spatially , we used just the four _ chandra _ data sets and performed simultaneous absorbed power law fits as outlined in [ sec : dpwn ] . for each region described below , we held @xmath40 fixed to 8.6@xmath51 @xmath43 , required the same @xmath41 for all the data sets , but allowed the normalization to vary between epochs . for the diffuse pwn , we extracted spectra from concentric 36@xmath1-wide annuli extending from a radius of 1@xmath813 to 4@xmath813 , centered on the pulsar as outlined in table [ radtable ] . the background region used for the diffuse pwn was a 36@xmath1 wide annulus just exterior to the outermost `` source '' annulus and interior to the rcw 89 region . the background annulus was exterior to most of the pwn except to the northwest and to the southeast where it includes faint pwn emission . the jet was excluded from the diffuse pwn analysis . we performed a linear least - squares fit to determine the rate at which the photon index changed with radius . the results are shown graphically in figure [ radfig ] . the diffuse pwn shows a steepening of [email protected] arcmin@xmath10 between @xmath82 and @xmath83 ( for a total @xmath84 of @xmath85 ) but flattens significantly thereafter . if we allow @xmath40 to also vary for the diffuse pwn fits , the photon index does not flatten at large radii and the best fit slope is [email protected] arcmin@xmath10 which translates to a total change in photon index between @xmath82 and @xmath86 of 0.25 . @xmath40 increases from 9.2@xmath51 @xmath43 in the annulus closest to the pulsar to 13.2@xmath51 @xmath43 in the annulus furthest from the pulsar . these values for the absorbing column are consistent with those derived by @xcite showing that there was about 1.5 times more absorption towards rcw 89 than towards the pulsar . lccc jet & 1.0 & 8.6 ( fixed ) & [email protected] + & 1.5 & & [email protected] + & 2.0 & & [email protected] + & 2.7 & & [email protected] + & 3.5 & & [email protected] + diffuse pwn & 1.3 & 8.6 ( fixed ) & [email protected] + & 1.9 & & [email protected] + & 2.5 & & [email protected] + & 3.1 & & [email protected] + & 3.7 & & [email protected] + & 4.3 & & [email protected] + diffuse pwn & 1.3 & [email protected] & [email protected] + & 1.9 & [email protected] & [email protected] + & 2.5 & [email protected] & [email protected] + & 3.1 & [email protected] & [email protected] + & 3.7 & [email protected] & [email protected] + & 4.3 & [email protected] & [email protected] + for the jet , we constructed five regions along the length of the jet as shown in figure [ allim ] . the separation from the pulsar was defined at the center of each region . the background for the jet analysis was chosen as the region immediately exterior to the total jet region . a linear least - squares fit was applied to the data and is shown in figure [ radfig ] . the jet shows a marginal photon index steepening of [email protected] arcmin@xmath10 between radii of @xmath87 to @xmath88 for a total change of 0.43 . suggested that the inner and outer arcs represent compressions downstream of the termination shock in an equatorial electron - positron pair outflow . the compressions are induced as the high energy heavy ions ( `` protons '' ) embedded in the flow enter the shock heated downstream pairs with the same velocity and lorentz factor as the upstream pair wind . this was an application of the model advanced by @xcite for the time variable wisps seen in the termination shock region of the crab nebula , which has been shown by ( * ? ? ? * hereafter sa04 ) to give a good representation of the structure and toroidally averaged variations seen in the inner x - ray ring as well as the propagating wave structures emerging from that ring seen by _ chandra _ @xcite . in the application of that model to [email protected] , the inner and outer arcs were interpreted as the turning points in the ions orbits as they do their first gyrations in the shock compressed magnetic field . fitting the model to the observations of [email protected] led to an estimate of the pair flux , upstream flow lorentz factor , ion flux and magnetic field in the equatorial flow . from that , derived an ion cyclotron time on the order of one to two years , which is the basic variability time of the ion induced compressions . if all of the variation is in the form of bulk motion , the predicted outflow velocity from is @xmath89 . although our observations do not show an outward velocity , we do confirm the theoretical prediction of the variability time scale expected in the inner and outer arcs . that model , being a toroidal average , did not predict the angular structure of the arcs uncovered by the new data presented here . this model depends on the existence of ultra high energy heavy ions being present in the equatorial flow , with the ions carrying a large enough fraction of the energy flux to be able to induce substantial compression in the radiating pair plasma . the inferred ion number flux , both in the crab and in [email protected] , is approximately what one expects on electrodynamic grounds , if the ion stream is the electric return current required to prevent the charging up of the neutron stars . it is not currently understood how the particles in the flow achieve their high energies ( @xmath90 ) , although it has been suggested that they are accelerated and heated by the mysterious dissipation processes that lead to the equatorial wind having magnetization as weak as has been inferred . these kinetic models focus on the time dependent morphology in and around the termination shock in the pairs in effect , the ions form part of a resolved shock structure whose basic form is an equatorial ring . the large scale structure of the nebular flow has been addressed in a series of mhd models and simulations , whose focus is on understanding the dramatic pole to equator asymmetry , especially the appearance of the jets @xcite . these models assume all the particles have small larmor radius , and thus are indifferent to composition and exclude the proposed high energy ions by assumption . with the premise that the injected energy flux decreases with increasing rotational latitude @xmath91 ( usually assumed proportional to @xmath92 ) , the energy is mostly injected through a toroidal belt shock , with some flow passing obliquely through a higher latitude `` arch shock '' which , because the shock is oblique to the flow , does not decelerate the plasma to subsonic speeds . these non - spherically symmetric shock structures can create the appearance of an inner ring of emission , with the weakly decelerated plasma emerging from the arch shocks contributing to form the appearance of an outer ring . doppler boosting may yield an appearance of partial arcs rather than complete rings . these models provide a plausible scenario for the formation of the jets , as the consequence of backflow and magnetic hoop stress at higher latitudes . the simulations of this model show no signs of short time variability corresponding to the wisps years ) with some possible shorter time variability when the magnetization is finite but not large the variability vanishes for unmagnetized flow models . the physical reasons for this variability are not understood ( bogovalov , private communication ) . ] . the simulations may lack sufficient resolution to find the kelvin - helmholtz instability between the outward flow emerging from the arch shock and the higher latitude backflow , an effect suggested by @xcite with a somewhat different flow geometry in mind as the origin of the wisps . alternatively , magnetization high enough to form the jet may suppress the shear flow instabilities . while the effects of high energy equatorial ions have not been explicitly included in the mhd simulations , if present they would be injected into the shock decelerated pairs through the equatorial ring shock seen in the mhd simulations . thus the formation of ion driven compressions is qualitatively consistent with the mhd models , and remains the only quantitatively elaborated model for the time variability seen in the crab wisps and perhaps seen in the data described here . based on this model , predicted arc motions of a few arcseconds per year . we see no outward motion of the outer arc and perhaps even an inward motion . noted that the positions of the ion driven compressions can move inward as well as outwards , with the apparent motion then depending on when one takes a snapshot of the structure the possible existence of apparent inward motion , perhaps seen in our data , is a prediction of the model . however , the apparent motion may also be due to the transverse structural variations in the outer arc between 2000 and 2003 non - axisymmetric variability , a 3d effect , has not been included in any of the models published to date . if such transverse change is the origin of the apparent motion , the arcs would then be quasi - stationary in the radial sense . or , we could be witnessing alaising as seen by @xcite in his observations of the wisps in the crab , due to the observations having undersampled the variations . we can test whether the quasi - stationary nature of the outer arc is due to aliasing . if we assume that the particles comprising the outer arc are relativistic , and thus doppler boosted , then we can determine a speed based on the brightness ratio of @xmath93 between the near and far sides . for a photon index of 1.6 , the doppler boosting formula reduces to @xmath94 , where @xmath95 and @xmath96 is the inclination angle to the line - of - sight . if the inclination of the outer arc is 60@xmath61 ( oriented 90@xmath61 from the jet ) , then @xmath97 . at this speed , we expect the outer arc to move nearly 16@xmath1 in 3 years , which is slightly greater than the width of the arc . over 6 months , the outer arc should move 2@xmath98 , which is not observed . suppose that the motion of particles _ through _ the arc is not indicative of the motion of the arc . in this case consider that the arc would need to move at 0.27c to move its width ( 0.25 pc ) in 3 years . at this speed , the arc would only move 1@xmath996 in 6 months . this amount of motion is difficult to rule in or out given the degree of structural change in the outer arc and the signal - to - noise ratio of the data . the nature of arc a and arc b is unknown . if indeed the arcs are outward - moving structures , then perhaps arc a and arc b are just more evolved versions of the inner and outer arcs . they could also represent a quasi - coherent oscillation of the flow in the outer pwn as observed in the simulations of @xcite . a more thorough analysis of these faint pwn structures will be performed at a later time . cccc 1 & 15 & 2.1 & 1.3 + 2 & 17 & 2.8 & 1.3 + 3 & 14 & 1.7 & 2.0 + 4 & 12 & 1.3 & 2.0 + 5 & 11 & 1.1 & 1.6 + mhd sausage or kink instabilities are viable candidates for the jet variability , as in the vela jet @xcite . the time scale for such phenomena is set by the alfvn crossing time @xmath100 , where @xmath101 is the cylindrical radius of the jet and @xmath102 here @xmath103 is the relativistic pressure , assumed to be isotropic , and @xmath104 is the energy density of the relativistic plasma . shown in table [ alfven ] are the magnetic fields , plasma energy densities , and alfvn crossing times for regions along the jet as shown in figure [ allim ] . position 1 is closest to the pulsar and we are assuming a cylindrical volume . the magnetic fields and energy densities are computed from equipartition arguments using integrated fluxes for the energy range 0.5 - 10 kev and assuming an uncooled spectrum @xcite . for all regions @xmath105 and the alfvn crossing times vary from 1.3@xmath02 years . thus the mhd instability of a magnetic pinch , the basic structure of the jet described in the mhd models , certainly is a viable candidate for the jet variations observed between 2000 and 2003 , and perhaps may be a candidate to explain the partial brightening of the jet from 1991/1992 to 1994 and 2000 . indeed , episodic jet outflows are observed in mhd simulations for certain magnetic field configurations @xcite . if we interpret the jet structural changes between 2000 and 2003 as clump motion along the jet , then the resulting velocity is @xmath4 0.5c . this velocity is consistent with the prediction of based on doppler boosting and also matches the outflow speeds observed in the crab nebula jet ( 0.4c , * ? ? ? * ) , the vela pwn jet ( @xmath106 , * ? ? ? * ) , and in the jet of [email protected] ( @xmath107 , * ? ? ? relativistic jet outflow velocities of @xmath3 are predicted by the mhd simulations described above @xcite . the jet brightnening from 1991/1992 to 1994 and 2000 can not simply be a result of material moving at @xmath108 down the jet causing the jet to lengthen and brighten with the inclusion of more material . this scenario is excluded for two reasons , the first is that it would take nearly 80 years for material moving at @xmath108 to travel the length of the jet , and the second is that the jet seems to brighten in a `` patchy '' manner and does not simply lengthen , as shown in figure [ rcjet ] . therefore , as discussed above , the brightening of the jet is more likely the result of an mhd instability such as a magnetic pinch . in this model , the instability causes compression and brightening of local regions , each of which has length no more than the jet diameter . there is no necessity for the energy to be injected into the jet at a varying rate , with the observed variations reflecting transport along the jet . this scenario works because the alfvn transit time _ across _ the jet is much shorter than the flow time along the jet . the model suggests that the clumps vary incoherently with respect to each other , on a time scale approximately equal to the alfvn crossing time ( @xmath109 years ) , consistent with the very limited time series available to us . the compact , small - scale knots near the pulsar are quite variable on 6 month timescales . although the knot to the southeast of the pulsar might represent material moving at 0.6c , given the startling knot variability to the northwest , a conclusion of knot motion is perhaps premature . we can estimate the magnetic fields in the knots in three ways : 1 . assume that the knot size is approximately the larmor radius of gyrating pairs , 2 . use equipartition / minimum energy , and 3 . use synchrotron lifetime if the knots are in motion at 0.6c and cool below detectability at a distance of the inner arc in time frames from 6 months to 1 year . for the first method , derived a magnetic field of 3@xmath110 g from the larmor radius for their knot 1 . for the second method , the magnetic field derived from minimum energy using a size of 0.1 pc and the integrated flux of knot 1 from for the energy range of [email protected] kev is @xmath111 g . for the third method , the synchrotron lifetime can be calculated using ( from equation 4 of ) @xmath112 where @xmath113 is the magnetic field in @xmath110 g and @xmath114 is the energy in kev . for 5 kev and @xmath115 year , @xmath116 g . given this high magnetic field , it seems unlikely that synchrotron cooling could be responsible for the transient nature of the knots . the knots may have a similar explanation as `` knot 2 '' in the crab nebula which has been interpreted as an unstable quasi - stationary shock in the polar jet outflow @xcite . in this scenario , one would expect the bulk of the knots to be in the approaching jet since the receding jet is doppler boosted down in brightness . however , in this case we observe most of the knot structure on the receding jet side . given that the knots are never observed beyond the inner arc radius and most of them occur on the approaching side of the arcs , it is likely that the knots are either associated with the equatorial outflow or the resulting backflow as described in [ sec : disarcs ] and shown in figure 3 of @xcite . two processes expected from the relativistic mhd simulations that could produce `` knot - like '' emission are kelvin - helmholtz ( kh ) instabilities arising in the shear layer between the equatorial outflow and the backflow and unsteady vortices that develop in the converging flow at the base of the jet . the kh instability creates waves that may grow and break , while traveling in the direction of the mass weighted jump in velocity across the interface between the outflow and backflow . doppler boosting as the breaking waves head toward us , and plasma and b field compression as the waves break , may lead to the formation of bright features that can appear to us as localized bright spots of emission . the main difficulty with this idea is that within the flow models , the mass weighted velocity jump is directed toward us only on the near ( visible jet ) side except very close in to the base of the jets . this is because flow convergence increases the mass density in the layer that converges on the axis at small distance from the axis , while the outflowing layer has low density at small radii . at larger radii , the relative densities of the layers reverses . thus this effect will work only if there is a lot of flow convergence . the converging flow at the base of the jet could also produce transient features due to the development of unstable vortices ( n. bucciantini , private communication ) . the backflow would need to focus quite well toward the base of the jet to produce doppler boosting of flows directed towards us and may be rather sensitive to the maintenance of axisymmetry . however , this hypothesis naturally explains the relative absence of knots on the near ( visible ) jet side since the converging flow at the base of the approaching jet would be directed away from us . regardless of the actual cause of the knots , they seem to be diagnostic of flow details in the pwn . to determine the true nature of the knots will require much higher resolution simulations and more analysis of the flow models . determined that the synchrotron cooling time for the jet at 5 kev and a magnetic field of @xmath117 g is 140 years . this cooling should result in a @xmath84 of 0.5 @xcite . indeed , there is a marginal indication of cooling in the jet with a total @xmath84 of 0.43 . we await analysis of the new , longer _ observation to verify the possible synchrotron cooling in the jet . the relativistic mhd simulations do predict that the jet should undergo synchrotron cooling , although there is mixing with the ambient pwn material ( s. komissarov , private communication ) . at the base of the jet , the photon index should match that of the equatorial outflow since the jet is constructed from purloined equatorial material . if we assume that the inner and outer arcs identify the equatorial outflow material , then their photon index of 1.6 does nearly match the photon index of 1.5 at the base of the jet . if the spectral steepening along the jet is real , it would be similar to that observed in the crab nebula jet ( @xmath118 ; * ? ? ? spectral steepening with increasing radius has been observed in a number of pwne and is generally attributed to both synchrotron cooling and expansion losses @xcite . the degree of steepening varies from @xmath119 to 1.3 with a typical value of 1 and the radial photon index profiles generally show a monotonic increase with radius . our spectral results for the diffuse component of the pwn of [email protected] also show a monotonic increase of photon index with radius if we allow the absorption to vary , but the degree of steepening ( @xmath120 ) is less than in other pwne . in contrast , models for synchrotron cooling in pwne based on the @xcite model predict a rapid increase in photon index at large radius @xcite . perhaps the discrepancy between the cooling model and the observations is due to large - scale mixing of recently accelerated material with `` older '' material as predicted by the relativistic mhd models @xcite . variability is observed in the x - ray pwn of psr b1509@xmath058 on time scales possibly as short as one week and up to twelve years . our primary results are as follows : 1 . the compact , small - scale knots appearing within 20@xmath1 of the pulsar exhibit transient behavior which may be attributed to turbulence in the flows surrounding the pulsar . possible knot motion is indicated with a velocity of @xmath121 . apparent outflow along the jet is observed with velocities of @xmath3 . this outflow alone can not account for the @xmath430% brightening of the jet between 1991/1992 and 2000 . the alfvn crossing time for the jet is 1.3@xmath02 years , therefore , mhd kink or sausage instabilities can account for the rapid morphological variations and perhaps the partial jet brightening . the outer arc has possibly moved inward with a velocity of @xmath80 , however the transverse structural changes seen in the outer arc may account for the apparent motion . we can not determine at this time if the outer arc is truly quasi - stationary or if we are witnessing aliasing . the diffuse pwn has not evolved significantly in structure or brightness over the 12-year time span . using the summed _ images , we identify two possible arc structures exterior to the outer arc . the photon indices of the diffuse pwn and possibly the jet steepen with increasing radius indicating synchrotron cooling at x - ray energies . although our imaging capabilities have improved substantially since the first optical observations of time variability in the crab nebula @xcite , our understanding of these variations in pwne is still quite limited . for instance , while we expect magnetic fields to play an important dynamical role in jets , and indeed we do see variations on the appropriate alfvn crossing times , we do not know for certain if mhd instabilities are the root cause of the observed variations . the arc structures we observe in [email protected] are equally enigmatic . we do not yet know if they are in steady motion or are quasi - stationary wave phenomena . the striking changes in the small - scale knots near the pulsar may simply be `` weather , '' diagnosing unimportant details in the pwn flow , or they may indicate important flow structure which is essential to understanding , for instance , diffusion of particles from the equatorial flow to higher latitudes , a loss essential to a post pair shock second - order fermi acceleration model . certainly deeper , and appropriately spaced , x - ray observations will help resolve some issues such as the possible spatial aliasing of the outer arc . also , the higher signal - to - noise will provide better constraints on the spatial spectral index variations and allow us to determine if and how much mixing has occurred in the diffuse pwn . finally , we are excited by the recent development of relativistic mhd models and we hope that some of the variability we observe here can eventually be observed in those simulations . we thank fred seward for assistance with _ rosat _ data analysis , elena amato , niccol bucciantini , and serguei komissarov for helpful discussions , and the anonymous referee for a careful reading of the manuscript . this research has made use of data obtained from the high energy astrophysics science archive research center ( heasarc ) , provided by nasa s goddard space flight center . t. d. & b. m. g. acknowledge the support of nasa through sao grant go3 - 4063a . j.a . acknowledges support from nasa chandra theory grant tm4 - 5005x , nasa atp grant nag5 - 12031 , and from the taxpayers of california . aharonian , f. et al . 2005 , , 435 , l17 bauciska - church , m. , & mccammon , d. 1992 , , 400 , 699 begelman , m. c. 1998 , , 493 , 291 begelman , m. c. 1999 , , 512 , 755 bietenholz , m. f. , hester , j. j. , frail , d. a. , & bartel , n. 2004 , , 615 , 794 bogovalov , s. v. , chechetkin , v. m. , koldoba , a. v. , & ustyugova , g. g. 2005 , , 358 , 705 bogovalov , s. v. & khangoulian , d. v. 2002 , , 336 , l53 brazier , k. & becker , w. 1997 , , 284 , 335 cusumano , g. , mineo , t. , massaro , e. , nicastro , l. , trussoni , e. , massaglia , s. , hermsen , w. , & kuiper , l. 2001 , , 375 , 397 davis , j. e. 2001 , , 562 , 575 del zanna , l. , amato , e. , & bucciantini , n. 2004 , , 421 , 1063 gaensler , b. m. , arons , j. , kaspi , v. m. , pivovaroff , m. j. , kawai , n. , & tamura , k. 2002 , , 569 , 878 ( g02 ) gaensler , b. m. , brazier , k. t. s. , manchester , r. n. , johnston , s. , & green , a. j. 1999 , , 305 , 724 gallant , y. a. & arons , j. 1994 , , 435 , 230 gonzalez , m. & safi - harb , s. 2003 , , 583 , l91 greiveldinger , c. , caucino , s. , massaglia , s. , gelman , h. , & trussoni , e. 1995 , , 454 , 855 helfand , d. j. , gotthelf , e. v. , & halpern , j. p. 2001 , , 556 , 380 hester , j. j. et al . 2002 , , 577 , l49 hester , j. j. et al . 1995 , , 448 , 240 jerius , d. , donnelly , r. h. , tibbetts , m. s. , edgar , r. j. , gaetz , t. j. , schwartz , d. a. , van speybroeck , l. p. , zhao , p. 2000 , proc spie , 4012 , 17 kaspi , v. , manchester , r. , siegman , b. , johnston , s. , & lyne , a. 1994 , , 422 , l83 kardashev , n. s. 1962 , soviet astron . , 6 , 317 kennel , c. f. & coroniti , f. v. 1984 , , 283 , 694 komissarov , s. s. & lyubarsky , y. e. 2003 , , 344 , l93 komissarov , s. s. & lyubarsky , y. e. 2004 , , 349 , 779 livingstone , m. a. , kaspi , v. m. , gavriil , f. p. , & manchester , r. n. 2005 , , 619 , 1046 lou , y. 1998 , , 294 , 443 lu , f. j. , wang , q. d. , aschenbach , b. , durouchoux , p. , & song , l. m. 2002 , , 568 , l49 lyubarsky , y. e. 2002 , , 329 , l34 melatos , a. , scheltus , d. , whiting , m. t. , eikenberry , s. s. , romani , r. w. , rigaut , f. , spitkovsky , a. , arons , j. , & payne , d. j. b. 2005 , , in press ( astro - ph/0509914 ) mori , k. , burrows , d. n. , hester , j. j. , pavlov , g. g. , shibata , s. , & tsunemi , h. 2004 , , 609 , 186 mori , k. , burrows , d. n. , pavlov , g. g. , hester , j. j. , shibata , shinpei , & tsunemi , h. 2004 , young neutron stars and their environments , iau symposiom vol . 218 muzzio , j. c. 1979 , , 84 , 639 ouyed , r. & pudritz , r. 1997 , , 484 , 794 ouyed , r. , pudritz , r. , & stone , m. 1997 , nature , 385 , 409 pacholczyk , a. g. 1970 , radio astrophysics ( san francisco : freeman ) pacini , f. & salvati , m. 1973 , apj , 186 , 249 pavlov , g. g. , kargaltsev , o. y. , sanwal , d. , & garmire , g. p. 2001 , , 554 , l189 pavlov , g. g. , teter , m. a. , kargaltsev , o. , & sanwal , d. 2003 , , 591 , 1157 pfeffermann , e. , et al . 1987 , proc spie , 733 , 519 reynolds , s. p. 2003 , in proc . iau colloquium 192 , supernovae : 10 years of sn 1993j , ed . j. m. marcaide & k. w. weiler ( berlin : springer ) , posters supplement , 161 roberts , m. s. et al . 2003 , , 588 , 992 rodgers , a. w. , campbell , c. t. , & whiteoak , j. b. 1960 , , 121 , 103 safi - harb , s. , harrus , i. m. , petre , r. , pavlov , g. g. , koptsevich , a. b. , & sanwal , d. 2001 , , 561 , 308 scargle , j. d. 1969 , , 156 , 401 seward , f. d. , harnden jr . , f. r. , murdin , p. , & clark , d. h. 1983 , , 267 , 698 slane , p. , helfand , d. j. , van der swaluw , e. , & murray , s. s. 2004 , , 616 , 403 spitkovsky , a. & arons , j. 2000 , in iau colloq . 177 , pulsar astronomy : 2000 and beyond , eds . m. kramer , n. wex , & r. wielebinski ( san francisco : asp ) , 507 spitkovsky , a. & arons , j. 2004 , , 603 , 669 ( sa04 ) tanvir , n. r. , thomson , r. c. , & tsikarishvili , e. g. 1997 , newa , 1 , 311 trmper , j. 1982 , adv . space res . , 2 , 241 trussoni , e. , massaglia , s. , caucino , s. , brinkman , w. , & aschenbach , b. 1996 , , 306 , 581 weisskopf m. c. et al . 2000 , , 536 , l81 yatsu , y. , kataoka , j. , kawai , n. , kotani , t. , tamura , k. , & brinkmann , w. 2005 , astro - ph/0507049

the properties of string networks at scales well below the horizon are poorly understood , but they enter critically into many observables . we argue that in some regimes , stretching will be the only relevant process governing the evolution . in this case , the string two - point function is determined up to normalization : the fractal dimension approaches one at short distance , but the rate of approach is characterized by an exponent that plays an essential role in network properties . the smoothness at short distance implies , for example , that cosmic string lensing images are little distorted . we then add in loop production as a perturbation and find that it diverges at small scales . this need not invalidate the stretching model , since the loop production occurs in localized regions , but it implies a complicated fragmentation process . our ability to model this process is limited , but we argue that loop production peaks a few orders of magnitude below the horizon scale , without the inclusion of gravitational radiation . we find agreement with some features of simulations , and interesting discrepancies that must be resolved by future work . * analytic study of small scale structure on cosmic strings * joseph polchinski _ kavli institute for theoretical physics _ _ university of california _ _ santa barbara , ca 93106 - 4030 _ [email protected] jorge v. rocha _ department of physics _ _ university of california _ santa barbara , ca 93106 _ [email protected] = 16pt

hep-ph0606205

the evolution of cosmic string networks is a challenging problem . the need to consider large ratios of length and time scales makes a complete numerical analysis impossible , while the nonlinearity of the process defeats a purely analytic treatment . thus a full understanding of most of the signatures of cosmic strings will require a careful combination of analytic and numerical approaches . this is needed both to establish precisely the current bounds on the dimensionless cosmic string tension @xmath0 , and to anticipate what will be the most sensitive future measurements . also , if cosmic strings are one day discovered , a precise understanding of the network properties will be needed in order to distinguish different microscopic models . on scales close to the horizon size , the networks are reasonably well understood from simulations @xcite . in particular , there are a few dozen long strings crossing any horizon volume . on shorter scales , however , the situation is far less clear . there is a nontrivial short distance structure on the strings , which arises because the intercommutation process produces kinks @xcite . there have been many previous analytic and numerical studies of this , but there is no general consensus as to its nature . in particular , the size at which typical loops form is uncertain to tens of orders of magnitude . one widely - held assumption has been that gravitational radiation is necessary in order for cosmic string networks to scale @xcite , and that it determines the size of loops . if so , the loops will be parametrically smaller than the horizon scale , by a power of @xmath0 @xcite ( even shorter scales have been suggested @xcite ) . more recent work appears to be converging on loop sizes at a fixed fraction of the horizon scale , but even here there are significant differences . ref . @xcite suggests that loops form at around @xmath1 times the horizon scale , whereas refs . @xcite ( and some portion of the conventional wisdom ) suggest a scale several orders of magnitude smaller . the larger loops would lead to enhanced signatures of several types , and tighter bounds on @xmath0 @xcite . in a system with a large hierarchy of scales , one might hope that analytic methods would be particularly useful in separating the processes occurring at different scales , while numerical methods would be needed to understand the nonlinearities at a given scale . this is the philosophy that we will attempt to implement . we focus on a microscopic description , similar in spirit to ref . @xcite and in particular ref . an important difference from ref . @xcite is that we are less ambitious : that work was largely directed at understanding the horizon - scale structure quantitatively , whereas we are only interested in shorter scales . also , we do not attempt to reduce the dynamics to a parameterized model but rather focus on the full two - point function ; this two - point function appears to be characterized by a critical exponent , which did not enter into previous work . finally , our ( tentative ) conclusions are opposite to ref . @xcite , in that we believe that the network scales even without gravitational radiation . the full microscopic equations for the string network @xcite appear to be too complicated to solve , and so we model what we hope are the essential physical processes . we have had some success in matching results from simulations , but there are also some discrepancies which may indicate additional processes that must be included . in section 2 we consider the evolution of a small segment on a long string . if the rate of string intercommutations is fixed in horizon units , then over a range of scales the only important effect would be stretching due to the expansion of the universe . we are then able to determine the two - point functions characterizing the small scale structure . we find that the string is actually rather smooth , in agreement with recent simulations @xcite : its fractal dimension approaches one as we go to smaller scales . there is a nontrivial power law , but this appears in the _ approach _ of the fractal dimension to one ; that is , the critical exponent @xmath2 is related to the power spectum of perturbations on the long string . the two - point functions depend on two parameters that must be taken from simulations . one of these is the mean @xmath3 of an element of the long string ; this determines the exponent @xmath2 . the second parameter is the normalization of the two - point function . our results for the two - point function match reasonably well with the simulations @xcite over a range of scales , but there is substantial deviation at the shortest scales . it remains to be seen whether this is due to transient effects in the simulations , or real effects that we have omitted . in section 3 we study the effects of the small scale structure on string lensing . because the string is fairly smooth , these effects are not as dramatic as has been considered in some previous work . however , there are calculable deviations from perfect lensing . we also consider , for lensing by a long string , the likely trajectory of the string relative to the axis of a given lensed object . we discuss some deviations from gaussianity in the small scale structure , due to kinks . in section 4 we add in loop production as a perturbation . even though the strings are relatively smooth at short distance , we find that the total rate of loop production diverges for small loops ; the rate of divergence appears to agree with recent simulations @xcite . this divergence does not invalidate our stretching model , because the loop production is localized to regions where the left- and right - moving tangents @xmath4 are approximately equal , but it points to a complicated fragmentation process , cascading to smaller and smaller loops . we are not able to follow the fragmentation process analytically at present , but we give an analytic argument as to why the cascade should terminate , leaving loops at some small but fixed multiple of the horizon scale . we also point out an interesting puzzle related to loop velocities , which again points to a complicated fragmentation process . in section 5 we discuss various applications and future directions . the behavior of string networks is notoriously complicated , and it appears to remain so even when one focusses on small scales . thus we view our work as part of an ongoing dialogue between analytic calculations and simulations , which we hope will lead to a more complete and precise picture . we consider `` vanilla '' cosmic strings , a single species of local string without superconducting or other extra internal degrees of freedom . the evolution of a network of such strings is dictated by three distinct processes . first , the expansion of the universe stretches the strings : on scales larger than the horizon scale @xmath5 irregularities on a string are just conformally amplified , but on smaller scales the string effectively straightens @xcite . this is described by the nambu action in curved spacetime . second , gravitational radiation also has the effect of straightening the strings , but it is significant only below a length scale proportional to @xmath5 and to a positive power of the dimensionless string coupling @xmath0 @xcite . since this is parametrically small at small @xmath0 , we will ignore this effect for the present purposes . finally , intercommutations play an important role in reaching a scaling solution , in particular through the formation of closed loops of string . at first we shall neglect this effect but we will be forced to return to this issue in section 4 . let us consider the evolution of a small segment on a long string . we take the segment to be very short compared to the horizon scale , but long compared to the scale at which gravitational radiation is relevant . the scaling property of the network implies that the probability per hubble time for this segment to be involved in a long string intercommutation event is proportional to its length divided by @xmath5 , and so for short segments the intercommutation rate per hubble time will be small . formation of a loop much larger than the segment might remove the entire segment from the long string , but this should have little correlation with the configuration of the segment itself , and so will not affect the probability distribution for the ensemble of short segments . formation of loops at the size of the segment and smaller could affect this distribution , and the results of section 4 will indicate that the production of small loops is large , but this process takes place only in localized regions where the left- and right - moving tangents are approximately equal . thus , there is a regime where stretching is the only relevant process . if we follow a segment forward in time , its length increases but certainly does so more slowly than the horizon scale @xmath5 , which is proportional to the frw time @xmath6 . thus the length divided by @xmath5 decreases , and therefore so does the rate of intercommutation . if we follow the segment backward in time , its length eventually begins to approach the horizon scale , and the probability becomes large that we encounter an intercommutation event . our strategy is therefore clear . for the highly nonlinear processes near the horizon scale we must trust simulations . at a somewhat lower scale we can read off the various correlators describing the behavior of the string , and then evolve them forward in time using the nambu action until we reach the gravitational radiation scale . the small probability of an intercommutation involving the short segment can be added as a perturbation . this approach is in the spirit of the renormalization group , though with long and short distances reversed . in an frw spacetime , @xmath7 the equation of motion governing the evolution of a cosmic string is @xcite @xmath8 here @xmath9 is given by @xmath10 these equations hold in the transverse gauge , where @xmath11 . dots and primes refer to derivatives relative to the conformal time @xmath12 and the spatial parameter @xmath13 along the string , respectively . the evolution of the parameter @xmath9 follows from equation ( [ eom ] ) , @xmath14 from the second derivative terms it follows that signals on the string propagate to the right and left with @xmath15 . thus the structure on a short piece of string at a given time is a superposition of left- and right - moving segments , and it is these that we follow in time . in an expanding spacetime the left- and right - moving waves interact they are not free as in flat spacetime . from eq . ( [ epsev ] ) it follows that the time scale of variation of @xmath9 is the hubble time , and so to good approximation we can replace @xmath16 with the time - averaged @xmath17 ( bars will always refer to rms averages ) , giving @xmath18 as a function of time only . reparameterizations . a convenient choice is to take @xmath9 to be independent of @xmath13 at the final time , and then @xmath9 will be @xmath13-independent to good approximation on any horizon length scale in the past . ] from the definition of @xmath9 it then follows that the energy of a segment of string of coordinate length @xmath13 is @xmath19 . for simplicity we will refer to @xmath20 as the length @xmath21 of a segment , @xmath22 though this is literally true only in the rest frame . the scale factor is @xmath23 where @xmath24 the mean rms velocities for points on long strings are taken from simulations @xcite . it follows that @xmath25 in the radiation era , @xmath26 and in the matter era @xmath27 . thus the physical length of the segment grows in time , but more slowly than the comoving length @xcite , and much more slowly than the horizon length @xmath28 . for illustration , consider a segment of length @xmath29 , as would be relevant for lensing at a separation of a few seconds and a redshift of the order of @xmath30 . according to the discussion above , @xmath31 depends on time as @xmath32 in the matter era . thus the length of the segment would have been around a hundredth of the horizon scale at the radiation - to - matter transition . in other words , it is the nonlinear horizon scale dynamics in the radiation epoch that produces the short - distance structure that is relevant for lensing today , in this model . this makes clear the limitation of simulations by themselves for studying the small scale structure on strings , as they are restricted to much smaller dynamical ranges . in terms of left- and right - moving unit vectors @xmath33 , the equation of motion ( [ eom ] ) can be written as @xcite @xmath34 . \label{eom2}\ ] ] we will study the time evolution of the left - moving product @xmath35 . for this it is useful to change variables from @xmath36 to @xmath37 where @xmath38 is constant along the left - moving characteristics , @xmath39 . then @xmath40 where @xmath41 . the equations of motion ( [ eom2 ] , [ bilinear ] ) are nonlinear and do not admit an analytic solution , but they simplify when we focus on the small scale structure . if @xmath42 were a smooth function on the unit sphere , we would have @xmath43 ^ 2)$ ] as @xmath44 approaches @xmath38 . we are interested in any structure that is less smooth than this , meaning that it goes to zero more slowly than @xmath45 ^ 2 $ ] . for this purpose we can drop any term of order @xmath45 ^ 2 $ ] or higher in the equation of motion ( smooth terms of order @xmath46 cancel because the function is even ) . consider the product @xmath47 . the right - moving characteristic through @xmath48 and the left - moving characteristic through @xmath49 meet at a point @xmath50 where @xmath51 is of order @xmath52 . we could choose coordinates where @xmath53 and @xmath54 ^ 2)$ ] , and then @xmath55 . ] eq . ( [ eom2 ] ) states that @xmath56 is slowly varying along left - moving characteristics ( that is , the time scale of its variation is the frw time @xmath6 ) , and @xmath57 is slowly varying along right - moving characteristics . thus we can approximate their product at nearby points by the local product where the two geodesics intersect , @xmath58 then @xmath59 ^ 2)\ .\end{aligned}\ ] ] when we integrate over a scale of order the hubble time , the @xmath51 shifts in the arguments have a negligible effect @xmath60 and so we ignore them . defining @xmath61 we have @xmath62 ^ 2 ) ] \ .\ ] ] thus @xmath63 ^ 2 ) ] \label{hppint } \biggr\ } \ .\ ] ] averaging over an ensemble of segments , and integrating over many hubble times ( and therefore a rather large number of correlation times ) the fluctuations in the exponent average out and we can replace @xmath64 with @xmath65 , @xmath66 note that in contrast to previous equations the approximation here is less controlled . we do not at present have a good means to estimate the error . it depends on the correlation between the small scale and large scale structure ( the latter determines the distribution of @xmath67 ) , and so would require an extension of our methods . we do expect that the error is numerically small ; note that if we were to consider instead @xmath68 then the product in eq . ( [ hppint ] ) would become a sum and the averaging would involve no approximation at all . averaging over a translationally invariant ensemble of solutions , we have @xmath69 we have used the fact that to good approximation ( again in the sense of eq . ( [ avexp ] ) ) , @xmath9 is only a function of time , and so we can choose @xmath70 and @xmath71 . equivalently , @xmath72 the ratio of the segment length to @xmath73 is @xmath74 the logic of our earlier discussion is that we use simulations to determine the value of @xmath75 at @xmath76 somewhat less than one , and then evolve to smaller scales using the nambu action . that is , @xmath77 for some constants @xmath78 and @xmath79 . we assume scaling behavior near horizon scale , so that @xmath79 is independent of time . using this as an initial condition for the solution ( [ soln ] ) gives @xmath80 in the last form we have expressed the correlator in terms of physical quantities , the segment length @xmath21 defined earlier and the frw time @xmath6 . and @xmath12 , or equivalently for @xmath9 and @xmath81 . the value of @xmath79 depends on this choice , but the value of @xmath82 does not . ] ( [ hpp ] ) is our main result . equivalently ( and using @xmath13 parity ) , @xmath83 in the radiation era @xmath84 and in the matter era @xmath85 . there can be no short distance structure in the correlator @xmath86 , because the left- and right - moving segments begin far separated , and the order @xmath87 interaction between them is too small to produce significant nonsmooth correlation . thus , from ( [ ppa ] ) we get @xmath88 before interpreting these results , let us present the derivation in a slightly different way . the exponent @xmath2 is positive , so for points close together the vectors @xmath89 and @xmath90 are nearly parallel . thus we can write the structure on a small segment as a large term that is constant along the segment and a small fluctuation : @xmath91 with @xmath92 and @xmath93 . inserting this into the equation of motion ( [ eom2 ] ) and expanding in powers of @xmath94 gives @xmath95\ , \\[6pt ] \dot{\bf w}_+ - \frac{1}{\epsilon } { \bf w}'_+ & = & \frac{\dot{a}}{a } \left [ ( { \bf w}_+ \cdot { \bf p}_-)\ , { \bf p}_+ + ( { \bf p}_+ \cdot { \bf p}_-)\ , { \bf w}_+ \right ] \nonumber\\ & = & - ( { \bf w}_+ \cdot \dot{\bf p}_+)\ , { \bf p}_+ + \frac{\dot{a}}{a } ( { \bf p}_+ \cdot { \bf p}_-)\ , { \bf w}_+\ . \label{eom3}\end{aligned}\ ] ] since the right - moving @xmath57 is essentially constant during the period when it crosses the short left - moving segment , we have replaced it in the first line of ( [ eom3 ] ) with a @xmath13-independent @xmath96 . in the second line of ( [ eom3 ] ) we have used @xmath97 . in the final equation for @xmath94 , the first term is simply a precession : @xmath98 rotates around an axis perpendicular to both @xmath98 and @xmath96 , and this term implies an equal rotation of @xmath94 so as to keep @xmath94 perpendicular to @xmath98 . ( [ eom3 ] ) then implies that in a coordinate system that rotates with @xmath98 , @xmath94 is simply proportional to @xmath99 . it follows that @xmath100 ^ 2 \right\rangle \label{upup}\ ] ] scales as @xmath101 as found above ( again we are approximating as in eq . ( [ avexp ] ) , and again this statement would be exact if we instead took the average of the logarithm ) . similarly the four - point function of @xmath94 scales as @xmath102 . we have not assumed that the field @xmath94 is gaussian ; the @xmath103-point functions , just like the two - point function , can be matched to simulations near the horizon scale . we can anticipate some degree of nongaussianity due to the kinked structure ; we will discuss this further in section 3.2 . now let us discuss our results for the two - point functions . we can also write them as @xmath104 these are determined up to two parameters @xmath17 and @xmath82 that must be obtained from simulations . a first observation is that these scale , they are functions only of the ratio of @xmath21 to the horizon scale . this is simply a consequence of our assumptions that the horizon scale structure scales and that stretching is the only relevant effect at shorter scales . we emphasize that these results are for segments on long strings ; we will discuss loops in sec . it is natural to characterize the distribution of long strings in terms of a fractal dimension . the rms spatial distance between two points separated by coordinate distance @xmath13 is @xmath105\ . \label{r2}\end{aligned}\ ] ] we can then define the fractal dimension @xmath106 ( which is 1 for a straight line and 2 for a random walk ) , @xmath107 the fractal dimension approaches 1 at small scales : the strings are rather smooth . there is a nontrivial scaling property , not in the fractal dimension but rather in the deviation of the string from straightness , @xmath108 we define the scaling dimension @xmath109 . note that @xmath110 is not large , roughly @xmath111 in the radiation era and @xmath112 in the matter era , so the approach to smoothness is rather slow . . ] our general conclusions are in agreement with the recent simulations of ref . @xcite , in that the fractal dimension approaches 1 at short distance . to make a more detailed comparison it is useful to consider a log - log plot of @xmath113 versus @xmath21 , as suggested by the scaling behavior ( [ crit ] ) ; we thank c. martins for replotting the results of ref . @xcite in this form . the comparison is interesting . at scales larger than @xmath114 ( which is @xmath115 in the radiation era and @xmath116 in the matter era ) the correlation goes to zero . rather abruptly below @xmath114 the slope changes and agrees reasonably well with our result . it is surprising to find agreement at such long scales where our approximations do not seem very precise . on the other hand , at shorter scales where our result should become more accurate , the model and the simulations diverge ; this is especially clear at the shortest scales in the radiation - dominated era ( fig . [ mvd rad ] ) . . ] one possible explanation for the discrepancy is transient behavior in the simulations . we have argued that the structure on the string is formed at the horizon scale and ` propagates ' to smaller scales ( in horizon units ) as the universe expands . in ref . @xcite the horizon size increases by a factor of order 3 , and so even if the horizon - scale structure forms essentially at once , the maximum length scale over which it can have propagated is @xmath117 , less than half an order of magnitude . at smaller scales , the small scale structure seen numerically would be almost entirely determined by the initial conditions . on the other hand , the authors of ref . @xcite ( private communication ) argue that their result appears to be an attractor , independent of the initial conditions , and that loop production may be the dominant effect . motivated by this we will examine loop production sec . 4 . indeed , we will find that this is in some ways a large perturbation , but we are still unable to identify a mechanism that would produce the two - point function seen in the simulations . this is an important issue to be resolved in future work . thus far we have discussed @xmath118 . our result ( [ corrs ] ) implies a linear relation between @xmath118 and @xmath119 . in fact , this holds more generally from the argument that there is no short - distance correlation between @xmath120 and @xmath57 , eq . ( [ pppm ] ) : @xmath121 inspection of fig . 2 of ref . @xcite indicates that this relation holds rather well at all scales below @xmath122 . the small scale structure on strings is sometimes parameterized in terms of an effective tension @xcite . for a segment of length @xmath21 the effective tension is given by @xmath123 where we have made use of result ( [ r2 ] ) . note that this is strongly dependent on the scale @xmath21 of the coarse - graining . in conclusion , let us emphasize the usefulness of the log - log plot of @xmath124 . in a plot of the fractal dimension , all the curves would approach one at short distance , though at slightly different rates . the difference is much more evident in figs . 1 and 2 , and gives a clear indication either of transient effects or of some physics omitted from the model . in this subsection we will assume that our stretching model is actually valid down the scale where gravitational radiation sets in . if loop production or other relatively rapid processes are actually determining the small scale structure then this subsection is moot . we have noted that at very short scales we see structure that actually emerged from the horizon dynamics during the radiation era . thus we should take the radiation - to - matter transition into account in our calculation of the small - scale structure . at the time of equal matter and radiation densities , @xmath125 where @xmath126 is the length of the segment between @xmath13 and @xmath127 at @xmath128 . assuming that the transition from radiation - dominated to matter - dominated behavior is sharp ( which is certainly an oversimplification ) , we evolve forward to today using the result ( [ soln ] ) . the right - hand side of eq . ( [ teq ] ) is then multiplied by a factor @xmath129 . in terms of the length today , @xmath130 , we have @xmath131 this expression applies to scales @xmath132 that , evolved backward in time , reached the horizon scale @xmath5 before the transition occurred , i.e. at a time @xmath133 defined by @xmath134 such that @xmath135 . for longer scales , which formed during the matter era ( for which @xmath136 ) , we have simply @xmath137 , as a function of the length @xmath21 at present time ( solid curve ) . on scales larger than the critical length @xmath138 the structure is determined by the matter era expression . on scales below @xmath139 the transition result ( [ radmat ] ) gives an enhanced effect . the dashed curves show the extrapolations of the two relevant expressions : on small scales the actual structure is enhanced relative to the pure matter era result . ] the transition between the two forms occurs along the curve determined by the intersection of the two surfaces ( [ radmat ] ) and ( [ radmat2 ] ) . this determines the critical length at the time of equal matter and radiation densities , @xmath140 . in terms of the length at some later time @xmath6 , the transition occurs at @xmath141 so that the transition scale at the present time is @xmath138 ( fig . [ transition ] ) . the transition result ( [ radmat ] ) implies more structure at the smallest scales than would be obtained from the matter era result , by a factor @xmath142 . of course , precise studies of the small scale structure must include also the effect of the recent transition to vacuum domination ; this period has been brief so the effect should be rather small . let us now consider the effect of the small scale structure on the images produced by a cosmic string lens . previous work has discussed possible dramatic effects @xcite , including multiple images and large distortions . we can anticipate that the rather smooth structure that we have found , which again we note is subject to our assumptions , will produce images with only small distortion . we will use our stretching model of the two - point function . if this proves incorrect one could apply the analysis using phenomenological values of @xmath2 and @xmath82 ; for example , the extrapolation of the results of ref . @xcite give a smoother string , and even less distorted images . we quote the result of ref . @xcite for the angular deflection of a light ray by a string , @xmath143 is the four - velocity of the unperturbed light ray , which we take to be @xmath144 as shown in fig . [ schematic ] , @xmath145 is a reference point on this ray , and subscripts @xmath146 and @xmath147 are with respect to the spatial direction of the ray . also , @xmath148 is the string coordinate , , @xmath149 , @xmath150 . ] in terms of which @xmath151 and @xmath152 is defined by @xmath153 . -axis and the distant source and the observer located at points @xmath154 and @xmath155 respectively . ] in keeping with sec . 2.3 we separate the string locally into a straight part and a fluctuation ; we will keep the deflection only to first order in the fluctuation . we consider here only the simplest geometry , in which the straight string is perpendicular to the light ray and at rest , so that @xmath156 . to first order in the fluctuation , @xmath157 . one then finds @xmath158 with @xmath159 being half the deficit angle of the string . to the order that we work @xmath160 is a constant , corresponding to the time when the light ray is perpendicular to the straight string . we can use our results for the small scale structure to calculate the two - point functions of the deflection . let us focus on the local magnifications parallel and perpendicular to the string , given by basic lensing theory as @xmath161 here @xmath162 and @xmath163 are the distances of the source from the observer and the lens respectively ( these would be the angular diameter distances on cosmological scales ) . it is particularly interesting to consider the differential magnifications for the two images produced by a string , @xmath164 we take for simplicity @xmath165 and @xmath166 for @xmath167 ; this corresponds to the symmetric images of an object directly behind the string . then @xmath168 from sec . 2 we obtain @xmath169/t)^{2\chi } \bigr\}\ . \label{u2p}\end{aligned}\ ] ] note that eq . ( [ upup ] ) does not fully determine the two - point functions ( [ u2p ] ) , and in fact that the latter can not be translation invariant . we have fixed the ambiguity by defining the expectation value to vanish when @xmath13 or @xmath127 vanishes ( that is , at the point on the string nearest to the light ray ) ; this amounts to a choice of how one splits @xmath170 into @xmath171 and @xmath172 . as is stands , equation ( [ u2p ] ) applies only when @xmath173 is larger than the critical length @xmath139 . if this is not the case , one must take into account the radiation - to - matter transition as discussed in section 2.5 . according to equation ( [ radmat ] ) , the rhs of ( [ u2p ] ) gets multiplied by a power of @xmath174 . from @xmath175 we obtain , is typically of the order of @xmath176 , thus much smaller than the critical length @xmath139 . we can then safely use the formula for @xmath175 valid for the smallest scales over the full range of integration . corrections from the longest scales will increase the final result , but not significantly . ] @xmath177 plugging in the numeric values for @xmath2 , @xmath82 and @xmath178 and using a representative value for the dimensionless parameter @xmath179 , we obtain a rms differential magnification slightly below @xmath180 : @xmath181 another question related to short - distance structure is the alignment of lenses . suppose we see a lens due to a long string , with a certain alignment . where should we look for additional lens candidates ? previous discussions @xcite have considered the two extreme cases of a string that is nearly straight , and a string that is a random walk on short scales ; clearly the networks that we are considering are very close to the first case . we keep the frame of the previous section , with the lens at the origin in @xmath182 and aligned along the 1-axis . then as we move along the string , the rms transverse deviation is @xmath183 the extension in the @xmath184 direction is just @xmath21 , so the rms angular deviation is @xmath185 if we put in representative numbers , looking at an apparent separation on the scale of arc - minutes for a lens at a redshift of order @xmath1 , we obtain with the matter era parameters a deviation @xmath186 radians . that is , any additional lenses should be rather well aligned with the axis of the first . if the string is tilted by an angle @xmath187 to the line of sight , then projection effects increase @xmath188 and @xmath189 by a factor @xmath190 . of course , for lensing by a _ loop _ , the bending will be large at lengths comparable to the size of the loop . lens alignment provides an interesting setting for discussing the nongaussianity of the structure on the string . if the fluctuations of @xmath191 were gaussian , then the probability of finding a second lens at an angle @xmath189 to the axis of the first would be proportional to @xmath192 , and therefore very small at large angles . however , we have considered thus far a typical string segment , which undergoes only stretching . there will be a small fraction of segments that contain a large kink , and one might expect that it is these that dominate the tail of the distribution of bending angles . let us work this out explicitly . consider a left - moving segment of coordinate length @xmath13 , and let @xmath193 be the probability that it contain a kink for which the discontinuity @xmath194 lies between @xmath195 and @xmath196 ( @xmath197 ) . there are two main contributions to the evolution of @xmath198 . intercommutations introduce kinks at a rate that we assume to scale , so that it is proportional to the world - sheet volume in horizon units , @xmath199 , and to some unknown function @xmath200 . also , the expansion of the universe straightens the kink , @xmath201 @xcite . then @xmath202 we set @xmath198 to zero at the matching time @xmath203 defined by @xmath204 as in eq . ( [ match ] ) : earlier kinks are treated as part of the typical distribution , while @xmath198 identifies kinks that form later . for simplicity we assume that @xmath78 is small enough that the probability of more than one kink can be neglected . to solve this , define @xmath205 then @xmath206 this can now be integrated to give @xmath207 here @xmath208 and @xmath209 . note that @xmath210 is just a constant times @xmath211 , so the probability distribution scales . the source @xmath200 vanishes by definition for @xmath212 , so @xmath213 is equivalent to @xmath214 . rather than the angle @xmath189 between the axis of the first lens and the position of the second lens , it is slightly simpler to consider the angle @xmath188 between the two axes . in the small fluctuation approximation this is just @xmath215 , and so the rms fluctuation is @xmath216 in the gaussian approximation , the probability distribution is @xmath217 and so is very small for large angles . on the other hand , a large angle might also arise from a segment that happens to contain a single recent kink . treating the segment as straight on each side of the kink , the probability density is then precisely the function @xmath193 just obtained , with @xmath218 . if we consider angles that are large compared to @xmath219 but still small compared to 1 , the range of integration in the solution ( [ psol ] ) extends essentially to the full range @xmath220 to @xmath221 and so the integral gives a constant . then @xmath222 thus the tail of the distribution is not gaussian but a power law , dominated by segments with a ` recent ' kink . one finds the same for the distribution of @xmath189 . it is notable however that the exponent in the distribution is rather large , roughly 4 in the matter era and 10 in the radiation era . thus the earlier conclusion that the string is rather straight still holds . the sharp falloff of the distribution also suggests that a gaussian model might work reasonably well , and indeed we will employ this in the next section . the model presented thus far assumes that stretching is the dominant mechanism governing the evolution of the small scale structure . motivated by the discrepancy with ref . @xcite ( and at the suggestion of those authors ) , we now consider the production of small loops . given the smoothness of the strings on short distances as found above , one might have thought that loop production on small scales would be suppressed , but we will see that this is not the case . we will ignore loop reconnection , based on the standard argument that this is rare for small loops @xcite . we will treat the stretching model as a leading approximation , and add in the loop production as a perturbation . however , we will find that the loop production diverges at small scales , and so the problem becomes nonlinear and our analytic approach breaks down . ultimately we would expect that some sort of improved scaling argument can be used to follow the structure to still smaller scales , but this is beyond our current methods ; it is necessary first to identify the relevant processes . in this section we aim to determine the rate of loop production , which occurs when a string self - intersects . that is , we want to compute the number of loops @xmath223 with invariant length between @xmath21 and @xmath224 originating from a self - intersection at a string coordinate between @xmath13 and @xmath225 during a time interval @xmath226 . loops much smaller than the horizon size evolve in an essentially flat space so we are able to use null coordinates @xmath227 and @xmath228 . in this case the left- and right - moving vectors @xmath170 depend only on @xmath229 or @xmath230 respectively and the condition for the formation of a loop of length @xmath21 is @xmath231 with @xmath232 hence the number of loops formed is given by @xmath233 where @xmath234 is the jacobian for the transformation @xmath235 . this formalism is as in refs . @xcite . expanding around the stretching result in the form of section 2.3 , we have the functional probability distribution @xmath236 \approx { \cal p}_0({\bf p}_+ , { \bf p}_- ) { \cal p}_+[{\bf w}_+ ] { \cal p}_-[{\bf w}_-]\ .\ ] ] that is , with stretching alone the left- and right - movers can not be correlated on small scales . we will further assume that @xmath237 are gaussian with variance given by the two - point function . as discussed in sec . 3.1 this two - point function is not completely determined by eq . ( [ upup ] ) , and in general we have @xmath238 the function @xmath239 is defined for later reference , with the factor of 2 representing the sum over transverse directions . to specify the function @xmath240 one needs to choose how to split @xmath170 into @xmath171 and @xmath172 . the natural choice here is to make @xmath171 parallel to @xmath241 , i.e. @xmath242 . the transverse fluctuations along the segment that forms the loop , say @xmath243 , then have vanishing average , @xmath244 integrating eq . ( [ wpwp ] ) @xmath245 and using the condition ( [ vanav ] ) to set this to zero gives @xmath246 where @xmath247 \ . \label{fsigma}\ ] ] similarly , @xmath248 with @xmath249 . the three - dimensional @xmath51-function in eq . ( [ dn ] ) can be expressed in spherical coordinates as @xmath250 we take the @xmath251-direction to be aligned along @xmath252 , in which case the jacobian matrix is given by @xmath253{ccc } w^x_+(l ) - w^x_+(0 ) & \ ; w^y_+(l ) - w^y_+(0 ) \ ; & \frac{1}{2}\!\left [ w^2_+(0 ) - w^2_+(l ) \right ] \\ \\ w^x_-(0 ) - w^x_-(l ) & \ ; w^y_-(0 ) - w^y_-(l ) \ ; & \frac{1}{2}\!\left [ w^2_-(l ) -w^2_-(0 ) \right ] \\ \\ w^x_+(l ) - w^x_-(0 ) & \ ; w^y_+(l ) - w^y_-(0 ) \ ; & \frac{1}{2}\!\left [ w^2_-(0 ) -w^2_+(l ) \right ] \end{array } \right ) \ .\ ] ] for notational simplicity we have , after differentiating , set @xmath254 and @xmath255 . the fluctuations @xmath172 effectively live in the @xmath184-@xmath256 plane due to the orthogonality condition @xmath257 . let us first estimate the scaling of @xmath258 with @xmath21 . we have @xmath259 the width of the distribution of @xmath260 is of order @xmath261 and so the average of @xmath262 is a density of order @xmath263 . the @xmath51-function ( [ deltas ] ) then scales as @xmath264 while the jacobian scales as @xmath265 ( the columns are respectively of orders @xmath266 , @xmath266 , @xmath267 ) , giving in all @xmath268 thus , for @xmath269 ( as found in both the matter and radiation eras ) , the total rate of string loss diverges at small @xmath21 : the calculation breaks down . we will discuss this cutoff further in the next subsection , but first we estimate the numerical coefficient . we see no reason for a strong correlation between the @xmath51-function and jacobian factors in @xmath258 , so we take the product of their averages . for the radial part of the @xmath51-function , a gaussian average gives @xmath270 the subscripts @xmath271 in the second line refer to the connected expectation values , obtained by contracting the gaussian fields @xmath172 with the propagator ( [ wpwp ] ) . we have defined @xmath272 numerically @xmath273 for @xmath274 . we now turn to the angular part of the probability distribution function . this must be a function of @xmath275 , and we will take it to have exponential form @xcite : @xmath276 the prefactor normalizes the distribution . the parameter @xmath277 can be determined from the requirement that @xmath278 , or equivalently @xmath279 thus , @xmath280 . more precisely , @xmath277 is @xmath281 in the radiation epoch and @xmath282 in the matter epoch . we are now able to calculate the average value of ( [ deltas ] ) . recognizing that @xmath283 we have @xmath284 we now consider @xmath285 . in the gaussian approximation the relevant probability distribution is @xmath286 \ , \ ] ] where the index @xmath287 is summed over the two coordinates @xmath184 and @xmath256 , and @xmath288 ( that is , the columns and rows of the @xmath289 matrix @xmath290 correspond to the points @xmath220 and @xmath21 , not the index @xmath287 . ) as usual , the whole distribution is determined solely by the two - point functions which have already been determined in ( [ wpwp ] ) and ( [ fsigma ] ) , @xmath291 . thus , one finds that @xmath292 we now have all we need to write down an expression for @xmath285 . before we do so , let us perform a simplifying change of variables : @xmath293 \ , \\ [ 7pt ] y_\pm \equiv \sqrt{\frac{1}{2a } } \left({t}/{l}\right)^\chi \left [ w^x_\pm(l ) - w^x_\pm(0 ) \right ] \ , \\ [ 7pt ] z_\pm \equiv \sqrt{\frac{1+\chi}{1-\chi } \frac{1}{2a } } \left({t}/{l}\right)^\chi \left [ w^y_\pm(l ) + w^y_\pm(0 ) \right ] \ , \\ [ 7pt ] w_\pm \equiv \sqrt{\frac{1}{2a } } \left({t}/{l}\right)^\chi \left [ w^y_\pm(l ) - w^y_\pm(0 ) \right ] \ , \end{array}\ ] ] under which the expectation value of the loop jacobian takes a compact form : @xmath294 here we have defined an 8-dimensional vector @xmath295 and two functions of it : @xmath296 \ . \nonumber\end{aligned}\ ] ] numerically the integral is @xmath297 for the radiation epoch and @xmath298 for the matter epoch ; analytic estimates agree . finally , we can combine results ( [ avdelta ] ) and ( [ integral ] ) . noticing that @xmath299 , we get @xmath300 with @xmath301 in the radiation epoch and @xmath302 in the matter epoch . the total rate of string loss @xmath303 diverges at small @xmath21 for @xmath304 , as is the case in both the radiation and matter eras . the fractal dimension approaches 1 at short distance , but the exponent @xmath2 , which characterizes the approach to this limit , indicates a relatively large amount of short distance structure . for example , if we consider the functions @xmath305 on the unit sphere , then @xmath306 : the effective fractal dimension is @xmath307 , though the path is not continuous , there are gaps due to kinks . of course , the total rate of string loss is bounded , and so for small loops we must be multiply counting . a precise account of this effect will be very complicated ; see for example ref . @xcite , which parameterizes it in terms of a modification of the exponential distribution ( [ expform ] ) taking the form of a ` hole ' in the forward direction @xmath308 . for the present we just determine the scale at which our calculation _ must _ break down . in a scaling solution , the total amount of long string in a comoving volume scales as @xmath309 , while stretching alone would give @xmath310 . the rate of string lost to loops must be @xmath311 our result for the same quantity , with a cutoff @xmath312 , is @xmath313 equating these gives @xmath314 in both eras . the cutoff ( [ lstar ] ) is rather large , just an order of magnitude below the horizon scale . however , this is just the beginning of the story . the formation of these large loops does not terminate the process of loop production on smaller scales ; this continues on the large loop , unaffected at least initially by the separation of the loop from the infinite string . ) may not be seen in the primary loops as usually defined . rather , we should follow null rays from the two world - sheet points of primary loop formation backwards to the point where they meet , and define the length of the ` causal primary ' loop as twice the time difference so as to enclose the entire causally disconnected region . ] thus we must expect a rather dramatic fragmentation process . the divergence of the rate of loop loss implies that at scales @xmath21 well below @xmath315 the nave number density of loops of size @xmath316 is greater than @xmath317 : these become dense on the string . thus we might come to the conclusion that the fragmentation process continues until we end up with all loops at the gravitational radiation scale , the opposite of the nave result ( [ lstar ] ) above . , where the loop production is only logarithmically divergent , while the more noisy is dominated by harmonics near the cutoff and does not resemble our distribution . ] in fact , we believe that this is not precisely the situation either . rather , a non - self - intersecting loop will occasionally form , likely an order of magnitude or more smaller than the primary loop , and this length of loop will then be lost to the fragmentation process on smaller scales . if a fraction @xmath318 of the string is lost in this way at each scale , then the total length of string in loops of size less than @xmath21 will scale as @xmath319 . the exponent may be rather small , so there could be a substantial production of small loops . the divergent rate of loop production implies that our initial attempt to use analytic methods to separate scales has been too simple , but we might hope that a more sophisticated scaling model along the lines just indicated will be successful . let us briefly describe an analytic model of the formation of non - self - intersecting loops . when the loop forms , we can define the distributions @xmath320 because of the high fractal dimensions of the curves @xmath321 , we might think of these distributions as reasonably smooth functions . the structure on the high harmonics of the loop will tend to smear these distributions into gaussians , but they will be skewed by the particular random values of the lowest harmonics . thus the distributions @xmath322 and @xmath323 will differ . the production of the smallest loops is roughly local in @xmath324 , and so it can only continue until whichever of the two densities is smaller has been depleted . this leaves residual distributions @xmath325 according to the sign of the difference at each point . this nonvanishing difference implies a cutoff on the production of the smallest loops , and allows non - self - intersecting loops of finite size to form . the results that we have found have some notable agreements , and disagreements , with simulations . one success is an apparent agreement with the recent simulations of ref . @xcite for the distribution of loop sizes : that reference finds a number density of loops per volume and per length @xmath326 with @xmath327 in the matter era and @xmath328 in the radiation era . we have exponents @xmath329 and 2.8 respectively . our exponent is for the production rate rather than the density , but in this regime the density is dominated by recently produced loops and so these are the same . let us verify this , and also obtain the relative normalizations for the two quantities . the number of loops per comoving volume of length between @xmath21 and @xmath224 is unaffected by the expansion of the universe , and we are assuming that we are at scales where gravitational radiation can be neglected . the number then changes only due to production : @xmath330 where we have inserted the string scaling density @xmath331 . this integral is dominated by recent times as long as @xmath332 , as holds in both eras ( again , @xmath333 ) . integrating over time then gives @xmath334 in the notation of ref . @xcite , @xmath335 where the last factor is from the conversion from @xmath5 to @xmath6 . using @xmath336 respectively in the matter and radiation eras @xcite , this gives @xmath337 , 10.3 . these are larger than found in ref . @xcite by factors of 20 and 260 respectively . a large difference is not surprising , as we have argued that there will be substantial fragmentation which will move the loops to smaller @xmath21 where the distributions are larger . however , it remains to be seen whether this can account for such an enormous difference . the fragmentation need not change the scaling as long as the long - string correlation functions remains a power law , because the shapes of the produced loops , and the resulting fragmentation , all scale ; thus the agreement of the loop production exponents may be real . as we have noted , the loop scaling must eventually break down due to conservation of string , but with these reduced normalizations this happens at much lower scales , @xmath338 and @xmath339 respectively . below these scales we expect the distribution to fall as discussed in the fragmentation discussion . thus we might expect the distribution to peak near these scales , but perhaps to be rather broad in both directions . a possible phenomenological formula for loop production after fragmentation would be @xmath340 the constant @xmath341 is the reduced value of our @xmath271 , determined by comparison with simulations such as ref . @xcite , the exponent @xmath318 , which is likely to be small , must be determined by improved simulations of fragmentation , and the value @xmath342 is determined by conservation of total string . specific signatures , however , might be dominated by the tails , either at large loops or small . the relatively large scale ( [ lstar ] ) is similar to that found in the recent paper @xcite . as discussed in sec . 4.2 , we expect that the final loop size is much smaller because of extensive fragmentation ; the divergence that we have found for small loop formation will be equally present in the calculation of fragmentation of a large loop into smaller ones . ref . @xcite finds a much less extensive fragmentation . this points up the need for more complete studies of this process . this should be possible within the context of flat spacetime simulations , beginning with long strings having the small scale structure as input ( this does depend on resolving first the issue of the long string two - point function , to which we return below ) . the production of very small loops takes place when @xmath343 : that is , near cusps on the long strings as proposed in refs . @xcite . indeed , all the loops that we consider are rather smaller than the correlation length , so the functions @xmath344 and @xmath345 are each somewhat localized on the unit sphere , and necessarily in the same region since @xmath231 . in this sense all of our loops are produced near cusps . it is possible that there is a second population of loops that form at scales much longer than the correlation length , though most of these will quickly reconnect with long strings . the picture of a complicated fragmentation process is consistent with the results of ref . now let us return to the two - point functions as found there . the partial agreement with our stretching model might now seem surprising , given the large production of small loops . if the loop formation were distributed on the long strings , there would be at least one additional effect , where the string shortens due to loop emission and more distant ( and therefore less correlated ) segments are brought closer together . however , the loop production is not distributed : it occurs when the functions @xmath321 , as they wander on the unit sphere , come close together . our picture is that whole segments of order @xmath315 ( [ lstar ] ) are then excised . thus , most of the segments that remain on a long string at given time would have been little affected by loop production . the discrepancy at short scales , which is particularly evident in the radiation era ( fig . 1 ) , remains a puzzle . let us first note that at the shortest scales in figs . 1 and 2 the slope of the curve approaches and possibly even exceeds unity , corresponding to the critical value @xmath346 : if the two - point function is of this form then the small - loop production converges . thus it is appealing to assume that it is feedback from production of short loops that accounts for the break in the curve in fig . 1 . corresponds to the functions @xmath170 being random walks on the unit sphere . this again suggests this value might be some dynamical fixed point . ] this is distinctly different from the cutoff via a @xmath347 hole as discussed in sec.4.2 and and in ref . @xcite : the latter is a modification of the @xmath348 correlations ( though one that would have little effect on the two - point function ) , whereas fig . 1 shows a modification of the @xmath349 correlation . we are unable to identify a physical mechanism that would account for the observed two - point function . consider a very short segment on a long string . with time , its size as a fraction of the horizon size decreases , so it moves to the left on figs . 1 and 2 . if it experiences only stretching , it will follow the slope of the dashed line . the loop production must have a strong bias toward removing segments with large fluctuations to account for the simulations . however , the general picture above is that the initial loop formation occurs at the scale @xmath315 , and so any small segments will be carried away without any bias on their own internal configuration . thus we would expect the two - point function to be given by the stretching power law down to arbitrarily small scales , until gravitational radiation enters . slightly is that segments that do not fall into the loop production ` hole ' at @xmath350 may experience on average a larger value of @xmath351 and so have a reduced two - point function see eq . ( [ hppint ] ) . however , it seems impossible that this could account for the factor of 5 discrepancy of slopes evident in fig . 1 . ] we will not attempt to connect with the flat spacetime simulations of refs . @xcite because our approach requires expansion to drive the fractal dimension to 1 at short distance . finally we would like to note a significant puzzle regarding the loop velocities . for a loop of length @xmath21 , the mean velocity is @xmath352 then @xmath353 looking for example at loops with @xmath354 , we obtain a typical velocity @xmath355 in the matter era and @xmath356 in the radiation era . these are significantly larger than are generally expected ; ref . @xcite gives values around @xmath357 and @xmath358 for loops of this size . we emphasize that this is not a consequence of our dynamical assumptions , but can be seen directly by using the two - point functions of ref . @xcite ( the discrepancy is then even greater in the radiation era ) . assuming that both simulations are correct , we must conclude that the two - point function on loops is very different from that on long strings , in fact less correlated . one possible explanation is that loop formation is biased in this way , but this seems unlikely : loops form when right- and left - moving segments of equal @xmath359 meet , and this will happen most often in the center of the distribution . rather , this seems to be another indication of the complicated nature of the fragmentation : the final non - self - intersecting loops must contain segments which began on the long string many times further separated than the final loop size . our attempt to isolate the physics on different scales has not been completely successful , but it suggests some interesting directions for future work . first of all , the two - point function on long strings must be better understood : is the discrepancy in fig . 1 a result of omitted physics or numerical transients ? this question can be settled numerically , by studying sensitivity to initial conditions , and by studying the time - dependence of an ensemble of small segments . a good understanding of the two - point function would then allow extrapolation to much smaller scales , for application to lensing ( as in section 3 ) , gravitational radiation @xcite , and possible interference between short distance structure and cusps @xcite . second , our work points to correlations among the two - point function , loop production , fragmentation , and loop velocity . in particular , our conjecture of a very complicated fragmentation process can be tested for example , the argument that non - self - intersecting loops must contain segments that originated far apart on the long string . also , the normalization difference between our loop production and that in ref . @xcite must presumably be understood in terms of fragmentation . in conclusion , cosmic string networks are probably not as complicated as turbulence , but they share the property that a rather simple set of classical equations leads to a complicated dynamics that challenges both numerical and analytical attacks . we hope that our work is a step toward a more unified understanding of the properties of these networks . we are grateful to carlos martins and paul shellard for providing the data from their numerical simulations and for valuable comments and suggestions . we also thank ed copeland and thibault damour for discussions and xavier siemens for comments . jvr acknowledges financial support from _ fundao para a cincia e a tecnologia _ , portugal , through grant sfrh / bd/12241/2003 . this work was supported in part by nsf grants phy99 - 07949 and phy04 - 56556 . a. albrecht and n. turok , `` evolution of cosmic string networks , '' phys . d * 40 * , 973 ( 1989 ) . d. p. bennett and f. r. bouchet , `` high resolution simulations of cosmic string evolution . network evolution , '' phys . d * 41 * , 2408 ( 1990 ) . b. allen and e. p. s. shellard , `` cosmic string evolution : a numerical simulation , '' phys . * 64 * , 119 ( 1990 ) . a. vilenkin and e. shellard , _ cosmic strings and other topological defects _ ( cambridge univ . press 1994 ) . m. b. hindmarsh and t. w. b. kibble , rept . phys . * 58 * , 477 ( 1995 ) , hep - ph/9411342 . d. p. bennett and f. r. bouchet , `` evidence for a scaling solution in cosmic string evolution , '' phys . * 60 * , 257 ( 1988 ) . d. austin , e. j. copeland and t. w. b. kibble , `` evolution of cosmic string configurations , '' phys . d * 48 * , 5594 ( 1993 ) [ arxiv : hep - ph/9307325 ] . x. siemens , k. d. olum and a. vilenkin , `` on the size of the smallest scales in cosmic string networks , '' phys . d * 66 * , 043501 ( 2002 ) [ arxiv : gr - qc/0203006 ] . g. r. vincent , m. hindmarsh and m. sakellariadou , `` scaling and small scale structure in cosmic string networks , '' phys . d * 56 * , 637 ( 1997 ) [ arxiv : astro - ph/9612135 ] . v. vanchurin , k. d. olum and a. vilenkin , `` scaling of cosmic string loops , '' arxiv : gr - qc/0511159 . c. ringeval , m. sakellariadou and f. bouchet , `` cosmological evolution of cosmic string loops , '' [ arxiv : astro - ph/0511646 ] . c. j. a. martins and e. p. s. shellard , `` fractal properties and small - scale structure of cosmic string networks , '' [ arxiv : astro - ph/0511792 ] . k. d. olum and a. vilenkin , `` reionization from cosmic string loops , '' arxiv : astro - ph/0605465 . f. embacher , `` cosmic string evolution by exact methods . 1 : model for loop production in flat space , '' b * 387 * , 129 ( 1992 ) ; `` cosmic string evolution by exact methods . 2 : gaussian model , '' nucl . b * 387 * , 163 ( 1992 ) . a. vilenkin , `` cosmic strings , '' phys . d * 24 * , 2082 ( 1981 ) . d. p. bennett and f. r. bouchet , `` high resolution simulations of cosmic string evolution : part 2 . small scale structure and loops , '' in _ the formationand evolution of cosmic strings , _ g. w. gibbons et al.(eds . ) , cambridge university press ( 1990 ) . m. sakellariadou , `` gravitational waves emitted from infinite strings , '' phys . d * 42 * , 354 ( 1990 ) [ erratum - ibid . d * 43 * , 4150 ( 1991 ) ] . m. hindmarsh , `` gravitational radiation from kinky infinite strings , '' phys . b * 251 * , 28 ( 1990 ) . j. m. quashnock and d. n. spergel , `` gravitational selfinteractions of cosmic strings , '' phys . d * 42 * , 2505 ( 1990 ) . n. turok and p. bhattacharjee , `` stretching cosmic strings , '' phys . d * 29 * , 1557 ( 1984 ) . b. allen and r. r. caldwell , `` small scale structure on a cosmic string network , '' phys . rev . d * 43 * , 3173 ( 1991 ) . a. vilenkin , `` effect of small scale structure on the dynamics of cosmic strings , '' phys . d * 41 * , 3038 ( 1990 ) . b. carter , `` integrable equation of state for noisy cosmic string , '' phys . d * 41 * , 3869 ( 1990 ) . a. a. de laix , `` observing long cosmic strings through gravitational lensing , '' phys . d * 56 * , 6193 ( 1997 ) [ arxiv : astro - ph/9705223 ] . f. bernardeau and j. p. uzan , `` cosmic string lens phenomenology : model of poisson energy distribution , '' phys . d * 63 * , 023005 ( 2001 ) [ arxiv : astro - ph/0004102 ] . a. a. de laix and t. vachaspati , `` gravitational lensing by cosmic string loops , '' phys . d * 54 * , 4780 ( 1996 ) [ arxiv : astro - ph/9605171 ] . d. huterer and t. vachaspati , `` gravitational lensing by cosmic strings in the era of wide - field surveys , '' phys . d * 68 * , 041301 ( 2003 ) [ arxiv : astro - ph/0305006 ] . m. oguri and k. takahashi , `` characterizing a cosmic string with the statistics of string lensing , '' phys . d * 72 * , 085013 ( 2005 ) [ arxiv : astro - ph/0509187 ] . r. j. scherrer and w. h. press , `` cosmic string loop fragmentation , '' phys . d * 39 * , 371 ( 1989 ) . a. albrecht , `` small scale structure on cosmic strings , '' fermilab - conf-89 - 221-a _ invited talk given at formation and evolution of cosmic strings , cambridge , eng . , jul 2 - 7 , 1989_. x. siemens and k. d. olum , `` cosmic string cusps with small - scale structure : their forms and gravitational waveforms , '' phys . d * 68 * , 085017 ( 2003 ) [ arxiv : gr - qc/0307113 ] . v. vanchurin , k. olum and a. vilenkin , `` cosmic string scaling in flat space , '' phys . d * 72 * , 063514 ( 2005 ) [ arxiv : gr - qc/0501040 ] . d. p. bennett and f. r. bouchet , `` the two point correlation function of cosmic string loops , '' phys . lett . * 63 * , 1334 ( 1989 ) .

the convergence of a recently proposed coupled cluster ( cc ) family of perturbation series [ eriksen _ et al . _ , j. chem . phys . * 140 * , 064108 ( 2014 ) ] , in which the energetic difference between a parent and a target cc model is expanded in orders of the mller - plesset ( mp ) fluctuation potential , is investigated for four prototypical closed - shell systems ( ne , singlet ch , distorted hf , and f ) in standard and augmented basis sets . in these investigations , energy corrections of the various series have been calculated to high orders and their convergence radii determined by probing for possible front- and back - door intruder states . in summary , we conclude how it is primarily the choice of target state , and not the choice of parent state , which ultimately governs the convergence behavior of a given series . for example , restricting the target state to , say , triple or quadruple excitations might remove intruders present in series that target the full configuration interaction ( fci ) limit , such as the standard mp series . furthermore , we find that whereas a cc perturbation series might converge within standard correlation consistent basis sets , it may start to diverge whenever these become augmented by diffuse functions , similar to the mp case . however , unlike for the mp case , such potential divergences are not found to invalidate the practical use of the low - order corrections of the cc perturbation series .

1609.03945

in mller - plesset ( mp ) perturbation theory @xcite , the hartree - fock ( hf ) one - determinant wave function is perturbatively corrected for the effect of single ( s ) , double ( d ) , etc . , excitations out of the state ( up to the level of @xmath2 excitations , where @xmath2 is the number of electrons in the system at hand ) . the target energy for the mp series is thus that of the full configuration interaction ( fci ) model . the electronic hamiltonian is in mp theory partitioned into the fock operator , @xmath3 ( a zeroth - order part ) , and the fluctuation potential , @xmath4 ( the perturbation ) , and at lowest ( second ) order , the resulting mp2 model marks a successful approach to the correlation problem in quantum chemistry , providing a useful , size - extensive doubles correction to the hf energy at a low computational cost . the higher - order corrections of the mp series , however , represent less successful treatments of the electron correlation problem , in part due to the fact that these models offer significantly worse compromises between cost and accuracy than does the mp2 model , and in part because of convergence problems that reflect inherent problems with the actual partitioning of the hamiltonian . we will now briefly revisit the reason why the mp series might occasionally diverge . made possible by an efficient fci implementation @xcite , a study appearing some twenty years ago numerically showed how , upon augmenting standard correlation - consistent basis sets of double-@xmath5 quality by diffuse functions @xcite , the mp series was at the risk of becoming oscillating and , ultimately , divergent for a few prototypical , single - reference dominated examples @xcite . these results were in stark contrast to the behavior of the series within non - augmented standard basis sets , for which the convergence behavior of the mp series had previously been found to reflect the dominance of the reference state in the exact solution @xcite ( at least for mp expansions formulated upon restricted hf ( rhf ) references @xcite ) . thus , since the convergence behavior of the mp series was found to exhibit an extreme dependence on the choice of one - electron basis set , even for simple examples dominated by a single determinant , for which no _ a priori _ reason for expecting divergences seemed to exist , the usefulness of the higher - order models of the mp series was immediately brought into question , for total energies as well as properties derived from these @xcite . the numerical findings mentioned above were subsequently complemented by an explicit determination of the _ radius of convergence _ for the mp series @xcite . in that study , the electronic schrdinger equation was solved for an electronic hamiltonian expressed as @xmath6 , where @xmath7 is a complex perturbational strength parameter , such that the zeroth - order ( hf ) and physical ( fci ) wave functions were recovered by the values @xmath8 and @xmath9 , respectively . when regarded in this way , the mp series is recognized as a power series expansion in @xmath7 of the fci eigenvalues . this expansion will then be divergent if the ground state becomes degenerate with some excited state at a _ point of degeneracy _ , @xmath10 , within the complex unit circle defined by @xmath11 @xcite . in practice , the authors of ref . replaced the above search for branch points of the ground state energy function by a search for avoided crossings between the ground state and excited states on the real axis only ( i.e. , they searched for real - valued points of avoided crossings within the interval @xmath12 $ ] ) , since such a search is far less complex than a search within the entire unit circle while still enabling the identification of degeneracies . in this terminology , an excited state is denoted a _ back - door _ intruder if an avoided crossing is observed in the interval from @xmath13 , while it is denoted a _ front - door _ intruder if the crossing falls within the interval from @xmath14 . a few years later , the analysis in ref . was supplemented by a simple two - state model capable of explaining the divergence of the mp series @xcite . in that study , it was illustrated how divergences are bound to occur whenever a basis set is sufficiently flexible to give a reasonable description of highly excited and diffuse back - door intruders , which couple only weakly to the ground state ( and , as such , are nearly invisible in the energy spectrum at either @xmath8 @xmath15 @xmath16 ) , leading the authors to state that _ divergence is the rule rather than exception in mp theory , which converges only in small basis sets_. this point has since then been repeatedly stressed and extended in related studies @xcite . + ] @xmath17 ) and cc[@xmath18([@xmath19@xmath17 ) families of series . ] we have recently presented a range of perturbation series @xcite that are all based on a coupled cluster @xcite ( cc ) wave function rather than an hf wave function as the zeroth order state , while maintaining the mp partitioning of the hamiltonian . in the present study , we will extend the type of investigation outlined for the mp series above to these perturbative cc series as well as truncated mp series . our investigation is centered around the bivariational cc perturbation series introduced in ref . , while a few results for the recently introduced energy - based cc perturbation series @xcite will also be presented for comparison . for the sake of brevity , we will introduce a general notation , cc[@xmath20 , for the cc model resulting from a truncation of the cluster operator at a specific level , @xmath21 ( i.e. , cc[@xmath22 is the cc singles and doubles @xcite ( ccsd ) model , cc[@xmath23 is the cc singles , doubles , and triples @xcite ( ccsdt ) model , etc . ) . in particular , we want to compare the convergence radii of * ( i ) * a family of so - called hf([@xmath19@xmath17 ) expansions , in which the correlation energy of a target cc[@xmath19 model is expanded in the fluctuation potential , with * ( ii ) * the bivariational cc[@xmath18([@xmath19@xmath17 ) expansions @xcite , which describe the energetic difference between a parent model cc[@xmath18 and a target model cc[@xmath19 , again in orders of the fluctuation potential , cf . [ parent_target_state_overview_figure ] . thus , the hf([@xmath19@xmath17 ) series are mp - like expansions , which theoretically converge from the hf energy , not towards the exact fci energy , but rather towards the energy of a cc target model , e.g. , the target energy of the hf(sdtq@xmath17 ) series is the ccsdtq @xcite ( cc singles , doubles , triples , and quadruples ) energy . these series are thus equivalent to the standard mp series whenever the latter is restricted to at most @xmath24 $ ] excitations . as examples of the second type , the ccsd(t@xmath17 ) , ccsd(tq@xmath17 ) , and ccsdt(q@xmath17 ) series have all recently been proposed @xcite , forming order expansions that converge from either the ccsd or ccsdt energies onto the ccsdt or ccsdtq energies , subject to the same partitioning of the electronic hamiltonian as that used in mp theory , i.e. , @xmath25 . for both families of expansions ( based on either an hf or a cc reference state ) , we will here probe for possible back- and front - door intruders in the real - valued interval @xmath12 $ ] by searching for those @xmath7-values , at which the ground state becomes degenerate with an excited intruder state . we consider four examples , all of which have previously been shown to have slowly convergent or divergent mp series the ne atom , the f@xmath1 anion , and the singlet ch@xmath0 and hf molecules at equilibrium and distorted geometries , respectively @xcite . furthermore , numerical results for the cc - based perturbation series for different choices of parent and target models will be reported in order to assess to what degree potential divergences might influence these . in this respect , we wish to distinguish between two types of convergence , namely * ( i ) * formal convergence , i.e. , _ does a given expansion indeed converge ? _ , and * ( ii ) * practical convergence , i.e. , _ are the lowest - order corrections of a given series physically meaningful ? _ thus , while the formal convergence may be assessed by explicitly determining the radius of convergence , the practical convergence may be assessed simply by inspecting the lowest - order energy corrections . for this purpose , we have implemented arbitrary - order energy corrections for all members of the cc[@xmath18([@xmath19@xmath17 ) family of cc perturbation series . we will now outline the theory behind the intruder state scan and the computation of cc[@xmath18([@xmath19@xmath17 ) energy corrections to arbitrary order . in [ theory_formal_subsection ] , we discuss how the radii of convergence for the perturbation expansions are defined by the largest perturbation parameter for which the cc jacobian remains non - singular . in [ theory_scan_section ] , we use this insight to scrutinize the theoretical origin of possible intruder states from a comparison of the jacobians for hf- and cc - based perturbation series . finally , in [ theory_order_corrections_section ] , we review our implementation of general - order cc[@xmath18([@xmath19@xmath17 ) energy corrections . the original analysis of the convergence of mp perturbation theory discussed in [ intro_section ] may not directly be applied to the present cc context , as the energies and wave function parameters are now obtained from a non - linear set of equations , rather than as the solutions to a standard hermitian eigenvalue problem for the hamiltonian operator @xmath26 in cc theory , the energy is in general an analytic algebraic function of the cluster parameters , @xmath27 , and the perturbation strength , @xmath7 , i.e. , @xmath28 . in searching for intruder states , the task boils down to determining the range of @xmath7 within which the energy may be expressed in terms of an analytic function of @xmath7 , or , equivalently , as a convergent taylor series in @xmath7 . in turn , this range of @xmath7 is identical to the one where the cluster amplitudes may be formulated as analytic functions of @xmath7 . in order to actually determine this range , we write the non - linear set of equations that define the cluster amplitudes on the generic form @xmath29 with one equation for each individual cluster amplitude . assuming that for @xmath30 , the amplitudes @xmath31 satisfy the equations @xmath32 , the implicit function theorem @xcite will ensure that the amplitudes are indeed analytic functions of @xmath7 in a neighborhood of @xmath33 , provided that the cc jacobian , i.e. , the derivative of [ eq_cceqs ] with respect to @xmath27 , is non - singular in the point @xmath34 . by examining the jacobian as a function of @xmath7 , we may determine a value @xmath35 as that value of @xmath7 that has the smallest norm while still giving rise to a singular jacobian . for @xmath36 , the cluster amplitudes and hence the energy is thus analytic , implying that @xmath37 may be expanded as a convergent taylor expansion around @xmath38 . in standard cc response theory , the eigenvalues of the jacobian are identified as excitation energies , which implicates that a singularity in the jacobian corresponds to a degeneracy between the ground state and an excited state . thus , although the convergence criteria for the standard mp series and the present cc perturbation series are obtained in different manners , the radius of convergence is in both cases identified as the lowest value of @xmath39 at which such a degeneracy occurs . phrased differently , the determination of the radius of convergence requires the determination of a complex variable @xmath7 that produces a zero - eigenvalue of the jacobian . as discussed in ref . , the imaginary part of this _ branch point _ is in general small for back - door intruders , whereas it may be large for front - door intruders . in both cases , its approximate location may be determined from a scan over the eigenvalues of the jacobian for real values of @xmath7 , and on par with the procedure in refs . and , we will therefore only examine the lowest eigenvalue of the jacobian as a function of real values of @xmath7 . thus , instead of exact degeneracies , we will probe for avoided crossings where the real value of the lowest jacobian eigenvalue reaches a minimum . for a back - door intruder , the smallness of the imaginary part of the branch point will often lead to very pronounced avoided crossings with the lowest eigenvalue approaching zero , whereas for front - door intruders , the large imaginary part of the branch point tends to produce less pronounced avoided crossings . in general , the cc jacobian is a non - symmetric matrix . this fact leads to two main differences compared to the standard approach of fci , for which eigenvalues of an hermitian matrix are determined ( assuming that @xmath40 ) . first , a non - symmetric matrix needs not have a complete set of eigenvalues and eigenvectors . thus , whereas the singularity of the jacobian matrix at the branch point trivially corresponds to a zero eigenvalue , it will not be ensured that this eigenvalue changes into another very small eigenvalue for a value of @xmath7 infinitesimally close to the branch point . however , we note that such abrupt dissolutions of zero eigenvalues were never encountered in the course of the present study . second , the eigenvalues of a non - symmetric matrix may be complex . thus , whereas for the hermitian case , the occurrence of an intruder state lower in energy than the reference state is a clear sign of a singularity ( and hence a branch point ) , this may not need be the case for a non - hermitian problem . however , since complex eigenvalues of a real matrix always occur in pairs with identical real parts and opposite imaginary parts , and due to the analyticity of the eigenvalues of a general non - degenerate matrix as functions of its elements , a real eigenvalue may only turn into a complex eigenvalue at a branch point . the immediate consequence of this is the practical convenience of being able to determine branch points in much the same way for cc perturbation theory as was previously realized for mp theory in refs . and . in formulating either of the hf- or cc - based series discussed in [ intro_section ] as order expansions in a perturbational strength parameter , @xmath7 , we partition the electronic hamiltonian according to [ hamiltonian_scaled ] . next , we define the amplitude equations ( [ eq_cceqs ] ) needed for the evaluation of the energy corrections in the series ( @xmath41 and @xmath42 amplitudes , respectively , for the hf- and cc - based series ) [ ampl_eqs_collected ] @xmath43 in [ ampl_eqs_collected ] , the @xmath44 , @xmath45 , and @xmath46 cluster operators are all generically defined as @xmath47 ( with @xmath48 and @xmath49 being the amplitude and excitation operator , respectively , for excitation @xmath50 within the manifold at level @xmath51 ) , and the @xmath52 amplitudes are those of parent model cc[@xmath18 ( for an hf parent state , these trivially vanish ) . furthermore , the p- and q - spaces define the so - called primary and secondary ( complementary ) excitation manifolds , such that the @xmath42 and @xmath52 amplitudes have components within the primary space , while the @xmath41 and @xmath42 amplitudes both exist in the complementary space . for example , for the mp - like hf(sdt@xmath17 ) series , the p - space will be empty ( as is the case for all hf([@xmath19@xmath17 ) series ) and the q - space will be the complete manifold of all single , double , and triple excitations , while for the ccsd(t@xmath17 ) series , the p - space will contain all single and double excitations whereas the q - space will be restricted to the triples manifold only . a schematic illustration of some of these models is given in [ parent_target_state_overview_figure ] . since we will search for intruder states by probing for avoided crossings between the ground state and an excited state in the interval @xmath12 $ ] , the degeneracies will show up as zero - valued excitation energies . as mentioned in [ theory_formal_subsection ] , excitation energies are defined within cc response theory as eigenvalues of the cc jacobian , which may be derived as the derivative of the amplitude equations in [ ampl_eqs_collected ] with respect to either of the @xmath41 or @xmath42 sets of amplitudes . the elements of the jacobians for the hf([@xmath19@xmath17 ) and cc[@xmath18([@xmath19@xmath17 ) series are thus given as [ jacobians_collected ] @xmath53 & = \langle \mu_i | \exp{(-t(z))}[\hat{h}(z),\hat{\tau}_{\nu_j}]\exp{(t(z ) ) } | \text{hf } \rangle \label{hf_cc_jacobian } \\ \tilde{j}_{\mu_i,\nu_j}[z,\tilde{t}(z ) ] & = \langle \mu_i | \exp{(-\tilde{t}(z))}[\hat{h}(z),\hat{\tau}_{\nu_j}]\exp{(\tilde{t}(z ) ) } | \text{hf } \rangle \label{e_cc_jacobian}\end{aligned}\ ] ] where @xmath54 may refer to any excitation level up to level [ @xmath19 . as is evident upon comparing the two jacobians for the hf([@xmath19@xmath17 ) and cc[@xmath18([@xmath19@xmath17 ) series in [ jacobians_collected ] , these have the same structure , differing only in the amplitudes , @xmath41 and @xmath42 . using the baker - campbell - hausdorff expansion @xcite , the difference between the matrix elements @xmath55 and @xmath56 becomes @xmath57,t(z)-\tilde{t}(z ) ] + \ldots | \text{hf } \rangle \label{jacobian_difference}\end{aligned}\ ] ] which , in general , makes the eigenvalues of the two jacobians , i.e. , the excitation energies , differ if the @xmath41 or @xmath42 amplitudes themselves differ notably . for @xmath8 @xmath58 @xmath9 , however , we note how the excitation energies obtained from a diagonalization of [ hf_cc_jacobian ] and [ e_cc_jacobian ] will be identical ; for @xmath8 , both of the @xmath59 and @xmath60 matrices are diagonal with orbital energy differences along the diagonal , and for @xmath9 , both return the excitation energies of target model cc[@xmath19 . to a first approximation and for a general @xmath7 , the jacobians in [ jacobians_collected ] are both dominated by the common contribution @xmath61 = \langle \mu_i | [ \hat{f}+z{\hat{\phi}},\hat{\tau}_{\nu_j } ] | \text{hf } \rangle \label{jacobian_common}\end{aligned}\ ] ] which is independent of either set of amplitudes . in particular , we note from [ jacobian_common ] how for @xmath62 , the unphysical action of the fluctuation potential will be to attract , rather than repel the electrons @xcite . to see what ramifications this effect might have on the resulting excitation energies , we focus on the dominating diagonal elements of @xmath63 in [ jacobian_common ] , which may be written as @xmath64 & = \epsilon_{\mu_i } + z \big(\langle \mu_i | \hat{\phi } | \mu_i \rangle - \langle \text{hf } | \hat{\phi } | \text{hf } \rangle \big ) \nonumber \\ & = ( 1-z)\epsilon_{\mu_i } + z \big(\langle \mu_i | \hat{h}(1 ) | \mu_i \rangle - \langle \text{hf } | \hat{h}(1 ) | \text{hf } \rangle \big ) \label{jacobian_common_diag}\end{aligned}\ ] ] where @xmath65 is the ( positive ) difference in energy between the virtual and occupied orbitals of excitation @xmath50 . for the diagonal elements in [ jacobian_common_diag ] , the second term is the difference in energy between the determinant for the @xmath50-th excited state and that for the hf ground state . thus , @xmath66 will be always be positive , and it will be large for a system with a very diffuse excited state ( high excited state energy ) and/or for an electron - rich system with a dense hf ground state ( low hf ground state energy ) . hence , assuming that the eigenvalues of @xmath63 will be dominated by the diagonal elements in [ jacobian_common_diag ] , we may expect these to approach zero for general combinations of electron - rich ground states and diffuse excited states at @xmath62 , if the difference in electron repulsion energy counterbalances the orbital energy difference , i.e. , if the following condition is met @xmath67 where @xmath65 increases as well when the virtual states involved are located high up in the energy spectrum . we will test this hypothesis in [ num_results_section ] by calculating excitation energies , not only from the two jacobians in [ jacobians_collected ] , but also from the dominant contribution to these , i.e. , @xmath63 in [ jacobian_common ] . in passing , however , we note that the above reasoning is in line with that of ref . , in which the presence of intruder states was also linked to differences in first - order mp corrections for the ground state and diffuse excited states , i.e. , @xmath68 . in calculating energy corrections to arbitrary order , the simplest procedure is to evaluate these using the standard @xmath69 rule of rayleigh - schrdinger perturbation theory , which allows for the energy correction at order @xmath69 to be calculated from the @xmath17th - order amplitude corrections . for the mp series , the resulting recursive algorithm , as formulated within an fci program , has previously been described in , e.g. , refs . and , and will not be repeated here . for the cc[@xmath18([@xmath19@xmath17 ) family of perturbation series , on the other hand , which was originally derived using wigner s @xmath70 and @xmath71 rules , arbitrary - order corrections are here computed using the same @xmath69 rule as used for the mp series . furthermore , we will also report numbers for the so - called e - cc[@xmath18([@xmath19@xmath17 ) family of series @xcite , which too form order expansions of the difference in energy between a parent and a target cc model in orders of the fluctuation potential . for a given choice of cc[@xmath18 and cc[@xmath19 , the e - cc[@xmath18([@xmath19@xmath17 ) and cc[@xmath18([@xmath19@xmath17 ) series can both be derived from a bivariational energy lagrangian for target model cc[@xmath19 , by using information exclusively of the right - hand cc state ( e - cc[@xmath18([@xmath19@xmath17 ) ) or a combination of information on the right- and left - hand ( @xmath72 ) states ( cc[@xmath18([@xmath19@xmath17 ) ) of the parent cc[@xmath18 model . for the physical system ( @xmath73 ) , the @xmath17th - order amplitude corrections , from which energy corrections to order @xmath69 may be evaluated in either of the two series , are given as @xcite [ cc_n_order_ampl_eqs ] @xmath74 + \tfrac{1}{2 } [ [ \hat{\phi}^{{^{\ast}}\hat{t}},\delta\hat{t}],\delta\hat{t } ] + \ldots | \text{hf } \rangle \big)^{(n ) } \label{cc_n_order_ampl_p_space } \\ \delta t_{\mu_{\text{q}}}^{(n ) } & = - \epsilon_{\mu_{\text{q}}}^{-1 } \big(\langle \mu_{\text{q } } | \hat{\phi}^{{^{\ast}}\hat{t } } + [ \hat{\phi}^{{^{\ast}}\hat{t}},\delta\hat{t } ] + \tfrac{1}{2 } [ [ \hat{\phi}^{{^{\ast}}\hat{t}},\delta\hat{t}],\delta\hat{t } ] + \ldots | \text{hf } \rangle \big)^{(n ) } \ . \label{cc_n_order_ampl_q_space}\end{aligned}\ ] ] where @xmath75 is the cc[@xmath18 similarity - transformed fluctuation potential . for the sake of notational brevity , we have partitioned the correction amplitudes in [ cc_n_order_ampl_eqs ] into two components , i.e. , the corrections to the parent state amplitudes , @xmath52 , in the p - space ( @xmath76 $ ] ) and the full amplitudes in the q - space ( @xmath77 $ ] ) @xmath78}\delta\hat{t}_{i } + \sum_{j\in[\text{q}]}\delta\hat{t}_{j } = \delta\hat{t}_{\text{p } } + \delta\hat{t}_{\text{q } } \ .\end{aligned}\ ] ] in the cc[@xmath18([@xmath19@xmath17 ) and e - cc[@xmath18([@xmath19@xmath17 ) series , the difference in energy between the cc[@xmath18 and cc[@xmath19 states are next expanded in terms of the @xmath79 and @xmath80 corrections , respectively , according to the following relations [ cc_and_e_cc_series ] @xmath81 } & = e^{\text{cc}[m_{\text{p } } ] } + \sum_{n=2}^{\infty}e^{(n ) } \qquad ( \text{cc}[m_{\text{p}}]([m_{\text{q}}]\text{--}n ) ) \label{cc_series_1 } \\ e^{\text{cc}[m_{\text{q } } ] } & = e^{\text{cc}[m_{\text{p } } ] } + \sum_{n=3}^{\infty}\bar{e}^{(n ) } \qquad ( \text{e - cc}[m_{\text{p}}]([m_{\text{q}}]\text{--}n ) ) \label{e_cc_series_1}\end{aligned}\ ] ] where the @xmath17th - order energy corrections are defined as [ cc_and_e_cc_series_2 ] @xmath82 | \text{hf } \rangle + \tfrac{1}{2}\sum_{m=1}^{n-2 } \langle { ^{\ast}}\lambda | [ [ \hat{\phi}^{{^{\ast}}\hat{t}},\delta\hat{t}^{(m)}],\delta\hat{t}^{(n - m-1 ) } ] + \ldots | \text{hf } \rangle \label{cc_series_2 } \\ \bar{e}^{(n ) } = \langle \text{hf } | [ \hat{\phi}^{{^{\ast}}\hat{t}},\delta\hat{t}_{\text{p}}^{(n-1 ) } ] | \text{hf } \rangle + \tfrac{1}{2}\sum_{m=2}^{n-3 } \langle \text{hf } | [ [ \hat{\phi}^{{^{\ast}}\hat{t}},\delta\hat{t}_{\text{p}}^{(m)}],\delta\hat{t}_{\text{p}}^{(n - m-1 ) } ] | \text{hf } \rangle \ . \label{e_cc_series_2}\end{aligned}\ ] ] in [ cc_series_2 ] , @xmath83 where @xmath84}\sum_{\nu_{p}}{^{\ast}}\lambda_{\nu_{p}}\langle \text{hf } | \hat{\tau}^{\dagger}_{\nu_{p}}$ ] , and in [ e_cc_series_2 ] , the summation range for the second contribution is limited by the fact that amplitude corrections within the p - space vanish at first order , cf . [ cc_n_order_ampl_p_space ] . finally , we note that the expansion of the cluster amplitudes in orders of the mp fluctuation potential is the same in the e - cc[@xmath18([@xmath19@xmath17 ) and cc[@xmath18([@xmath19@xmath17 ) series , see [ cc_n_order_ampl_eqs ] , and the two series therefore share the same jacobian in [ e_cc_jacobian ] . hence , the e - cc[@xmath18([@xmath19@xmath17 ) and cc[@xmath18([@xmath19@xmath17 ) series have identical convergence radii , albeit significantly different rates of convergence , as exemplified by the comparison of the e - ccsd(t@xmath17 ) and ccsd(t@xmath17 ) triples series in ref . . this point will also be highlighted by the numerical results . + in the present section , we will report the position of the nearest avoided crossing along the real axis within ( or outside ) the unit circle , as calculated using either @xmath59 in [ hf_cc_jacobian ] for the hf([@xmath19@xmath17 ) series or @xmath85 in [ e_cc_jacobian ] for the ccsd([@xmath19@xmath17 ) series ( [ @xmath18 = 2 ) . as mentioned in the closing paragraph of [ intro_section ] , the investigations will be made for four prototypical closed - shell examples , all of which are known to have slowly convergent or even divergent mp series upon moving to higher orders in the perturbation . the four examples are : the ne atom in an aug - cc - pvdz basis @xcite ( [ neon_subsection ] ) , hf ( at a distorted geometry of twice the equilibrium bond length , @xmath86 pm ) and ch@xmath0 ( @xmath87 ) ( @xmath88 pm , @xmath89 ) in a cc - pvdz basis @xcite ( [ hf_dist_subsection ] and [ methylene_subsection ] , respectively ) , followed by the f@xmath1 anion in an aug - cc - pvdz basis ( [ f_anion_subsection ] ) . the frozen - core fci level is [ @xmath19 = 6 for ch@xmath0 and [ @xmath19 = 8 for ne , f@xmath1 , and hf . the main results are summarized in [ convergence_table ] , which reports whether a given series is convergent ( con . ) or divergent ( div . ) for a given truncation level , [ @xmath19 . furthermore , we will report total deviations from frozen - core fci results for all four systems . the mp and cc[@xmath18([@xmath19@xmath17 ) series start at second order , while for the e - cc[@xmath18([@xmath19@xmath17 ) series , the leading - order correction is of third order , cf . [ cc_and_e_cc_series_2 ] . all of the involved calculations have been performed using the program @xcite with verification of the low - order results for the simplest expansions done using the program @xcite . ) , ccsd(tq@xmath17 ) , ccsd(tq5@xmath17 ) , and ccsd(tq56@xmath17 ) series from the frozen - core / aug - cc - pvdz fci correlation energy for ne . ] in [ neon_figure ] , we consider the convergence of the neon atom for various perturbation series . in spite of having a substantial weight of the hf determinant in the fci state ( with a resulting homo - lumo energy gap in excess of 30 ev ) , the mp series for ne was previously in ref . found to be divergent in an aug - cc - pvdz basis , due to a weakly coupled back - door intruder at @xmath90 , cf . [ convergence_table ] . if this intruder is decomposed into contributions from individual excitation levels , it is observed ( in the mp series , i.e. , with [ @xmath19 = 8) to have more than @xmath91 of the weight of its wave function assigned to hextuple and higher - level ( seven- and eightfold ) excited determinants . thus , the intruder is clearly an unphysical and diffuse state , which will only appear in augmented basis sets that allow for such continuum states . however , by truncating either of the hf- and cc - based families of series at the level of hextuple excitations , the position of the avoided crossing is observed to move towards a lower @xmath7-value ( @xmath92 for [ @xmath19 @xmath93 ) , and by further truncating them at the level of pentuple ( or lower - level ) excitations , the crossing eventually falls outside the unit circle ( i.e. , @xmath94 for [ @xmath19 @xmath95 ) . thus , while the mp series does not converge for ne in an aug - cc - pvdz basis , for [ @xmath19 @xmath95 , any of its truncated hf([@xmath96@xmath17 ) series will , as will the corresponding cc[@xmath18([@xmath19@xmath17 ) ( and e - cc[@xmath18([@xmath19@xmath17 ) , cf . [ neon_cc_and_ecc_series_subsubsection ] ) series . the equivalence between the hf- and cc - based perturbation series in this case is further confirmed by calculations of excitation energies from the @xmath63 jacobian in [ jacobian_common ] , for which the crossing is again observed to enter the unit circle upon an inclusion of hextuple excitations in the total excitation manifold , i.e. , whenever [ @xmath97 \geq 6 $ ] . ) series from the frozen - core / aug - cc - pvdz fci correlation energy for ne . ] in the present section , we want to directly assess the effect of higher - level excitations and basis set diffuseness on the convergence behaviour of the cc[@xmath18([@xmath19@xmath17 ) series . first , in order to quantify that the description of hextuple excitations is indeed at the root of the divergence problems for ne in the aug - cc - pvdz basis , we probe for avoided crossings in the ccsdtq5(6@xmath17 ) series , that is , a series with a cc[5 ] parent state and a cc[6 ] target state , for which the difference in energy is a mere @xmath98 kcal / mol . these results are presented in [ neon_ccsdtq5_6_figure ] . in summary , an identical crossing with the same intruder state as in the case of the ccsd - based series in [ neon_figure ] is observed within the unit circle ( at @xmath92 ) , making the series divergent in spite of the minuscule magnitude of the energy difference which the series aims at expanding in orders of the perturbation . ) series from the frozen - core fci correlation energy in the aug - cc - pvdz and d - aug - cc - pvdz basis sets for ne . ] next , we assess whether other intruders but the hextuples - dominated state in the aug - cc - pvdz basis might enter the unit circle upon an augmentation of the basis by an additional set of diffuse functions . in [ neon_d_aug_ccpvdz_figure ] , the convergence of the ccsd(tq@xmath17 ) series is shown in the aug - cc - pvdz and d - aug - cc - pvdz basis sets . as is clear , the convergent behaviour of the series in the former of the two basis sets is deteriorated in the latter , as a new intruder dominated by quadruple excitations now enters the unit circle . thus , in general , a convergent hf- or cc - based series will likely become divergent if the excitation level of the target state is increased and/or the diffuseness of the basis set is increased . ] @xmath17 ) and e - ccsd([@xmath19@xmath17 ) series from the frozen - core / aug - cc - pvdz fci correlation energy for ne . ] as discussed in [ theory_order_corrections_section ] , the cc[@xmath18([@xmath19@xmath17 ) and e - cc[@xmath18([@xmath19@xmath17 ) series will exhibit identical convergence radii , due to their common jacobian in [ e_cc_jacobian ] . however , as initially shown in ref . through results for the lowest - order corrections of the ccsd(t@xmath17 ) and e - ccsd(t@xmath17 ) series , the rate of convergence towards the cc[@xmath19 target energy will be different . in [ neon_cc_and_ecc_series_figure ] , this is exemplified for ne not only for the ccsd(t@xmath17 ) and e - ccsd(t@xmath17 ) series through higher orders , but also for differing choices of cc[@xmath19 target states . from these results , we note how the oscillations through the lowest orders are considerably dampened in the ccsd([@xmath19@xmath17 ) series over the corresponding e - ccsd([@xmath19@xmath17 ) series , a direct consequence of the fact that the order expansions of the former family of series are markedly more balanced than those of the latter , cf . the discussions in [ theory_order_corrections_section ] and , in particular , ref . . thus , in terms of practical convergence , the cc[@xmath18([@xmath19@xmath17 ) variants will always be preferable . ) and ccsdt(q@xmath17 ) series from the frozen - core / cc - pvdz fci correlation energy for stretched hf . ] in ref . , the convergence of the mp series for hydrogen fluoride was investigated for combinations of two different geometries and two different basis sets . at the equilibrium geometry , the mp series was found to be rapidly convergent in the cc - pvdz basis set and excessively oscillating ( ultimately divergent ) in the aug - cc - pvdz basis . upon distorting the geometry , however , the mp series was found to diverge in the standard cc - pvdz basis as well , with an avoided crossing between the ground state and a back - door intruder well within the unit circle . ) , and ccsdt(q@xmath17 ) series from the frozen - core / cc - pvdz fci correlation energy for stretched hf . ] as opposed to the intruder state in the case of the neon atom ( [ neon_subsection ] ) , the back - door intruder for distorted hf is not dominated by an exorbitant amount of high - level excitations , but rather by a mixture of double , triple , and , in particular , quadruple excitations . in fact , the quadruples contribution to the description of the intruder state is so pronounced that if these are left out of the cluster operators in the cc - based series , the position of the avoided crossing is observed to shift from just within the unit circle , as for the ccsd(tq@xmath17 ) series in [ convergence_table ] and the ccsdt(q@xmath17 ) series in [ hf_ccpvdz_figure ] , to a value slightly below @xmath99 for the ccsd(t@xmath17 ) series , also shown in [ hf_ccpvdz_figure ] . thus , whereas both hf- and cc - based expansions are in general divergent for distorted hydrogen fluoride , it is possible to form truncated expansions that do converge by choosing a target state that does not include quadruple ( and higher - level ) excitations . this conclusion is thus equivalent to the one for the neon atom in [ neon_divergence_subsubsection ] , for which the omission of hextuple and higher - level excitations in the cluster operator leads to a convergent series . for both of the cc - based series in [ hf_ccpvdz_figure ] , however , a _ practical _ convergence towards the respective target energy ( e.g. , ccsdt or ccsdtq ) is observed , cf . [ hf_ccpvdz_2_figure ] . specifically , the divergence of the ccsdt(q@xmath17 ) series through the first 70 corrections is manifested only in very small energy oscillations of the order of 0.001 kcal / mol , which are negligible for all practical purposes and not even visible on the energy scale of [ hf_ccpvdz_2_figure ] . ) , ccsd(tq@xmath17 ) , and ccsd(tq56@xmath17 ) series from the frozen - core / cc - pvdz fci correlation energy for ch@xmath0 . ] turning next to the case of singlet methylene , we note how the radius of convergence of the mp series was previously found in ref . to be larger than @xmath100 in the cc - pvdz basis ( as well as in the aug - cc - pvdz basis ) , thereby validating a convergent mp series . however , an avoided crossing between the ground state and a low - lying doubly excited state of the same spin and spatial symmetry was found at a positive @xmath7-value of about @xmath101 ( that is , in the vicinity of , but still outside the boundary of the unit circle ) , which is confirmed for all of the tested perturbation series in [ convergence_table ] . in particular , the position of this crossing is not observed to shift significantly with changes in the truncation level , [ @xmath19 . however , despite the fact that this excited state does not represent a front - door intruder state , the presence of the avoided crossing near the boundary of the unit circle is seen in [ ch2_figure ] to cause a monotonic , but distinctly slow rate of convergence for all of the tested series , regardless of the choice of target state . in particular , we note how the shapes of the energy curves for singlet methylene in [ ch2_figure ] are completely different from the ones for ne and hf , cf . [ neon_figure ] and [ hf_ccpvdz_figure ] , differences which are intimately related to those of the energy profiles of front- and back - door intruders as described in , for instance , ref . . finally , we close the present section with the most challenging of the four examples studied here , namely the fluoride anion in an aug - cc - pvdz basis set , for which the mp series was previously found in refs . and to diverge from the very onset of the expansion ( from the mp3 model and onwards ) , essentially exploding up through higher orders . although f@xmath1 is isoelectronic with ne , and the weight of the hf determinant is approximately the same for the two species in the cc - pvdz and aug - cc - pvdz basis sets ( @xmath102 ) , being an anion , the need for augmented basis sets in the accurate description of f@xmath1 is of utmost importance . this is also recognized from the fact that the homo - lumo gap in f@xmath1 is considerably lowered upon augmenting the cc - pvdz basis by diffuse functions , from @xmath103 ev in the cc - pvdz basis to less than @xmath104 ev in the aug - cc - pvdz basis . in ref . , an avoided crossing between the ground state and a relatively strongly coupled back - door intruder was observed at @xmath105 for the mp series , and the position of the crossing closer to @xmath38as well as the larger coupling between the two states were used to explain the more rapid divergence for f@xmath1 over that for ne . ) , ccsd(t@xmath17 ) , and ccsdt(q@xmath17 ) series from the frozen - core / aug - cc - pvdz fci correlation energy for f@xmath1 . please note the difference in scales in the individual plots . ] in analogy with the studies of the neon atom in [ neon_subsection ] and distorted hf in [ hf_dist_subsection ] , the wave function for the intruder state may be decomposed into contributions from individual excitation levels . by doing so , we find that more than @xmath106 of the wave function consists of singly , doubly , and triply excited determinants . furthermore , for an hf(sd@xmath17 ) order expansion of the ccsd correlation energy , no back - door intruder is observed . collectively , this indicates how the inclusion of triple excitations is the primary cause for the divergence of the hf(sdt@xmath17 ) series , see [ f_anion_1_figure ] . however , in contrast to this result , we note from [ f_anion_1_figure ] how the ccsd(t@xmath17 ) triples series converges . this difference between the hf(sdt@xmath17 ) and ccsd(t@xmath17 ) series hence indicates that the choice of parent state may also impact the convergence behavior of cc perturbation series , despite being a minor overall factor compared to the influence of basis set diffuseness and level of target state . relating this back to the analysis in [ theory_scan_section ] , this implies that if pronounced differences exist between the amplitudes of @xmath107 and @xmath108 ( i.e. , the difference in [ jacobian_difference ] is substantial ) , this may too impact the radius of convergence . the actual convergence of the ccsd(t@xmath17 ) series for f@xmath1 , though , is observed to be highly irregular and almost pulse - like , as may be realized from the detailed view of the order corrections in [ f_anion_2_figure ] . for the sake of completeness , it should also be noted that the lowest jacobian eigenvalue obtained during the ccsd(t@xmath17 ) @xmath7-scan for f@xmath1 contained a very small imaginary component for @xmath7-values in the interval @xmath109 $ ] . however , based on the ccsd(t@xmath17 ) energy corrections in [ f_anion_1_figure ] and [ f_anion_2_figure ] alongside the fact that this imaginary component was found to be a full six orders of magnitude smaller than the ( positive ) real component , we conclude that the ccsd(t@xmath17 ) series for f@xmath1 is indeed convergent . it should also be mentioned that the @xmath7-scan of the jacobian for the ccsd(tq@xmath17 ) series was the only case in which the eigenvalue with the smallest real part had a large imaginary component . in this case , it is therefore not possible to use the @xmath7-scan on the real axis to determine whether the series is convergent or divergent . however , the perturbation expansion ( not shown ) was observed to diverge rapidly . ) series from the frozen - core / aug - cc - pvdz fci correlation energy for f@xmath1 . ] of all the examples considered in the present work , we only encountered a single example namely , the hf(sdt@xmath17 ) and ccsd(t@xmath17 ) series for f@xmath1for which the choice of parent state could be attributed as having an effect on the convergence behaviour , while the choice of target state is generally a much more important factor ( see [ neon_subsection ] and [ hf_dist_subsection ] ) . this contrast in significance is also illustrated by the fact that the ccsd(tq@xmath17 ) and ccsdt(q@xmath17 ) series for f@xmath1 both diverge , see [ convergence_table ] and [ f_anion_1_figure ] . it is , however , worth noting by pragmatically comparing the corrections of the various series in [ f_anion_1_figure]how the use of the lowest - order corrections of any of the cc - based series is not invalidated in the case of f@xmath1 , as opposed to the situation for hf - based analogues , for which only the second - order ( mp2 ) correction is typically considered meaningful . now , the special case of the convergent ccsd(t@xmath17 ) series aside , the results in [ f_anion_1_figure ] highlight how the majority of the problems for f@xmath1 originates from the underlying mp partitioning of the electronic hamiltonian used in both the hf- and cc - based series , that is , the division of @xmath110 into the zeroth - order fock operator and a sizeable perturbation in the form of the fluctuation potential . this is so because the combination of an electron - rich ground state and an exceedingly diffuse excited state ( archetypal for anions in augmented basis sets ) makes for dominant contributions to either of the @xmath59 and @xmath85 jacobians that induce avoided crossings between such states for negative values of @xmath7 . this notion is further substantiated by probing for negative excitation energies using the @xmath63 jacobian in [ jacobian_common ] , for which the same crossing as in the hf(sdt@xmath17 ) and ccsdt(q@xmath17 ) series is again observed . the radius of convergence of various hf- and cc - based mp perturbation expansions with different cc target states has been determined for a selection of prototypical closed - shell examples that are known from the literature to have slowly convergent or even divergent mp series . in particular , we have probed for the presence of potential intruder states in truncated mp series as well as in the recently proposed ccsd(t@xmath17 ) triples and ccsd(tq@xmath17)/ccsdt(q@xmath17 ) quadruples series , in turn by probing for zero- or negative - valued excitation energies as calculated from the cc jacobian for each of the series . the similarities between the intruder states encountered in hf- and cc - based perturbation theory have been interpreted in terms of structural similarities between the resulting jacobians for the different series . by focusing on the common and dominating contribution to these , we have been able to confirm and supplement the main conclusions of the original mp convergence study in ref . . we have detailed how perturbation theory formulated around an mp partitioning of the electronic hamiltonian , be that based on an hf or a cc ground state wave function , will be prone to back - door intruder states whenever a description of diffuse continuum states is possible , as , for instance , is the case when an augmented basis set is used or the target state contains highly excited determinants . thus , in general , a divergent hf- or cc - based series for a given combination of molecular structure and basis set can likely be made convergent by reducing the excitation level of the target state and/or reducing the diffuseness of the basis set . on the other hand , the choice of parent state has typically ( but not always ) no effect on the convergence behaviour . furthermore , we have reiterated the statement saying that back - door intruders are indeed prevalent for electron - rich systems with dense ground state wave functions . from a comparison of the jacobians of hf- and cc - based perturbation theory , both of these conclusions have been traced back to the fundamental mp partitioning of the hamiltonian , namely the fact that the perturbation within this framework , i.e. , the fluctuation potential , is substantial in size , in contrast with the general premise of perturbation theory where the perturbation in itself is assumed small . thus , we argue that the observed divergences are artefacts of the mp partitioning and , as such , inherent to any perturbation expansion that makes use of this . perhaps somewhat counterintuitive , we have also illustrated how the magnitude of the actual energy difference , which is expanded in orders of the perturbation in any given perturbation series , is completely irrelevant for the actual convergence behaviour of said series . however , divergences aside , the cc - based perturbation series have all been observed to behave significantly more stably than the mp series or any of its truncated variants . furthermore , we have found the bivariational ccsd(t@xmath17 ) and ccsdt(q@xmath17 ) series to be capable of remedying the disordered oscillations at lower orders in the perturbation which is traditionally one of the premier characteristics of divergent perturbation expansions even in cases where these series formally diverge at higher orders . as such , despite the fact that the present analysis has been limited to minimal systems ( atoms and diatomics ) , we find that even when potential divergences exist , as , for instance , in the case of the ccsdt(q@xmath17 ) series for stretched hydrogen fluoride and the fluoride anion , these are not found to invalidate the use of lower - order models such as , e.g. , the ccsdt(q3 ) and ccsdt(q4 ) models . this is indeed an important point to stress , as formal divergences remain the norm rather than the exception for perturbation expansions in orders of the mp fluctuation potential , irrespective of whether these are based on an uncorrelated hf state or a correlated cc state . j. j. e. , k. k. , and p. j. acknowledge support from the european research council under the european union s seventh framework programme ( fp/2007 - 2013)/erc grant agreement no . 291371 , d. a. m. acknowledges support from the us national science foundation ( nsf ) under grant number aci-1148125/1340293 and from the arnold and mabel beckman foundation as an arnold o. beckman postdoctoral fellow , and j. o. acknowledges support from the danish council for independent research , dff-4181 - 00537 . mller , c. ; plesset , m. s. _ phys . rev . _ * 1934 * , _ 46 _ , 618 shavitt , i. ; bartlett , r. j. _ many - body methods in chemistry and physics : many - body perturbation theory and coupled - cluster theory _ ; cambridge university press : cambridge , uk , 2009 helgaker , t. ; jrgensen , p. ; olsen , j. _ molecular electronic - structure theory _ , 1st ed . ; wiley & sons , ltd . : west sussex , uk , 2000 olsen , j. ; roos , b. o. ; jrgensen , p. ; jensen , h. j. a. _ j. chem . phys . _ * 1988 * , _ 89 _ , 2185 olsen , j. ; jrgensen , p. ; simons , j. _ chem . phys . lett . _ * 1990 * , _ 169 _ , 463 dunning jr . , t. h. _ j. chem . phys . _ * 1989 * , _ 90 _ , 1007 woon , d. e. ; dunning jr . , t. h. _ j. chem . phys . _ * 1995 * , _ 103 _ , 4572 olsen , j. ; christiansen , o. ; koch , h. ; jrgensen , p. _ j. chem . phys . _ * 1996 * , _ 105 _ , 5082 knowles , p. j. ; somasundram , k. ; handy , n. c. ; hirao , k. _ chem . phys . lett . _ * 1985 * , _ 113 _ , 8 handy , n. c. ; knowles , p. j. ; somasundram , k. _ theor . chim . acta _ * 1985 * , _ 68 _ , 87 gill , p. m. w. ; radom , l. _ chem . lett . _ * 1986 * , _ 132 _ , 16 nobes , r. h. ; pople , j. a. ; radom , l. ; handy , n. c. ; knowles , p. j. _ chem . phys . lett . _ * 1987 * , _ 138 _ , 481 halkier , a. ; larsen , h. ; olsen , j. ; jrgensen , p. _ j. chem . phys . _ * 1999 * , _ 110 _ , 7127 christiansen , o. ; olsen , j. ; jrgensen , p. ; koch , h. ; malmqvist , p .- . lett . _ * 1996 * , _ 261 _ , 369 kato , t. _ perturbation theory for linear operators _ ; springer - verlag : berlin , germany , 1966 schucan , t. h. ; weidenmuller , h. a. _ ann . ( ny ) _ * 1972 * , _ 73 _ , 108 olsen , j. ; jrgensen , p. ; helgaker , t. ; christiansen , o. _ j. chem _ * 2000 * , _ 112 _ , 9736 larsen , h. ; halkier , a. ; olsen , j. ; jrgensen , p. _ j. chem . _ * 2000 * , _ 112 _ , 1107 stillinger , f. h. _ j. chem _ * 2000 * , _ 112 _ , 9711 leininger , m. l. ; allen , w. d. ; schaefer iii , h. f. ; sherrill , c. d. _ j. chem . phys . _ * 2000 * , _ 112 _ , 9213 olsen , j. ; flscher , m. p. _ chem . _ * 2000 * , _ 326 _ , 225 sergeev , a. v. ; goodson , d. z. ; wheeler , s. e. ; allen , w. d. _ j. chem . phys . _ * 2005 * , _ 123 _ , 064105 sergeev , a. v. ; goodson , d. z. _ j. chem . phys . _ * 2006 * , _ 124 _ , 094111 herman , m. s. ; hagedorn , g. a. _ int . j. quantum chem . _ * 2009 * , _ 109 _ , 210 eriksen , j. j. ; kristensen , k. ; kjrgaard , t. ; jrgensen , p. ; gauss , j. _ j. chem . phys . _ * 2014 * , _ 140 _ , 064108 eriksen , j. j. ; jrgensen , p. ; olsen , j. ; gauss , j. _ j. chem . phys . _ * 2014 * , _ 140 _ , 174114 kristensen , k. ; eriksen , j. j. ; matthews , d. a. ; olsen , j. ; jrgensen , p. _ j. chem . * 2016 * , _ 144 _ , 064103ek , j. _ j. chem . phys . _ * 1966 * , _ 45 _ , 4256ek , j. _ adv . * 1969 * , _ 14 _ , 35 paldus , j. ; ek , j. ; shavitt , i. _ phys . a _ * 1972 * , _ 5 _ , 50 purvis , g. d. ; bartlett , r. j. _ j. chem . phys . _ * 1982 * , _ 76 _ , 1910 noga , j. ; bartlett , r. j. _ j. chem . phys . _ * 1987 * , _ 86 _ , 7041 scuseria , g. e. ; schaefer , h. f. _ chem . lett . _ * 1988 * , _ 152 _ , 382 oliphant , n. ; adamowicz , l. _ j. chem . phys . _ * 1991 * , _ 95 _ , 6645 kucharski , s. a. ; bartlett , r. j. _ j. chem . phys . _ * 1992 * , _ 97 _ , 4282 eriksen , j. j. ; jrgensen , p. ; gauss , j. _ j. chem . * 2015 * , _ 142 _ , 014102 eriksen , j. j. ; matthews , d. a. ; jrgensen , p. ; gauss , j. _ j. chem . phys . _ * 2016 * , _ 144 _ , 194102 eriksen , j. j. ; matthews , d. a. ; jrgensen , p. ; gauss , j. _ j. chem . * 2015 * , _ 143 _ , 041101 eriksen , j. j. ; matthews , d. a. ; jrgensen , p. ; gauss , j. _ j. chem . phys . _ * 2016 * , _ 144 _ , 194103 fritzsche , k. ; grauert , h. _ from holomorphic functions to complex manifolds _ , 1st ed . ; springer - verlag new york : new york , usa , 2002 ; p 34 merzbacher , e. _ quantum mechanics _ , 2nd ed . ; wiley : new york , usa , 1970 kendall , r. a. ; dunning jr . , t. h. ; harrison , r. j. _ j. chem . phys . _ * 1992 * , _ 96 _ , 6796 @xmath111 , a correlation program written by j. olsen with contributions from a. khn solomonik , e. ; matthews , d. ; hammond , j. r. ; stanton , j. f. ; demmel , j. _ j. parallel distrib . comput . _ * 2014 * , _ 74 _ , 3176

recent discoveries on topological characterization of gapless systems have attracted interest in both theoretical studies and experimental realizations . examples of such gapless topological phases are weyl semimetals , which exhibit 3d dirac cones ( weyl points ) , and nodal line semimetals , which are characterized by line nodes ( two bands touching along a line ) . inspired by our previous discoveries that the kicked harper model exhibits many fascinating features of floquet topological phases , in this manuscript we consider a generalization of the model , where two additional periodic system parameters are introduced into the hamiltonian to serve as artificial dimensions , so as to simulate ad system . we observe that by increasing the hopping strength and the kicking strength of the system , many new band touching points at floquet quasienergies and will start to appear . some of them are weyl points , while the others form line nodes in the parameter space . by taking open boundary conditions along the physical dimension , edge states analogues to fermi arcs in static weyl semimetal systems are observed . finally , by designing an adiabatic pumping scheme , the chirality of the weyl points and the berry phase around line nodes can be manifested .

1512.00202

one paradigm in studies of quantum chaos is the kicked harper model ( khm ) . it was originally introduced in ref . @xcite to model a kicked charge in a magnetic field . later a whole class of generalized versions of khm were introduced , which may also be realized via a periodically kicked charge in a magnetic field @xcite . remarkably , more than two decades ago , the khm was already used to study topological phase transitions in floquet states and to understand possible connections between large floquet band chern numbers , delocalization of floquet eigenstates , and the underlying classical chaos @xcite . @xcite were perhaps the first two studies illuminating on topological phases of matter in periodically driven systems , a topic of wide interest at present due to the finding of floquet topological insulators @xcite . studies of topological phase transitions in the khm opened up stimulating questions of interest to different communities , connecting concepts from different areas such as condensed matter physics and nonlinear dynamics . the relevance of the khm to studies of topological phases becomes even more evident in recent years . in particular , it was shown that the khm and a variant of another paradigm of quantum chaos , namely , the double - kicked rotor model ( dkr ) , have a certain kind of topological equivalence though their spectrum has many distinctively different features @xcite . furthermore , it was shown that the khm has drastically different edge states as compared with that associated with the dkr , even though all their bulk floquet bands have exactly the same chern numbers @xcite . this result is useful for ongoing investigations of the issue of bulk - edge correspondence in floquet topological states as well as the issue of weak topological insulators . in previous studies @xcite , we already showed that adding one periodic system parameter to the conventional khm hamiltonian can serve the purpose of introducing an artificial dimension , thus enabling us to study the properties of two - dimensional ( @xmath3d ) systems from the perspective of one - dimensional ( @xmath4d ) systems . it was found that the quasienergy spectrum of this effective @xmath3d khm is able to host as many @xmath3d dirac cones as possible at both quasienergy @xmath1 and quasienergy @xmath2 by simply increasing the kicking strength . this suggests that the khm might be a good platform to study gapless topological phases , such as weyl semimetal @xcite and nodal line semimetal @xcite in floquet systems . motivated by these results regarding the khm , in this paper we propose to study an extended khm by introducing two periodic system parameters into the hamiltonian . with two periodic system parameters available , together with the conserved quasimomentum variable due to the translational invariance in the sole physical ( lattice ) dimension , it is possible to obtain floquet bands that can be effectively associated with three - dimensional ( @xmath0d ) systems . this thus creates an interesting opportunity to explore the khm from a perspective related to @xmath0d gapless topological phases of matter , which has recently become an interesting topic @xcite . in order to make both periodic system parameters nontrivial , we propose to introduce modulations to both the lattice hopping term and the kicking potential in the khm . these modulations are characterized by two phase shift parameters playing the role of two periodic system parameters . the model hamiltonian can thus also be deemed as a kicked version of a recently proposed off - diagonal harper model @xcite . having two periodic system parameters leads to one direct advantage in understanding the implications of various topological phases for the dynamics . specifically , it is now possible to execute adiabatic loops in the @xmath3d parameter space spanned by the two phase shift parameters , by directly and slowly tuning the two system parameters . therefore , even for a single quasimomentum component of the system , it is possible to realize closed adiabatic loops to manifest the topological phases . the purpose of this paper is to present a variety of interesting and motivating results for the simplest two - band case of the above - mentioned extended khm . first of all , we observe the emergence of isolated @xmath0d dirac cones ( weyl points ) as well as line nodes ( two bands touching along a line ) in the floquet bands , at quasienergies @xmath1 and @xmath2 . this strongly suggests that even @xmath4d driven systems can be useful to study @xmath0d floquet topological phases . in particular , the emergence of weyl points and line nodes at quasienergy @xmath2 is a finding that is absent in static systems . secondly , by taking advantage of the simplicity of the two - band case , we also analytically show how it is possible to obtain as many weyl points and line nodes as possible in our extended khm . thirdly , when we take open boundary conditions ( obc ) along the only physical dimension , we observe a line connecting two weyl points with opposite chiralities , which is analogues to the so - called fermi arc @xcite in static weyl semimetal systems , at both quasienergies @xmath1 and @xmath2 . fourthly , we also perform dynamics studies and demonstrate how the chirality of the ( floquet ) weyl points and the @xmath2 berry phase around the ( floquet ) line nodes can be manifested in the adiabatic dynamics of the system . by considering an adiabatic loop with an appropriate initial state in the @xmath3d parameter space enclosing a weyl point , its chirality can be captured . furthermore , by adiabatically driving a floquet eigenstate at a fixed quasimomentum over one adiabatic cycle , the @xmath2 berry phase can be observed . these results should be of experimental interest , especially for waveguide and optical lattice realizations of @xmath4d driven systems . this paper is organized as follows . in sec . [ ii ] , we start by describing the hamiltonian studied in this paper , which is referred to as the off - diagonal modulated kicked harper model ( odkhm ) , and a possible experimental realization . by focusing on the situation where the floquet operator associated with this hamiltonian consists of only two bands , we analytically show the emergence of band touching points in a form of isolated weyl points or line nodes as the hopping strength and kicking strength are varied . in sec . [ iii ] , we present numerically the emergence of localized edge states associated with the weyl points and line nodes by taking obc along the physical or artificial dimension of the system . in sec . [ iv ] a , we design an adiabatic pumping scheme surrounding a weyl point and show how its chirality can be captured by choosing an appropriate initial state . we further show that by using the same adiabatic pumping scheme and initial state , the chirality of weyl points at quasienergy @xmath1 and quasienergy @xmath2 are captured in different manners , which may suggest the distinction between the two types of weyl point . in sec . [ iv ] b , we also show a possible way to capture the @xmath2 berry phase of the line node by adiabatically pumping a floquet eigenstate at a fixed value of its quasimomentum . finally , we summarize our results and present possible future directions in sec . we start by considering the following hamiltonian , @xmath5\ , |n+1\rangle \langle n |+h.c.\right\rbrace \\ & + \sum_{n=1}^{n-1}\sum_{j } v \cos(2\pi \beta_2 n+\phi_z ) \ , |n \rangle \langle n | \ , \delta(t - jt)\ ; , \end{split } \label{fund}\ ] ] where @xmath6 and @xmath7 are the parameters controlling the hopping strength , @xmath8 is the kicking strength , @xmath9 is the kicking period , @xmath10 is the lattice site index , and @xmath11 counts the total number of lattice sites . if @xmath12 , the above hamiltonian reduces to the conventional khm . when @xmath13 , however , eq . ( [ fund ] ) depends on two periodic parameters @xmath14 and @xmath15 , which can take any value in @xmath16 . physically , @xmath14 and @xmath15 represent the phase shift in the hopping and kicking strength modulations , which can be independently tuned . in particular , because of the extra parameter @xmath14 which modulates the hopping strength , eq . ( [ fund ] ) shall be referred to as the off - diagonal kicked harper model ( odkhm ) . due to the range of values they can take , the two parameters @xmath14 and @xmath15 can play the role of quasimomenta along perpendicular directions to simulate a @xmath0d system . by taking @xmath17 , eq . ( [ fund ] ) becomes , @xmath18\ , |n+1\rangle \langle n |+h.c.\right\rbrace \\ & + \sum_{n=1}^{n-1}\sum_{j } ( -1)^nv \cos(\phi_z ) \ , |n \rangle \langle n | \ , \delta(t - jt)\;. \end{split } \label{fund2}\ ] ] since eq . ( [ fund2 ] ) is time periodic , we can define a floquet operator as a one period time - evolution operator from @xmath19 to @xmath20 , which is given by , @xmath21 \ , |n+1\rangle \langle n |+h.c.\right\rbrace } e^{-\mathrm{i } \sum_{n=1}^{n-1 } ( -1)^n v \cos(\phi_z)\ , |n \rangle \langle n |}\ ; , \label{tri}\ ] ] where we have assumed @xmath22 for simplicity , and hence @xmath6 , @xmath7 , and @xmath8 are dimensionless quantities . under periodic boundary conditions ( pbc ) , eq . ( [ tri ] ) will be invariant under translation by two lattice sites . as a consequence , we can perform a fourier transform to express eq . ( [ tri ] ) in terms of quasimomentum @xmath23 . the reduced floquet operator can then be written as ( see appendix [ derive ] for details ) , @xmath24 } e^{-\mathrm{i}v \cos(\phi_z ) \sigma_z}\ ; , \label{mod}\ ] ] where it satisfies @xmath25 . in order to realize eq . ( [ tri ] ) experimentally , we could consider a two - stage process defined as follows in the framework of @xmath4d lattice . when @xmath26 , with @xmath27 , we turn off any hopping amplitude between two lattice sites while turning on the on - site potential of the lattice sites given by @xmath28 . when @xmath29 , we turn off all the on - site potentials , while turning on the hopping amplitude between two lattice sites given by @xmath30 . experimentally , this non - uniform hopping amplitude can be realized by using waveguide arrays in the framework of photonic quasicrystals setup @xcite . by diagonalizing eq . ( [ mod ] ) at given values of @xmath23 , @xmath14 , and @xmath15 and scan these parameters over the whole brillouin zone " , i.e. , @xmath31 , we will be able to get an effective quasienergy spectrum of eq . ( [ mod ] ) , where quasienergy @xmath32 is defined as the phase of the eigenvalue of eq . ( [ mod ] ) , satisfying @xmath33 with @xmath34 be an eigenvector of @xmath35 . ( [ spect1 ] ) shows the quasienergy spectrum at @xmath36 , @xmath37 , and @xmath38 . it suggests that the two bands touch at @xmath39 . furthermore , near each of these touching points , the quasienergy dispersion is linear with respect to any two parameters when the other parameter is fixed . in fact , a detailed calculation in the appendix [ exp ] shows that by expanding eq . ( [ mod ] ) around each of these touching points , the effective hamiltonian is found to be in the form of a weyl hamiltonian with chirality @xmath40 @xcite . , ( b ) @xmath41 , and ( c ) @xmath42 , under @xmath36 , @xmath37 , @xmath38 . here and in all other figures , plotted quantities are in dimensionless units . ] by increasing the value of parameters @xmath6 , @xmath7 , and @xmath8 , we found that more band touching points start to appear at both quasienergy @xmath1 and quasienergy @xmath43 , as illustrated in fig . ( [ spect2 ] ) , an observation which is absent in static systems such as the non - kicked version of the model @xcite . in order to characterize all the band touching points , we shall first introduce the following lemma , whose proof is provided in appendix [ proof ] . ( a ) , @xmath41 ( b ) , and @xmath42 ( c ) , under @xmath44 , @xmath45 , @xmath46 . ] let @xmath47 and @xmath48 be two perpendicular unit vectors , and @xmath49 and @xmath50 be functions of three periodic variables @xmath51 . the band touching points of the quasienergy spectrum of @xmath52 can only occur if @xmath49 and @xmath50 are integer multiples of @xmath2 . [ lem ] in order to use lemma [ lem ] , we need to first turn eq . ( [ mod ] ) into the following form , @xmath53 where @xmath54 , @xmath55 , and @xmath56 , @xmath57 , @xmath58 are unit vectors along @xmath59 , @xmath60 , and @xmath61 directions , respectively . since @xmath62 , lemma [ lem ] is applicable , and it suggests that the band touching points appear when @xmath63 and @xmath64 , where @xmath65 . all possible band touching points that appear in this model can be classified into three types . since the quasienergy is only defined modulo @xmath68 , eq . ( [ mod2 ] ) implies that if @xmath69 is even , the two bands touch at quasienergy @xmath1 , while if @xmath69 is odd , they touch at quasienergy @xmath2 . furthermore , @xmath66 implies that @xmath70 , and therefore @xmath71 and @xmath72 . in order for the solutions to exist , @xmath73 , so that @xmath74 . it is therefore clear from this analysis that new band touching points will start to appear as @xmath8 is increased by @xmath2 . therefore , by controlling the value of @xmath8 , we will be able to generate as many band touching points of this type as possible . by expanding eq . ( [ mod ] ) around one of these band touching points up to first order in @xmath23 , @xmath14 , and @xmath15 , we obtain the following , @xmath75\right\rbrace}\ ; , \label{ham1}\ ] ] where we have defined @xmath76 , @xmath77 , @xmath78 , with @xmath79 . the details of this calculation is presented in appendix [ exp ] . according to eq . ( [ ham1 ] ) , the effective hamiltonian @xmath80 associated with the floquet operator eq . ( [ mod ] ) , which is defined according to @xmath81 , is in the form of a weyl hamiltonian with chirality @xmath82 $ ] . the analysis therefore shows that this type of band touching points corresponds to a weyl point . in sec . [ iii ] a , we will show that under obc along the physical dimension , a line connecting two weyl points associated with this type of band touching points ( fermi arc ) will appear . by referring to the proof of lemma [ lem ] presented in appendix [ proof ] , it follows that the two bands touch at quasienergy @xmath2 when either @xmath85 or @xmath69 is odd and the other is even , and they touch at quasienergy @xmath1 when both @xmath85 and @xmath69 are even or odd . furthermore , similar to type @xmath4 , in order for the solutions to exist , @xmath73 , so that @xmath74 . however , since @xmath83 , @xmath23 and @xmath14 can in general take a continuous range of values satisfying @xmath86 . by treating @xmath14 as an independent variable , the two bands touch if @xmath23 satisfies the following condition , @xmath87\;. \label{line}\ ] ] since @xmath88 , a solution to @xmath23 exists for any @xmath89 . as a result , the two bands will touch along a curve as a function of @xmath14 and @xmath23 . fig . [ type23](b ) shows a typical example of the line nodes of this type at @xmath44 , @xmath90 , and @xmath38 , i.e. , @xmath91 and @xmath92 . by expanding eq . ( [ mod ] ) around @xmath93 , where @xmath94 , which corresponds to a point where the two bands touch , up to first order in @xmath23 , @xmath14 , and @xmath15 , the effective hamiltonian is found to be ( see appendix [ exp ] for the calculation details ) , @xmath95\hat{n}\cdot \sigma + \frac{2j k_x \sin(k_0')}{\phi(k_x , k_z)}\left[-2 \lambda k_y\sin(k_0')+\pi v k_z \right ] \sigma_y \ ; , \end{split } \label{lineff}\ ] ] where @xmath96 , @xmath77 , @xmath97 , @xmath98 , and @xmath99 . the quasienergy associated with this effective hamiltonian is @xmath100 $ ] . according to ref . @xcite , line nodes can be topologically characterized by a topological winding number , which is constructed as follows . on @xmath101 plane , the effective hamiltonian , as defined in type @xmath4 , depends only on two pauli matrices and can be written in the following form , @xmath102 define a complexified hamiltonian @xmath103 . the topological winding number associated with this line node is defined as , @xmath104 which can in general take values @xmath1 or @xmath105 ( where the sign depends on the direction of the line integral ) , and is closely related to the berry phase around a loop surrounding the line node @xcite . by considering a loop parametrized by @xmath106 according to @xmath107 and @xmath108 , it follows that @xmath109 . this suggests that as @xmath106 goes from @xmath1 to @xmath68 , @xmath110 goes from @xmath1 to @xmath111 , so that @xmath112 and @xmath113 . therefore , @xmath114= - 1\;,\ ] ] which shows that the line nodes of this type are topologically nontrivial . similar to type @xmath3 , the two bands touch at quasienergy @xmath1 when both @xmath85 and @xmath69 are even or odd , and at quasienergy @xmath2 when either @xmath85 or @xmath69 is even , while the other is odd . also , similar to both type @xmath4 and @xmath3 , in order for the solutions to exist , @xmath73 , so that @xmath74 . by treating @xmath14 as an independent variable , @xmath23 will need to satisfy eq . ( [ line ] ) in order for the two bands to touch . however , since a cosine function can only take values between @xmath116 and @xmath4 , a solution to @xmath23 will only exist at some values of @xmath14 in @xmath16 . if @xmath117 , a solution to @xmath23 exists as long as the numerator of eq . ( [ line ] ) is @xmath118 , or @xmath119\cup [ \phi_0 , \pi-\phi_0]$ ] , where @xmath120 . if @xmath121 , a solution to @xmath23 exists as long as the numerator of eq . ( [ line ] ) is @xmath122 . in both cases , the two bands will touch along a closed curve within @xmath23 and @xmath14 brillouin zone . as an illustration , fig . [ type23](a ) shows a typical line node of this type at @xmath44 , @xmath45 , and @xmath38 , i.e. , @xmath91 and @xmath92 . , @xmath38 , ( a ) @xmath45 and ( b ) @xmath90.,title="fig:",scaledwidth=50.0% ] , @xmath38 , ( a ) @xmath45 and ( b ) @xmath90.,title="fig:",scaledwidth=50.0% ] by expanding eq . ( [ mod ] ) around @xmath123 up to first order in @xmath23 , @xmath14 , and @xmath15 , i.e. , a point where the two bands touch , the effective hamiltonian is found to be , @xmath124\hat{m}\cdot \sigma -\frac{2\lambda k_y \sin(\phi_0)}{\phi'(k_y , k_z)}\left ( \pi v k_z+ 2 j k_x\right ) \sigma_x \ ; , \label{lineff2}\ ] ] where @xmath76 , @xmath125 , @xmath97 , @xmath126 , and @xmath127 . the quasienergy associated with this effective hamiltonian is @xmath128 $ ] . we can also calculate the topological winding number in a similar fashion as in type @xmath3 . on @xmath129 plane , the effective hamiltonian depends only on two pauli matrices which can be written as , @xmath130 as before , we define a complexified hamiltonian @xmath131 . by parametrizing @xmath132 and @xmath133 , it follows that @xmath134 . the topological winding number can then be shown to be @xmath135 , which shows that the line nodes of this type are also topologically nontrivial . one characterized signature of a topological phase is the existence of localized edge states when the system takes obc . for example , the band structure of a 2d quantum hall insulator supports localized chiral edge states which traverse the gap between two bands @xcite , while that of a spin hall insulator supports localized helical edge states @xcite . localized edge states can also appear in gapless systems , e.g. fermi arc in weyl semimetal @xcite and surface states in nodal line semimetal @xcite , which indicates that such systems can also be topologically characterized . we have shown in section [ ii ] that the odkhm hosts a number of weyl points and topologically non - trivial line nodes at quasienergy @xmath1 and quasienergy @xmath2 . in this section , we will investigate the emergence of localized edge states associated with these weyl points and line nodes . the results are hoped to motivate more studies of bulk - edge correspondence in floquet systems . [ 2 ] shows the quasienergy spectrum of the floquet operator eq . ( [ tri ] ) under obc with @xmath136 , @xmath37 , @xmath36 , and @xmath46 . under the given parameters , it confirms our analysis in section [ ii ] that there exists some weyl points at both quasienergy @xmath1 and quasienergy @xmath2 , corresponding to @xmath92 and @xmath137 . a slice of the full quasienergy spectrum at a fixed @xmath138 as shown in fig . [ 2](b ) shows the existence of a line connecting two weyl points with opposite chiralities , which is analogous to the fermi arc in static weyl semimetal systems . the emergence of this fermi arc shares a similar topological origin as the edge states of the su - schrieffer - heeger ( ssh ) model @xcite . at a fixed @xmath138 , the effective hamiltonian takes the form , @xmath139 which can be mapped to a static ssh model in the form of @xmath140 $ ] , where @xmath141 , by a suitable fourier transform . ssh model can be characterized by a @xmath142 topological invariant . it is topologically nontrivial when @xmath143 and @xmath144 are less than @xmath1 , while it is topologically trivial when @xmath143 and @xmath144 are greater than @xmath1 . in this case , the signs of @xmath143 and @xmath144 are controlled by the parameter @xmath14 . this explains why edge states exist only when @xmath145 , and thus appear to form a line connecting two weyl points at @xmath146 . ) under obc at a fixed ( a ) @xmath147 and ( b ) @xmath138 . red and green marks denote edge states localized at the right and left edges , respectively . the full quasienergy spectrum is presented in ( c ) . the parameters used are @xmath136 , @xmath37 , @xmath36 , and @xmath46.,title="fig:",scaledwidth=50.0% ] ) under obc at a fixed ( a ) @xmath147 and ( b ) @xmath138 . red and green marks denote edge states localized at the right and left edges , respectively . the full quasienergy spectrum is presented in ( c ) . the parameters used are @xmath136 , @xmath37 , @xmath36 , and @xmath46.,title="fig:",scaledwidth=60.0% ] fig . [ 2](a ) shows a slice of the full quasienergy spectrum at a fixed @xmath147 , where we observe counter - propagating chiral edge states joining the two bands . these counter - propagating edge states exist for any fixed value of @xmath14 in @xmath148 . when @xmath149 , the two gaps at quasienergy @xmath1 and @xmath2 close . by increasing @xmath150 further , the gaps reopen and the edge states disappear . furthermore , for any value of @xmath14 where the edge states appear , they cross at @xmath151\right\rbrace$ ] , where the effective hamiltonian has inversion symmetry and can be written in the form of eq . ( [ inv ] ) . these edge states crossings are what appears as a fermi arc joining two weyl points . [ 3 ] shows another quasienergy spectrum of eq . ( [ tri ] ) under obc with @xmath36 , @xmath152 , and @xmath38 , where the line nodes are predicted to appear . apart from the counter - propagating edge states and the fermi arcs we mentioned earlier , there does not seem to exist any additional edge states arising as a consequence of the line nodes . therefore , in order to be able to capture the topology of the line nodes , we take a look at the system under obc along an artificial dimension , which will be presented in section [ iii ] b. ) under obc at a fixed ( a ) @xmath153 and ( b ) @xmath41 . red and green marks denote edge states localized at the right and left edges , respectively . the full quasienergy spectrum is presented in ( c ) . the parameters used are @xmath136 , @xmath152 , @xmath36 , and @xmath38.,title="fig:",scaledwidth=50.0% ] ) under obc at a fixed ( a ) @xmath153 and ( b ) @xmath41 . red and green marks denote edge states localized at the right and left edges , respectively . the full quasienergy spectrum is presented in ( c ) . the parameters used are @xmath136 , @xmath152 , @xmath36 , and @xmath38.,title="fig:",scaledwidth=50.0% ] obc along an artificial dimension may be experimentally realized by decoupling the atomic internal degrees of freedom of the lattice that generate the phase shift @xmath14 or @xmath15 @xcite . mathematically , this can be done by performing a fourier transform to eq . ( [ mod ] ) with respect to @xmath14 or @xmath15 to get an effective lattice model , then taking obc along this dimension by considering a finite number of lattice sites . here , the fourier transformed lattice site corresponding to @xmath14 and @xmath15 can be imagined as @xmath60 and @xmath61 directions , respectively . we found that even under obc along @xmath60 or @xmath61 direction , there will still be no additional edge states associated with the line nodes . however , by opening the system along a diagonal direction in @xmath60 and @xmath61 , edge states associated with the line nodes become visible , as shown in fig . [ 4](c ) . to be more explicit , we define @xmath154 and @xmath155 . by performing a fourier transform with respect to @xmath156 , eq . ( [ mod ] ) can be transformed into the following form , @xmath157 } e^{-\mathrm{i } \sum_{n_y'=1}^{n_y'-1}\left[\frac{v}{2}e^{-\mathrm{i}\phi_z'}\sigma_z\otimes |n_y'+1 \rangle \langle n_y ' | + h.c.\right]}\ ; , \label{uy}\ ] ] where @xmath158 represents a lattice site along the diagonal direction . [ 4](c ) is then obtained by plotting the quasienergy spectrum of eq . ( [ uy ] ) with @xmath159 , @xmath36 , @xmath45 , and @xmath38 , i.e. , when line nodes of type @xmath0 are expected to appear . at a fixed @xmath42 , fig . [ 4](a ) shows the emergence of degenerate @xmath2-modes inside each line node . these degenerate @xmath2-modes are very similar to the fermi arc , since they correspond to the crossing between two edge states localized at opposite edges , which is illustrated in fig . [ 4](b ) . ) under obc at a fixed ( a ) @xmath42 and ( b ) @xmath160 . red and green marks denote edge states localized at the right and left edge , respectively . the full quasienergy spectrum is presented in ( c ) . the parameters used are @xmath159 , @xmath45 , @xmath36 , and @xmath38.,title="fig:",scaledwidth=50.0% ] ) under obc at a fixed ( a ) @xmath42 and ( b ) @xmath160 . red and green marks denote edge states localized at the right and left edge , respectively . the full quasienergy spectrum is presented in ( c ) . the parameters used are @xmath159 , @xmath45 , @xmath36 , and @xmath38.,title="fig:",scaledwidth=50.0% ] as another illustration , fig . [ 5](a ) and [ 5](b ) show the quasienergy spectrum of eq . ( [ uy ] ) under obc at another set of values of the parameters @xmath6 , @xmath7 , and @xmath8 , where line nodes of type @xmath3 are expected to appear , at a fixed @xmath161 and @xmath162 , respectively . the edge states appearing from a point on a line node and disappearing at another point on the same line node are visible , albeit not as clear as the emergence of edge states when line nodes of type @xmath0 appear . nevertheless , the existence of these edge states confirms the fact that both line nodes of type @xmath3 and type @xmath0 have topological origins , although this might be associated with a weak topological effect @xcite , as the emergence of edge states depends on the direction in which we cut " the system . ) under obc at a fixed ( a ) @xmath161 and ( b ) @xmath162 . red and green marks denote edge states localized at the right and left edge , respectively . the parameters used are @xmath159 , @xmath44 , @xmath163 , and @xmath164 . ] another way to capture the floquet band topology is through the quantization of particle transport by means of an adiabatic pumping protocol @xcite . in this section , we will show that the topology of the weyl points and line nodes appearing in the odkhm can be manifested by designing certain adiabatic pumping schemes , which are relatively easy to realize experimentally since they only involve tuning the experimental parameters @xmath14 and @xmath15 without the need to accelerate the particle by changing @xmath23 . according to ref . @xcite , the chirality characterizing a weyl point represents a source of berry curvature in quasimomentum space . it means that if we form a closed @xmath3d surface enclosing a weyl point in quasimomentum space parametrized by two parameters @xmath165 and @xmath166 , the chern number associated with the brillouin zone formed by @xmath165 and @xmath166 equals the chirality of the weyl point enclosed . therefore , in order to capture the chirality of a weyl point , we first prepare an initial state as a uniform superposition of the lower - band floquet eigenstates over all @xmath23 in the brillouin zone ( wannier state centered around @xmath167 ) @xcite . that is , @xmath168 where @xmath49 is the lattice constant . next , an adiabatic pumping protocol is realized by tuning @xmath14 and @xmath15 adiabatically according to @xmath169 $ ] and @xmath170 $ ] , where @xmath171 is a periodic function of time with a period much longer than the kicking period . in this work , we consider a discrete linear sweeping protocol by changing @xmath106 by @xmath172 every time step @xmath173 , i.e. , @xmath174 for @xmath175 and @xmath176 . our objective is to evaluate the shift in position expectation value after one adiabatic cycle @xmath177 . analytically , eq . ( [ chiral ] ) derived in appendix [ quantized ] implies the following two observations : 1 . the change in position expectation value of the wave packet ( @xmath178 ) after one adiabatic cycle is quantized by the total chirality of the weyl points enclosed . 2 . under the same initial state and adiabatic pumping scheme , the chirality of the weyl points at quasienergy @xmath2 and quasienergy @xmath1 contribute to @xmath179 in an opposite manner . the above observations can also be verified numerically , as shown in fig . [ pump](a ) is obtained by setting @xmath180 and @xmath181 so that the adiabatic loop does not enclose any weyl points , and @xmath179 after one adiabatic cycle is zero . in fig . [ pump](b ) , @xmath182 and @xmath181 so that the adiabatic loop encloses two weyl points at quasienergy @xmath1 at @xmath183 , and @xmath184 @xcite , which is equal to the total chirality of the weyl points enclosed . in fig . [ pump](c ) , @xmath185 , @xmath186 , and @xmath187 so that the adiabatic loop encloses two weyl points at quasienergy @xmath2 at @xmath188 , and @xmath189 , which is equal to the total chirality of the weyl points enclosed multiplied by minus one . an interesting result is observed in fig . [ pump](d ) . under @xmath185 , @xmath190 , and @xmath191 , the adiabatic loop encloses two weyl points at quasienergy @xmath1 at @xmath183 and two other weyl points at quasienergy @xmath2 at @xmath188 . although all the weyl points enclosed have the same chirality @xmath116 , @xmath192 rather than @xmath193 . this suggests that the adiabatic pumping scheme we considered responds to the chirality of the weyl points at quasienergy @xmath1 and quasienergy @xmath2 in the opposite manner . these numerical results agree with the two observations earlier based on analytical derivation . , ( c ) encloses two weyl points at quasienergy @xmath2 and ( d ) encloses two weyl points at quasienergy @xmath1 and two other weyl points at quasienergy @xmath2 . the parameters used are @xmath36 , @xmath37 , @xmath38 for ( a ) and ( b ) and @xmath46 for ( c ) and ( d ) . ] as suggested in sec . [ ii ] b , the line nodes appearing in the odkhm can be characterized by a topological winding number , which is closely related to the berry phase around a loop enclosing a line node . this @xmath2 berry phase can also be manifested via the following adiabatic pumping scheme . first , we prepare a floquet eigenstate at a fixed @xmath42 as an initial state . then , we drive it adiabatically along a closed loop in the artificial dimensions parametrized by @xmath106 according to @xmath194 $ ] and @xmath195 $ ] . [ piberry ] shows the berry phase at @xmath36 and @xmath38 as @xmath7 is varied from @xmath196 to @xmath197 . it shows that when @xmath198 , the berry phase is @xmath1 as the loop does not enclose any line node at these values of @xmath7 . when @xmath199 and therefore line nodes start to appear , the berry phase jumps to @xmath2 . we note also that the @xmath2 berry phase obtained above when @xmath199 is contributed only due to the singularity of the line node , i.e. , the point where the two quasienergies become degenerate , and hence due to its topology . this is because the contribution from the integral of the berry curvature over the area of the loop , excluding the singularity due to the line node , is @xmath1 . to show this , we start by rewriting eq . ( [ mod ] ) as follows , @xmath200 where @xmath201 and @xmath202 . we then apply a unitary transformation to eq . ( [ med ] ) to obtain , @xmath203 since @xmath204 and @xmath205 are related by a unitary transformation , they share the same eigenvalues , while the eigenvectors are related by @xmath206 furthermore , eq . ( [ med2 ] ) implies that @xmath207 . it therefore follows that if @xmath208 and @xmath209 are eigenvectors to @xmath205 that correspond to quasienergy @xmath210 and @xmath211 respectively , @xmath212 in terms of the eigenvectors of @xmath204 , @xmath213 therefore , the berry curvature at @xmath214 can be shown to be @xmath215 which is odd in @xmath216 and @xmath217 , and thus its integral over the area , excluding the singularity due to the line node , enclosed by the loop we defined earlier is @xmath1 . ) adiabatically driven along a closed loop in the artificial dimensions as a function of @xmath7 at @xmath36 and @xmath38 . ] in this study , we have shown that even by focusing only on the two - band case , the floquet states of the odkhm are able to host as many weyl points and line nodes as possible . by taking obc along the physical dimension of the system , dispersionless edge states in a form of a line connecting two weyl points with opposite chiralities are observed , which is analogue to the fermi arc in static systems . the emergence of these edge states can be connected with a type of @xmath4d topological invariant , such as that used to characterize the topology of the ssh model . although the edge states associated with the line nodes can not be observed when the system is open along its physical dimension , they become visible if instead we open the system along its artificial dimension . by preparing a wannier state centered around @xmath167 as the initial state , and driving it adiabatically along a closed loop in its artificial dimensions , which can be achieved by simply changing the experimental parameters @xmath14 and @xmath15 , the change in position expectation value @xmath179 of the wave packet is quantized by the total chirality of the weyl points enclosed . we further observed that under the same initial state and adiabatic pumping scheme , the chirality of the weyl points at quasienergy @xmath1 and quasienergy @xmath2 contributes to @xmath179 in an opposite manner . lastly , the @xmath2 berry phase of the line nodes is verified by adiabatically driving a floquet eigenstate along a closed loop in the artificial dimensions . this paper should therefore add new knowledge on the studies of the topological semimetal phases in floquet systems . following this study , there are many possible future directions which deserve further explorations . for example , given that the two - band odkhm has a very rich floquet band structure , it is expected that considering the odkhm with more than two bands may give rise to more interesting novel phenomena to observe . furthermore , we might also consider a continuous driven version of this model by replacing the kicking strength in eq . ( [ fund ] ) with a harmonic driving amplitude @xcite , which is more feasible to realize experimentally . in addition , it will be interesting to further explore the difference between weyl points at quasienergy @xmath1 and quasienergy @xmath2 , for example in terms of their edge states behaviour or chiral anomaly @xcite . lastly , since we can practically control the number of edge states in the two - band odkhm , it will be a good platform to study interaction effects and edge state transfer @xcite in floquet systems . * acknowledgements : * we thank hailong wang and dario poletti for helpful discussions . this work is supported by singapore ministry of education academic research fund tier i ( wbs no . r-144 - 000 - 353 - 112 ) . * appendix * this appendix has four sections . in appendix a , we present detailed derivations of eq . ( 4 ) in the main text . this is followed by appendix b , where lemma @xmath4 is proved . in appendix c , we derive the approximate floquet operator around different types of band touching points . in appendix d , we establish the relation between the quantized adiabatic pumping of wannier states and the chirality of the weyl nodes . whenever possible , we use the same notation as in the main text . define @xmath218 and @xmath219 . in terms of @xmath220 and @xmath221 , eq . ( [ tri ] ) can be written as , @xmath222 where , @xmath223|b_{n-1 } \rangle \langle a_n |+ \left[j-\lambda \cos(\phi_y)\right]|a_n \rangle \langle b_n | + h.c\right\rbrace , \\ h_\mathrm{on } & = & \sum_{n } \left [ v \cos(\phi_z ) |a_n\rangle \langle a_n | - v \cos(\phi_z ) |b_n\rangle \langle b_n |\right ] . \label{on}\end{aligned}\ ] ] next , introduce fourier transformed quantities , @xmath224 where @xmath225 is the lattice constant , which can be taken to be @xmath4 for simplicity . ( [ off ] ) and ( [ on ] ) become , @xmath226 \nonumber \\ & \hat{= } & \sum_k \left[\begin{array}{cc } 0 & 2 j \cos(k)-2\mathrm{i}\lambda \sin(k ) \cos(\phi_y ) \\ 2 j \cos(k)+2\mathrm{i}\lambda \sin(k ) \cos(\phi_y ) & 0 \\ \end{array}\right ] \otimes |k \rangle \langle k |\nonumber\\ & = & \sum_k \left(2 j \cos(k ) \sigma_x+2\lambda \sin(k)\cos(\phi_y ) \sigma_y \right)\otimes |k\rangle \langle k | \nonumber \\ & = & \sum_k h_\mathrm{off}(k ) \otimes |k \rangle \langle k | , \\ h_\mathrm{on}&= & \sum_{k } v \cos(\phi_z ) |a_k\rangle \langle a_k | - v \cos(\phi_z ) |b_k\rangle \langle b_k | \nonumber \\ & \hat{= } & \sum_k \left[\begin{array}{cc } v\cos(\phi_z ) & 0 \\ 0 & -v \cos(\phi_z ) \\ \end{array}\right ] \otimes |k\rangle \langle k | \nonumber \\ & = & \sum_k v \cos(\phi_z ) \sigma_z\otimes |k \rangle \langle k | \nonumber \\ & = & \sum_k h_\mathrm{on}(k ) \otimes |k\rangle \langle k |.\end{aligned}\ ] ] @xmath23 is therefore a conserved quantity , and eq . ( [ tri ] ) becomes , @xmath227\otimes |k\rangle \langle k | \nonumber \\ & = & \sum_k \mathcal{u}(k,\phi_y,\phi_z ) \otimes |k \rangle \langle k |,\end{aligned}\ ] ] where @xmath228 is given by eq . ( [ mod ] ) . we start by using euler s formula , @xmath229 \left[\cos(b)+\mathrm{i}\sin(b ) \vec{m}\cdot \sigma \right ] \\ & = & \cos(a)\cos(b)-\sin(a)\sin(b)\vec{n}\cdot\vec{m } + \mathrm{i } \left [ \sin(a)\cos(b ) \vec{n}\cdot \sigma\right . \\ & & \left.+\sin(b)\cos(a ) \vec{m}\cdot \sigma + \sin(a)\sin(b)\vec{m}\times\vec{n}\cdot \sigma \right ] \\ & = & e^{\mathrm{i}\theta \vec{r}\cdot \sigma } , \end{aligned}\ ] ] where @xmath230 is some unit vector and @xmath231 $ ] . since @xmath47 and @xmath48 are perpendicular , @xmath232 $ ] . band touching points correspond to @xmath106 being an integer multiple of @xmath2 . since cosine function is bounded in @xmath233 $ ] , this is only possible if both @xmath234 and @xmath235 are @xmath105 . this proves lemma [ lem ] . as explained in the main text , one of the band touching points occurs at @xmath236 , where @xmath79 . let @xmath76 , @xmath77 , and @xmath78 . ( [ mod ] ) then becomes , @xmath237\sigma_z } \;. \label{c1}\ ] ] the second exponential can be written as @xmath238 since @xmath239 commutes with itself . furthermore , according to euler s formula , @xmath240 , where the sign depends on whether @xmath69 is an even or odd integer . in both cases , the resulting exponential commutes with all pauli matrices , and eq . ( [ c1 ] ) becomes , @xmath241 } \\ & = & e^{-\mathrm{i } \left\lbrace l_2\pi- \left[2j k_x \sigma_x + 2\lambda k_y\sigma_y+v\sin(\phi_1)k_z\sigma_z\right]\right\rbrace } \\\end{aligned}\ ] ] since the derivation of the effective hamiltonian near a line node of type @xmath3 and @xmath0 is similar , we will present the derivation of eq . ( [ lineff ] ) only in this section . let @xmath96 , @xmath77 , and @xmath97 , where @xmath94 . ( [ mod ] ) then becomes , @xmath242 \sigma_x + 2\lambda k_y\sin(k_0')\sigma_y\right\rbrace } e^{\mathrm{i}v\sin(\phi_1)k_z\sigma_z } \;. \label{c2}\ ] ] by using euler s formula , eq . ( [ c2 ] ) can be written as , @xmath243\cos(vk_z)\right\rbrace\sigma_x\nonumber \\ & & + \frac{\sin(\theta)}{\theta}\left\lbrace 2\lambda k_y\sin(k_0 ' ) \cos(vk_z)+\left[2jk_x\sin(k_0')-\pi\right]\sin(vk_z)\right\rbrace \sigma_y\left.\right ] \nonumber \\ & = & e^{\mathrm{i}\frac{\theta'}{\sin(\theta')}\vec{r}\cdot \sigma } \nonumber \\ & = & e^{-\mathrm{i}h_\mathrm{eff } } , \end{aligned}\ ] ] where @xmath244 ^ 2 + 4\lambda^2k_y^2\sin^2(k_0')}$ ] and @xmath245 $ ] . up to first order in @xmath246 , @xmath216 , and @xmath217 , @xmath247 , @xmath248 , @xmath249 , and @xmath250 . it therefore follows that , @xmath251\left[-\frac{vk_z}{\phi(k_x , k_z)}\sigma_z+\frac{2jk_x\sin(k_0')}{\phi(k_x , k_z)}\sigma_x\right ] \nonumber \\ & & + \frac{2jk_x\sin(k_0')}{\phi(k_x , k_z)}\left[-2\lambda k_y\sin(k_0')+\pi vk_z\right]\sigma_y \nonumber \\ & = & \left[\pi-\phi(k_x , k_z)\right]\hat{n}\cdot\sigma + \frac{2jk_x\sin(k_0')}{\phi(k_x , k_z)}[-2\lambda k_y\sin(k_0')+\pi vk_z]\sigma_y , \end{aligned}\ ] ] as given in the main text . the derivation of eq . ( [ lineff2 ] ) follows exactly the same step as above with @xmath252 , @xmath253 , @xmath254 , and @xmath255 . the purpose of this section is to show how chirality emerges when an adiabatic pumping scheme elucidated in sec . [ iv ] a around a weyl point is performed . a typical effective hamiltonian around a weyl point in the odkhm takes the following form , since quasienergy is only defined mod @xmath68 , odd @xmath257 corresponds to a weyl point at quasienergy @xmath2 while even @xmath257 corresponds to a weyl point at quasienergy @xmath1 . by parametrizing @xmath132 and @xmath258 , the lower - band eigenstates as a function of @xmath259 associated with the above effective hamiltonian at quasienergy @xmath2 and @xmath1 respectively can be explicitly calculated in the @xmath260 representation as follow , @xmath261 and @xmath262 . by preparing an initial state in a form of eq . ( [ is ] ) with @xmath246 , @xmath216 , and @xmath217 replace @xmath23 , @xmath14 , and @xmath15 respectively , both the population weight of the initial state and the quasienergy spectrum of eq . ( [ weyl ] ) are invariant under reflection about @xmath263 . ref . @xcite then suggests that the shift in position expectation value after one adiabatic cycle is given by , @xmath264 \nonumber \\ & = & \frac{a}{2\pi } \int_{-\frac{\pi}{a}}^\frac{\pi}{a } dk_x \int_0^{2\pi } d\theta \frac{d\phi}{d\theta } \frac{d}{dk_x}\left[\frac{r_0 ^ 2}{2r_0 ^ 2 + 2v_x^2 k_x^2+(- ) 2v_x k_x \sqrt{r_0 ^ 2+v_x^2 k_x^2}}\right]\nonumber \\ & = & \frac{a}{2\pi } \int_0^{2\pi } d\theta \frac{d\phi}{d\theta } \left\lbrace \int_{-\frac{\pi}{a}}^\frac{\pi}{a } dk_x \frac{d}{dk_x}\left[\frac{r_0 ^ 2}{2r_0 ^ 2 + 2v_x^2 k_x^2+(- ) 2v_x k_x \sqrt{r_0 ^ 2+v_x^2 k_x^2}}\right ] \right\rbrace . \label{save}\end{aligned}\ ] ] let s first focus on the term inside the square bracket . since we can take @xmath230 as small as possible , and hence @xmath265 as small as possible , we can assume @xmath266 , so that upon performing the integration , @xmath267 & = & \frac{r_0 ^ 2}{2r_0 ^ 2 + 2v_x^2 \frac{\pi^2}{a^2}+(- ) 2v_x \frac{\pi}{a } \sqrt{r_0 ^ 2+v_x^2 \frac{\pi^2}{a}}}\\ & & -\frac{r_0 ^ 2}{2r_0 ^ 2 + 2v_x^2 \frac{\pi^2}{a^2}-(+ ) 2v_x \frac{\pi}{a } \sqrt{r_0 ^ 2+v_x^2 \frac{\pi^2}{a } } } \\ & \approx & -(+ ) \frac{4 r_0 ^ 2 \frac{v_x\pi}{a } \sqrt{r_0 ^ 2+v_x^2 \frac{\pi^2}{a}}}{\left(2r_0 ^ 2 + 2\frac{v_x^2\pi^2}{a}\right)^2 - 4\frac{v_x^2\pi^2}{a}\left(r_0 ^ 2+v_x^2 \frac{\pi^2}{a}\right ) } \\ & \approx & -(+ ) \mathrm{sgn}(v_x)\end{aligned}\ ] ] we have thus showed that the change in position expectation value after one adiabatic cycle is quantized by the lattice constant times the chirality of the weyl point enclosed . although in the derivation above we assumed that the radius @xmath230 of the loop in the artificial dimensions is very small , the above result holds for arbitrary @xmath230 or even arbitrary shape of the loop enclosing the weyl point , provided there are no other singularities inside the loop we constructed . this is because chirality can also be interpreted as the total berry flux passing through a given closed surface . in analogy to gauss law in electrostatics , the total flux should not depend on the shape of the closed surface . therefore , as long as the initial state fully populates all quasimomenta @xmath23 ( so that @xmath23 and @xmath106 form a closed surface ) and has a reflection symmetry about @xmath263 , the above result is expected to hold . 99 g. m. zaslavskii , m. yu . zakharov , and r. z. sagdeev , d. a. usikov and a. a. chernikov , sov . jetp * 64 * , 294 ( 1986 ) . i. dana , phys . a * 197 * , 413 ( 1995 ) . p. leboeuf , j. kurchan , m. feingold and d. p. arovas , * 65 * , 3076 ( 1990 ) . p. leboeuf , j. kurchan , m. feingold and d. p. arovas , chaos * 2 * , 125 ( 1992 ) . n. h. lindner , g. refael and v. galitski , nat . * 7 * , 490 - 495 ( 2011 ) . h. wang , d. y. h. ho , w. lawton , j. wang and j. b. gong , * 88 * , 052920 ( 2013 ) . d. y. h. ho and j. b. gong , * 90 * , 195419 ( 2014 ) . a. a. burkov , m. d. hook and l. balents , * 84 * , 235126 ( 2011 ) . x. wan , a. m. turner , a. vishwanath and s. y. savrasov , * 83 * , 205101 ( 2011 ) . p. hosur , s. a. parameswaran and a. vishwanath , * 108 * , 046602 ( 2012 ) . s. y. xu , c. liu , s. k. kushawa , r. sankar , j. w. krizan , i. belopolski , m. neupane , g. bian , n. alidoust , t. r. chang , h. t. jeng , c. y. huang , w. f. tsai , h. lin , p. p. shibayev , f. c. chou , r. j. cava and m. z. hasan , science * 347 * , 294 - 298 ( 2015 ) . m. phillips and v. aji , * 90 * , 115111 ( 2014 ) . k. mullen , b. uchoa and d. t. glatzhofer , * 115 * , 026403 ( 2015 ) . r. yu , h. weng , z. fang , x. dai and x. hu , * 115 * , 036807 ( 2015 ) . y. kim , b. j. wieder , c. l. kane and a. m. rappe , * 115 * 036806 ( 2015 ) . p. bonderson and c. nayak , * 87 * , 195451 ( 2013 ) . s. matsuura , p. y. chang , a. p. schnyder and s. ryu , new j. phys . * 15 * 065001 ( 2013 ) . s. ganeshan , k. sun and s. d. sarma , * 110 * , 180403 ( 2013 ) . m. verbin , o. zilberberg , y. e. kraus , y. lahini and y. silberberg , * 110 * , 076403 ( 2013 ) . p. hosur and x. qi , c. r. phys . * 14 * , 857 - 870 ( 2013 ) . s. ganeshan and s. d. sarma , * 91 * , 125438 ( 2015 ) . b. i. halperin , * 25 * , 2185 - 2190 ( 1982 ) . a. h. macdonald and p. streda , * 29 * , 1616 - 1619 ( 1984 ) . y. hatsugai , * 71 * , 3697 - 3700 ( 1993 ) . c. l. kane and e. j. mele , * 95 * , 226801 ( 2005 ) . b. a. bernevig , t. l. hughes and s. c. zhang , science * 314 * , 1757 ( 2006 ) . su , j. r. schrieffer and a. j. heeger , * 42 * , 1698 ( 1979 ) . a. celi , p. massignan , j. ruseckas , n. goldman , i. b. spielman , g. juzeliunas and m. lewenstein , * 112 * , 043001 ( 2014 ) . t. -s . zeng , c. wang and h. zhai , * 115 * , 095302 ( 2015 ) . k. kobayashi , t. ohtsuki and k. i. imura , * 110 * , 236803 ( 2013 ) . y. yoshimura , a. matsumoto , y. takane and k. i. imura , * 88 * , 045408 ( 2013 ) . h. guo , y. lin and s. -q . shen , * 90 * , 085413 ( 2014 ) . d. j. thouless , * 27 * , 6083 ( 1983 ) . d. y. h. ho and j. b. gong , * 109 * , 010601 ( 2012 ) . since the quasi - energy spectrum is only defined mod @xmath68 , it is not clear which band is lower or higher . here , by lower band i m referring to the values of quasi - energy that belong to the same equivalence class as quasi - energy smaller than @xmath1 . the factor @xmath3 in the denominator of @xmath269 comes from the fact that eq . ( [ tri ] ) is periodic by two lattice sites . the effective lattice constant of the system appearing in eq . ( [ chiral ] ) is therefore @xmath270 . l. zhou , h. wang , d. y. h. ho and j. b. gong , eur . . j. b * 87 * , 204 ( 2014 ) . a. a. burkov , j. phys . condens . matter * 27 * , 113201 ( 2015 ) . m. bello , c. e. creffield and g. platero , arxiv:1510.01379v1 ( 2015 ) . h. wang , l. zhou and j. b. gong , * 91 * , 085420 ( 2015 ) .

we study the origin of the ultraluminous x - ray source m82-x1 in the nearby starburst galaxy m82 . this x - ray source is of particular interest as it is currently the best candidate intermediate mass black hole ; it is associated with a 54mhz quasi periodic oscillations with a relatively low (kev ) black - body temperature . we perform detailed binary evolution calculations of 215 stars which transfer mass to a 1002000 black hole and present an empirical model for the x - ray characteristics expected for such binaries . based on the binary evolution calculations and the assumption in our simulations we conclude that the most likely candidate for the bright x - ray source m82-x1 is a 10 - 15 star near the end of its main - sequence or slightly evolved , which transfers mass to a black hole . we expect the system to be in the high / soft state . in that case the binary will not be visible as a source of gravitational wave radiation , but other transient x - ray binaries with lower mass donors way be rather bright sources of gravitational wave radiation . [ firstpage ] black hole physics binaries : close galaxies : individual : m82 x - ray : binaries x - rays : individual m82-x1

astro-ph0408402

black holes in the universe have either masses less than about 20 m@xmath2 ( orosz 2003 ) , or masses exceeding about @xmath3 ( see kormendy & richstone , 1995 , for some exceptions see fillipenko & ho 2003 ; greene & ho 2004 ) . the first category of black holes are typically the member of a binary system in a normal galactic environment , the other type of supermassive black holes can be very bright in x , but always live in the nuclei of galaxies . the discovery of bright x - ray sources which are not in the nuclei of galaxies has led to the suggestion that these sources may hide black holes , with a mass intermediate between 20 and @xmath3 ; we call them intermediate mass black holes ( imbh ) . bright non - nuclear x - ray sources have been known to exist since the 1980s ( watson , stanger & griffiths 1984 ) , recently with the advent of chandra and xmm these sources have attracted considerably more attention . the conventional definiton of an ultraluminous x - ray source is @xmath4 ergs / s , but for the purpose of this paper we consider only luminosities of @xmath5 ergs / s as true ulx . this definition corresponds to the eddington luminosity of an accreting compact object of about 80 times the chandrasekhar mass . the brightest galactic x - ray binary , grs 1915 + 105 , occasionally reaches @xmath6 ergs / s , and should then , according to the earlier definition be classified as a ulx . grs 1915 + 105 , however , contains a rather ordinary @xmath7 black hole ( harlaftis & greiner 2004 ) , which has an eddington limit of 3.5@xmath8 ergs / s if it accretes helium , and a factor of two lower if it accretes hydrogen rich material . we therefore argue that the term ultraluminous is inflationary , and opt for a more stringent definition ( see also irwin , bregman & athey 2004 ; irwin et al . 2004 ) . the definition with @xmath9erg / s is thus chosen because it is difficult to explain these sources with conventional x - ray binaries containing stellar mass black holes ( see also kalogera et al . 2004 ) . just the x - ray luminosity is not sufficient to identify a ulx as an imbh . a few studies have claimed spectral evidence for intermediate mass black holes , in the form of relatively low temperature accretion disks ( i.e. @xmath10@xmath11 kev ) which can not produce the observed high x - ray luminosities unless the accretion radius is quite large , and hence flowing into massive black holes ( miller et al . 2003a , b ) . on the other hand , such spectral results are generally based on fitting the disk emission ( shakura & sunayev 1973 ) using the models of mistuda et al . ( 1984 ) and gierlinski et al . ( 1999 ) including a power law , which presumably fits the compton up - scattered emission at high energies ( thorne & price 1975 ; shapiro , lightman & eardley 1976 ; sunyaev & trmper 1979 ) . inferring the black hole mass from the inner disk radius measured by spectral fits typically works well for sources whose x - ray fluxes are dominated by a quasi - thermal component , but less well for sources dominated by the power law component ( see e.g. sobczak et al . as the ulxs for the spectral mass measurement technique has been applied have spectra dominated by the power law component , these results should be taken with some caution . a small subset of extragalactic ulxs show properties that are unlike x - ray sources in the milky way . these objects provide some indication for the existence of either intermediate mass black holes or for unusual modes of accretion . many of these bright x - ray sources are variable , which helps provide confidence that the emission comes from a single accreting object , and some also show evidence of associations with counterparts in other wavelengths , such as star clusters ( zezas & fabbiano 2002 ; smith & wilson 2003 ) or optical emission nebulae ( shih , iwasawa & fabian 2003 ) or even bright stars ( liu , bregman , seitzer 2004 ) , which makes them unlikely to be background objects . there are a number of theoretical models for these sources which do not require an intermediate mass black hole , but rather use the fact that the eddington limit applies strictly to spherical accretion . the non - imbh explanations include mechanical beaming ( king et al . 2001 ) , relativistic beaming from a jet ( krding , falcke & markoff 2002 ) or photon bubble instabilities ( begelman 2002 ) ; many of these configurations can be obtained from a rather ordinary 10 black hole accreting from a donor during case b or c mass transfer ( podsiadlowski et al . a review of the theory of ulxs can be found in king ( 2003 ) . one particular source , m 82 x-1 , represents an especially compelling candidate for hosting an intermediate mass black hole ( ebisuzaki et al 2001 ; portegies zwart et al . it is located sufficiently close to a massive young star cluster that a chance superposition is unlikely , so it can be claimed to truly reside in m 82 with considerable confidence . it has an x - ray luminosity of about 5@xmath12 ergs / s ( i.e. well in excess of the luminosity of any galactic x - ray binary ) , and shows a 54 mhz quasi - periodic oscillation ( qpo ) as well as a broad iron line ( strohmayer & mushotzky 2003 ) . they argue that in a mechanical beaming scenario , the qpo should be broadened by scatterings which increase the photons light travel times , while the iron line should be narrow , since lines are typically broadened due to projected rotation effects . in a relativistic beaming scenario , neither qpos nor iron lines are expected . furthermore , they note that the observed qpo and noise spectrum are reminiscent of the galactic x - ray binaries in their very high state ( see van der klis 1995 for a review ) , except that the qpo frequency is a factor of about 10 - 100 lower than those seen in the galactic x - ray binaries . applying a simple linear mass scaling to the qpo frequency implies then a mass of about 100 - 1000 @xmath13 for the black hole . for a characterization of the bright x - ray sources it is important to identify models for their origin and evolution . for the remainder of the paper we adopt the intermediate mass black hole hypothesis and study the characteristics of an x - ray binary in which such a black hole accretes from a stellar companion . our assumptions are motivated by the compelling case of m 82 x-1 . the configuration in mind can be compared to low - mass x - ray binaries in which a main - sequence star or giant fills its roche - lobe to a black hole , in this case , however , the accreting object is a black hole o(1000 ) . in this paper we discuss the consequence of a @xmath14 black hole that accretes from a stellar companion . we envision that such systems may form in the dense cores for young star clusters . we discuss the formation of such systems in sect.[sect : fande ] . we further evolve the binary system through several phases of roche - lobe overflow ( rlof ) . for this purpose we perform detailed binary evolution calculations for such systems and demonstrate that if the binary can form with sufficiently small orbital period a stable low - mass x - ray binary forms in which the donor is slowly consumed by the black hole . the x - ray emission while the @xmath15 donor is on the main - sequence reaches luminosities above the eddington limit for a 1 accreting compact object . in the shorter post main - sequence phase less massive donors also exhibit super - eddington luminosities , while the higher mass donors may become ultraluminous for a few million years . for larger orbital periods , rlof starts while the donor is on the giant branch , which can lead to very high x - ray luminosities for a short while . intermediate mass black holes with a lower mass donor may contribute to the emission of gravitational waves in the lisa band and simultaneously be visible as bright x - ray sources . several theoretical models exist for producing black holes of intermediate mass , ( between @xmath16 ; portegies zwart & mcmillian 2002 ; miller & hamilton 2002 ; madau & rees 2001 ) . young and dense star clusters may experience an evolutionary phase in which the stellar density in the core becomes so high that physical collisions between stars become frequent ( quinland & shapiro 1987 ; portegies zwart et al 1999 ; rasio , freitag & grkan , 2003 ) . portegies zwart & mcmillan ( 2002 ) derive a mass accumulation rate due to collisions of @xmath17 , with @xmath18 the mass of the star cluster and its half - mass relaxation time . the requirement for this mass coagulation is that the initial relaxation time of the cluster is less than @xmath19myr , otherwise the most massive stars which drive the collision runaway experience supernovae before the star cluster experiences core collapse . if the cluster is born with sufficiently concentrated , a longer relaxation time up to about 100myr can still lead to a phase of collision runaway ( portegies zwart et al 2004 ) . the final mass of such a episode of coagulation is about 0.1 percent of the initial cluster mass portegies zwart & mcmillan ( 2002 , see also grkan et al 2004 ) . following this scenario , intermediate mass black holes populate the central regions of massive and dense star clusters . but the presence of an intermediate mass black hole is no guarantee for x - ray emission . the ambient gas density in such clusters is generally rather low and is insufficient to power an ultraluminous x - ray source . another possibility would be the tidal disruption of a main sequence star that smashes into the black hole or grazes it near the tidal radius . as a result , the star may be completely destroyed and accreted by the black hole . tidal disruption with a supermassive black hole leads to short flares ( lasting about a year , rees , 1988 ; ulmer , 1999 ; ayal , livio & piran , 2000 ) . for lower mass black holes the duration of accretion will longer inversely proportional with the mass of the compact object . from this point of view , such a scenario is consistent with the properties of ulx , but the event rate must be impractically high to explain the number of ultraluminous x - ray sources which have been observed , like m82-x1 . a more attractive model was recently proposed by hopman , portegies zwart & alexander ( 2004 ) . they argue that in a young and dense cluster a star may be captured by the intermediate mass black hole in a tidal event . further tidal interaction between the black hole and the captured star then circularize the orbit . they further assume that the captured star is on the main - sequence , but the same argumentation can be made for evolved stars . as a orbit has been fully circularized the captured star under - fills its roche - lobe only slightly ( hopman et al 2004 ) . during the remainder of the main - sequence lifetime of the captured star it grows in size by about a factor of two and gravitational wave radiation reduces the orbital separation . ultimately the star fills its roche lobe and mass transfer to the intermediate mass black hole starts . an alternative scenario for the formation of a binary is by the dynamical capture of a main - sequence star by the intermediate mass black hole . such events will typically result in rather wide ( 100 1000 day ) binaries ( baumgardt et al . 2004 ) . the companion to the intermediate mass black hole is then be among the most massive stars in the cluster . at young age ( @xmath20myr ) the companion and may well be a massive ( @xmath21 ) main - sequence star , but for older cluster the companion is more likely to be another black hole . the star clusters in which either tidal or dynamical capture can take place are very young ( @xmath22myr ) , extremely dense @xmath23pc@xmath24 and rich @xmath25 stars ( portegies zwart et al . the relatively recently identified class of young dense clusters ( sometimes pronounced as yodec ) tend to populate active star forming regions in interacting galaxies , like m82 and the antennae ( zezas & fabbiano 2002 ) . hopman et at . ( 2004 ) argue that in about half of the star clusters with are able to produce an intermediate mass black hole via collision runaway , the back hole is able to capture a companion star in a sufficiently short orbital period binary that roche - lobe overflow starts some time on the main sequence . together with the possibility of dynamical capture , wide binaries with an intermediate mass black hole may be comparatively common as short period binaries . we do not further consider the details of dynamical or tidal capture , nor do we assess the probability of forming the type of binary systems discussed in this paper , as we mainly concentrate on the phase in the lifetime of the intermediate mass black hole at which it is visible in x - rays or as a gravitational wave source . here we assume that an intermedate mass black hole was formed and has acquired a stellar companion via one of the mechanisms discussed above . mass transfer from a lower mass ( secondary ) star to a much more massive ( primary ) star is driven by the expansion of the donor and the loss of angular momentum from the binary system . in our evolutionary model , we initialize the binary systems with a black hole mass ( typically 1000 ) and a stellar mass secondary of 2 5 , 10 or 15with some selected orbital period at birth ( @xmath26 ) . apart from the previous model description in [ sect : formation ] we do not further discuss how such parameters are obtained . the evolution of each selected secondary mass is calculated with an initial orbital period such that roche - lobe overflow starts at zero - age , which could be representative if the system was formed via tidal caputre . this choice seems somewhat arbitrary , as it leaves no time for the imbh to form and to acquire roche - lobe contact ; to first order however , this is a good assumption as it takes less than about three million years to form the imbh ( portegies zwart et al 2004 ) , and the capture of a companion star followed by tidal secularization takes only a few million years ( hopman et al , 2004 ) . for our 2 donors this time lag is negligible , while for the 15 donor it is roughly half the main - sequence lifetime . we still argue , however , that the exact moment of rlof has little effect for the further evolution of the binary system , as long as it starts somewhere on the main - sequence . some binaries start roche - lobe overflow at a later evolutionary stage , on the giant or super giant branches . such larger initial orbital period could be representative in the case that the binary was formed by a dynamical capture . an overview of the selected initial conditions is presented in tab.[tab : models ] . once initialized , we evolve the binary through various stages using the binary evolution code of eggleton ( see pols et al . 1995 and references therein for a description ) , assuming a population i chemical composition ( y = 0.98 , z = 0.02 ) and mixing - length parameter of @xmath27 . convective overshooting is taken into account using an overshooting constant @xmath28 ( pols et al . 1998 ) . the orbital evolution of the system is affected by the emission of gravitational waves ( landau & lifshitz 1958 ) , roche - lobe overflow by non - conservative mass transfer or mass loss via a wind ( soberman et al . the evolution of the orbital separation , @xmath29 , can then be computed by the sum of three terms @xmath30}{1 + q } \frac{\dot{m}_{{\rm d}}}{m_{{\rm d } } } \nonumber \\ & & - 2 \frac{\dot{m}_{{\rm bh}}}{m_{{\rm bh } } } + \frac{\dot{m}_{{\rm t}}}{m_{{\rm t}}}. \label{doublens : eq : orbit } \end{aligned}\ ] ] here @xmath31 is the constant of gravity , @xmath32 is the speed of light in vacuum , @xmath33 and @xmath34 are the masses of the donor star and the black hole , respectively , @xmath35 and @xmath36 . the parameter @xmath37 gives the fraction of mass lost from the donor star in the form of fast isotropic wind ( de jager et al . 1988 ) and @xmath38 is the fraction of the wind mass accreted by the black hole . during the detached phase , we assume that @xmath39 and that the black hole does not accrete mass from the stellar wind ( @xmath40 ) . matter leaves the system carrying the specific angular momentum of the donor . during mass - transfer we assume that the black hole accretes matter up to its eddington limit ( see e.g. king 2000 ) as @xmath41 the remaining mass is lost from the system with the specific angular momentum of the black hole . a more sophisticated eddington mass accretion is given by podsiadlowski et al . ( 2003 , see their eq.9 ) , which depends on the hydrogen abundance of the donor and the initial and present mass of the black hole . because in our calculations @xmath42 , throughout the evolution the present mass of the black hole does not differ much from its initial mass , and therefore our adopted eddington limit is about a factor @xmath43 4.3 lower than that of podsiadlowski et al . as the consequence , the black holes in our case accrete less mass and the orbital period expands more strongly . table[tab : models ] gives an overview of the initial conditions and results of our calculations . in fig . [ fig : tmdotedd_zams ] we present the evolution of mass - transfer rate for our calculations where mass transfer started at the zero - age main sequence . binaries which start roche - lobe contact at the zams undergo case a mass transfer . this prolongs their main - sequence lifetime considerably . the most peculiar evolution in our sample is arguably that of the 2 donor . after transferring about 1 the star starts to develop a convective envelope which initiates chemical mixing , which again leads to magnetic activity . as a result orbital evolution will become dominated by magnetic braking . this mechanism ceases when the whole star becomes convective with a homogeneous composition of @xmath44 . some time later , the central temperature drops below the hydrogen burning limit and thermonuclear fusion stops : this happens at about @xmath45myr . at that point the system remains detached for about 8myr before it undergoes another phase of mass transfer . at @xmath46gyr the code breaks while the donor still fills its roche lobe ; at this point @xmath47 840 k and @xmath48 0.011 ( about 10 @xmath49 ) , which can be considered a brown dwarf . under the influence of gravitational - wave radiation alone , the donor spirals in toward the black hole until they merge after about 74 myr ( peters 1964 ) . for the more massive stars , case a mass transfer is followed by mass transfer when the star ascends the giant branch during hydrogen shell burning ( case ab ) , lasting for 8.3 myr ( for @xmath50 ) , 0.26myr ( for 10 ) and 0.05myr ( for @xmath51 ) . the mass - transfer rate increases dramatically in this phase to 5.2 @xmath52 , @xmath53 and 2.3 @xmath54 /yr for the 5 , 10 and 15 donor , respectively ; as shown by the first spike in the evolution in fig.[fig : tmdotedd_zams ] . these mass transfer rates result in x - ray luminosities in excess of @xmath55 erg / s . the 10 donor experiencing case a mass transfer ends as a helium star . this star experiences helium shell burning , lasting for about 40000 years , during which mass transfer starts again at a rate of @xmath56 /yr , regardless of the black - hole mass . these phases of mass transfer are shown by the second spikes in figs.[fig : tmdotedd_zams ] for 5 and 10 donors . though seemingly unresolved , some of these spikes last for several million years and are well resolved in our simulations ( see also tab.[tab : models ] ) . binaries initialized with a larger orbital period experience mass transfer when the donor is evolved to a giant ( case b ) or supergiant ( case c ) . the case b systems remain detached for only 0.29myr ( 5 ) 0.05myr ( 10 ) at mass - transfer rate of 2.4 @xmath57 ( 5 ) 7 @xmath54 /yr ( 10 ) . the 10 donors which experience case b mass transfer ascends the super giant branch at a later instant , leading to case bc mass transfer . the mass - transfer rate during this stage is of the same order as in the preceding phase . the evolution of the 10 donor in a @xmath58 day orbit around a 1000 black hole is quite peculiar ; mass transfer , in this system , ensues in a relatively advanced stage ( late case b ) such that only about 0.5 is transferred before the hydrogen shell burning is terminated , causing the star to shrinks inside its roche lobe . another 2 is transferred during case bc mass transfer before the donor turns into a helium star , which experiences helium shell burning , during which it expands quickly to overfill its roche lobe , initiating a short phases of high ( @xmath59 /yr ) mass transfer . also the 5 donor ultimately turns in a degenerate co core and experiences several carbon flashes during its remaining lifetime during which the mass transfer rate varies between @xmath60 /yr and @xmath59 /yr . only one of our simulations experienced case c mass transfer , which occurred for the 5 donor with a 600 day orbital period . this resulted in a short phase of mass - transfer at a rates of 1.4 @xmath57 /yr . only the binaries with a 10 donor for case b , and 15 for case a , leave neutron stars at the end of their evolution ; all other evolutions leave carbon - oxygen white dwarfs . [ cols="<,<,<,<,<,<,>,>,>,>,>,^,^ " , ] [ tab : models ] the durations of mass transfer are presented in columns 710 of tab.[tab : models ] . only if roche - lobe overflow occurs at an early stage , near the zams , can mass transfer be sustained at high rates for a long time . high accretion rates , however are only achieved on the giant ( case b ) and supergiant ( case c ) branches . we also find that mass transfer caused by helium shell burning can result in mass transfer rates as high as @xmath59 /yr , which is sufficient to power an @xmath9 erg / s x - ray source , but only for a short while . in this section we apply and discuss an empirical observational model based on our understanding of low - mass x - ray binaries in the galaxy , but then applied to a system where the donor is a 215 star and the accretor is an intermediate mass black hole . we then use the results of our binary evolution calculations as input for the empirical model . in appendix[sect : appendix ] we present the empirical evidence on which this model is based . the binary evolution simulations provide us with the donor mass , accretor mass , orbital period and a secular mass accretion rate , all as a function of time . we use these information to calculate the x - ray luminosity of the simulated binaries ( below ) ; the gravitational wave signatures are discussed in [ sect : gwr ] . our binary evolution calculations provide secular accretion rates , rather than instantaneously observable ones . this difference can be quite important , considering that a large fraction of the stallar mass accreting black holes are transients . since the theoretical understanding of disk instabilities is not presently well enough established to make quantitative predictions , we develop an _ ad hoc _ model based on phenomenology ( discussed in appendix[sect : appendix ] ) . the parameters we require in order to compare our simulations with observations are the recurrence time ( @xmath61 ) and the decay time ( @xmath62 ) , via which we calculate the time @xmath63 that the source is brighter than a certain limiting x - ray luminosity @xmath64 . we then compute the duty cycle ( @xmath65 ) . after applying the empirical model for x - ray emission , we study the observational characteristics of these sources . we require that to recognize the source as a ulx it has to be brighter than @xmath64 , for which we adopt @xmath66 erg / s . this is done in the following way : first we calculate the total energy from a fit to the empirical data in fig.[fig : porb_e ] and the decay time from fig.[fig : ltr ] . the recurrence time is subsequently a function of the total energy @xmath67 and the rate of mass transfer via , @xmath68 . the bright time is subsequently calculated by integrating the luminosity curve , for which we assume an exponential decaying function ( see appendix[sect : appendix ] ) from the start to the moment it drops below @xmath64 . this results in @xmath69 , where we adopt the factor two in the denominator as seems reasonable from fig.[fig : ltr ] . the duty cycle is subsequently determined as @xmath70 . in figures[fig : bright_time ] and[fig : duty_cycle ] we present the resulting bright time ( @xmath63 ) and duty cycle ( @xmath71 ) for a selection of binaries with 2 , 5and 10 donors which start rlof at the zams . the binaries which start at wider orbits and the simulations for a 15 donor which starts rlof at the zams are all persistent x - ray source with luminosities well above @xmath64 . the binaries with a relatively wide initial orbit , however , remain bright for only a short while . our empirical prescription is supported by the qualitative similarities with the proposed limit for transient and persistent sources of dubus et al . ( 1999 ) , which yields that binaries with accretion rates exceeding @xmath72 are persistent x - ray emitters . with this prescription , lower mass donors ( @xmath73 ) show transient behavior . we have analyzed the results of evolutionary calculations in which a donor star of 2 15 transfers mass to a 100 2000 black hole . in this section we discuss our simulation results with respect to the observed bright x - ray source m82-x1 in the starburst galaxy m82 . this source is one of the best candidate intermediate mass black holes , and we argue that such a black hole may produce x - rays as it receives mass from a companions star . binaries which start rlof at the zams remain bright only for a few weeks continuously . near the end of the main sequence the bright time increases to several years only for the 5 and 10 donors . our calculation for a 2 donor only experiences a short ( few days ) bright ( @xmath74 ) flash every few months . since the ultraluminous x - ray source m82-x1 was observed for an extended period ( over 20 years ) we conclude that it is unlikely to host a low mass ( @xmath75 ) donor . of course , it could be that the x - ray source was caught near peak luminosity each time it was observed , in which case a lower mass donor would prove satisfactory , but this is unlikely . our simulations for the 5 and 10 donor on the main - sequence exhibit similar episodes of bright time but with shorter duty cycle , making these sources still rather unlikely candidates for the ulx in m82 . however , near the end of the main sequence , @xmath63 increases considerable to several years . the binaries with a 10 donor then become steady sources , making these suitable candidates for the ulx in m82 . the 15 donor which starts rlof at the zams is also persistently bright . strictly speaking , the systems with an evolved donor are transients , but for as long as rlof is sustained the bright time and recurrence time become so long that these sources could rather easily be identified as persistent ultraluminous x - ray emitters . and though these sources are rather short lived they should not be excluded as suitable candidates ( see also kalogera et al 2004 ) . based on the above analysis and the figures[fig : bright_time ] and [ fig : duty_cycle ] we conclude that within the limitations of our model the most likely configuration for the bright x - ray source m82-x1 is a 1000 with a 10 - 15 donor , probably near the end of its main - sequence lifetime . note that for the star cluster mgg-11 this would also be consistent with the observed age of the cluster , being 7 - 12myr . the turn - off age of a 15 star is about 11.5myr . in a phase of mass transfer the stellar lifetime can be extended considerably possibly allowing even more massive donors ( see section[sect : binev ] ) . a binary emits gravitational waves , and as long as the stars are far apart we can assume that they behave as point masses . in the case we consider here , the donor star fills its roche - lobe and can hardly be described as a point mass . the amplitude of the gravitational wave signal is given by ( evens et al . 1987 ) @xmath76 here @xmath77 is the chirp mass . the gravitational wave frequency is twice the orbital frequency . the resulting gravitational wave signal is presented in fig.[fig : tmdotedd_gwr ] , for binaries which started roche - lobe overflow on the zero - age main - sequence . binaries which do not start rlof during the main - sequence phase of the donor will not become detectable gravitational wave sources . figure[fig : tmdotedd_gwr ] presents the evolution of the gravitational wave signal for several systems which start rlof at the zams . the last column of tab.[tab : models ] presents the time that the binary is visible above the noise curve of the lisa gravitational wave detector . assuming a standard distance of 10kpc , the binaries which start mass transfer at birth emit gravitational waves at a strain of about @xmath78 ( almost independent of the donor mass ) . the frequency of the gravitational waves is barely detectable for the _ lisa _ space based antennae ( see figure[fig : tmdotedd_gwr ] ) . during mass transfer the binary moves out of the detectable frequency band , as the orbital period increases . once the donors has turned into a white dwarf , as happens for the 5 and 10 donors , the emission of gravitational waves brings the two stars black into the relatively high frequency regime and it becomes detectable again for the _ lisa _ antennae . this process , however takes far longer than a hubble time , unless the orbit has an eccentricity @xmath79 . such high eccentricities , after the phase of mass transfer , can only be achieved in the cases where the initial donor collapses to a compact object in a supernova explosion , or if the binary is perturbed by external influences . we conclude that the here discussed class of binaries with relatively massive donors @xmath15 are probably not important sources of gravitational waves . the binary in which a 2 main - sequence star starts to transfer to a 1000 black hole , however , remains visible as a bright source of gravitational waves for its entire lifetime ( see figure[fig : tmdotedd_gwr ] ) . the reason for this striking result is the curious evolution of the donor ( see sect.[sect : binev ] ) . mass transfer in this evolutionary stage is rather slow causing the x - ray source to be transient . we argue that such a transient could result in an interesting synchronous detection of x - rays and gravitational waves . the known ulxs are at mpc distance and unlikely to be seen with lisa . we performed detailed simulations of the evolution of close binaries with an intermediate mass ( 1001000 ) black hole and a stellar companion . if the initial orbital period is of the order of a day , the main - sequence star starts filling its roche lobe before the terminal age . in this case the binary remains in contact for it s remaining main - sequence lifetime , which is considerably extended compared to the main - sequence lifetime of an isolated star of the same mass . relatively low mass ( @xmath80 ) donors which start mass transfer at zero age are likely to be transient , with high ( @xmath81erg / s ) peak luminosity . the recurrence time for these sources , however , is generally rather long ( several weeks ) and the duration of the bright phase is short ( few days ) . more massive donors are bright for a longer time and have shorter recurrence time . the sources become persistent for donor stars of about 10 . our simulation with a 15 zams donor resulted in a persistent bright x - ray source for its entire main - sequence lifetime . the duty cycle also increases near the end of the main - sequence . stars which start roche - lobe overflow after the main - sequence phase remain persistent for their mass transferring episode , which , for a 5 donor lasts for @xmath82 million years . based on these findings , and within the limitations of our analysis and assumptions , we argue that low mass ( @xmath80 ) donors and donors near the zero - age main - sequence are unlikely candidates for the ultraluminous x - ray source m82-x1 . a 1015 donor near the terminal - age main sequence or on the early giant branch produces recurrence times and luminosities consistent with the observed ultra luminous x - ray source in m82 . an interesting possibility is provided by a @xmath83 donor which starts to transfer mass to an intermediate mass black hole at birth . such a binary is likely to be visible as a bright transient x - ray source and simultaneously as gravitational wave source in the _ lisa _ band to a distance of several mpc . we are grateful to chris belczynski , clovis hopman , peter jonker , vicky kalogera , emmi meyer - hofmeister , onno pols , ed van den heuvel , lev yungelson , andrzej zdziarski , steve zepf and andreas zezas for discussions and for providing useful data . this work was supported by the royal netherlands academy of sciences ( knaw ) , netherlands research school for astronomy ( nova ) and the dutch organization for scientific research ( nwo ) for the spinoza grant 08 - 0 to e. p. j. van den heuvel and a talent fellowship . we are grateful to northwestern university for hospitality . in this appendix we discuss the characteristics of low - mass black hole x - ray binaries in the galaxy , in order to compare them with the x - ray binaries we have been simulating in the main body of the paper . the main reasons to use the galactic population of soft x - ray transients is because of the similarities in the geometry and composition of these systems with accreting intermediate mass black holes . most accreting black holes in the milky way are transient systems . in fact , only cygnus x-1 is a persistent emitter among the dynamically confirmed black hole candidates ( see mcclintock & remillard 2003 for a review ) , and the only other persistent emitters considered to be plausible black hole candidates are 4u 1957 + 115 , ss 433 and cygnus x-3 , along with the large magellanic cloud sources lmc x-1 and lmc x-3 . of these , only 4u 1957 + 115 has a low mass ( i.e. @xmath84 ) donor star , and the case for its being a black hole is by far the weakest ( see nowak & wilms 1999 ) . on the other hand , there are many persistent neutron star accretors . since standard accretion disk theory suggests that disk instabilities occur due to changes in the disk s hydrogen ionization state , we expect stable disks only when the disk is hot enough for its outer edge to be above the ionization temperature of hydrogen . given that more massive accretors have lower inner disk temperatures ( shakura & sunyaev 1973 ) and larger orbital separations ( and hence bigger disks ) for the same orbital period ( kepler 1609 ) , it is then not too surprising that neutron stars can more easily be persistent accretors than black holes . it thus becomes rather likely that intermediate mass black holes will be transient accretors ( see the discussion in dubus et al . 1999 ) . the most comprehensive study to date of the outburst properties of soft x - ray transients is that of chen , shrader & livio ( 1997 , hereafter csl97 ) . this work compiled the outburst durations , total fluencies , and peak luminosities for all the observed x - ray transient outbursts in the literature at the time of its writing , considering both optical and x - ray properties of the outbursts . we will borrow a few key results from that paper , while also making a few simplifications , to allow for a prescription for converting secular accretion rate into x - ray luminosities that is suitable for porting to accreting intermediate mass black holes . we will supplement these data with some more recent work , including that by garcia et al . ( 2003 ) who tabulated the peak outbursts luminosities of several x - ray transients , and other data sets which will be specified in more detail . the frequencies of outbursts are poorly constrained both by observations and by theory . a reasonable approximation , then , is to assume that the outbursts we see are `` typical '' , and hence that roughly periodic , identical outbursts will be seen for sources , with the periodicity and the properties of the outburst varying from source to source . to estimate the recurrence timescale ( @xmath61 ) , we compare the secular mass accretion rate from our binary evolution simulations ( [ sect : binev ] ) with an estimate of the total amount of mass accreted during an outburst based on the energy emitted during the outburst , @xmath67 , plus an estimate of the radiative efficiency . we then apply the observational characteristics to our binary evolution calculations , which is done in [ sect : observations ] . the first assumption we make is that the outburst follow a fast - rise , exponential decay ( fred ) luminosity law as a function of time . this approximation is supported by the shape of the light curves for about half the outbursts studied in chen , shrader & livio ( 1997 ) . the other half , and in particular grs 1915 + 105 the galactic system most analogous to a ulx , show rather `` flat - topped '' outburst light curves ( chen , shrader & livio 1997 ) . the next assumption is that the orbital period of a system is the key parameter in determining the properties of its outbursts . the radial size of the accretion disk is likely to be the actual driver of the outburst properties , but it is hard to measure . given simple scaling laws , though , and the assumption that the compact accretor dominates the mass of the binary system , which is here the case , it can be shown that the mass content of an accretion disk depends primarily on the orbital period , with a linear dependence on mass of the accretor . the outer radius of the disk , @xmath85 then scales with the orbital separation , @xmath29 , since @xmath86@xmath87 ; and the mass of the disc is proportional to @xmath88 ; then the mass of the accretion disk is proportional to the mass of the accretor . if we then scale all observational quantities in eddington units , the mass of the accretor has relatively weak effects on the properties of the outbursts . fig.[fig : ltr ] shows the peak luminosity at outbursts as a function of the orbital period for several black hole x - ray binaries ( data from garcia et al . there are relatively few systems , but the systems in short orbital periods ( i.e. less than about 10 hours ) fit approximately to the relation @xmath89 , with the sources above 10 hours fitting approximately to @xmath90 , which is also supported by theoretical studies , which suggests that the amount of mass accumulated in the disc in an outburst cycle is a function of the orbital period with a cutoff around 10 hours ( see meyer - hofmeister & meyer , 2000 ; meyer - hofmeister , 2004 ) . we note that the systems used to make the correlations between outburst properties and orbital periods appear to be prediminantly at most only mildly evolved , so that the applications of these results to highly evolved systems ( i.e. giant donors ) could be dangerous . the two outliers are v4641 sgr , at 67.6 hours orbital period , and @xmath91 and xte j 1118 + 480 , with @xmath92 hours and @xmath93 . since v4641 sgr shows a radio jet with an apparent proper motion of more than 10 times the speed of light ( hjellming et al . 1999 ; orosz et al . 2001 ) its jet must be pointed very close to the line of sight , and if its jet has a substantial x - ray spectral component , then this beamed component may be responsible for the strong x - ray emission , in contrast to what is expected for the other sources . the other outlier , xte j 1118 + 480 , is one of the x - ray transients that never reached a high luminosity state ( see e.g. brocksopp , bandyopadhyay & fender , 2004 , for a discussion of such sources ) . since these low luminosity systems are thought to be radiatively inefficient , it may be that the instantaneous mass transfer rate did reach levels comparable to the level of the correlation ( indicating that the disk outburst physics was essentially the same for this source as the others ) . alternatively , since the outbursts of individual recurrent x - ray transients are not all identical , it may simply be that we saw a fainter than typical outburst of xte j 1118 + 480 . given that the primary goal of this paper is to develop a model for _ bright _ x - ray binaries , this outlier is not particularly troublesome . next , we consider the outburst durations . taking the data from csl97 for black hole binaries outburst energies , we find a similar relation with orbital period , that @xmath94 , as shown in fig.[fig : porb_e ] . we note that the strong outlier is gro j 1655 - 40 , which has shown multiple outbursts with a short recurrence time , highlighting the limitations of the `` periodic outburst '' hypothesis , which we use here as a simplification . we also note that the highest energy , longest period system , grs 1915 + 105 has been in outburst continuously since about 1992 , but the data from chen et al . ( 1997 ) obviously covers only a fraction of that timespan . using the value from chen et al . ( 1997 ) , it falls a factor of about 3 below the relation we propose between orbital period and outburst energy . combining the two empirical relations from the data , with the added assumption of a `` fred '' lightcurve shape , we then find that the exponential decay timescale @xmath62 for an outburst should be approximately @xmath95 we emphasize that the prescription presented here is a crude approximation to the poorly understood phenomenology of transient disks ; however , we feel it is a better approximation that the standard ones made in converting secular accretion rates into instantaneous mass accretion rates in the x - ray producing regions of the accretion flow . the theoretical model recently proposed by meyer - hofmeister ( 2004 ) for outbursts also predict a correlation between orbital period and peak output luminosity . we note especially that the fred shape may be a poor approximation for the ulxs , considering that grs 1915 + 105 , the galactic system most analogous to the ulxs , has shown a rather `` flat - topped '' outburst light curve . based on this phenomenological analysis we conclude that the peak luminosity for a @xmath1black hole binaries easily exceed @xmath55 erg / s , but these high luminosities are generally sustained for a short while only ( see also [ sect : observations ] ) . next , we provide a brief review of spectral states of x - ray binaries , which will be used to justify bolometric corrections to the data . between three and five distinct spectral states exist for black hole x - ray binaries , depending on the classifying author . these states are essentially accretion modes with distinct spectral and variability characteristics . the state of an object is , to first order , correlated with its mass accretion rate , with often some hysteresis ( miyamoto & kitamoto , 1995 ; nowak , wilms & dove 2002 ; smith , heindl & swank 2002 ; maccarone & coppi 2003 ) . for a more detailed review , we refer to nowak ( 1995 ) which concentrates more on the x - ray spectroscopic features and van der klis ( 1995 ) which concentrates more on the variability features . the three canonical spectral states are the low / hard state ( typically seen above below 2% of the eddington luminosity maccarone 2003 ) , the high / soft state ( typically seen between about 2 and 30% of the eddington luminosity ) , and the very high state ( typically seen above about 30% of the eddington luminosity nowak 1995 ) . the two additional states are the quiescent state , which may just be the low luminosity extension of the low / hard state , and the intermediate or transition state , which shows spectral and timing properties essentially the same as the very high state , but occurs at lower luminosities , most notably as the source moves between the low / hard state and the high / soft state ( see homan et al . 2001 for a discussion of the intermediate state ) . active galactic nuclei seem to show similar spectral states as black hole binaries at the same fractions of the eddington luminosity ( maccarone , gallo & fender 2003 ) , so the extension of these results to intermediate mass black holes seems to be a reasonable assumption . for illustrative purposes we present figure[fig : hardsoft ] , which shows the x - ray spectrum of cygnus x-1 in the soft state ( top solid curve for low @xmath22kev and high energies @xmath96kev ) and in the hard state ( bottom curve in the _ chandra _ band ) . the dashed curves are corrected for absorption . the two lowest luminosity states are the quiescent state and the low / hard state . no clear indications exist that there are fundamental differences in the accretion mode between these two states , and it has been suggested that the quiescent state may be merely the lowest luminosity low / hard state . the x - ray spectra of low / hard states are typically well fit by thermal comptonization models with photon indices , @xmath97 ( defined such that the number of photons per unit energy , @xmath98 of 1.52.0 and roughly exponential cutoffs at about 200 kev . the quiescent state objects are typically so faint that their spectral states can not be well constrained but they seem to have power law slopes consistent with those of the low / hard state objects ( kong et al . additionally , since their spectra have been measured only with soft x - ray imaging telescopes , no spectral cutoffs have been seen from the quiescent black holes . taking typical black hole low / hard state spectral parameters ( e.g. nowak et al . 2002 ; mcclintock et al . 2001 ) the bolometric corrections from the _ chandra _ band are factors of @xmath43 515 . if a x - ray source is in the low / hard state an observed luminosity of @xmath99ergs@xmath100 in the _ chandra _ bandpass indicates that the total luminosity should exceed @xmath101ergs@xmath100 ; the black hole then should exceed about 2000 in order for this luminosity to be less than 2% of the eddington luminosity . another key issue for converting mass accretion rate into observed luminosity for the low luminosity states is the radiative efficiency correction . geometrically thin accretion disks are generally believed to be radiatively efficient . defining the radiative efficiency @xmath102 , it is generally found that @xmath103 for non - rotating black holes accretion disks , and a factor of @xmath104 higher for disks around maximally rotating kerr black holes . on the other hand , thick accretion disks , such as those thought to be responsible for low / hard states , may be quite radiatively inefficient . in the context of advection dominated accretion disks , it is found that @xmath105 ( esin , mcclintock & narayan 1997 ) , where @xmath106 is the mass accretion rate at the luminosity of the state transition between the low / hard and high / soft states . the correlation between radio and x - ray luminosity in low / hard state black hole binaries ( gallo , fender & pooley 2003 ) has been interpreted that the radio jet takes kinetic power from the accretion flow in such a manner as to also produce a @xmath107@xmath108 ( fender , gallo & jonker 2003 ) , although the coefficient of proportionality was not well constrained by the data . the other two ( or three ) states all have similar spectral properties . they show dominant strong thermal emission at low energies , consistent with a standard thermal accretion disk spectrum with a temperature of @xmath109 kev at the inner edge , and a weaker power law component with a spectral index of @xmath110 and typically no observable cutoff . the very high state and the intermediate state typically show a slightly higher fraction of their luminosity coming out in the power law component , but in reality the difference between the two states is best seen from the variability properties , as a strong @xmath111 hz qpo is seen in the very high state , while the high / soft state typically shows root - mean square fractional variability less than 5% . we take typical source parameters from miller et al . ( 2001 ) who presented tabulated fits to the rxte spectra for an entire outburst of the x - ray binary xte j1748 - 288 . we then impose a low energy cutoff to the power law at 2 kev , where the flux is dominated by the thermal component , in order to keep the spectral model from diverging at low energies ( the exact location of this cutoff is unimportant so long as it is in the photon energy range where the luminosity is dominated by the thermal component ) . we then compare the bolometric luminosity from the spectral model to the bolometric luminosity in the _ chandra _ bandpass and find that about 75% of the luminosity is within the _ chandra _ bandpass for the very high state and about 90% for the high / soft state . bolometric corrections are therefore not a major source of uncertainty in converting between accretion rates and luminosity for the high luminosity sources . additionally , we wish to take the opportunity to make a few points about the spectral states of x - ray binaries that may help explain certain phenomenology of ulxs , and especially some points which seem to contradict this picture , but in fact do not . in particular , there has been discussion of low / soft and high / hard spectral state phenomenology . we note first that with the _ chandra _ bandpass , a low / hard state will usually fit to a steeper power law spectral index than will a high / soft state ( e.g. liu et al . this is because the peak of a 1 - 2 kev disk blackbody spectrum in power per decade will be at about 3 - 6 kev . the exponential cutoff of the spectrum will be in the part of the _ chandra _ bandpass where the effective area is relatively low ; with a few hundred total photons , as is typical for ulx studies , the statistical weight of the spectral fit will be dominated by the softer x - rays , where the spectrum is quite flat . with very high signal - to - noise spectra , where one can actually rule out a power law model for the high / soft states or a disk blackbody model for the low / hard states ; only then is the spectral state identification useful for comparison with the black hole binaries . we note that kubota , done & makishima ( 2002 ) have also pointed out that ulx spectra should not be compared with the low / hard states of accreting black holes , but rather with the more high luminosity states . a second key related point is that in the high / soft state , the accretion disk seems to extend in to a constant inner radius ( see e.g. sobczak et al . 2000 ) , which is consistent with being the marginally stable orbit around the black hole . the luminosity variations are then determined almost entirely by changes in the disk temperature . in the very high state , the disk may change size a bit more , but still the luminosity rises are correlated primarily with increases in the temperature of the disk and increases in the ratio of the power law component to the blackbody component . that is to say , even in normal x - ray binaries , at high x - ray luminosities , the spectra frequently harden as the luminosity rises , but without the qualitative change in behavior associated with state transitions . some recent work on bright well - studied ulxs show slight deviations from the simplistic picture presented for ulxs . in particular , the hardness ratios of a bright source in the antennae galaxies are sometimes positively correlated with flux and sometimes inversely correlated with flux , with some indications that loops are traced out in a hardness versus intensity diagram ( zezas et al . this is consistent with behavior seen in the galactic microquasar grs 1915 + 105 ( fender & belloni 2004 ) , where within the very high state , such loops are quite often seen . this may be revealing some quite interesting physics , as discussed in the reference above , but is unlikely to have a major affect on the bolometric corrections we have described here . we note further that in the outbursts of bright x - ray transients , the bulk of the mass accretion will occur during the high luminosity phases of the outburst . computations of the outburst duty cycle , then can assume radiatively efficient accretion , and can assume the exponential decay continues down to arbitrarily small luminosity ( in reality , soft x - ray transients seem not to drop below about @xmath112 of the eddington limit garcia et al . 2001 ) without seriously affecting the parameter values . angelini , loewenstein & mushotzky , 2001 , apjl , 557 , l35 , s. , livio , m. , piran , t. 2000 , , 545 , 772 baumgardt , h. , portegies zwart , s.f . , mcmillan , s.l.w . , makino , j. , ebisuzaki , t. , 2004 , to appear in the proceedings of the `` formation and evolution of masive young clusters '' conference , cancun , mexico , 17 - 21 november 2003 ( astro - ph/0403140 ) . begelman , m.c . , 2002 , apjl , 568 , 97 brocksopp , c. , bandyopadhyay , r. m. , fender , r. p. , 2004 , newa , 9 , 249 chen , w. , shrader , c.r . , livio , m. , 1997 , apj , 491 , 312 coppi , p.s . , 1999 , asp conf . 161 , p. 375 ( poutanen & svenson editors ) , c. , nieuwenhuijzen , h. , van der hucht , k. a. 1988 , , 72 , 259 dubus , g. , lasota , j - p . , jameury , j - m . , charles , p. , 1999 , mnras , 303 , 139 , t. , makino , j. , tsuru , t. g. , funato , y. , portegies zwart , s. , hut , p. , mcmillan , s. , matsushita , s. , matsumoto , h. , kawabe , r. 2001 , , 562 , l19 esin , a.a . , mcclintock , j. e. , narayan , r , 1997 , apj 489 , 865 , g. , baldi , a. , king , a. r. , ponman , t. j. , raymond , j. , read , a. , rots , a. , schweizer , f. , zezas , a. 2004 , , 605 , l21 fabbiano , g. , 1998 , apj , 325 , 544 fender , r.p . , gallo , e. jonker , p.g . , 2003 , mnras , 343 , l99 fender , r.p . , belloni , t. , 2004 , araa in press ( astro - ph/0406483 ) fillipenko & ho 2003 , apjl , 588 , 13 foschini , l. et al . , 2002,a&a , 392 , 817 gallo , e. , fender , r.p . , pooley , g.g . , 2003 , mnras , 344 , 60 garcia , m. r. , mcclintock , j. e. , narayan , r. , callanan , p. , barret , d. , murray , s. s. 2001 , apj , 553 , l47 garcia , m. r. , miller , j. m. , mcclintock , j. e. , king , a. r. , orosz , j. , 2003 , apj , 591 , 388 gierlinski , m. , zdziarski , a.a . , poutanen , j. , coppi , p.s . , ebisawa , k. & johnson , w.n . , 1999 , mnras , 309 , 496 green , j. , ho , l , 2004 , submitted to apj ( astro - ph/0404110 ) , m. a. , freitag , m. , rasio , f. a. 2003 , arxiv astrophysics e - prints harlaftis , e.t . & greiner , j. , 2004 , a&a let . , 414 , 13 hjellming , r. m. , rupen , m. p. , mioduszewski , a. j , et al . , 1999 , apj , 514 , 383 liu , j - f , bregman , j.n . , seitzer , p. , 2004 , apj , 602 , l249 , c. , portegies zwart , s. f. , alexander , t. 2004 , , 604 , l101 irwin , j.a . , bregman , j.n . & athey , a. , 2004 , apjl , 601 , 143 jeltema , t.e . , canizares , c.r . , buote , d.a . & garmire , g.p . , 2003 , apj , 585 , 756 , v. , henninger , m. , ivanova , n. , king , a. r. , 2004 , apj l41 , a. r. 2000 , , 312 , l39 king , a.r . , 2003 , in _ compact stellar x - ray sources _ king , a.r.,davies , m. b. , ward , m. j. , fabbiano , g. , elvis , m. , 2001 , apj , 552 , l109 krding , e. , falcke , h. , markoff , s. , 2002 , a&a , 382 , l13 kormendy , j. & richstone , d. 1995 , araa 33 , 581 , l. d. , lifshitz , m. 1958 , , pergamon press , oxford , london , new york , paris , p. , rees , m. j. 2001 , , 551 , l27 maccarone , t.j . , gallo , e. , fender , r.p . , 2003 , mnras , 345 , l19 maccarone , t.j . , coppi , p.s . & taam , r. , 2000 , head , 32.2507 maccarone , t.j . , 2001 , ph.d . thesis , yale university masetti , n. , foschini , l. , ho , l.c . , dadina , m. , di cocco , g. , malaguti , g. & palazzi , e. , 2003 , a&a let . , 406 , 27 mcclintock , j. e. and remillard , r. a. , 2003 astro - ph/0306213 meyer - hofmeister , e. , meyer , f. , 2000 , a&a 355 , 1073 meyer - hofmeister , e. , 2004 , submitted to a&a . mcconnell , m. et al . , 2000,apj , 543 , 928 miller , j.m . , fabbiano , g. , miller , m.c . & fabian , a.c . , 2003a , apjl , miller , j.m . , fabian , a.c . & miller , m.c . , 2003b , apj , submitted ( astro - ph/03 ... ) , m. c. , hamilton , d. p. 2002 , , 330 , 232 mitsuda , et al . , 1984 , pasj , 36 , 741 muno , m.p . , morgan , e.h . & remillard , r.a . , 1999 , apj , 527 , 321 nowak , m.a . , 1995 , pasp , 107 , 1207 nowak , m.a . , wilms , j , 1999 , apj , 522 , 476 nowak , m.a . , wilms , j , dove , j.b . , 2002 , mnras , 332 , 856 nowak , m.a . , heinz , s. , begelman , m. , 2002 , apj , 573 , 778 miyamoto , s. , kitamoto , s. , 1999 , an , 320 , 355 orosz , j. , kuulkers , e. , van der klis , m. , mcclintock , j.e . , garcia , m.r . , callanan , p.j . , bailyn , c.d . , jain , r.k . , remillard , r.a . , 2001 , apj , 555 , 489 pakull , m. , 2003 , iauc 194 peters , p. c. 1964 , phys . , 136 , 1224 podsiadlowsky , ph . , rappaport , s. , han , z. , 2003 mnras 341 , 385 podsiadlowsky , ph . , rappaport , s. , pfahl , e. , 2002 , apj 565 , 1107 , o. r. , schroder , k. , hurley , j. r. , tout , c. a. , eggleton , p. p. 1998 , , 298 , 525 , o. r. , tout , c. a. , eggleton , p. p. , han , z. 1995 , , 274 , , s. f. , makino , j. , mcmillan , s. l. w. , hut , p. 1999 , , 348 , 117 , s. f. , mcmillan , s. l. w. 2002 , , 576 , 899 , s. f. , baumgardt , h. , hut , p. , makino , j. , mcmillan , s. l. w. 2004 , nature , 428 , 724 , g. d. , shapiro , s. l. 1987 , , 321 , 199 , m. j. 1988 , , 333 , 523 shakura , n.i . & sunyaev , r.a . , 1973 , a&a , 24 , 337 shapiro , s.l . , lightman a.b . & eardley , d.p . , 1976 , apj , 204 , 187 shih , d.c . , iwasawa , k. , fabian a.c , 2003 , mnras , 341 , 973 smith , d.m . , heindl , w.a . , swank j.h . , 579 , 2002 , apj , l129 smith , d.m . , wilson , 2003 , apj , 591,138 , g. e. , phinney , e. s. , van den heuvel , e. p. j. 1997 , , 327 , 620 spergel , d. et al . , 2003 , apjs , 148 , 175 strohmayer , t.e . mushotzky , r.f . , 2003 , apj , 586 , l61 sunyaev , r.a . & trmper , j. 1979 , nature , 279 , 506 thorne , k.s . & price , r.h . 1975 , apjl , 195 , 101 , a. 1999 , , 514 , 180 van der klis , m. , 1995 , proceedings of the iau colloquium no . 151 ; held in sonneberg ; germany ; 5 - 9 december 1994 ; xxii ; 477 pp 321 , m. g. , stanger , v. , griffiths , r. e. 1984 , , 286 , 144 white , n.e . & angelini , l. , 2001 , apjl , 561 , 101 , m. g. , stanger , v. , griffiths , r. e. 1984 , , 286 , 144 zdziarski , a.a . , grove , j.e . , poutanen , j. , rao , a.r . & vadawale , s.v . , 2001 , apjl , 554 , 45 zdziarski , a.a . & gierlinski , m. , 2004 , astro - ph/0403683 zezas , a. , fabbiano , g. , 2002 , apj , 577 , 723 zezas , a. , fabbiano , g. , schweizer , f. , king , a. r. , baldi , a. , ponman , t. j. , raymond , j. c. , read , a. m. , rots , a. h , 2003 , baas 203 , 126.04 in the paper `` intermediate mass black holes in accreting binaries : formation , evolution and observational appearance '' by simon f. portegies zwart , jasinta dewi and tom maccarone ( mnras october [ 2004 ] ) contains an error causing the frequency of the gravitational wave sources to be too large by a factor @xmath113 . this affects the discussion in 4.1 and the numbers in the right most column of tab . 1 . if instead of a distance of 10kpc one assumes a distance of 1kpc the new numbers in the right most column of tab . 1 become 1200myr , 0 , 10 , 0 , 0 , 0 , 0 , 0 , 2 , 0 , 0 , 0 , 0 , 1myr , from the top row to the bottom row .

we provide x - ray constraints and perform the first x - ray spectral analyses for bright sources ( 5 mjy ; s / n ) in an area of the 2 ms deep field - north survey containing the hubble deep field - north . x - ray emission is detected from seven of the ten bright submm sources in this region down to 0.58.0 kev fluxes of erg s , corresponding to an x - ray detected submm source density of deg ; our analyses suggest that this equates to an x - ray detected fraction of the bright submm source population of 36% , although systematic effects may be present . two of the x - ray detected sources have nearby ( within 3 ) x - ray companions , suggesting merging / interacting sources or gravitational lensing effects , and three of the x - ray detected sources lie within the approximate extent of the proto - cluster candidate . five of the x - ray detected sources have flat effective x - ray spectral slopes ( ) , suggesting obscured agn activity . x - ray spectral analyses suggest that one of these agns may be a compton - thick source ; of the other four agns , three appear to be compton - thin sources and one has poor constraints . the rest - frame unabsorbed x - ray luminosities of these agns are more consistent with those of seyfert galaxies than qsos ( i.e. , erg s ) . thus , the low x - ray detection rate of bright submm sources by moderately deep x - ray surveys appears to be due to the relatively low luminosities of the agns in these sources rather than compton - thick absorption . a comparison of these sources to the well - studied heavily obscured agn ngc 6240 shows that the average agn contribution is negligible at submm wavelengths . the x - ray properties of the other two x - ray detected sources are consistent with those expected from luminous star formation ; however , we can not rule out the possibility that low - luminosity agns are present . the three x - ray undetected sources appear to lie at high redshift ( ) and could be either agns or starbust galaxies .

astro-ph0211267

due to the strong negative @xmath14-correction for galaxies at submillimeter ( submm ; @xmath15 = 3001000 @xmath16 m ) wavelengths , submm surveys facilitate observations of the high - redshift universe ( e.g. , blain & longair 1993 ) . for typical starburst galaxies and active galactic nuclei ( agns ) , the expected submm flux density is approximately insensitive to redshift for @xmath17 110 . indeed , the majority of the submm - detected sources in ( holland 1999 ) surveys appear to be extremely luminous infrared galaxies at @xmath17 16 ( e.g. , smail , ivison , & blain 1997 ; hughes 1998 ; ivison 1998 ; smail 2000 , 2002 ; dunlop 2001 ; blain 2002 ) . it is generally believed that these submm - detected sources are proto - galaxies undergoing intense dust - enshrouded star formation ( e.g. , blain 1999 ; lilly 1999 ) . however , evidence for agns has also been found using a variety of techniques . for example , optical spectroscopic observations have revealed agns in a number of optically bright submm sources ( e.g. , ivison 1998 ; barger 1999a ; soucail 1999 ) , and observations of known optically bright quasi - stellar objects ( qsos ; i.e. , luminous type 1 agn ) have shown that luminous agns comprise a fraction of the submm source population ( e.g. , mcmahon 1999 ; page 2001 ; isaak 2002 ; willott 2002 ) . , it is probably of the order of @xmath18 510% . ] although clearly some sources host agns , the ubiquity of agns in the submm source population , and their contribution to the submm emission , are poorly constrained . arguably the best discriminator of agn activity is the detection of hard x - ray emission ( i.e. , @xmath19 2 kev ) . hard x - ray observations are particularly useful in identifying agns in sources where the optical signatures are weak ( e.g. , optically faint agns or agns without strong emission lines ; vignati 1999 ; mushotzky 2000 ; alexander 2001 ; hornschemeier 2001 ) . due to their high - energy x - ray coverage , high x - ray sensitivity , and excellent positional accuracy , the ( weisskopf 2000 ) and ( jansen 2001 ) observatories offer the best opportunities for the x - ray investigation of sources . although sensitive enough to detect luminous agns out to high redshift , the cross - correlation of moderately deep x - ray surveys with surveys has yielded little overlap between the x - ray and submm detected source populations ( fabian 2000 ; hornschemeier 2000 , 2001 ; severgnini 2000 ; barger 2001a ; almaini 2002 ; however , see also bautz 2000 and ivison 2002 ) . these studies have suggested that a large number of bolometrically important agns ( i.e. , agns that can dominate the submm emission ) can only be present in the submm source population if they are obscured by compton - thick ( i.e. , @xmath20 @xmath4 ) material . [ sourceexp ] targeting x - ray sources detected in the 1 ms deep field - north ( cdf - n ; brandt 2001 ) survey with jiggle - map observations , barger ( 2001b ) showed that submm counterparts can be found for faint x - ray sources . five of the six x - ray detected submm sources with good x - ray constraints had flat ( @xmath10 ) effective x - ray spectral slopes , an indicator of obscured agn activity . however , while clearly showing that many of the agns found in submm sources are obscured , barger ( 2001b ) did not investigate the possibility that the absorption toward the agns is compton thick . thus strong constraints on the origin of the submm emission were not placed . the primary aim of this paper is to place strong constraints on the origin of the submm emission from agn - classified submm sources by determining if their agns are obscured by compton - thin or compton - thick material . in this study we focus on an @xmath2 region within the 2 ms cdf - n survey with excellent multi - wavelength data , including deep radio ( richards 1998 ; richards 2000 ) and optical / near - infrared ( near - ir ) observations ( barger 1999b ) . due to the extremely high sensitivity of the 2 ms survey , we have the potential to detect x - ray emission from high - luminosity starburst galaxies ( i.e. , @xmath11 erg s@xmath5 ) to @xmath17 2.5 . in this study we place the tightest constraints to date on the x - ray emission from star formation in submm - detected galaxies . the galactic column density along this line of sight is @xmath21 @xmath4 ( stark 1992 ) , and @xmath22 km s@xmath5 mpc@xmath5 , @xmath23 , and @xmath24 are adopted throughout . the x - ray results reported in this paper were obtained with acis - i ( the imaging array of the advanced ccd imaging spectrometer ; garmire et al . 2002 ) onboard . the 2 ms cdf - n observations were centered on the hubble deep field - north ( hdf - n ; williams 1996 ) and cover @xmath25 arcmin@xmath26 . the region investigated in this study extends over an @xmath2 area that has its center near to the aim - point of the cdf - n observations . this 70.3 arcmin@xmath26 region is defined by the optical and near - ir observations of barger ( 1999b ) and is referred to here as the reduced hawaii flanking - field area ( see alexander 2001 ) . the x - ray data processing of these observations was similar to that described in brandt ( 2001 ) for the 1 ms exposure . the full 2 ms source catalog is presented in alexander ( 2003 ) . one hundred and ninety - three ( 193 ) x - ray sources are detected in the reduced hawaii flanking - field area with a wavdetect ( freeman 2002 ) false - positive probability threshold of 10@xmath27 down to on - axis 0.52.0 kev ( soft - band ) and 28 kev ( hard - band ) flux limits of @xmath28 erg @xmath4 s@xmath5 and @xmath29 erg @xmath4 s@xmath5 , respectively . the x - ray coverage in this region is fairly uniform , with 0.58.0 kev ( full - band ) effective exposure times of the 193 x - ray detected sources covering 1.281.94 ms ( with a median time of 1.82 ms ) . the adaptively smoothed full - band 2 ms image is shown in figure 1 . the positional uncertainties of the x - ray sources are @xmath30 10 , depending on off - axis angle ( see alexander 2003 ) , and the median positional uncertainty is 04 . these small positional uncertainties allow for accurate cross - correlation with multi - wavelength counterparts . in this study we focus on bright submm ( @xmath0 5 mjy ) sources detected with a signal - to - noise ratio ( s / n ) @xmath31 4 . this threshold was chosen to guard against possible spurious sources ( e.g. , see the estimated number of spurious sources at different significance thresholds for the observations of the elais - n2 and lockman hole east regions ; scott 2002 ) . the submm data used here was taken from a number of different studies ( hughes 1998 ; barger 2000 , 2002 ; borys 2002 ) . ten bright submm sources are detected with s / n@xmath31 4 ; see table 1 and figure 1 . if we relaxed the detection threshold to s / n@xmath31 3.5 another five sources would be added to our sample ; however , @xmath18 20% of the sources in the total sample could be spurious ( e.g. , scott 2002 ) . although we have utilized all of the published data currently available , this region only has complete coverage down to @xmath32 12 mjy ( @xmath18 75% of the region has coverage down to @xmath32 5 mjy ; e.g. , see figure 1 of barger 2002 ) . therefore , there are likely to be further bright submm sources in this region that have not yet been detected by . two main observing techniques were employed in these studies : hughes ( 1998 ) and borys ( 2002 ) performed scan - map observations while barger ( 2000 , 2002 ) performed jiggle - map observations of radio and x - ray sources . the positional uncertainty of a source is dependent on the signal - to - noise ratio and is typically @xmath18 24@xmath9 ( e.g. , serjeant 2002 ; smail 2002 ; although see hogg 2001 ) . however , in the cases where radio and x - ray sources have been targeted with jiggle - map observations , @xmath331@xmath9 positions can be inferred from the radio and x - ray data . seven of the ten submm sources were detected in jiggle - map observations of radio and x - ray sources , and have positional uncertainties of @xmath331@xmath9 . for these sources we identified optical and near - ir counterparts in the photometric source catalogs produced by alexander ( 2001 ) using a 1@xmath9 matching radius ( see table 1 ) ; these source catalogs were produced from the barger ( 1999b ) @xmath34-band and @xmath35-band images which reach @xmath36 magnitude limits of @xmath37 and @xmath38 , respectively.cowie/hdflank/hdflank.html . the relationship between the @xmath14 band and @xmath39 band is @xmath40 ( barger 1999b ) . ] three of the ten submm sources were detected in scan maps and have positional uncertainties of @xmath18 24@xmath9 . however , in the case of one source ( 123652.0 + 621225 , which is often referred to as hdf850.1 , the brightest submm source in the hdf - n ; hughes 1998 ) we can adopt the sub - arcsecond 1.3 mm source position obtained with the iram interferometer on the plateau de bure ( iram pdbi ) telescope ( i.e. , downes 1999 ) . dunlop ( 2002 ) used these data in combination with extremely sensitive merlin+vla 1.4 ghz radio observations to identify accurately the optical and near - ir counterpart for 123652.0 + 621225 . since this source is located in the hdf - n , the optical and near - ir constraints are more sensitive than for the other submm sources investigated here ( 123652.0 + 621225 has @xmath41 and @xmath42 ; dunlop 2002 ) . we must rely on the positions for the other two sources since neither source has a radio or x - ray counterpart within 4@xmath9 . the @xmath18 24@xmath9 positional uncertainties for these sources are too large to be able to identify optical and near - ir counterparts relibably . only one of the submm sources has an optical spectroscopic redshift ( 123629.1 + 621045 has @xmath43 ; cohen 2000 ; hornschemeier 2001 ) and another source has a redshift estimate based on its multi - wavelength spectral energy distribution ( sed ; 123652.0 + 621225 has @xmath44 ; dunlop 2002 ) . however , on the assumption that the other sources obey the radio - to - far - infrared ( far - ir ; @xmath15 = 40120 @xmath16 m ) correlation found for local star - forming galaxies ( e.g. , helou , soifer , & rowan - robinson 1985 ) , we can estimate redshifts for all of the radio - detected sources , and place redshift lower limits on the radio - undetected sources , using the millimetric redshift technique ( e.g. , carilli & yun 1999 ) . since the submm emission corresponds to the longer wavelength rayleigh - jeans tail of the thermal dust emission detected at far - ir wavelengths , redshifts are directly determined ( albeit with considerable uncertainty ) on the basis of the radio - to - submm spectral slope . as noted by smail ( 2000 ) , this technique should give a lower limit on the source redshift when a component of the radio emission is produced by an agn . we determined millimetric redshifts using the redshift estimators of carilli & yun ( 2000 ) , eales ( 2000 ) , and rengarajan & takeuchi ( 2001 ) . we found that the eales ( 2000 ) estimator gave the lowest redshift determinations ( a minimum redshift of @xmath45 and a maximum redshift of @xmath46 ) , while the rengarajan & takeuchi ( 2001 ) estimator gave the highest redshift determinations ( a minimum redshift of @xmath47 and a maximum redshift of @xmath48 ) . the differences in the redshift determinations are due to the assumed sed and the temperature of the dust that produces the far - ir - to - submm emission . the millimetric redshifts of the source with an optical spectroscopic redshift ( 123629.1 + 621045 ; @xmath43 ) cover @xmath49 1.42.3 ( @xmath49 1.03.0 when the 1@xmath50 millimetric redshift uncertainties are taken into account ) . clearly the optical spectroscopic redshift of 123629.1 + 621045 is marginally in agreement with the millimetric redshift estimates ( we note that larger studies find a @xmath8 50% agreement between the optical spectroscopic redshifts and millimetric redshifts of submm sources ; e.g. , smail 2000 , 2002 ) . however , in the absence of better redshift estimations we must determine the redshifts for the majority of these sources using the millimetric redshift technique . the adopted millimetric redshifts for our sources are determined using the carilli & yun ( 2000 ) redshift estimator , which provides a good compromise between the eales ( 2000 ) and rengarajan & takeuchi ( 2001 ) millimetric redshift estimates ( see table 1 ) . we note that since the typical 1@xmath50 millimetric redshift uncertainty for a source is @xmath51@xmath49 0.8 ( the range of redshift uncertainties is @xmath51@xmath49 0.31.1 ) , the differences in these redshift determinations are not critical . however , even given these uncertainties , all sources have @xmath52 with all millimetric redshift estimators . five of the ten submm sources were detected by barger ( 2002 ) in jiggle - map observations of x - ray sources and are thus x - ray detected submm sources ( see table 2 ) . we searched for x - ray counterparts for the other five submm sources using a 1@xmath9 matching radius for the three sources with a radio counterpart and a 4@xmath9 matching radius for the two sources without a radio counterpart . one further x - ray counterpart was found for one of the radio - detected submm sources ( 123622.6 + 621629 ; see table 2 ) . given the low surface density of the submm sources , we can also search for lower significance x - ray sources associated with submm sources without introducing a significant number of spurious x - ray sources . we ran wavdetect with a false - positive probability threshold of 10@xmath53 and found one further match to a radio - detected submm source ( 123618.3 + 621551 ; see table 2 ) ; the probability of this x - ray - submm match being spurious is . this source was also detected in the 1 ms exposure at a 10@xmath53 false - positive probability threshold ( alexander 2001 ) and is probably not detected here at a higher probability threshold since it lies within the bright extended x - ray source ( e.g. , bauer 2002a ) . thus , seven ( @xmath54% ) of the ten submm sources have x - ray counterparts . level ; these were calculated assuming poisson statistics . ] three submm sources are x - ray undetected . there is no evidence for excess x - ray emission in the images for two of these sources ( 123620.3 + 621701 and 123621.1 + 621250 ) ; however , there is a suggestion of soft - band emission at the location of 123652.0 + 621225 . indeed , 123652.0 + 621225 lies within 02 of the position of an x - ray source detected by wavdetect at a 10@xmath55 false - positive probability threshold in the 1 ms cdf - n study of very red objects ( vros , @xmath56 ; alexander 2002b ) . since 123652.0 + 621225 lies closer to the x - ray source than the vro reported in alexander ( 2002b ) , it may actually be the correct counterpart to this extremely faint low - significance x - ray source . however , since the x - ray source is not detected at a higher significance level in the 2 ms exposure , we consider 123652.0 + 621225 undetected here . examining the association between and sources , almaini ( 2002 ) found little overlap between the source populations but detected a strong clustering signal . they suggested that and sources are generally unrelated ( at least down to the depths of their 75 ks survey ) but trace the same large - scale structure . although there may be some large - scale structure effects present in our sample , the cross - correlation distances between our and sources are much smaller than the up to 100 @xmath9 clustering distances found by almaini ( 2002 ) . thus , we expect all of our x - ray - submm matches to be legitimate . [ fig : redshift1 ] the derived density of x - ray detected submm sources in this region with @xmath0 5 mjy is @xmath6 deg@xmath7 . this source density should be considered a lower limit since there may be further bright submm sources with x - ray counterparts in this region that have not yet been detected by ( see 2.2 ) . we can compare our source density to that found in other studies . the survey of the elais - n2 and lockman hole east regions ( i.e. , scott 2002 ) contains sensitive and observations ( down to 0.58.0 kev flux limits of @xmath57 erg @xmath4 s@xmath5 ; i.e. , manners 2002 ; hasinger 2001 ) . one counterpart was matched to a submm source in the elais - n2 region ( almaini 2002 ) , corresponding to an x - ray detected submm source density of @xmath58 deg@xmath7 . four counterparts were found for submm sources in the lockman hole east region ( ivison 2002 ) , corresponding to a higher x - ray detected submm source density of @xmath59 deg@xmath7 . the x - ray detected submm source densities found in our study are @xmath18 310 times higher than those found by almaini ( 2002 ) and ivison ( 2002 ) due to the greater sensitivity of the 2 ms observations . three x - ray detected submm sources lie within 1 of the x - ray source ( see figure 1 ) . this extended x - ray source could be a high - redshift cluster or proto - cluster ( bauer 2002a ; chapman 2002b ) , and the similar millimetric redshifts of the three nearby x - ray detected submm sources suggest that they may be directly associated with cxohdfn j123620.0 + 621554 . the statistics are extremely poor on how common such sources are . however , if they are rare , then our x - ray detected submm source density is likely to be higher than average at this x - ray depth ; thus our sample may be affected by `` cosmic variance '' . the reduced hawaii flanking - field area does not have complete coverage down to @xmath32 5 mjy ( see 2.2 ) , and hence we can not directly place constraints on the fraction of the bright submm source population with x - ray counterparts . however , based on the submm source density results of eales ( 2000 ) , and the compilations of blain ( 2002 ) , cowie , barger , & kneib ( 2002 ) , and smail ( 2002 ) , we would expect 3001000 submm sources per deg@xmath26 with @xmath0 5 mjy . this suggests that the fraction of bright submm sources in this region with x - ray counterparts is likely to be @xmath60% . note that this source fraction was not directly measured in our field and systematic effects could be present . two of the x - ray detected submm sources ( 123616.1 + 621513 and 123646.0 + 621448 ) have x - ray counterparts within 3@xmath9 ( see figure 2 ; alexander 2003 ) . in both cases one x - ray source has a radio counterpart and one does not . in addition , the radio - detected sources have @xmath61 2425 while the radio - undetected sources are optically blank . given the density of x - ray sources in this region , the probability of two x - ray sources lying within 3@xmath9 of each other is @xmath30 1% and therefore these associations appear to be significant . there are two obvious scenarios that can explain the close proximity of the sources in these two pairs : ( 1 ) these sources are merging or interacting , or ( 2 ) these sources are gravitationally lensed . unfortunately , since these sources are optically faint , we can not search for the features of merging , interacting , or gravitational lensing in the current optical images . however , in the gravitational - lensing scenario , each member of each pair would generally have similar fluxes at all wavelengths ( e.g. , see review by blandford & narayan 1992 ) . therefore , the fact that one member of each pair is detected at both optical and radio wavelengths , while the other is not , provides some evidence against gravitational lensing . in both cases we have identified the radio - detected x - ray source as the counterpart to the submm source since the radio and submm emission are more closely matched in frequency space than the x - ray and submm emission . however , it is possible that the radio - undetected x - ray sources could contribute to a fraction of the submm emission . we show the submm flux density versus full - band x - ray flux of all the sources in figure 3 . as a comparison we also show the submm - detected -selected optically bright qsos ( i.e. , luminous type 1 agns ) of page ( 2001 ) . there is a clear distinction in x - ray flux between the qsos and our x - ray detected submm sources . the former would be detectable in @xmath62 ks exposures , while the majority of the latter are only detectable in @xmath8100 ks exposures . since optical spectroscopic observations of submm sources have shown that optically bright qsos probably comprise only a small fraction of the submm source population ( i.e. , @xmath18 510% , see footnote 1 ; e.g. , barger 1999a ; ivison 2000 ; smail 2002 ) , @xmath8100 ks exposures are clearly required to detect x - ray counterparts for typical bright submm sources . [ fig : redshift1 ] figure 3 shows evidence for an anti - correlation between the full - band flux and the submm flux density for our sources . while this anti - correlation could be real , we caution that it may be an artifact caused by small number statistics and sample incompleteness . the effective x - ray photon index ( @xmath63 ) of a source can provide a simple constraint on its nature . = @xmath64 + 1 where @xmath65 . ] within the context of agn - classified sources , obscured agns have considerably flatter effective x - ray spectral slopes than the canonical @xmath66 photon index of unobscured agns ( e.g. , george 2000 ) , due to the energy - dependent photo - electric absorption of the x - ray emission ( e.g. , risaliti , maiolino , & salvati 1999 ) . star - forming galaxies are also distinguishable from obscured agns since their rest frame @xmath18 28 kev emission is often consistent with that of a @xmath66 power law ( e.g. , kim , fabbiano , & trinchieri 1992a ; ptak 1999 ) . 0.7 kev thermal component ; ptak 1999 ) ; however , at the probable redshifts of our sources the very soft thermal component will be redshifted out of the energy band . ] the x - ray luminosity of a source can provide a further check on the nature of the x - ray emission since very few starburst galaxies have full - band luminosities in excess of , even when including luminous sources at moderate redshifts ( e.g. , moran , lehnert , & helfand [ fig : redshift1 ] 1999 ; zezas , alonso - herrero , & ward 2001 ; alexander 2002a ) . while this would usually be a reasonable upper limit to the expected x - ray emission from star - formation activity , given the extreme infrared luminosities of submm sources ( e.g. , ivison 1998 ; smail 2002 ) , we might expect full - band luminosities up to an order of magnitude larger ( i.e. , up to ) for this source class.@xmath67 galaxies are probably the best candidates for detecting @xmath11 erg s@xmath5 starburst galaxies , none have been found to date . ] the x - ray band ratio versus redshift for the six sources with band ratio constraints is shown in figure 4 . five of the x - ray detected submm sources have effective photon indices of @xmath10 . such flat x - ray spectral slopes are only likely to be produced by obscured agn activity and , under the assumption that the underlying continuum follows a @xmath68 power law , these sources have intrinsic column densities of @xmath69 @xmath70 @xmath4 . the rest - frame x - ray luminosities of these sources are calculated as @xmath71 where @xmath72 is the luminosity distance ( cm ) , @xmath73 is the x - ray flux ( erg @xmath4 s@xmath5 ) , and @xmath68 is the assumed photon index . assuming the underlying continuum follows a @xmath68 power law , the unabsorbed rest - frame full - band luminosities ( ) of the five sources with flat x - ray spectral slopes are also consistent with obscured agn activity ( the intrinsic absorption is calculated from the band ratios ; i.e. , see figure 4 ) . these sources are classified hereafter as agns ( see table 2 ) . the x - ray spectral constraints are weak for the other five submm sources . the rest - frame full - band luminosities or upper limits ( ) for the three sources with luminosity constraints ( see table 2 ) suggest that their x - ray emission could be produced by luminous star - formation activity . however , since we can not rule out agns in any of these five sources ( e.g. , they could be extremely highly obscured or of low luminosity ) , they are classified here as unknown ( see table 2 ) . from a comparison of the x - ray and radio luminosities of starburst galaxies detected in the 1 ms exposure of the cdf - n , bauer ( 2002b ) showed that the radio emission can be used to predict the x - ray emission from star formation ( see also ranalli , comastri , & setti 2002 ) . the main assumption in this prediction is that the radio emission is dominated by star formation ; the presence of an agn component to the radio emission will lead to an overprediction of the contribution from star formation at x - ray energies . none of the radio - detected submm sources is classified by richards ( 1999 ) as an agn on the basis of its radio emission although four sources are classified as unknown ( see table 1 ) . since only two of these four sources are classified as agns at x - ray energies , and all but one have steep radio spectral slopes , the radio emission is likely to be dominated by star - formation activity in most cases ( e.g. , condon 1992 ; richards 2000 ) . the rest - frame radio luminosity densities of these sources are calculated as @xmath74 where @xmath72 is the luminosity distance ( cm ) , @xmath75 is the 1.4 ghz flux density ( @xmath16jy ) , and @xmath64 corresponds to the radio spectral index . in determining the radio luminosity density we have assumed @xmath76 , the average spectral index for star - forming galaxies ( e.g. , yun , reddy , & condon 2001 ) . a comparison of the rest - frame full - band x - ray and radio luminosities of the radio - detected submm sources is shown in figure 5 . the predicted x - ray luminosity from star - formation activity for all of the radio - detected submm sources is @xmath77@xmath78 erg s@xmath5 . the agn - classified sources are on average @xmath18 10 times more luminous at x - ray energies than that predicted by star formation . however , the x - ray emission from the other three radio - detected submm sources could be completely dominated by star - formation activity . our interpretation of the x - ray spectral slopes and luminosities for the five agn - classified sources ( see 3.2.1 ) was based on the assumption that the obscuration to the x - ray emitting source is not optically thick to compton scattering . however , in some cases the obscuration may be compton thick ( i.e. , @xmath79 @xmath4 ) , and the observed x - ray emission will be predominantly produced by reflection and scattering processes ( e.g. , matt 1997 ; turner 1997a ; matt [ fig : redshift1 ] 2000 ) . the most direct discrimination between compton - thin and compton - thick absorption is made with x - ray spectral analysis . the x - ray spectrum of a compton - thick agn is generally characterized by a large equivalent width iron k@xmath64 emission line ( @xmath80 0.5 kev ; e.g. , matt , brandt , & fabian 1996 ; maiolino 1998 ; bassani 1999 ; matt 2000 ) and a flat or inverted ( @xmath81 ) x - ray spectral slope , due to pure reflection . by contrast , the x - ray spectrum of a compton - thin agn is usually well fitted by an absorbed power - law model and a smaller equivalent width iron k@xmath64 emission line ( generally @xmath82 0.10.5 kev ; nandra 1997 ; turner 1997b ; bassani 1999 ) . to account for the range of roll angles and aim points in the 20 separate observations that comprise the 2 ms cdf - n , we used the acis source extraction code ( acis extract ) described in broos ( 2002 ) . briefly , for each source this code extracts the counts from each of the observations taking into account the changing shape and size of the psf with off - axis angle as given in the x - ray center ( cxc ) psf library . a local background is extracted after all sources are excluded from the x - ray event file , and the spectra and response matrices are summed using standard ftools routines ( blackburn 1995 ) . the latest version of xspec ( v11.2.0i ; arnaud 1996 , 2002 ) was used for all of the model fitting , and the fit parameter uncertainties are quoted at the 90% confidence level ( i.e. , avni 1976 ) . given the limited counting statistics for the agn - classified sources ( @xmath18 30200 full - band counts ) , we performed all the x - ray spectral analysis using the @xmath83-statistic ( cash 1979 ) . the advantage of using the @xmath83-statistic is that the data can be fitted unbinned , making it well suited to low - count sources ( e.g. , nousek & shue 1989 ) . in the latest version of xspec , the @xmath83-statistic can be used on background - subtracted data . we carried out several checks of the background - subtraction method to verify that no spurious residual features were present in the background - subtracted data . we only considered two simple models for the x - ray spectral fitting : ( 1 ) a power - law model with galactic absorption , and ( 2 ) a power - law with both galactic and additional absorption . the results are given in table 3 . as expected , the photon indices derived using the galactic absorbed power - law emission model are in good agreement ( within @xmath84 ) with those calculated from the band ratio . when fit with an absorbed power - law model , the best - fitting photon indices and observed absorption for three of the sources are in broad agreement with those expected for compton - thin agns ( e.g. , turner 1997b ; bassani 1999 ) , while two sources have particularly flat photon indices ( 123622.6 + 621629 and 123712.0 + 621211 ) , possibly indicative of reflection and compton - thick absorption . however , with only @xmath18 30 full - band counts , 123712.0 + 621211 is probably too faint to provide reliable constraints . to investigate whether any of these sources could be compton - thick agns we searched for emission - line features in the 2 @xmath50 binned x - ray spectra . the only source that showed evidence for an emission - line feature was 123622.6 + 621629 ( see figure 6 ) ; this was also one of the sources with a flat photon index . to assess whether this @xmath18 4.4 kev emission - line feature was real , we simulated 10,000 low - count ( @xmath18 90 full - band counts ) x - ray spectra using the _ fakeit _ routine within xspec ; the input model was a galactic absorbed @xmath85 0.7 power law . thirty - four ( 0.34% ) of the simulations produced a feature with an equivalent width as large as or larger than that measured in the data , corresponding to a significance level of 2.9 @xmath50 . while this provides an estimation of the significance of this feature if we were expecting it to appear at @xmath18 4.4 kev , it does not take into account of the fact that we simply identified the most significant feature over the observed energy range of @xmath18 1.06.4 kev . to take this into account we must determine the number of possible distinguishable features over this energy range . given the average spectral resolution of acis - i ( @xmath18 150 ev ) we would expect @xmath18 30 distinguishable features . however , by adding unresolved gaussian lines we estimate that we would only be able to distinguish six features in the x - ray spectrum of 123622.6 + 621629 ; this demonstrates that photon statistics are more important than spectral resolution in low - count sources . thus , the adjusted significance level for this feature is @xmath18 2.3 @xmath50 ( i.e. , @xmath86% ) . assuming that the @xmath18 2.3 @xmath50 emission feature is a redshifted , unresolved 6.4 kev iron k@xmath64 emission line , the observed energy ( @xmath87 kev ) implies @xmath49 0.46@xmath88 and a rest - frame equivalent width of @xmath89 kev . the moderately large equivalent width of this possible 6.4 kev iron k@xmath64 emission line , and the inverted x - ray spectral slope ( @xmath85 0.7 ; see table 3 ) , provide evidence that 123622.6 + 621629 may be a compton - thick agn . under the assumption that the underlying agn model is [ fig : redshift1 ] the same as that fitted to the compton - thick agn ngc 6240 ( vignati 1999 ) , the unabsorbed rest - frame full - band luminosity at @xmath17 0.46 is @xmath11 erg s@xmath5 . the inferred redshift ( @xmath49 0.46@xmath88 ) is in considerable disagreement with the millimetric redshift ( @xmath49 2.4@xmath90 ) . however , we note that since 123622.6 + 621629 is detected in both the @xmath91 band ( barger 2000 ) and the @xmath92 band ( hogg 2000 ) , it is actually likely to lie at @xmath93 ( to avoid the lyman break ) . with a faint optical counterpart ( @xmath94 ) , the absolute optical magnitude of 123622.6 + 621629 at @xmath17 0.46 would correspond to that expected from a dwarf galaxy ( i.e. , @xmath18 34 mags below @xmath95 ) . there are no known examples of luminous agns in dwarf galaxies in the local universe ( e.g. , mcleod & rieke 1995 ) , although it is possible that 123622.6 + 621629 is an @xmath95 galaxy with @xmath18 34 mags of dust extinction . if 123622.6 + 621629 lies at redshift of @xmath17 0.46 then its properties are atypical for a submm source ( e.g. , ivison 2000 ; smail 2002 ) . however , we note that 123622.6 + 621629 shares many similar characteristics with fn1 - 40 , a @xmath96 submm source with an @xmath97 optical counterpart ( chapman 2002a ) . although good x - ray constraints do not exist for fn1 - 40 , the lack of broad optical emission lines led chapman ( 2002a ) to claim no evidence for agn activity . while fn1 - 40 is clearly not a broad - line agn , we note that with an [ oiii ] @xmath155007/h@xmath98 emission line ratio of @xmath8 3 ( see figure 2 of chapman 2002a ) , it could be a narrow - line agn ( e.g. , baldwin , phillips , & terlevich 1981 ; veilleux & osterbrock 1987 ) . hence , fn1 - 40 and 123622.6 + 621629 may be similar objects . optical spectroscopy is clearly required to determine the redshift of 123622.6 + 621629 and help shed light on the origin of its @xmath18 2.3 @xmath50 emission line feature . [ fig : redshift1 ] it is often assumed that the submm emission from sources detected in surveys can provide an unobscured indicator of the star - formation rate ( e.g. , blain 1999 ) . however , five of the ten submm sources investigated here clearly host obscured agns . while the agns obviously dominate the x - ray emission in these sources , this does not necessarily imply that the submm emission is also powered by the agns . unfortunately , the broad - band submm emission itself can not provide an unambiguous diagnostic of the nature of the dominant physical mechanism since it is likely to be reprocessed thermal emission produced via the absorption of the primary emission by dust grains . co observations can provide some insight into the origin of the submm emission by measuring the molecular gas mass and hence constrain the expected dust mass and emission from star formation ( e.g. , young & scoville 1991 ) . for instance , co observations have been used to show that star - formation activity may account for @xmath18 50% of the submm emission from the @xmath99 balqso smm j023990136 ( e.g. , frayer 1998 ) and perhaps all of the submm emission from the @xmath100 galaxy smm j14011 + 0252 ( e.g. , frayer 1999 ; ivison 2001 ) . however , in the absence of co observations we must adopt an indirect strategy in estimating the origin of the submm emission from our submm sources . the relative importance of agn and star - formation activity in our submm sources can be estimated by comparing their submm - to - x - ray spectral slopes ( @xmath101 ) to those of well - studied nearby galaxies . we calculate @xmath101 as @xmath102~\times~0.163,\ ] ] where @xmath103 is the observed flux density at 2 kev ( kev @xmath4 s@xmath5 kev@xmath5 ) , calculated from the full - band flux using the estimated photon index , and @xmath104 is the observed flux density at 850 @xmath16 m ( mjy ) . larger values of @xmath101 indicate stronger submm emission relative to the x - ray emission . following fabian ( 2000 ) we have compared our sources to the qso 3c273 , the compton - thick agn ngc 6240 , and the archetypal ultra - luminous starburst galaxy arp 220 . while 3c273 is clearly a powerful unobscured qso , ngc 6240 contains components of obscured agn and star - formation activity . by contrast , the emission from arp 220 appears to be entirely dominated by star formation . if an agn is present in arp 220 , then to remain undetected at x - ray energies it must be either of low luminosity , or have extreme absorption ( i.e. , @xmath105@xmath106 @xmath4 ) and a small component of scattered emission ( e.g. , iwasawa 2001 , 2002 ; clements 2002 ) . since ngc 6240 has a considerably flatter x - ray spectral slope than 3c273 and arp 220 , the analysis of the submm - to - x - ray spectral slopes requires knowledge of the x - ray photon index for each source . for instance , at a given submm luminosity , the 2 kev flux density of a @xmath107 source will be @xmath18 5 times lower than that of a @xmath68 source . therefore , when interpreting the data for an unobscured qso , the comparison should be made to 3c273 , while when interpreting the data for an obscured agn , the comparison should be made to ngc 6240 . the submm - to - x - ray spectral slope versus redshift of all the sources is shown in figure 7 . the agn - classified sources have larger @xmath101 values than ngc 6240 , suggesting that their agns are comparitively weaker at x - ray energies ( with respect to their submm emission ) . under the assumption that the obscuration in these sources is compton thin ( as expected for the majority of these sources ; see 3.3 ) , they are @xmath18 15150 times less luminous at x - ray energies ( for a given submm luminosity ) than ngc 6240 ( assuming a column density of @xmath108 @xmath4 ; i.e. , the top dotted line in figure 7 ) . taking all of the agn - classified sources together , they are on average @xmath18 70 times less luminous at x - ray energies ( for a given submm luminosity ) than ngc 6240 . as can be seen from figure 7 , these results are relatively insensitive to redshift so long as the agn - classified sources lie at @xmath52 . the only source where we have evidence that this may not be the case is the possible compton - thick agn 123622.6 + 621629 ( see 3.3.1 and figure 6 ) . under the assumption that the obscuration to this source is the same as that found for ngc 6240 ( i.e. , compton - thick absorption with a 5% scattering fraction ; vignati 1999 ) , at @xmath17 0.46 it would be @xmath18 80 times less luminous at x - ray energies ( for a given submm luminosity ) than ngc 6240 ( see figure 7 ) . we suggested in 3 that the x - ray emission from the submm sources without good x - ray constraints may be produced by star - formation activity in some cases . this hypothesis is further supported for the two sources with millimetric redshifts of @xmath109 since their submm - to - x - ray spectral slopes are comparable to that expected from arp 220 at @xmath52 ( see figure 7 ) . however , we can not distinguish between star - formation and agn scenarios for the three @xmath13 sources . although the individual constraints on the origin of the submm emission from the agn - classified sources are weak , they all clearly have larger submm - to - x - ray spectral slopes than ngc 6240 . even though ngc 6240 is a nearby source ( @xmath18 115 mpc in our chosen cosmology ) , the origin of its submm emission is poorly known . for instance , the application of radiative transfer models shows that the far - ir - to - submm emission can be almost completely explained by starburst activity ( rowan - robinson & crawford 1989 ) , and the infrared ( imaging and spectroscopy ) characteristics of ngc 6240 suggest that star formation dominates the bolometric luminosity ( genzel 1998 ) . furthermore , its nuclear optical spectrum shows the characteristics of a liner ( e.g. , veilleux 1995 ) . conversely , consideration of the energetics of the agn derived from x - ray observations suggests that it may contribute @xmath18 50% of the far - ir - to - submm luminosity ( e.g. , iwasawa & comastri 1998 ; vignati 1999 ; lira 2002 ) . assuming that the agn completely dominates the submm emission in ngc 6240 ( an extreme hypothesis ) and the x - ray emission from the agn scales linearly with submm luminosity , the contribution to the submm emission from agn activity in the agn - classified sources would be 0.66.6% ( as determined from the @xmath18 15150 times difference in x - ray luminosity for a given submm luminosity ; see 4.1 ) . following the same assumptions , the average contribution to the submm emission from agn activity would be @xmath18 1.4% . hence , although there is considerable uncertainty in the agn contribution at submm wavelengths , these analyses strongly suggest that the submm emission is dominated by star - formation activity . this conclusion is likely to be valid even if the @xmath13 sources are also found to contain agns ( see figure 7 ) . it has been known for some time that agns comprise a fraction of the submm source population . indeed , the first identified source in a blank - field survey was found to be an agn ( smm j023990136 ; ivison 1998 ) , and at least three of the 15 submm - detected sources in the -lens survey are agns ( smail 2002 ) . in all these cases the evidence for agn activity was found via optical spectroscopy of optically bright sources . the advantage of searching for agns with x - ray observations is that the limitations of optical spectroscopy are removed , allowing optically faint agns and agns without strong emission lines to be identified ( e.g. , vignati 1999 ; mushotzky 2000 ; alexander 2001 ; hornschemeier 2001 ) . however , obviously the lack of optical spectroscopy for the majority of our sources leads to considerable uncertainties in their source redshifts . in this final section , we discuss the dependence of our results on the assumed source redshifts , the fraction and properties of agns in the submm source population , and the agn contribution to the submm emission in sources . the redshifts for the majority of the submm sources are derived from the radio - to - submm spectral slope and are uncertain . as mentioned in 4.1 these uncertainties should not change our main result ( i.e. , that the average agn contribution to the submm emission is of the order of @xmath18 1.4% ) so long as the sources lie at @xmath52 . however , we note here that since the redshift tracks for the template galaxies rise to lower values of @xmath101 at @xmath110 ( see figure 7 ) , the case for star formation dominating the submm emission is even stronger for sources at lower redshifts ( as may be the case for ; see 3.3.1 ) . finally , we also note that although the implied column densities and x - ray luminosities of the agn - classified sources will be lower if they lie at @xmath110 , since @xmath10 power - law emission is only likely to be produced by obscured agn activity , their classification as agns is insensitive to redshift . in this study we have shown that five of the seven x - ray detected submm sources appear to be obscured agns , placing a lower limit on the source density of obscured agns with @xmath0 5 mjy of @xmath111 deg@xmath7 . under the same assumptions as those used in determining the fraction of the bright submm source population with x - ray counterparts ( see 2.4.2 ) , this suggests that @xmath8 26% of the bright submm source population host agns ( in reasonable agreement with model predictions ; e.g. , almaini , lawrence , & boyle 1999 ; fabian & iwasawa 1999 ) . ivison ( 2002 ) found a similar agn fraction to that derived here using the moderately deep observations of the lockman hole east region ( e.g. , hasinger 2001 ) . however , their submm - detected agn source density is @xmath18 2 times smaller ( i.e. , @xmath59 deg@xmath7 versus @xmath111 deg@xmath7 ) . the difference in the submm - detected agn source densities is mostly due to the higher sensitivity of our x - ray observations . hence the similarity between the agn fraction found by ivison ( 2002 ) and our lower limit may indicate that our estimated agn fraction is conservative . the unabsorbed rest - frame full - band luminosities of the agn - classified sources are @xmath11@xmath12 erg s@xmath5 . this range of x - ray luminosity is typical of luminous seyfert galaxies . therefore , although the majority of the submm sources are likely to be extremely luminous infrared galaxies ( e.g. , ivison 1998 ; smail 2002 ) , their agns are comparitively weak . for example , typical qsos of similar infrared luminosity ( e.g. , isaak 2002 ; page 2001 ) are @xmath18 10100 times more luminous at x - ray energies ( e.g. , page 2001 ; vignali 2001 , 2003 ) . under the assumption that bright submm sources are massive proto galaxies ( e.g. , blain 1999 ; lilly 1999 ) , the luminosities of these agns are consistent with those expected for growing supermassive black holes ( e.g. , archibald 2002 ) . the current constraints suggest that the total contribution to the submm emission from agn activity is unlikely to be more than @xmath18 1.4% ; however , as noted in 2.2 , there are likely to be further bright submm sources in our region that have not yet been detected by . in addition , we have not considered the possible contribution from optically bright submm - detected qsos . although optically bright submm - detected qsos may comprise only @xmath18 510% of the bright submm source population ( see footnote 1 ) , their submm emission could have a large agn component ( perhaps of the order of @xmath18 50% ; frayer 1998 ; bautz 2000 ) . hence , although optically bright submm - detected qsos may not be numerically significant , they may provide the bulk of the agn emission at submm wavelengths . we have used the 2 ms exposure of the cdf - n to constrain the x - ray properties of bright sources ( @xmath0 5 mjy ; s / n@xmath1 ) . in this study we have focused on the x - ray spectral properties of the x - ray detected submm sources to determine whether the agn - classified sources are compton thick or compton thin . we have used these results to constrain the contribution to the submm emission from agn activity . our main results are the following : * seven of the ten bright submm sources are detected with x - ray emission . the corresponding source density of bright submm sources with x - ray counterparts ( @xmath6 deg@xmath7 ) suggests that @xmath8 36% of bright submm sources have x - ray counterparts at this x - ray depth ; however we note that this fraction is somewhat uncertain since this region does not have complete coverage down to @xmath32 5 mjy . see 2 . * five of the x - ray detected submm sources have flat x - ray spectral slopes ( @xmath10 ) and luminous x - ray emission ( @xmath11@xmath12 erg s@xmath5 ) , suggesting obscured agn activity with @xmath112 @xmath4 . one source is possibly a compton - thick agn since it has an extremely flat x - ray spectral slope ( @xmath113 0.7 ) and shows possible evidence ( @xmath18 2.3 @xmath50 ) for a redshifted 6.4 kev iron k@xmath64 emission line ; however , the inferred redshift ( @xmath49 0.46@xmath88 ) is in considerable disagreement with the millimetric redshift ( @xmath114 ) . see 3 . * a comparison of the five agn - classified sources to the well - studied heavily obscured agn ngc 6240 suggests that the agns in these sources contribute , on average , a negligible fraction ( i.e. , @xmath18 1.4% ) of the submm emission . hence , the submm emission from these sources appears to be dominated by star formation . this result is relatively insensitive to redshift . see 4 . * the x - ray constraints are weak for the other five submm sources . we find that the x - ray properties for the two sources with millimetric redshifts of @xmath109 are consistent with those expected from luminous star - formation activity ( @xmath77@xmath78 erg s@xmath5 ) . the submm - to - x - ray spectral slopes of these sources are also similar to those expected from arp 220 ( the archetypal ultra - luminous dusty starburst galaxy ) at @xmath52 ; however , we can not rule out the possibility that low - luminosity agns are present in these sources . the three @xmath13 sources could be either agns or starbust galaxies . see 3 and 4 . * the fraction of the bright submm source population with agn activity ( i.e. , @xmath8 26% ) is in reasonable agreement with model predictions . the comparitively low x - ray luminosities of the agn - classified sources are more consistent with seyfert galaxies than qsos . we suggest that optically bright submm - detected qsos ( i.e. , luminous type 1 agns ) , although possibly not numerically significant , may provide the bulk of the agn emission at submm wavelengths . see 5 . these results show that the low x - ray detection rate of bright submm sources by moderately deep x - ray surveys is probably due to the relatively low luminosities of the agns in these sources rather than compton - thick absorption . the current constraints therefore suggest that the total contribution from agns to the submm emission in the bright submm source population is negligible . hence , the submm emission of bright submm sources appears to be predominantly powered by star - formation activity , and therefore can be used to determine star - formation rates . this work would not have been possible without the support of the entire and acis teams ; we particularly thank pat broos and leisa townsley for data analysis software , cti correction support , and the development of acis extract . special thanks go to rob ivison and ian smail for detailed and insightful comments on various drafts of this paper . we also thank omar almaini , mike eracleous , konstantin getman , giorgio matt , matt page , susie scott , paola severgnini , chris willott , and the attendees of the durham - edinburgh extragalactic workshop on `` the first galaxies and agn '' for fruitful discussions . we acknowledge the financial support of nasa grants nas 8 - 38252 and nas 8 - 01128 ( gpg , pi ) , nsf career award ast-9983783 ( dma , feb , wnb , cv ) , cxc grant g02 - 3187a ( dma , feb , wnb , cv ) , nasa gsrp grant ngt5 - 50247 ( aeh ) , nsf grant ast-9900703 ( dps ) , nasa gsrp grant ngt5 - 50277 ( scg ) , and the pennsylvania space grant consortium ( aeh , scg ) . llccccccccccc 12 36 16.15 & + 62 15 13.7 & y & y & 54@xmath1158 & @xmath116 & s & 24.3 & @xmath333.2 & 5.7@xmath117 & b02 & 2.4@xmath90 & @xmath118 + 12 36 18.33 & + 62 15 50.5 & y & y & 151@xmath11511 & @xmath190.63 & u & @xmath1925.3 & @xmath116 & 7.8@xmath119 & b00 & 1.7@xmath120 & @xmath121 + 12 36 20.3 & + 62 17 01 & n & n & @xmath122 & @xmath116 & @xmath116 & @xmath116 & @xmath116 & 13.2@xmath123 & by & @xmath194.8 & @xmath116 + 12 36 21.1 & + 62 12 50 & n & n & @xmath122 & @xmath116 & @xmath116 & @xmath116 & @xmath116 & 11.4@xmath124 & by & @xmath194.3 & @xmath116 + 12 36 22.65 & + 62 16 29.7 & y & y & 71@xmath1159 & @xmath116 & s & 23.4 & @xmath332.5 & 7.1@xmath125 & b00 & 2.4@xmath90 & @xmath126 + 12 36 29.13 & + 62 10 45.8 & y & y & 81@xmath1159 & @xmath190.80 & u & 22.2 & 3.8 & 5.0@xmath127 & b02 & 1.013 & 0.4 + 12 36 46.05 & + 62 14 48.7 & y & y & 124@xmath11510 & [email protected] & u & 24.9 & @xmath333.8 & 10.7@xmath128 & b02 & 2.3@xmath90 & @xmath129 + 12 36 52.06 & + 62 12 25.7 & y & n@xmath130 & 16@xmath1154 & 0.42@xmath131 & @xmath116 & @xmath1928.6 & @xmath195.2 & 7.0@xmath132 & hu & 4.1@xmath133 & @xmath134 + 12 37 07.21 & + 62 14 08.1 & y & y & 45@xmath1158 & [email protected] & u & 25.0 & 5.0 & 6.2@xmath135 & b02 & 2.9@xmath136 & @xmath137 + 12 37 12.07 & + 62 12 11.3 & n & y & @xmath122 & @xmath116 & @xmath116 & @xmath1925.3 & @xmath116 & 8.0@xmath138 & b02 & @xmath193.6 & @xmath116 + llccccccccccc 12 36 16.15 & + 62 15 13.7 & 130.4@xmath139 & 68.8@xmath140 & 57.4@xmath141 & @xmath142 & @xmath143 & 1.01 & 0.16 & 0.80 & 1.30@xmath144 & @xmath145 & agn + 12 36 18.33 & + 62 15 50.5@xmath130 & 29.4@xmath146 & @xmath3317.5 & @xmath3317.2 & @xmath116 & 1.4 & 0.17 & @xmath330.04 & @xmath330.19 & 1.41@xmath147 & @xmath148 & u + 12 36 20.3 & + 62 17 01 & @xmath3315.5 & @xmath3310.4 & @xmath3314.1 & & & @xmath330.10 & @xmath330.03 & @xmath330.19 & @xmath191.49 & @xmath116 & u + 12 36 21.1 & + 62 12 50 & @xmath3311.1 & @xmath335.3 & @xmath3310.3 & & & @xmath330.07 & @xmath330.01 & @xmath330.14 & @xmath191.50 & @xmath116 & u + 12 36 22.65 & + 62 16 29.7 & 63.4@xmath149 & @xmath3313.2 & 56.9@xmath150 & @xmath194.39 & @xmath33 0.6 & 1.03 & @xmath330.03 & 1.15 & 1.41@xmath147 & @xmath151 & agn + 12 36 29.13 & + 62 10 45.8 & 198.4@xmath152 & 66.3@xmath153 & 134.7@xmath154 & @xmath155 & @xmath156 & 2.32 & 0.16 & 2.26 & 1.28@xmath147 & @xmath157 & agn + 12 36 46.05 & + 62 14 48.7 & 13.8@xmath158 & 12.6@xmath159 & @xmath337.5 & @xmath330.60 & 1.4 & 0.08 & 0.03 & @xmath330.09 & 1.49@xmath160 & @xmath148 & u + 12 36 52.06 & + 62 12 25.7 & @xmath339.8 & @xmath3310.8 & @xmath334.0 & & & @xmath330.06 & @xmath330.03 & @xmath330.05 & @xmath191.48 & @xmath331.1 & u + 12 37 07.21 & + 62 14 08.1 & 84.4@xmath161 & 32.6@xmath162 & 52.3@xmath163 & @xmath164 & @xmath165 & 0.98 & 0.09 & 0.92 & 1.34@xmath147 & @xmath166 & agn + 12 37 12.07 & + 62 12 11.3 & 30.7@xmath167 & @xmath168 & @xmath169 & @xmath170 & @xmath171 & 0.39 & 0.03 & 0.38 & 1.44@xmath172 & @xmath1911.3 & agn + 123616.1 + 621513 & @xmath176 & @xmath177 & @xmath178 + 123622.6 + 621629 & @xmath179 & @xmath180 & @xmath181 + 123629.1 + 621045 & @xmath182 & @xmath183 & @xmath184 + 123707.2 + 621408 & @xmath185 & @xmath186 & @xmath187 + 123712.0 + 621211 & @xmath188 & @xmath189 & @xmath330.8 + the uncertainties refer to the 90% confidence level ( for one interesting parameter ) . + @xmath190 best - fit photon index for galactic absorption model . + @xmath191 best - fit photon index for absorbed power - law model . + @xmath192 best - fit column density in units of @xmath193 @xmath4 at @xmath194 ; the column density at the source redshift ( @xmath195 ) is related to the best - fit column density at @xmath194 by @xmath196 .

we have observed a large sample of spectroscopic binary stars in the hyades cluster , using high resolution infrared spectroscopy to detect low mass companions . we combine our double - lined infrared measurements with well constrained orbital parameters from visible light single - lined observations to derive dynamical mass ratios . using these results , along with photometry and theoretical mass - luminosity relationships , we estimate the masses of the individual components in our binaries . in this paper we present double - lined solutions for 25 binaries in our sample , with mass ratios from . this corresponds to secondary masses as small as . we include here our preliminary detection of the companion to vb 142 , with a very small mass ratio of ; this indicates that the companion may be a brown dwarf . this paper is an initial step in a program to produce distributions of mass ratio and secondary mass for hyades cluster binaries with a wide range of periods , in order to better understand binary star formation . as such , our emphasis is on measuring these distributions , not on measuring precise orbital parameters for individual binaries .

0808.3393

observations of spectroscopic binary stars provide dynamical measurements of stellar mass and binary mass ratio that are important inputs to theories of binary star formation @xcite . measuring the radial velocity versus orbital phase for a single - lined spectroscopic binary ( sb1 ) yields orbital parameters and the mass function , @xmath3 , in which the mass ratio , @xmath4 , is inseparable from the orbital inclination . for an ensemble of sb1s , a statistical approach can be used to derive the distribution of mass ratios ( e.g. , * ? ? ? if , however , a binary is observed as a double - lined system ( sb2 ) , then the dynamical mass ratio follows directly from the radial velocity measurements @xcite . with an estimate of the primary mass , say based on the spectral type , the mass ratio gives the secondary mass , and in the case of an ensemble of sb2s , the secondary mass distribution . obtaining a purely dynamical mass ratio distribution using sb2s is challenging : a binary with a mass ratio of much less than one has a small flux ratio , @xmath5 , so the secondary component is difficult to detect . this flux ratio problem is particularly inhibitive at visible wavelengths where much of the long term monitoring of spectroscopic binaries has occurred . for this reason , most identified sbs are sb1s , and there is a strong selection effect favoring the detection of sb2s with @xmath6 near 1 . a binary composed of main sequence stars with unequal masses will have an @xmath7 that increases towards longer wavelengths , making the companion easier to detect using infrared spectroscopy . such observations have allowed the measurement of sb2s with @xmath6 s as small as @xmath8 ( e.g. , * ? ? ? * ; * ? ? ? @xcite examined a sample of 164 nearby sb1s , sb2s , visual binaries , and common proper motion pairs , with f- and g - type primary stars , and found that the mass ratio distribution for medium to long period binaries increases towards small mass ratios . @xcite used a well defined sample of 129 sb1s and sb2s from the carney - latham sample of high proper motion stars @xcite to derive a bimodal mass ratio distribution with peaks at @xmath9 and @xmath10 . they found somewhat different distributions for halo versus disk stars , and for binaries with primary masses greater than and less than 0.67@xmath11 . by working in the infrared , @xcite observed 32 binaries from the carney - latham sample as sb2s , with mass ratios as small as @xmath120.20 . when combined with sb2s from visible light spectroscopy , they found a mass ratio distribution that is approximately flat from @xmath13 . the samples studied by @xcite , @xcite , and , for @xmath6 s near 1 by @xcite , represent averages of star formation outcomes in the solar neighborhood over billions of years . most star formation occurs in localized regions within molecular clouds . studies of the binary population in diverse star forming regions can isolate the physical parameters that determine the properties of the binaries formed . for example , the competing effects of fragmentation , dynamical interactions , and accretion are not fully understood @xcite . observations of binaries in open clusters can clarify the important processes because they share a common star formation origin and history . ideally , such observations would be carried out on young clusters that have an easily identifiable membership and still retain their most massive members . however , it is difficult to measure the orbital velocities of the youngest stars because they tend to have high rotation velocities . stars in older open clusters still retain much of their cluster identity yet are rotationally spun down . the hyades is one such nearby open cluster : it is well studied and presents an excellent laboratory for measuring a well defined sample of binary stars in order to investigate binary star formation . the cluster has an age of @xmath14myr @xcite , old enough that stars later than spectral type f3 have slowed to @xmath15 @xcite , and g - type and later to @xmath16 @xcite . most mid k - type and earlier cluster members have individual _ hipparcos _ distances , and with a mean cluster distance of @xmath17pc @xcite the known binaries are bright . additionally , with metallicity [ fe / h]=0.14 @xcite the spectral lines of hyades members are deep and well suited to spectroscopic analysis . radial velocity surveys of the hyades ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) have identified a sufficient number of sbs to facilitate a statistical analysis of their physical properties . @xcite carried out a study of hyades `` visual binaries '' using speckle interferometry . these wide binaries have periods longer than most sbs and several are the wide components of hierarchical triples with known sbs . from their observations , @xcite derived a photometric mass ratio distribution for long period systems that appears to increase towards small @xmath6 . at the harvard - smithsonian center for astrophysics ( cfa ) , robert p. stefanik ( rps ) and david w. latham ( dwl ) have monitored hyades members for sbs by visible light spectroscopy since 1979 ( e.g. , * ? ? ? * ; * ? ? ? their observations have refined the cluster membership and yielded precise parameters for many sb1s and sb2s . in 2003 , the current authors began a collaboration with the cfa group to extend , through infrared observations , the detection of binary companions to smaller masses than possible by visible light spectroscopy alone . to this end , rps and dwl made available to us their unpublished parameters for many hyades sb1s . in this paper we present infrared sb2 detections that span the range of mass ratios from @xmath18 to @xmath19 . we verified these results using the sample of hyades sb1s available in the literature ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? when combined with available surveys ( both present and future ) of hyades binaries , our sb2 detections will enable better determinations of the distribution of secondary masses . in [ sample ] , we present our hyades binary sample ; [ infrared ] describes our infrared observations and velocity measurement techniques ; [ sb2solutions ] includes the sb1 and sb2 solutions for our sample binaries ; [ discussion ] discusses the quality of our sb2 solutions and derives the secondary masses ; and [ summary ] presents a brief summary and comments on our future papers . because the hyades cluster is nearby , it covers a large angular extent on the sky , so determining the cluster membership is difficult . considerable effort has been applied to identify true hyades members using measurements of proper motion , radial velocity , parallax , and color - magnitude relationships ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? nevertheless , these surveys contain many candidates that are non - members or have uncertain membership status . in particular the membership of sbs is often ambiguous until their center - of - mass velocities can be calculated from orbital solutions . the multi - decade long campaign at cfa to monitor candidate hyades binaries and multiple systems ( [ intro ] ) has , for many systems , resulted in precise measurements of both the orbital parameters ( the period , @xmath20 , the eccentricity , @xmath21 , the time of periastron passage , @xmath22 , the longitude of periastron , @xmath23 , the semi - major amplitude of the primary , @xmath24 , and the center - of - mass velocity , @xmath25 ) and the mass function , @xmath3 . in 2003 , we used the preliminary cfa sb1 solutions to select a sample of 32 binaries suitable for further observations in the infrared as sb2s . we will refer to this as the _ infrared sample _ , to distinguish from the entire cfa sample . all infrared sample members are confirmed as cluster members through the methods listed above . to avoid early type stars with large rotational velocities ( [ intro ] ) we restricted the sample to targets with primaries of spectral type f and later . despite this , however , five targets do have @xmath26 and required a modified analysis ( [ irrapidrotators ] ) . most of the infrared observations were obtained with the cshell spectrometer at the irtf , which , for hyades members , has an effective magnitude limit of _ _ h__@xmath129 in a 90 minute integration ; this corresponds to targets with early - m primary stars , and serves as the faint cutoff for our sample . we intended the infrared observing campaign to last at most a few years , during which the orbital phase of systems with long periods would not change significantly . using @xmath3 from the sb1 solutions , an estimate of the primary mass based on spectral type , and letting @xmath27 , we derived the minimum mass ratio , @xmath28 , allowed for each binary . from @xmath28 we predicted the velocity separation of long period systems , and excluded those with separations of only a few @xmath29 or smaller . long period systems with large predicted velocity separations were retained , with the expectation that observations at multiple phases would not be possible . table [ table : irsample ] lists the infrared sample . columns 15 give the name and hd number of the primary star , the j2000 coordinates , and the primary spectral type as listed in the simbad database and confirmed with the analysis described in [ infrared ] . columns 6 and 7 give _ v _ magnitude ( _ hipparcos _ where available and simbad otherwise ) and _ h _ magnitude ( 2mass ) , respectively . figure [ fig : hmag ] shows the distribution of _ h_-magnitude , from 2mass , for the infrared sample . the hashed regions indicate sample members for which we did not detect a companion . ( see [ sb2nondetections ] ) . we observed the infrared sample at the nasa infrared telescope facility ( irtf ) with cshell , the facility near - infrared echelle spectrograph @xcite ; table [ table : obslog ] contains a log of the observations . cshell uses a 256x256 pixel insb detector with a plate scale of @xmath30 , and operates under natural seeing at the irtf . echelle orders are isolated with an order sorting circular variable filter . all of the observations used a single grating setting centered at 1.5548@xmath31 because this spectral region contains several deep lines , a requirement for precisely measuring radial velocities . we used a @xmath32 slit which provided a spectral resolution of @xmath33 . the spectra we obtained have a free spectral range of @xmath34 . we observed each binary as a series of short integrations , from 60 to 300s in duration , nodding the telescope @xmath35 along the slit between integrations in an abba pattern . by differencing each set of a and b frames we removed contributions from detector bias , dark current , and sky background . we took a series of flat and dark frames at the beginning of each observing session , which were median filtered and normalized to provide a flat field correction for each target `` a - b '' image . to measure the dispersion solution , we observed ar and kr arc - lamp spectra from cshell s internal lamps . we extracted each `` a - b '' image using a custom optimal extraction code written in idl , patterned after the algorithms described by @xcite , and averaged the set of resulting spectra for each target . we adjusted the total integration time for each target to provide spectra with a signal - to - noise of @xmath36 , although we relaxed this condition for targets with a large @xmath28 , and also for the faintest targets to avoid very long integrations that were an inefficient use of telescope time . observations in october 2003 suffered from poor weather conditions and did not result in any usable spectra . during the 2005 and 2006 observations the cshell detector experienced brief , uncontrolled temperature warm - ups that increased the thermal background by as much as two orders of magnitude over the nominal few tens of adus , affecting @xmath37 of the integrations . spectra obtained under these circumstances have decreased s / n , but most are still usable . the cshell observations from 2004 suggested that high s / n observations of h69 and vb 142 with a larger telescope would be particularly useful . h69 , with @xmath38 mag , is a faint target for the irtf . the sb1 solution for vb 142 yields a mass function with a very small @xmath28 of 0.04 , and the 2004 october 1 sb2 measurement indicated that @xmath6 was indeed small and the companion possibly a brown dwarf . we subsequently observed these binaries on 2005 february 22 at the w. m. keck observatory with nirspec , the facility near - infrared spectrograph @xcite . we also observed vb 43 , which at that time was nearing a phase with a large velocity separation . these observations are included in table [ table : obslog ] . we used nirspec in its cross - dispersed echelle mode with the 2-pixel slit , which provided a spectral resolution of @xmath39 . each target was observed as a sequence of abba integrations , using a nod of @xmath35 . a sequence of flat and dark frames was obtained at the beginning of the night and median filtered . night sky oh emission lines , identified from the catalog of @xcite , provided a simultaneous wavelength reference to determine the dispersion solution . we extracted the spectra in idl using redspec , the nirspec facility software package . in addition to normal image processing , redspec rectifies the images in both the spatial and spectral dimensions to extract each order . the extracted nod pairs were averaged for each target . we worked only with nirspec order 49 , centered at @xmath40 and spanning @xmath41 , because it is nearly completely free of telluric absorption ; the other orders provided by our grating setting are contaminated to varying degrees ( see * ? ? ? * fig . 1 ) . we measure the radial velocities of a binary spectrum with reference to template stellar spectra having spectral type , metallicity , and rotational velocity similar to the binary components . we have two libraries of observed single star templates . the first was obtained with nirspec and covers spectral types from g0 through m9 ( * ? ? ? * fig 36 ) . the template velocities are tied to the @xcite reference frame ; this procedure and the individual template velocities are reported in @xcite . the second library was obtained with cshell and covers spectral types from g0 through m1 ( * ? ? ? * fig 2 ) . this library is considerably more sparse in spectral type compared with the nirspec library and some cshell templates have a low s / n . both template libraries are comprised of stars with small rotational velocities . we used a non - linear limb darkening model @xcite to rotationally broaden the templates as needed , following the procedure outlined by @xcite . in the analysis presented here , we relied primarily on the nirspec templates because of their high quality , but also utilized the cshell templates where they matched the spectral types of the targets . we analyzed the infrared spectra with a two - dimensional correlation algorithm similar to the todcor algorithm described by @xcite . for each infrared observation , we used the sb1 parameters to calculate the orbital phase and the radial velocity of the primary . we then estimated the range of possible secondary velocities at that phase , corresponding to @xmath6 ranging from @xmath28 to 1 , with some allowance for the uncertainty in the primary mass estimate ( [ sample ] ) . we analyzed each target with two or three templates that matched the known characteristics of the primary and a range of four to eight plausible secondary templates , while leaving the flux ratio unconstrained . this resulted in sets of measured velocities and flux ratios for each combination of templates , which we averaged to obtain the final measured velocities reported in table [ table : obslog ] . we did not detect the companion for seven of the binaries in our sample , and for three others were successful for only part of our observations . we indicate these observations in table [ table : obslog ] as missing data , and address these systems in detail in [ sb2nondetections ] . we estimated the uncertainties of both the primary and secondary velocity measurements as a combination of two factors : ( 1 ) the velocity uncertainty of the templates and ( 2 ) the uncertainty in measuring the center of the correlation peak due to noise in the target spectrum . we took the contribution from the templates simply as the variance of the velocities measured over the set of template pairs . we measured the contribution from noise in each target spectrum as follows . first , we created a model binary from a pair of templates used in the target spectrum analysis , with the same velocity separation and flux ratio as measured in the target . then , we added random noise to the model and analyzed it using the correlation algorithm with the same pair of templates . we adjusted the amplitude of the noise until the peak correlation from the model equaled the peak correlation measured in the target spectrum . we then generated 100 copies of this model , each using a unique noise vector at the prescribed amplitude . analyzing these models resulted in distributions of primary and secondary velocities , which we fit as normal distributions . we repeated this with four different template pairs , and took the average of the distribution widths to be the uncertainty contributed by noise in that target spectrum . by combining this average in quadrature with the uncertainty due to the templates , we arrived at our best estimate of the uncertainty for an individual velocity measurement . we used this procedure to empirically determine the uncertainties on both our primary and secondary velocity measurements for each of our observed target spectra . the relative contribution to the total uncertainties from ( 1 ) and ( 2 ) varied for each target spectrum . templates with later spectral type have a worse velocity precision that those with an earlier spectral type @xcite . so , for example , an analysis using m - star templates would have a large portion of the total uncertainty contributed by the templates . conversely , a target spectrum with low s / n , often the case for the fainter binaries in our sample , would derive most of its total uncertainty from the measurement of the correlation peak . an additional source of noise that has effected previous correlation analyses ( e.g. , * ? ? ? * ) comes from a mismatch between the target primary and the template primary . if these spectra have different physical properties , particularly the metallicity of various species , then the precision of both the measured velocities and flux ratio is effected . we examined this on several of our target spectra where the velocity uncertainties obtained using the above procedure seemed inadequate ( see [ sb2solutions ] ) . we reanalyzed the model binaries from ( 2 ) using alternate template spectra of similar spectral type , thereby simulating potential mismatch between the templates and the target spectra . we found that the increase the velocity uncertainty over that obtained in ( 2 ) was negligible . this is likely due to the small spectral range of our cshell templates , which limits the possibility of metallicity mismatch . consequently , the velocity uncertainties reported in table [ table : obslog ] do not include any contribution from this effect . to verify that our visible and infrared velocity reference frames were in good agreement , we compared the primary velocities measured in the infrared spectra with those predicted from the visible light sb1 solutions . figure [ fig : vonediff ] shows the distribution of the difference between the predicted velocities and the measured velocities . the distribution has a mean of @xmath42 and is fit by a gaussian with a width of @xmath43 . this difference is smaller than our infrared velocity precision @xcite and demonstrates that the two reference frames are indeed consistent to within the measurement uncertainties . the samples in figure [ fig : vonediff ] with a velocity difference larger than @xmath44 are attributable to infrared spectra with a low s / n . the targets vb 8 , vb 30 , vb 68 , vb 77 , and bd+02 1102 have f - type primary stars that are rapidly rotating , with @xmath45 from @xmath46 , and so the spectral lines of these stars are strongly rotationally broadened . the limited spectral range of the cshell spectra did not contain enough strong features to measure the primary velocities in the infrared spectra of these binaries . we considered , however , that lower mass companions will have spun down and might therefore be sensitive to a correlation analysis . to examine this we modified our correlation analysis to use an artificial primary template comprised of a featureless , flat spectrum . we analyzed each target spectrum with a range of plausible secondary templates , allowing the flux ratio to vary , to measure the secondary velocity . this procedure was successful for vb 68 , vb 77 , and bd+02 1102 , and the velocities we measured for these systems are included in table [ table : obslog ] , along with a note indicating the altered analysis . the associated uncertainties were modeled using the procedure described in [ irccorr ] and the artificial primary template described above . we did not detect the secondaries for vb 8 or vb 30 ; we discuss both systems in [ sb2nondetections ] . four binaries in our infrared sample , l20 , vb 102 , vb 151 , and vb 40 , are the inner pair of hierarchical triple systems . speckle imaging from @xcite identified the wide companions in these systems with orbital periods from a few tens to a few hundreds of years , assuming circular orbits . we briefly describe how the presence of a third spectrum in our observations affected the analysis . the wide companion to l20 has similar brightness to the primary of the inner binary . its spectrum is clearly apparent in our infrared observations , with a radial velocity of @xmath47 . when left unaccounted for , this contribution limited the precision of our radial velocity and flux ratio measurements for the inner binary . to correct for this , we used a one - dimensional correlation to determine that the wide companion spectrum was well matched with a k5 type template and contributed about half of the total flux . we then used the template spectrum to subtract out this contribution prior to the normal two - dimensional cross - correlation step . in this manner , we increased the precision of our inner binary velocity measurements and decreased the uncertainty on @xmath48 by a factor of two over that obtained when not accounting for the wide companion . using the component masses derived by @xcite and the ( * ? ? ? * bcah ) stellar models , we estimate that the wide companion to vb 102 contributes only @xmath49 of the total flux at 1.6@xmath31 . additionally , its position angle @xcite placed it near the edge or outside of the cshell slit during our observations . if the outer binary is in a circular orbit , its period is @xmath50 years , and the velocity separation from the inner binary @xmath25-velocity is only a few @xmath29 . additionally , the radial velocity of the inner binary primary was only a few @xmath29 from the @xmath25-velocity during our observations . consequently , any contamination in our spectra from the outer companion was masked by the primary and we were unable to detect it ; we therefore solved vb 102 as a normal sb2 . the wide companions of vb 151 and vb 40 have large angular separations and were at position angles such that they fell outside of the cshell slit during our observations . even had they fallen within the slit , their contributions to the total flux are small at 1.6@xmath31 , @xmath49 and @xmath51 , respectively . we did not detect any contribution from the wide companion in either system , and concluded that our sb2 correlation analyses of the inner binaries were unaffected . with the permission of rps and dwl , we list in columns 27 of table [ table : orbitparams ] the cfa orbital parameters for the 25 sb1s that we succeeded in turning into sb2s with infrared detections ; the seven sample members for which we did not detect the secondary ( [ irccorr ] ) are not included . the cfa parameters are based on preliminary orbital solutions that may change slightly when the final results are published , but these changes are expected to be insignificant for our present purposes because the uncertainties in our final mass ratios are dominated by the infrared velocities for the secondaries . we have from one to three infrared spectra for most our binaries , and none were observed on more than five occasions . this contrasts with the several tens or more visible light observations for each system . consequently , the primary velocities that we measured in the infrared do not further improve the precision of the sb1 parameters . to solve each system as an sb2 , we took , without modification , the sb1 parameters from table [ table : orbitparams ] , and used a least - squares fitting routine to solve the infrared @xmath52 measurements for @xmath48 . for several of our binaries ( vb 62 , vb 69 , h532 , vb102 , vb 142 , vb 121 , and vb 151 ) , this procedure yielded reduced chi - squared , @xmath53 , much greater than one , indicating that the corresponding @xmath52 uncertainties ( 3.3 ) are underestimated . assuming that the missing uncertainty contribution in @xmath52 is normally distributed , we account for it by scaling the measured @xmath48 uncertainty by @xmath54 . because we do not understand the source of this additional velocity uncertainty , we choose to retain in table [ table : obslog ] the uncertainties derived in [ irccorr ] . to maintain consistent results throughout our sample , we applied this correction to all of the binaries ; column 8 of table [ table : orbitparams ] lists the derived @xmath48 along with the adjusted uncertainties . we list in column 9 of table [ table : orbitparams ] the mass ratio for each system , calculated from @xmath24 and @xmath48 as @xmath55 . we verified @xmath6 for the seven sb2s with large @xmath53 using the technique of @xcite ; in each case , these results agree to within @xmath56 of those from the least - squares fitting . the uncertainty we report for @xmath6 includes the unlisted uncertainty on @xmath24 from the preliminary cfa sb1 solution . this contribution , however , is generally negligible when compared with the large uncertainties on @xmath48 . for example , @xmath24 for vb 9 is determined to @xmath57 , which contributes only @xmath49 of the total uncertainty on @xmath6 ; most of the infrared sample members have @xmath24 determined much more precisely . l20 has @xmath48 determined to better than @xmath58 and is one of our most precisely measured infrared systems ; @xmath24 here contributes only @xmath59 to the total uncertainty on @xmath6 . columns 1013 in table [ table : orbitparams ] give the semi - major axes and component masses , combined with the unknown @xmath60 . figures [ fig : sb2plots1 ] and [ fig : sb2plots2 ] show the sb2 velocity curves and measured secondary velocities , plotted against orbital phase for the binaries in table [ table : orbitparams ] . the primary velocity curves shown come directly from the sb1 solutions , while the only free parameter in the secondary velocity curve is @xmath48 . we show the secondary velocity uncertainties listed in table [ table : obslog ] , without the scaling correction described earlier , because these values represent our best understanding of the measurements . as a consequence , some of the velocities are shown with underestimated uncertainties . figure [ fig : qhist ] shows two different representations of the mass ratio distribution for these binaries : ( _ a _ ) shows the case where each @xmath6 is determined with equal precision ; ( _ b _ ) distributes each measured @xmath6 over its corresponding uncertainty . the similarity of the distributions , within the reported @xmath61 uncertainties , suggests that our underestimate of the @xmath52 uncertainties has a small effect on the overall distribution . these distributions are obviously not complete for the cluster , most importantly because it does not include the large sample of binaries with @xmath62 that are detected as sb2s in the cfa visible light spectroscopy . figure [ fig : qhist ] does , however , clearly demonstrate the applicability of infrared observations to binaries with small mass - ratios . alternate sb1 orbits are available in the literature for nine of our infrared binaries : eight by r. f. griffin and colleagues @xcite and one by @xcite . the parameters published by these authors provide additional verification of our assertion that the uncertainties we report for @xmath48 and @xmath6 are dominated by our infrared measurements of @xmath52 . for each of these nine systems we used the alternate sb1 parameters without modification to derive @xmath48 and @xmath6 , following exactly the procedure describe above for the cfa parameters . however , we did not attempt to reconcile the various velocity zero points , which have small differences of order @xmath63 or less ( e.g. , * ? ? ? table [ table : altorbits ] lists these sb1 parameters , their references , and our alternate sb2 results . l20 , vb 43 , vb 62 , vb 69 , vb 77 , h509 , and vb 121 have @xmath6 listed in tables [ table : orbitparams ] and [ table : altorbits ] that agree to @xmath64 or better . the parameters @xmath22 and @xmath23 given for vb 40 by @xcite have a @xmath65 phase difference from the cfa parameters , which does not affect the derivation of @xmath6 . the slightly greater than @xmath56 difference in the derived @xmath6 for this binary could result from differences in the velocity zero point ; alternatively , vb 40 is a triple , and orbital motion of the wide pair over the @xmath66years between sanford s measurement and our own could account for the discrepancy . the @xmath6 calculated for l57 using the griffin and cfa sb1 parameters demonstrate the importance of a consistent velocity reference frame when combining sb1 and sb2 observations ( [ irccorr ] ) . due to its long period , l57 has a small @xmath24 , which amplifies the effect of any offset in the velocity reference frames . solving this system using the sb1 parameters of @xcite , but with the cfa @xmath25 , yields @xmath67 , nearly identical to the result in table [ table : orbitparams ] . this issue is independent of the large uncertainties we report on @xmath6 for l57 , which result from our observing it only once in the infrared . the results shown in table [ table : altorbits ] confirm that the uncertainties in our reported sb2 parameters originate primarily from our infrared measurements , not from the underlying sb1 orbits . the unpublished cfa orbits provided a sample of sb1s that is approximately three times larger than that available in the literature , and that has a velocity reference frame which is not only self - consistent , but also consistent with our infrared frame . four of the sb2 solutions shown in figures [ fig : sb2plots1 ] and [ fig : sb2plots2 ] depend on a single measurement of the secondary velocity , and another nine on only two measurements . consequently , many of the derived @xmath48 s , and the resulting @xmath6 s , have large uncertainties . figure [ fig : qvsqmin ] plots @xmath28 against the measured @xmath6 s to evaluate the plausibility of the measured values . as expected , @xmath6 is greater than or within @xmath56 of @xmath28 for all of the binaries , except vb 59 and h382 which are consistent with @xmath28to better than @xmath68 . both vb 59 and h382 have long periods and were observed at orbital phases where the velocity separation was small , resulting in low precision measurements of @xmath48 . however , we have two observations of each system at slightly different phases and the measured secondary velocities are consistent . therefore , we are confident that we are detecting the secondary in both of these systems . observations of vb 59 in @xmath69 and of h382 as soon as @xmath70 , when the orbital phases will have changed significantly , would further improve the sb2 solutions for these systems . the flux ratio , @xmath7 , measured by our cross - correlation routine offers an additional validation of each sb2 solution . figure [ fig : qvsalpha ] plots the @xmath7 measured for each binary , averaged over all observations , against the measured mass ratio . also shown are theoretical _ h_-band curves calculated from bcah for binaries with primary masses from 0.6@xmath11 to 1.2@xmath11 . two factors complicate a direct comparison between our measured flux ratios and the theoretical values . first , the small wavelength range of cshell spectra severely limits our ability to measure precise flux ratios because individual spectral lines vary only a small amount relative to each other with changing spectral type . considerably more accurate flux ratios can be measured with spectra covering a larger wavelength range , and thereby having many more spectral lines with differing dependencies on spectral type . second , the theoretical curves represent the integrated flux over the entire _ h_-band , of which our cshell spectra only sample @xmath71 . nonetheless , our measured values are well grouped along the curves . the two obvious outliers are l79 and vb 142 . we have only a single , low s / n observation of the long period binary l79 . our measured flux ratio is poorly constrained , varying significantly depending on the exact pair of templates used ; the uncertainty shown in the figure probably underestimates the actual uncertainty . however , our measured secondary velocity is well constrained , independent of the template pair , and we are confident that the secondary detection is real . vb 142 has a very small mass ratio , making the measurement of the companion particularly difficult ; [ vb142 ] addresses this system in detail . our infrared observations failed to detect the secondary in seven systems : vb 8 , vb 30 , vb 39 , h411 , l77 , l90 , and vb 114 . figure [ fig : qminhist ] shows the distribution of @xmath28 for the infrared sample binaries ; the systems not detected as sb2s are indicated by the hashed region . we expected that our sensitivity to binary companions would be incomplete for systems with the smallest mass ratios because these systems also have small flux ratios . however , several of the systems for which we did not detect the secondary have large @xmath28 . for three additional systems , vb 43 , l57 , and h509 , we did not detect the secondary in a subset of our observations . we address each of these ten systems below . vb 8 and vb 30 have rapidly rotating f - type primary stars and both have small @xmath28 , 0.15 and 0.19 , respectively . we observed both spectra in the infrared at multiple epochs , with orbital phases where the predicted velocity separation was large . we propose two possible explanations for our failure to detected these companions . the true value of @xmath6 may actually be close to @xmath28 . for a given mass ratio , the _ h_-band flux ratio of a binary decreases as the mass of the primary increases . because vb 8 and vb 30 have primaries more massive than those of a typical binary in our sample , their flux ratio s may be too small to detect the secondaries at the s / n of our spectra . alternatively , if the companions are also rotating rapidly , their spectral lines would be too broad to measure with cshell s limited wavelength coverage , even with high s / n observations . vb 114 and vb 39 have orbital periods of 4578 days and 5083 days , respectively . our observations of both systems occurred at orbital phases such that the primary velocities were near @xmath25 , and the predicted velocity separations were only a few @xmath29 . velocity measurements under such conditions are inherently difficult , and even had we detected the secondary components , the resulting @xmath48 s would be poorly constrained . vb 114 will have a more favorable orbital phase in @xmath122010 ; the phase of vb 39 will not improve until @xmath122013 . both have moderate @xmath28 , 0.44 for vb 39 and 0.29 for vb 114 , so the flux ratios should not impede in detecting these secondaries . h411 , l77 , and l90 fall towards the faint edge of the distribution shown in fig . [ fig : hmag ] , and our observations of these targets have s / n insufficient to detect their secondaries . h411 has a small @xmath28 , @xmath72 , which corresponds to a minimum h - band flux ratio of only a few percent , and is at the limit of our best observation of this binary , with s / n@xmath1250 . because h411 is faint , obtaining a spectrum with cshell that has better s / n would require many hours of integration ; such an observation could be carried out with a high resolution spectrometer at an @xmath73 m telescope in a relatively small amount of time . l77 has a large @xmath28 of @xmath74 and we expected to be sensitive to its secondary . however , our single observation of l77 has s / n@xmath1225 , and occurred at an orbital phase when the velocity separation was small . the orbit of this system is such that even had we detected the secondary , the resulting sb2 orbit would be largely unconstrained . l77 will be at a more favorable orbital phase in 2009 . l90 has @xmath75 ; we observed it on two occasions , each with s / n@xmath1250 . however , its long period and current orbital phase indicate a small velocity separation . this system will not be at a more favorable orbital phase until @xmath122011 . we did not detect a companion in our observations of vb 43 on 2005 sep 30 , l57 on 2005 nov 25 and 2006 feb 2 , and h509 on 2005 oct 2 . each of these observations were carried out at orbital phases corresponding to primary velocities near @xmath25 . the observation of vb 43 was compromised further by a low s / n , relative to our other observations of it . l57 has small @xmath24 and @xmath48 , which when combined with it s k2 primary spectral type make it a difficult target to observe with cshell : future observations would benefit from higher spectral resolution or a larger free spectral range . our four infrared observations of vb 142 provided good phase coverage , and included measurements near the maximum velocity separation and on both sides of the @xmath25-velocity . despite this , the sb2 solution that we derive is poorly determined and the residuals , @xmath76 , are large . we do not fully understand the reasons behind this poor solution , but we include it here for the following reasons . our observations , except for that on 2005 november 28 , return plausible velocities for the companion , albeit with large uncertainties . the 2005 november 28 observation occurred at phase @xmath77 , and with small velocity separation , so our inability to accurately measure the secondary in this spectrum is not surprising . the average @xmath7 that we measure for all of the observations , @xmath78 , is , however , much larger than that predicted by bcah ( figure [ fig : qvsalpha ] ) . to investigate this discrepancy we used our m - type template lhs2351 to introduce an additional spectrum into our observed vb 142 spectra from 2004 october 1 and 2005 february 22 . we added in this component with a `` true '' flux ratio , @xmath79 , ranging from 0.05 to 0.0001 , and , to avoid confusion with the actual vb 142 companion , at a radial velocity of @xmath80 . we then attempted to recover this signal with our correlation procedure and the set of templates used in the original vb 142 analyses , excluding lhs2351 . we recovered the lhs2351 spectrum at the proper velocity with @xmath79 as small as 0.0005 . however , the uncertainty of the measured velocity increased as @xmath79 decreased , and the measured flux ratio , @xmath81 became unreliable for @xmath82 . consider , for example , the case of lhs2351 introduced into the 2005 february 22 spectrum with a radial velocity of -20@xmath29 and @xmath83 . our correlation routine recovered this signal with a velocity of @xmath84 and @xmath85 . the larger than expected @xmath81 is consistent with the results obtained for the vb 142 companion , and may arise because our primary templates do not precisely match the vb 142 primary spectrum . for small flux ratios , the correlation routine tries to correct for this mismatch by scaling the primary using @xmath7 ; we have previously reported this behavior @xcite . our modeling with lhs2351 gives us confidence that we can detect a companion with a very small flux ratio . the @xmath6 that we derive for vb 142 , @xmath86 , is consistent with @xmath87 from the sb1 solution . whatever the true value of @xmath6 may be , our @xmath88 upper limit of @xmath89 is very small . our primary goal in this endeavor is to measure the binary mass ratio distribution for the hyades ( 1 ) : @xmath6 for vb 142 is sufficiently well determined for this purpose . of additional interest , the primary of vb 142 is a g5 star , and so a companion with @xmath2 could be a brown dwarf , which would be an important discovery in the hyades @xcite . while spectroscopic observations of an sb2 yield its dynamical mass ratio , they do not measure its orbital inclination and so alone they can not provide a dynamical measurement of the individual component masses . observations of the visual orbit measure the inclination and the total mass , and when combined with the mass ratio result in the individual masses . all of the binaries in our sample have components with a small angular separation and their visual orbits are not currently available . those with periods longer than a few hundred days are resolvable with adaptive optics imaging at a large aperture telescope . in the absence of visual orbits , we can still obtain good estimates of the individual masses if the distance is known . _ hipparcos _ measured the parallax of most of our sample binaries @xcite , and all have precise photometric measurements of their total flux from 2mass at j , h , and k. we combined these with our measured mass ratios and a theoretical mass - luminosity isochrone from bcah to calculate the individual component masses . we chose the bcah models for several reasons : they show a good , albeit not perfect , agreement with measured dynamical masses @xcite ; they include the effects of atmospheres , which are important in the low mass regime that applies to most of our secondaries ; and lastly , they are provided in a convenient form that specifies magnitudes in the standard photometric bands used by observers . we used the 625 myr isochrone , while noting that at such an old age the mass - luminosity relationship is mostly insensitive to age . table [ table : componentmasses ] lists the calculated component masses . the uncertainties given in table [ table : componentmasses ] include contributions from the parallax , photometry , and mass ratio ; they do not include any uncertainties from the bcah models . all of the 2mass j , h , and k photometric uncertainties are small , @xmath90 mag . because our binaries have small flux ratios , the precision with which we determine the primary masses is mostly dependent on the precision of the parallax measurements . the uncertainty for the secondaries strongly depends on the precision of the mass ratios . most of the primaries have masses determined to better than 10% , while some of the secondaries approach this level . _ hipparcos _ did not measure the parallax of vb 59 or l57 , so for these systems we used values reported in the tycho catalog , and the resulting uncertainties on both the primary and secondary masses are large . the parallaxes of h69 , h441 , and l79 have not been measured , so we estimated their primary masses directly from their spectral type and assumed an uncertainty of @xmath91 . the secondary masses then follow directly from our measured mass ratios . finally , the primary of vb 68 is more massive than the range covered by bcah , so for this system only we used the empirical isochrone determined by @xcite and note that the resulting masses are consistent with measured spectral types . we have obtained high resolution infrared spectroscopy of 32 sbs in the hyades , whose sb1 orbital parameters have been measured by rps and dwl at the cfa , in order to detect their companions and thereby study the binary mass ratio distribution in this young cluster . we detected the companion in 25 of these systems . for these , we combined our results with the sb1 parameters to determine their solutions as sb2s . some of the sb2 solutions we report have low precision for @xmath48 . however , obtaining precise orbital parameters for individual systems was not our objective here . instead , our intent was to constrain the _ distribution _ of mass ratios in binaries with low mass companions , and our results are sufficient for this purpose . we also estimated the primary and secondary masses of our sample binaries using 2mass photometry , _ hipparcos _ parallax measurements , and our measured mass ratios . the mass ratios of the binaries with the most reliable sb2 solutions span the range from @xmath92 , corresponding to secondary masses as small as @xmath1 . we also detect a very low mass companion to vb 142 . the solution for its mass ratio is not yet reliable , but it appears to be @xmath89 at the @xmath88 level , and may represent the detection of a brown dwarf companion . the precision of our derived primary masses is limited by the uncertainties in the _ hipparcos _ parallax measurements . the secondary mass measurements , however , can be improved significantly by reducing the uncertainties on the measured mass ratios through additional infrared observations . direct observations of sample members as visual binaries would measure their orbital inclinations and total masses , and when combined with the spectroscopy would yield dynamical component masses . by utilizing the _ hipparcos _ distances , such measurements would contribute a test of the theoretical mass - luminosity relationships ( e.g. , * ? ? ? improving the sb2 solutions or measuring the visual orbits of our sample would require a significant commitment of observing time and analysis resources . the visual orbit mapping may require technological improvements in interferometry and adaptive optics techniques . the results presented here are an initial step in a program to produce distributions of mass - ratio and secondary mass for hyades cluster binaries with periods from a few days to a few thousand days . future papers in this series will combine our new determinations of mass ratios with available orbital solutions from other hyades binary surveys , including the the spatially resolved systems studied by @xcite , to present the mass ratio and secondary mass distributions for the cluster . we also intend to address more fully the set of hierarchical triple systems , for which , when combined with the speckle observations of @xcite , we have information on both the inner and outer orbits . we are grateful to dwl and rps for providing the cfa sb1 parameters and for numerous discussions that improved the manuscript . we also thank the referee for several educational suggestions concerning the secondary velocity precision . we thank l. prato for providing the nirspec observations , t. mazeh for suggesting the procedure used to estimate the velocity uncertainties , and the telescope operators and staff at the irtf for their support during our many observing runs . the authors are visiting astronomers at the infrared telescope facility , which is operated by the university of hawaii under cooperative agreement no . ncc 5 - 538 with the national aeronautics and space administration , science mission directorate , planetary astronomy program . cb is supported by an nrc research associateship award at nrl . basic research in infrared astronomy at nrl is supported by 6.1 base funding . the authors were supported at stony brook in part by nsf grants 02 - 05427 and 06 - 07612 . data presented herein were obtained at the w.m . keck observatory , which is operated as a scientific partnership among the california institute of technology , the university of california and the national aeronautics and space administration . the observatory was made possible by the generous financial support of the w.m . keck foundation . this research made use of the simbad database , operated at cds , strasbourg , france , data products from 2mass , which is a joint project of the university of massachusetts and ipac at the california institute of technology , funded by nasa and nsf , and the hipparcos and tycho catalogues , esa sp-1200 . the authors wish to extend special thanks to those of hawaiian ancestry on whose sacred mountain we are privileged to be guests . lcccccc vb 8 & 25102 & 03 59 40.49 & + 10 19 49.4 & f5 & 6.4 & 5.4 + vb 9 & & 04 00 39.54 & + 20 22 49.5 & g4 & 8.7 & 7.0 + l20 & 284163 & 04 11 56.22 & + 23 38 10.8 & k0 & 9.4 & 6.6 + h69 & & 04 12 21.44 & + 16 15 03.5 & m1 & 14.0 & 9.1 + vb 30 & 27397 & 04 19 57.70 & + 14 02 06.7 & f0 & 5.6 & 4.9 + l33 & 286770 & 04 22 25.69 & + 11 18 20.6 & k8 & 9.8 & 7.1 + vb 40 & 27691 & 04 22 44.17 & + 15 03 21.9 & g0 & 7.0 & 5.6 + vb 39 & 27685 & 04 22 44.78 & + 16 47 27.7 & g4 & 7.8 & 6.3 + vb 43 & 284414 & 04 23 22.85 & + 19 39 31.2 & k2 & 9.4 & 7.3 + vb 59 & 28034 & 04 26 05.86 & + 15 31 27.6 & g8 & 7.5 & 6.2 + vb 62 & 28033 & 04 26 18.50 & + 21 28 13.6 & f8 & 7.4 & 6.1 + h382 & 28068 & 04 26 24.61 & + 16 51 12.0 & g1 & 8.0 & 6.6 + h411 & 285828 & 04 27 25.34 & + 14 15 38.5 & k2 & 10.3 & 7.8 + l57 & 285766 & 04 27 58.96 & + 18 30 00.9 & k2 & 10.2 & 7.7 + vb 68 & 28294 & 04 28 23.40 & + 14 44 27.5 & f0 & 5.9 & 5.1 + vb 69 & 28291 & 04 28 37.21 & + 19 44 26.5 & g5 & 8.6 & 7.0 + h441 & 285806 & 04 28 50.81 & + 16 17 20.3 & k7 & 10.7 & 7.6 + vb 77 & 28394 & 04 29 20.55 & + 17 32 41.8 & f7 & 7.0 & 5.8 + h509 & 28634 & 04 31 37.10 & + 17 42 35.2 & k2 & 9.5 & 7.3 + h532 & 286839 & 04 32 25.65 & + 13 06 47.6 & k0 & 11.0 & 7.8 + vb 96 & 285931 & 04 33 58.54 & + 15 09 49.0 & k0 & 8.5 & 6.6 + l79 & & 04 34 10.73 & + 11 33 29.6 & k7 & 11.7 & 8.3 + l77 & & 04 34 49.76 & + 20 23 41.6 & k7 & 11.1 & 8.0 + vb 102 & 29310 & 04 37 31.98 & + 15 08 47.2 & g1 & 7.5 & 6.1 + l90 & 29896 & 04 43 15.70 & + 17 04 08.8 & k0 & 9.9 & 7.5 + vb 142 & 30246 & 04 46 30.39 & + 15 28 19.4 & g5 & 8.3 & 6.8 + vb 113 & 30311 & 04 46 45.58 & + 09 01 02.7 & f5 & 7.2 & 5.9 + vb 114 & 30355 & 04 47 37.57 & + 18 15 31.4 & g0 & 8.5 & 6.9 + vb 115 & 284787 & 04 48 42.12 & + 21 06 03.6 & g5 & 9.1 & 7.2 + vb 121 & 30738 & 04 50 48.54 & + 16 12 37.6 & f8 & 7.3 & 6.2 + vb 151 & 240692 & 05 05 40.38 & + 06 27 54.6 & k2 & 9.9 & 7.6 + bd+02 1102 & 40512 & 05 59 29.92 & + 02 28 34.2 & f5 & 7.8 & 6.7 + llccc vb 8 & 2004 oct 4 & 53282.967 & c & + & 2005 sep 30 & 53643.982 & c & + vb 9 & 2004 oct 2 & 53280.978 & c & @xmath93 + & 2004 oct 4 & 53282.993 & c & @xmath94 + l20 & 2004 oct 1 & 53279.959 & c & @xmath95 + & 2004 oct 2 & 53280.950 & c & @xmath96 + & 2004 oct 3 & 53282.148 & c & @xmath97 + & 2005 oct 10 & 53645.985 & c & @xmath98 + & 2005 oct 3 & 53647.112 & c & @xmath99 + h69 & 2004 oct 3 & 53282.999 & c & @xmath100 + & 2005 feb 22 & 53423.715 & n & @xmath101 + & 2006 feb 3 & 53769.734 & c & @xmath102 + vb 30 & 2004 oct 3 & 53282.035 & c & + & 2004 oct 4 & 53283.036 & c & + l33 & 2004 oct 3 & 53281.959 & c & @xmath103 + & 2005 oct 2 & 53646.018 & c & @xmath104 + & 2005 nov 26 & 53700.826 & c & @xmath105 + vb 40 & 2004 oct 1 & 53280.012 & c & @xmath106 + & 2004 oct 3 & 53282.123 & c & @xmath107 + & 2005 oct 2 & 53645.958 & c & @xmath108 + & 2005 oct 3 & 53647.020 & c & @xmath109 + vb 39 & 2005 nov 26 & 53700.897 & c & + & 2006 feb 2 & 53768.727 & c & + vb 43 & 2004 oct 3 & 53282.052 & c & @xmath110 + & 2005 feb 22 & 53423.754 & n & @xmath111 + & 2005 sep 30 & 53644.008 & c & + & 2005 nov 27 & 53701.755 & c & @xmath112 + vb 59 & 2005 nov 27 & 53701.813 & c & @xmath113 + & 2006 feb 2 & 53768.812 & c & @xmath114 + vb 62 & 2004 oct 1 & 53280.052 & c & @xmath115 + & 2005 nov 25 & 53699.778 & c & @xmath116 + & 2005 nov 28 & 53702.758 & c & @xmath117 + h382 & 2005 nov 28 & 53702.801 & c & @xmath118 + & 2006 feb 2 & 53768.746 & c & @xmath119 + h411 & 2004 oct 3 & 53282.078 & c & + & 2005 oct 2 & 53646.060 & c & + l57 & 2004 oct 3 & 53282.104 & c & @xmath120 + & 2005 nov 25 & 53699.828 & c & + & 2006 feb 2 & 53768.889 & c & + vb 68 & 2005 oct 3 & 53657.074 & c & @xmath121 + vb 69 & 2004 oct 1 & 53280.082 & c & @xmath122 + & 2005 sep 30 & 53644.039 & c & @xmath123 + & 2005 nov 27 & 53701.787 & c & @xmath124 + h441 & 2005 nov 28 & 53702.851 & c & @xmath125 + & 2006 feb 2 & 53768.845 & c & @xmath126 + vb 77 & 2004 oct 1 & 53280.010 & c & @xmath127 + & 2005 nov 30 & 53644.122 & c & @xmath128 + h509 & 2004 oct 2 & 53281.117 & c & @xmath129 + & 2005 oct 2 & 53646.099 & c & + & 2006 feb 2 & 53768.786 & c & @xmath130 + h532 & 2004 oct 2 & 53281.015 & c & @xmath131 + & 2004 oct 4 & 53283.055 & c & @xmath132 + & 2005 nov 26 & 53700.916 & c & @xmath133 + vb 96 & 2006 feb 3 & 53769.922 & c & @xmath134 + l79 & 2006 feb 3 & 53769.792 & c & @xmath135 + l77 & 2006 feb 3 & 53769.840 & c & + vb 102 & 2004 oct 2 & 53281.041 & c & @xmath136 + & 2005 sep 30 & 53644.068 & c & @xmath137 + & 2006 feb 2 & 53768.921 & c & @xmath138 + l90 & 2005 nov 25 & 53699.881 & c & + & 2006 feb 3 & 53769.889 & c & + vb 142 & 2004 oct 1 & 53280.118 & c & @xmath139 + & 2005 feb 22 & 53423.860 & n & @xmath140 + & 2005 oct 1 & 53645.099 & c & @xmath141 + & 2005 nov 28 & 53702.910 & c & @xmath142 + vb 113 & 2004 oct 2 & 53281.066 & c & @xmath143 + & 2005 oct 1 & 53645.061 & c & @xmath144 + vb 114 & 2005 nov 27 & 53701.930 & c & + vb 115 & 2004 oct 4 & 53283.082 & c & @xmath145 + & 2005 nov 27 & 53701.894 & c & @xmath146 + vb 121 & 2004 oct 1 & 53280.135 & c & @xmath147 + & 2004 oct 3 & 53282.134 & c & @xmath148 + & 2005 nov 28 & 53702.941 & c & @xmath149 + vb 151 & 2004 oct 4 & 53283.112 & c & @xmath150 + & 2005 oct 2 & 53646.135 & c & @xmath151 + & 2005 nov 25 & 53699.929 & c & @xmath152 + bd+02 1102 & 2004 oct 4 & 53283.136 & c & @xmath153 + & 2005 oct 1 & 53645.135 & c & @xmath154 + lcccccccccccc vb 9 & 5070 & 0.22 & 107 & 48747 & 36.87 & 2.74 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & @xmath159 & @xmath160 + l20 & 2.394358 & 0.053 & 286.4 & 50553.507 & 40.86 & 66.96 & @xmath161 & @xmath162 & @xmath163 & @xmath164 & @xmath165 & @xmath166 + h69 & 128.114 & 0.062 & 337 & 49110.9 & 37.96 & 13.80 & @xmath167 & @xmath168 & @xmath169 & @xmath170 & @xmath171 & @xmath172 + l33 & 1044.9 & 0.250 & 295.5 & 48902 & 40.11 & 5.67 & @xmath173 & @xmath174 & @xmath175 & @xmath176 & @xmath177 & @xmath178 + vb 40 & 4.000177 & 0.0043 & 118 & 48353.24 & 37.920 & 39.88 & @xmath179 & @xmath180 & @xmath181 & @xmath182 & @xmath183 & @xmath184 + vb 43 & 589.76 & 0.619 & 304.7 & 50002.65 & 38.925 & 9.61 & @xmath185 & @xmath186 & @xmath187 & @xmath188 & @xmath189 & @xmath190 + vb 59 & 5724 & 0.975 & 224.2 & 51385 & 39.349 & 15.1 & @xmath191 & @xmath192 & @xmath193 & @xmath194 & @xmath195 & @xmath196 + vb 62 & 8.550647 & 0.212 & 41.0 & 49925.248 & 38.247 & 16.73 & @xmath197 & @xmath198 & @xmath199 & @xmath200 & @xmath201 & @xmath202 + h382 & 2657 & 0.682 & 248.4 & 48884.7 & 38.73 & 7.79 & @xmath203 & @xmath204 & @xmath205 & @xmath206 & @xmath207 & @xmath208 + l57 & 1911.3 & 0.486 & 94.1 & 49211.6 & 39.124 & 6.63 & @xmath209 & @xmath210 & @xmath211 & @xmath212 & @xmath213 & @xmath214 + vb 68 & 331.66 & 0.288 & 319 & 50290.4 & 40.00 & 11.19 & @xmath215 & @xmath216 & @xmath217 & @xmath218 & @xmath219 & @xmath220 + vb 69 & 41.6729 & 0.643 & 328.4 & 49443.176 & 38.993 & 7.03 & @xmath221 & @xmath222 & @xmath223 & @xmath224 & @xmath225 & @xmath226 + h441 & 7494 & 0.186 & 324 & 52730 & 40.44 & 3.62 & @xmath227 & @xmath228 & @xmath229 & @xmath230 & @xmath231 & @xmath232 + vb 77 & 238.86 & 0.200 & 132 & 48557.3 & 39.22 & 6.58 & @xmath233 & @xmath234 & @xmath235 & @xmath236 & @xmath237 & @xmath238 + h509 & 849.95 & 0.174 & 316.9 & 48643 & 39.553 & 6.35 & @xmath239 & @xmath240 & @xmath241 & @xmath242 & @xmath243 & @xmath244 + h532 & 1.484698 & 0.0035 & 267 & 48170.37 & 40.31 & 68.94 & @xmath245 & @xmath246 & @xmath247 & @xmath248 & @xmath159 & @xmath249 + vb 96 & 5100 & 0.664 & 310.4 & 50106 & 41.14 & 4.77 & @xmath250 & @xmath251 & @xmath252 & @xmath253 & @xmath254 & @xmath255 + l79 & 3688 & 0.765 & 341.8 & 48662.1 & 41.454 & 4.31 & @xmath256 & @xmath257 & @xmath258 & @xmath259 & @xmath260 & @xmath261 + vb 102 & 734.79 & 0.513 & 337.2 & 50015.8 & 40.192 & 3.92 & @xmath262 & @xmath263 & @xmath264 & @xmath265 & @xmath266 & @xmath267 + vb 142 & 975.7 & 0.675 & 271.9 & 50383.6 & 41.684 & 1.19 & @xmath268 & @xmath86 & @xmath269 & @xmath270 & @xmath271 & @xmath272 + vb 113 & 2429 & 0.327 & 294.1 & 51404 & 41.544 & 3.36 & @xmath273 & @xmath274 & @xmath275 & @xmath276 & @xmath277 & @xmath278 + vb 115 & 1208.2 & 0.480 & 129.1 & 50731.6 & 40.730 & 5.65 & @xmath279 & @xmath280 & @xmath281 & @xmath282 & @xmath283 & @xmath284 + vb 121 & 5.750872 & 0.361 & 42.0 & 49743.257 & 41.28 & 20.11 & @xmath285 & @xmath222 & @xmath286 & @xmath287 & @xmath288 & @xmath289 + vb 151 & 629.37 & 0.297 & 203.8 & 50544.0 & 42.167 & 5.22 & @xmath290 & @xmath291 & @xmath292 & @xmath293 & @xmath294 & @xmath295 + bd+02 1102 & 32.5121 & 0.344 & 228.7 & 51056.33 & 47.60 & 21.31 & @xmath296 & @xmath297 & @xmath298 & @xmath299 & @xmath300 & @xmath301 + lccccccccc l20 & 2.394357 & 0.057 & 279.2 & 43892.36 & 36.76 & 66.2 & @xmath302 & @xmath303 & g81 + vb 40 & 4.00050 & 0.060 & 13 & 22274.81 & 37.4 & 36.1 & @xmath304 & @xmath305 & s21 + vb 43 & 590.6 & 0.638 & 303.1 & 43512.9 & 39.81 & 9.91 & @xmath306 & @xmath307 & g85 + vb 62 & 8.55089 & 0.233 & 38.0 & 42588.22 & 38.77 & 16.46 & @xmath308 & @xmath309 & g78 + l57 & 1907 & 0.485 & 95 & 43470.5 & 40.23 & 6.83 & @xmath310 & @xmath311 & g85 + vb 69 & 41.6625 & 0.662 & 326.9 & 43650.67 & 39.81 & 7.28 & @xmath312 & @xmath313 & g85 + vb 77 & 238.87 & 0.242 & 127 & 43298 & 39.81 & 6.53 & @xmath314 & @xmath315 & g85 + h509 & 844.6 & 0.148 & 325 & 44413 & 40.32 & 6.20 & @xmath316 & @xmath317 & g85 + vb 121 & 5.75096 & 0.354 & 54.9 & 42192.06 & 42.74 & 19.70 & @xmath318 & @xmath319 & g78 + lcc vb 9 & @xmath320 & @xmath321 + l20 & @xmath322 & @xmath323 + h69 & @xmath324 & @xmath325 + l33 & @xmath326 & @xmath327 + vb 40 & @xmath328 & @xmath329 + vb 43 & @xmath330 & @xmath331 + vb 59 & @xmath332 & @xmath333 + vb 62 & @xmath334 & @xmath335 + h382 & @xmath336 & @xmath337 + l57 & @xmath338 & @xmath339 + vb 68 & @xmath340 & @xmath341 + vb 69 & @xmath342 & @xmath222 + h441 & @xmath343 & @xmath344 + vb 77 & @xmath345 & @xmath346 + h509 & @xmath347 & @xmath348 + h532 & @xmath349 & @xmath350 + vb 96 & @xmath351 & @xmath352 + l79 & @xmath343 & @xmath353 + vb 102 & @xmath354 & @xmath355 + vb 142 & @xmath356 & @xmath357 + vb 113 & @xmath358 & @xmath359 + vb 115 & @xmath360 & @xmath251 + vb 121 & @xmath361 & @xmath362 + vb 151 & @xmath363 & @xmath364 + bd+02 1102 & @xmath365 & @xmath366 +

we carry out a systematic search for extremely metal poor ( xmp ) galaxies in the spectroscopic sample of sloan digital sky survey ( sdss ) data release 7 ( dr7 ) . the xmp candidates are found by classifying all the galaxies according to the form of their spectra in a region 80 wide around h . due to the data size , the method requires an automatic classification algorithm . we use k - means . our systematic search renders 32 galaxies having negligible [ nii ] lines , as expected in xmp galaxy spectra . twenty one of them have been previously identified as xmp galaxies in the literature the remaining eleven are new . this was established after a thorough bibliographic search that yielded only some 130 galaxies known to have an oxygen metallicity ten times smaller than the sun ( explicitly , with ) . xmp galaxies are rare ; they represent 0.01% of the galaxies with emission lines in sdss / dr7 . although the final metallicity estimate of all candidates remains pending , strong - line empirical calibrations indicate a metallicity about one - tenth solar , with the oxygen metallicity of the twenty one known targets being . since the sdss catalog is limited in apparent magnitude , we have been able to estimate the volume number density of xmp galaxies in the local universe , which turns out to be . the xmp galaxies constitute 0.1% of the galaxies in the local volume , or 0.2% considering only emission line galaxies . all but four of our candidates are blue compact dwarf galaxies ( bcds ) , and 24 of them have either cometary shape or are formed by chained knots .

1109.0235

the evolution of galaxies involves the birth and death of stars , therefore , galaxies unavoidably produce metals as they live . thus , galaxies with very low metallic content are probably unevolved objects , and if we find them nearby , they provide a readily accessible fossil record from the early universe . these objects are to be expected according to the paradigm of hierarchical galaxy formation , where large galaxies arise through the assembly of smaller ones in an inefficient process leaving many dwarf galaxies as remnant ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? they seem to be materialized as the extremely metal - poor ( xmp ) dwarf galaxies observed today which , consequently , would be the closest examples we can find of these elementary primordial units from which larger galaxies assembled ( e.g. , * ? ? ? * ; * ? ? ? those units must have been extremely common in the past , but they can not be directly observed at high redshift . nearby low metallicity galaxies offer a chance for detailed studies otherwise impossible . studies of their interstellar medium ( ism ) can shed light on the properties of the primordial ism at the time of galaxy formation @xcite . for example , even the most metal - deficient galaxies in the local universe formed from matter already enriched by an early star formation episode , and the determination of the minimum galactic metallicity seems to be the best constraint available on these first stars ( e.g. , * ? ? ? * ; * ? ? ? because they have not undergone much chemical evolution , these galaxies are also the best objects for the determination of the primordial he abundance that constrains cosmological models ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . the class of blue compact dwarf galaxies ( bcds ; e.g. , * ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) contains the galaxies with the lowest known gas - phase metallicity . it is so far unclear whether such preference for xmp galaxies to be bcds is genuine or if the association results from an observational bias that systematically disfavors low surface brightness objects . the best xmp candidates in the local universe are bcds , but metal poor galaxies are found among other types of dwarf galaxies as well ( see * ? ? ? * ) . unfortunately , xmp galaxies are rare . the review by @xcite cites only 31 targets with metallicity below one tenth the solar value , which is the threshold customary used to define xmp galaxies . for decades izw18 held the record of lowest metallicity @xcite , and although a few other examples have been recently found ( e.g. , * ? ? ? * ; * ? ? ? * ) , there is a minimum metallicity close to that of i zw 18 , which corresponds to a few hundredths the solar value . the existence of such metallicity threshold is suggestive of the pre - galactic origin of metals as it happens with halo stars ( e.g. , * ? ? ? * ) , but it may also be due to other effects like the early self contamination of the ongoing starburst that rises any original level to a minimum metallicity ( e.g. , * ? ? ? * ) , or even the technical difficulty of metallicity determinations below a threshold . the number of low metallicity galaxies has significantly increased since the work by @xcite , but they are still rare objects . the thorough bibliographic compilation described in [ biblio_search ] shows only 129 such targets . the shortage of low metallicity galaxies is partly a consequence of their low luminosity as expected from the luminosity - metallicity relationship ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? . they must be faint and so detectable only within a very local volume . in order to enlarge the list of this rare yet interesting objects , we have carried out an automatic search for low metallicity galaxies in the seventh sloan digital sky survey data release ( sdss / dr7 ) . the work is reported here . so far as we are aware of , this is the first systematic search of this kind on sdss / dr7 , even though extensive searches in earlier sdss data releases have been reported . the advantage of an orderly search rather than the more traditional serendipitous discovery is twofold . first , it maximizes the number of potential candidates . second , the bias of the selection is quantifiable , allowing us to estimate for the first time the volume number density of xmp galaxies in the local universe . low metallicity galaxies are characterized by having a [ nii]@xmath56583 line negligibly small as compared to h@xmath0 . thus , the ratio between [ nii]@xmath56583 and h@xmath0 is used to measure metallicities through the appropriate calibration ( e.g. , * ? ? ? * ; * ? ? ? a low value of this ratio has been imposed as a necessary condition in classical works seeking for xmp galaxies . building on this classical approach , we address the problem in an original way by automatically classifying the galaxies according to the shape of their spectra in a region around h@xmath0 . we expected the classification to automatically separate classes of galaxies whose spectra present this property , and those targets would be regarded as metal poor candidates . note that our approach does not require the detailed knowledge of the spectral properties of the xmp galaxies , e.g. , we do not have to specify a particular ratio [ nii]@xmath56583 over h@xmath0 for a galaxy to be included . they are determined by the classification algorithm in view of the existing spectra . this minimum need of prior knowledge makes the search novel and robust against uncertainties in the selection criteria . the above conjecture turned out to work , and the result of the study is presented here . we employ a robust classification algorithm called _ k - means _ , commonly used in data - mining ( e.g. , * ? ? ? * ) , and which we have already successfully applied to sort out different types of astronomical spectra spanning from polarization profiles in the sun to galaxy spectra @xcite . the paper is organized as follows . first , we describe the systematic search for low metallicity galaxies ( [ kmeans ] ) . sdss / dr7 renders 32 xmp candidates . the physical properties of the galaxies thus selected are analyzed in [ properties ] . these candidates , together with the rest of sdss / dr7 galaxies , allow us to compute the volume number density of low metallicity galaxies in the local universe ( [ number_density ] ) . in order to contextualize our work , we carried out a comprehensive search for xmp in the literature . the results are given in [ biblio_search ] . a summary with conclusions and follow up work is provided in [ conclusions ] . as we mention in the introduction , the spectrum around h@xmath0 is very sensitive to metallicity . the ratio between the equivalent widths of [ nii]@xmath56583 and h@xmath0 goes to zero with decreasing metallicity , as calibrated by , e.g. , @xcite or @xcite . we make use of this sensitivity to select low metallicity galaxies classifying all the galaxies with spectra in sdss / dr7 according to their shape around h@xmath0 . those classes containing spectra where [ nii]@xmath56583 turns out to be negligible small with respect to h@xmath0 will be saved as candidates . for detailed information on the sdss spectral catalog , see @xcite , @xcite , as well as the comprehensive sdss / dr7 web site . the main properties of the catalog affecting our analysis are the spectral resolution of the spectra , some @xmath6 at h@xmath0 , and the fact that it includes all galaxies up to an integrated @xmath7 magnitude of @xmath8 . since the classification must be based on the shape of the spectrum rather that on other incidental property ( e.g. , the galaxy luminosity ) , the spectra must go through a previous normalization procedure . the original spectra are shifted to restframe wavelength using sdss redshifts , and then normalized to the flux in the @xmath9 color filter . in addition , the continuum around h@xmath0 is subtracted , and this spectrum is re - normalized to the peak intensity of [ nii]@xmath56583 . specifically , if @xmath10 stands for the spectrum in restframe wavelengths and normalized to @xmath9 , then the spectrum to be classified is @xmath11 , @xmath12 where @xmath5 stands for the wavelength , and @xmath13 is the intensity at the extreme of [ nii]@xmath56583 , i.e. , the maximum if the line is in emission and the minimum if it is in absorption . the continuum intensity @xmath14 is derived by linear interpolation of the spectrum in two clean windows at both sides of the spectral region of interest , namely , from 6400 to 6460 , and from 6610 to 6670 . note that the denominator of equation ( [ def1 ] ) is always positive so that the lines in absorption remain in absorption after normalization . we came across this normalization after some alternative trials . the normalization to [ nii]@xmath56583 turns out to maximize the contrast since it goes to zero for the low metallicity galaxies we are trying to identify . figure [ normalization ] summarizes the normalization procedure . it includes the original spectrum ( the black solid line ) as well as the two continuum windows used to determine the continuum intensity . after normalization ( i.e. , after continuum removal plus division by the maximum intensity at [ nii]@xmath56583 ) it becomes the dashed line , which is the 80 wide spectral range undergoing classification . @xmath56583 small compared to the nearby h@xmath0 line . the original spectrum ( the black solid line ) is normalized before classification rendering the ( red ) dashed line . the normalization includes removing the continuum , and dividing the residual by the maximum intensity of [ nii]@xmath56583 . the blue solid bars at the extremes mark the intervals used to determine the continuum . the central 80 wide bar embraces the spectral region of interest . wavelengths @xmath5 are in , and the spectra are given in arbitrary units . , scaledwidth=45.0% ] we employ the algorithm k - means for classification . it treats the spectra as vectors in a @xmath15-dimensional space , with @xmath15 the number of wavelengths . it is a rather standard technique in data - mining , machine learning , and artificial intelligence ( e.g. , * ? ? ? * ; * ? ? ? * ) , and we have already successfully employed it for massive classification of galaxy spectra @xcite . it works iteratively , starting from randomly chosen cluster centers . each galaxy spectrum is assigned to the cluster center that is closest in a least - squares sense . then the cluster centers are recomputed as the average spectrum of all the spectra in the class , and the assignation begins again . the algorithm ends when no spectra are re - classified in two successive iterations . the main advantages are : ( 1 ) it is fast , simple , and robust , as requited to classify large data sets , ( 2 ) it guarantees that similar spectra end up in the same class , ( 3 ) it automatically yields the number of classes , and ( 4 ) it provides spectra that are characteristic of the classes . these template spectra are just the average of all the spectra in the class , and they can be analyzed and interpreted as individual galaxies under the same assumptions followed when applying the popular _ stacking _ technique ( e.g. , * ? ? ? the drawbacks are : ( 1 ) the starting random seeds influence the classification , and such effect must be followed up and , eventually , corrected for , and ( 2 ) the classes do not necessarily represent individual clusters in the classification space , but may be parts of clusters . the latter downside is not a problem in our particular application since we are not interested in clustering but in separating spectra with different shapes . as for the former , one have to carry out several independent trials to test the robustness of the classes on the initialization . we apply k - means to the galaxies with spectra having redshift @xmath16 , which is equivalent to classifying all the @xmath17 galaxies with spectra in sdss / dr7 since the low metallicity targets are expected to be faint ( [ intro ] ) and , consequently they can not be observed at high redshift given the apparent magnitude threshold imposed by sdss . ( for a galaxy at redshift @xmath18 0.25 to be observed spectroscopically , the absolute @xmath7 magnitude must be smaller than @xmath19 . ) the k - means procedure is applied in two successive steps , because a single application renders classes too coarse to separate low metallicity galaxies . in the first application , we remove classes having h@xmath0 in absorption , as well as those whose [ nii]@xmath56583 is much too large . actually the limit is not imposed directly on [ nii]@xmath56583 , but on the equivalent oxygen abundance deduced via the so - called n2-method which depends , exclusively , on the ratio between the equivalent widths of [ nii]@xmath56583 and h@xmath0 , @xmath20\lambda6583}\big / w_{\rm h\alpha}\big ) . \label{n2def}\ ] ] in the calibration by @xcite , the oxygen abundance is , @xmath21 or alternatively , @xmath22 galaxies belonging to classes whose template spectrum has @xmath23 were used for a second k - means run . this arbitrary threshold was chosen as a trade off that removes enough high metallicity objects , yet allows the second classification to choose from a broad enough pool . the subset undergoing the second classification contains some 5000 galaxies , which correspond to only 0.6% of the original set . this second classification renders classes with a spectrum characteristic of low metallicity , i.e. , with h@xmath24}\lambda$]6583 . several templates of the classes thus obtained are included in fig . [ second_class ] . it shows genuine low metallicity classes ( e.g. , class # 14 ) as well as some others where the sdss skyline removal pipeline has artificially cut out [ nii]@xmath56583 ( e.g. , class # 15 ) . the finding of these fake low metallicity galaxies shows a weakness of the searching technique some sdss / dr7 galaxies with negligible [ nii]@xmath56583 may not be low metallicity after all , and the selection must be refined even further . however , this downside incidentally proves the procedure to be working properly , since we know it managed to identify spectra without [ nii]@xmath56583 . the problem arises as to how to separate genuine from fake low metallicity galaxies . fortunately , it can be sorted out easily as we explain below . } \lambda$]6583 , which may correspond to classes of low metallicity galaxies . the individual spectra have their continua removed , and have been normalized to the peak intensity of [ nii]@xmath56583 . in order to be plotted in a logarithm scale , the spectra have been artificially uplift by one unit . these classes include spectra where the sdss pipeline has artificially removed [ nii]@xmath56583 creating a mock low metallicity galaxy ( e.g. , class#15 is mostly formed by these spectra ) . wavelengths are given in . , scaledwidth=45.0% ] figure [ oxy_abu ] shows the oxygen abundance corresponding to the templates of the classes inferred from this second step , discarding those classes where the template shows obvious malfunctioning of the sdss pipeline ( i.e. , like class # 15 in fig . [ second_class ] ) . they are based on the two slightly different calibrations by @xcite given in equations ( [ pet1 ] ) and ( [ pet2 ] ) . we use it to select the galaxies in classes # 11 and # 14 as xmp galaxy candidates . the selection criteria are somewhat arbitrary but we choose these two classes because ( 1 ) the metallicity estimated for the template using the n2-method are smaller than one - tenth solar\lambda6583}\ge 140 $ ] , using the calibration in equation ( [ pet2 ] ) and assuming @xmath25 @xcite , as invoked by @xcite . ] , which is the reference value for xmp galaxies ( [ intro ] ) , ( 2 ) there seems to be a gap in metallicity between these two classes and the rest ( see fig . [ oxy_abu ] ) , and ( 3 ) the number of galaxies the classes contain is small enough to allow a detailed inspection of the individual spectra . the original 49 spectra included in these two classes were visually inspected to discover that 20 of them were faults produced by the sdss pipeline . these spectra are easily recognized because the pipeline linearly interpolates the spectrum around @xmath26\lambda6583}$ ] , and the presence of such a straight line boldly contrasts with the rest of the spectrum shape . they were excluded . since faulty spectra ended up in low - metal classes , we checked the faulty classes for truly xmp spectra sneaked in . most of the spectra were indeed failures from the pipeline , but we rescued three particularly low [ nii]@xmath56583 spectra included in these classes . all in all , our selection rendered 32 galaxies . their coordinates and the unique identifier of the galaxy spectrum in sdss / dr7 are listed in table [ list1 ] . the table also includes a column with the oxygen metallicity inferred applying equation ( [ pet2 ] ) to the equivalent widths measured on the individual sdss spectra . these values are only tentative since the prescription has an intrinsic scatter as large as 0.3 dex for individual galaxies . keeping this caveat in mind , the mean @xmath27 of the candidates is 7.60 , with a standard deviation of 0.22 . as we will see in [ biblio_search ] , this estimate is in agreement with the metallicity of the subset of candidates whose abundance has been derived with more precise means . llccccccc 1 & sdss j003630.40 + 005234.6 & 18.8 & 0.028 & 7.64 & & 194442194734546944 & single knot + 2@xmath28 & sdss j012534.19 + 075924.4 & 18.4 & 0.010 & 7.60 & 7.60 & 655785970125242368 & knotted cometary + 3 & sdss j015809.39 + 000637.2 & 18.1 & 0.012 & 7.75 & & 302812946306695168 & cometary + 4@xmath28 & sdss j030331.27 - 010947.1 & 19.7 & 0.030 & 7.48 & 7.22 & 225967511814799360 & cometary + 5 & sdss j031300.05 + 000612.1 & 19.2 & 0.029 & 7.82 & & 299996706848636928 & cometary + 6@xmath28 & sdss j080840.85 + 172856.4 & 19.2 & 0.044 & 7.36 & 7.48 & 585978593608204288 & single knot + 7 & sdss j082540.45 + 184617.2 & 19.0 & 0.038 & 7.75 & & 640023302407979008 & single knot + 8 & sdss j084236.58 + 103313.9 & 17.7 & 0.011 & 7.58 & & 724467307371298816 & cometary + 9@xmath28 & sdss j093402.03 + 551427.7 & 16.4 & 0.002 & 6.88 & 7.17 & 156443095175528448 & 2-knot cometary + 10 & sdss j094254.27 + 340411.8 & 19.1 & 0.023 & 7.67 & & 547698126572486656 & cometary ? + 11 & sdss j100348.66 + 450457.7 & 17.5 & 0.009 & 7.65 & & 265375788289753088 & single knot + 12@xmath28 & sdss j101624.52 + 375445.9 & 15.9 & 0.004 & 7.61 & 7.58 & 401892408779866112 & cometary + 13@xmath28 & sdss j103137.28 + 043422.0 & 16.2 & 0.004 & 7.52 & 7.70 & 162635977557278720 & cometary + 14@xmath28 & sdss j104457.80 + 035313.1 & 17.5 & 0.013 & 7.01 & 7.44 & 162917331083722752 & cometary + 15@xmath28 & sdss j111934.34 + 513012.1 & 16.8 & 0.004 & 7.75 & 7.51 & 247359886311030784 & 2-knot cometary + 16 & sdss j114506.25 + 501802.4 & 17.8 & 0.006 & 7.71 & & 272412374642720768 & cometary + 17@xmath28 & sdss j115132.94 - 022222.0 & 16.7 & 0.004 & 7.58 & 7.78 & 93111671593107456 & 2-knot cometary + 18 & sdss j115754.18 + 563816.7 & 16.9 & 0.001 & 7.83 & & 369803376499621888 & cometary + 19@xmath28 & sdss j120122.31 + 021108.3 & 17.6 & 0.003 & 7.69 & 7.49 & 145464500043120640 & cometary + 20@xmath28 & sdss j121402.48 + 534517.4 & 17.3 & 0.003 & 7.54 & 7.64 & 287049377199947776 & cometary + 21@xmath28 & sdss j123048.60 + 120242.8 & 16.7 & 0.004 & 7.78 & 7.73 & 454810434122285056 & cometary + 22@xmath28 & sdss j125526.07 - 021334.0 & 19.1 & 0.052 & 7.73 & 7.83 & 95080395053203456 & single knot + 23@xmath28 & sdss j132347.46 - 013252.0 & 18.1 & 0.023 & 7.21 & 7.78 & 96204976459612160 & single knot + 24 & sdss j132723.29 + 402204.1 & 19.0 & 0.012 & 7.67 & & 412307391283986432 & single knot + 25@xmath28 & sdss j133126.91 + 415148.3 & 17.1 & 0.012 & 7.64 & 7.75 & 412307391565004800 & cometary ? + 26@xmath28 & sdss j135525.66 + 465151.3 & 19.2 & 0.028 & 7.86 & 7.63 & 361921789577658368 & cometary ? + 27 & sdss j141851.13 + 210239.7 & 17.6 & 0.009 & 7.64 & & 784423532984532992 & cometary + 28@xmath28 & sdss j142342.88 + 225728.7 & 17.8 & 0.033 & 7.67 & 7.72 & 600334402043510784 & single knot + 29@xmath28 & sdss j150934.17 + 373146.1 & 17.3 & 0.033 & 7.74 & 7.85 & 394011865673367552 & cometary ? + 30@xmath28 & sdss j164710.66 + 210514.5 & 17.2 & 0.009 & 7.74 & 7.75 & 442143986740625408 & knotted cometary + 31@xmath28 & sdss j223831.12 + 140029.7 & 18.9 & 0.021 & 7.64 & 7.45 & 533060792673632256 & 2-knot cometary + 32@xmath28 & sdss j230210.00 + 004938.8 & 18.7 & 0.033 & 7.57 & 7.71 & 190784502518251520 & 2-knot cometary + [ list1 ] ) and boxes for equation ( [ pet2 ] ) . the horizontal dashed line represents one - tenth of the solar metallicity @xcite . , scaledwidth=50.0% ] k - means renders final classes that depend on the random initialization . this problem is not particularly severe in our application , since we are not trying to find clusters . we just try to separate spectra with a particular shape , independently on whether they appear on a single class or in several classes . therefore , we did not expected the random initialization to represent a serious problem , yet we study the impact on the selected spectra by repeating the k - means classification 5 times . classes with one - tenth solar metallicity are always present . they contain most of the 49 spectra in the original classification purged later on . specifically , the two classes of lowest metallicity always include between 79% and 93% of the spectra in the classification . our search is systematic . if sdss / dr7 has spectra with h@xmath24}\lambda$]6583 , they will appear in one of the classes . the assumption that we detect all the spectra with this property is used in [ number_density ] to estimate the number density of xmp galaxies in the local universe . deviations from this assumption are also analyzed in [ number_density ] . as pointed out in [ intro ] , the known xmp galaxies tend to be bcd galaxies , which are characterized by blue colors , high surface brightness , and low luminosity . obviously , the metal - poor galaxies represent only a small fraction of the bcds . in order to explore the physical properties of our candidates , we use the diagnostics developed to identify bcds in large galaxy samples , to see whether our candidates conform to the properties of known xmp galaxies . following @xcite , which condense the criteria by @xcite and @xcite , bcds can be observationally characterized by , [ cols="^,^,^ " , ] the various new symbols have their usual meaning @xmath29 and @xmath30 stand for the mean surface brightness in the sdss filters @xmath9 and @xmath7 , @xmath31 and @xmath32 represent the absolute magnitudes in these two filters , and @xmath33 is the radius containing 50% of the galaxy light . the criteria above assure that ( a ) bcds are blue , ( b ) they have high surface brightness , ( c ) they are dwarf , ( d ) they have a large star - formation rate , ( e ) they are metal poor , ( f ) they are not confused with active galactic nuclei ( agn ) , and ( g ) they have no close companions . figure [ diag_bcd ] shows various scatter plots relating the physical parameters involved in bcd characterization . the required colors and sizes have been taken directly from the sdss / dr7 database using petrosian magnitudes @xcite . the color - magnitude plot , fig . [ diag_bcd]b , shows that all our xmp candidates are dwarf the dashed line represents constraint ( c ) . their h@xmath0 equivalent width exceeds the required threshold ( constraint d ) , and the galaxies reside in a region of the bpt diagram ( after * ? ? ? * ) discarding the agn nature the dashed line in fig . [ diag_bcd]d divide agn activity and star - forming activity as worked out by @xcite . the xmp candidates obviously fulfill criterion ( e ) ( see fig . [ oxy_abu ] ) , and they are also blue - enough , staying below the horizontal dashed line in fig . [ diag_bcd]a corresponding to criterion ( a ) . as far as the compactness criterion ( b ) is concerned , most galaxies comply with it , being to the left of the slanted dashed line in fig . [ diag_bcd]a . four of them do not they are represented as asterisks in fig . [ diag_bcd]a , as well as in the remaining panels of the figure . the low surface brightness of these galaxies is to some extent apparent . visual inspection of their sdss composite - color images show two of them to present a marked cometary shape or double nucleus , and the sdss reduction pipeline has probably overestimated the apparent size of those galaxies . spurred by the unusual shape of these non - bcd galaxies , we carry out an eyeball inspection of all the 32 targets . many of them turned out to have distorted morphologies , often with cometary shape , knots and/or chains of knots . only 8 appear as a single knot without obvious substructure . the results of this crude morphological classification are given in table [ list1 ] . figure [ mugshot ] includes sdss mugshots of all xmp candidates not marked as _ single knot _ in table . [ list1 ] . those galaxies tagged with a question mark in table [ list1 ] represent ambiguous cases where the cometary shape is less obvious , however , even in theses cases the images show elongated asymmetrical substructure ( e.g. , # 29 ) . + + + + the color magnitude diagram in fig . [ diag_bcd]b deserves a separate comment . first , it shows how several xmp candidates are extremely blue , reaching up to @xmath34 . these extreme colors are rare , but they have been quoted in connection with luminous star - forming galaxies ( e.g. , * ? ? ? * ) , and with blue compact dwarfs ( e.g. , * ? ? ? second , the figure suggests a trend that contrasts with usual behavior where brighter galaxies tend to be redder the brighter the xmp candidate the bluer . this unusual trend can not be ascribed to photometric errors ( the error bars provided by sdss are included in fig . [ diag_bcd]b ) . the behavior remains independently of the type of magnitude ( fig . [ diag_bcd]b uses petrosian magnitudes , but we also tried the other magnitudes provided by sdss ; see * ? ? ? moreover , the trend disappears when colors other than @xmath35 are used . this unusual trend calls for an explanation but , so far , we can only offer conjectures . it may be a random fluctuation , since only a handful of galaxies define the trend . alternatively , it may be due to a subtle effect on the integrated colors of the large emission lines that dominate the spectrum of these galaxies . in short , all but four of the 32 candidates fulfill the criteria to be bcds , and 24 of them show cometary or knotted shape . the fact that the xmp candidates have these properties is by no means trivial . we have selected our sample according to the form of their spectra in an narrow spectral window around h@xmath0 , and this narrow bit of spectrum turns out to determine many global properties of the galaxy such as color , compactness , star formation rate , and even morphology . ccccccc ugc 12894 & 00 00 22 & + 39 29 44 & & & 7.64 & @xcite + j0004 + 0025 & 00 04 22 & + 00 25 36 & 19.4 & 433417380224303104 & 7.37 & + j0014 - 0044 & 00 14 29 & -00 44 44 & 18.7 & 306753510077104128 & 7.63 & + j0015 + 0104 & 00 15 21 & + 01 04 37 & 18.3 & 433983718529433600 & 7.07 & + j0016 + 0108 & 00 16 28 & + 01 08 02 & 18.9 & 193598873954942976 & 7.53 & + hs 0017 + 1055 & 00 20 21 & + 11 12 21 & & & 7.63 & + j0029 - 0108 & 00 29 05 & -01 08 26 & 19.2 & 434546692474273792 & 7.35 & + j0029 - 0025 & 00 29 49 & -00 25 40 & 20.4 & 434546692394582016 & 7.29 & + eso 473- g024 & 00 31 22 & -22 45 57 & & & 7.45 & @xcite + andromeda iv & 00 42 32 & + 40 34 19 & & & 7.49 & @xcite + j0057 - 0022 & 00 57 13 & -00 21 58 & 19.1 & 305062972235972608 & 7.60 & + ic 1613 & 01 04 48 & + 02 07 04 & & & 7.64 & + j0107 + 0001 & 01 07 51 & + 00 01 28 & 19.4 & 422158628955357184 & 7.23 & + am 0106 - 382 & 01 08 22 & -38 12 34 & & & 7.56 & + j0113 + 0052 & 01 13 40 & + 00 52 39 & 20.1 & 422158629852938240 & 7.24 & @xcite + j0119 - 0935 & 01 19 14 & -09 35 46 & 19.5 & 185997585616470016 & 7.31 & @xcite + hs 0122 + 0743@xmath28 & 01 25 34 & + 07 59 24 & 15.7 & 655785970125242368 & 7.60 & @xcite + j0126 - 0038 & 01 26 46 & -00 38 39 & 18.4 & 422724752210132992 & 7.51 & + j0133 + 1342 & 01 33 53 & + 13 42 09 & 18.1 & 120131172889001984 & 7.56 & + j0135 - 0023 & 01 35 44 & -00 23 17 & 18.9 & 303656125474013184 & 7.38 & + ugca20 & 01 43 15 & + 19 58 32 & 18.0 & & 7.60 & k , @xcite + um133 & 01 44 42 & + 04 53 42 & 15.4 & & 7.63 & k , + hkk97l14 & 02 00 10 & + 28 49 53 & & & 7.56 & @xcite + j0204 - 1009 & 02 04 26 & -10 09 35 & 17.1 & 187686313107914752 & 7.36 & @xcite + j0205 - 0949 & 02 05 49 & -09 49 18 & 15.3 & 187967849057222656 & 7.61 & @xcite + j0216 + 0115 & 02 16 29 & + 01 15 21 & 17.4 & 424413702440091648 & 7.44 & + 096632 & 02 51 47 & -30 06 32 & 16.3 & & 7.51 & + j0254 + 0035 & 02 54 29 & + 00 35 50 & 19.8 & 425817950054776832 & 7.28 & @xcite + j0301 - 0059 & 03 01 26 & -00 59 26 & 21.5 & 300559785177120768 & 7.64 & + j0301 - 0052 & 03 01 49 & -00 52 57 & 18.8 & 300559785005154304 & 7.52 & @xcite + j0303 - 0109@xmath28 & 03 03 31 & -01 09 47 & 19.8 & 225967511814799360 & 7.22 & + j0313 + 0010 & 03 13 02 & + 00 10 40 & 18.9 & 226248880952442880 & 7.44 & @xcite + j0315 - 0024 & 03 16 00 & -00 24 26 & 20.2 & 426661932058017792 & 7.41 & + ugc2684 & 03 20 24 & + 17 17 45 & 22.8 & & 7.60 & k , @xcite + sbs0335 - 052w & 03 37 38 & -05 02 37 & 19.0 & & 7.11 & k , @xcite + sbs0335 - 052e & 03 37 44 & -05 02 40 & 16.3 & & 7.31 & k , @xcite + j0338 + 0013 & 03 38 12 & + 00 13 13 & 24.4 & 227094714526990336 & 7.64 & + j0341 - 0026 & 03 41 18 & -00 26 28 & 18.8 & 325892648806121472 & 7.26 & + eso 358- g 060 & 03 45 12 & -35 34 15 & & & 7.26 & + g0405 - 3648 & 04 05 19 & -36 48 49 & & & 7.25 & + j0519 + 0007 & 05 19 03 & + 00 07 29 & 18.4 & & 7.44 & + to0618 - 402 & 06 20 02 & -40 18 09 & & & 7.56 & k , @xcite + eso489-g56 & 06 26 17 & -26 15 56 & 15.6 & & 7.49 & k , + j0808 + 1728@xmath28 & 08 08 41 & + 17 28 56 & 19.2 & 585978593608204288 & 7.48 & @xcite + j0812 + 4836 & 08 12 39 & + 48 36 46 & 16.0 & 124071834843873280 & 7.28 & @xcite + ugc 4305 & 08 19 05 & + 70 43 12 & & & 7.65 & + hs0822 + 03542 & 08 25 55 & + 35 32 31 & 17.8 & & 7.35 & k , + ddo53 & 08 34 07 & + 66 10 54 & 20.3 & & 7.62 & k , @xcite + ugc4483 & 08 37 03 & + 69 46 31 & 15.1 & & 7.58 & k , @xcite + hs0837 + 4717 & 08 40 30 & + 47 07 10 & 17.6 & & 7.64 & + hs 0846 + 3522 & 08 49 40 & + 35 11 39 & 18.2 & & 7.65 & + j0859 + 3923 & 08 59 47 & + 39 23 06 & 17.2 & & 7.57 & @xcite + j0910 + 0711 & 09 10 29 & + 07 11 18 & 16.9 & 336307481519063040 & 7.63 & @xcite + j0911 + 3135 & 09 11 59 & + 31 35 36 & 17.8 & 448054218540974080 & 7.51 & @xcite + j0926 + 3343 & 09 26 09 & + 33 43 04 & 17.8 & 448617233443192832 & 7.12 & @xcite + izw18@xmath28 & 09 34 02 & + 55 14 25 & 16.4 & 156443095175528448 & 7.17 & k , @xcite + j0940 + 2935 & 09 40 13 & + 29 35 30 & 16.5 & 546853820680896512 & 7.65 & @xcite + cgcg 007 - 025 & 09 44 02 & -00 38 32 & 16.0 & 75094093385957376 & 7.65 & + sbs940 + 544 & 09 44 17 & + 54 11 34 & 19.1 & & 7.46 & k , + cs 0953 - 174 & 09 55 00 & -17 00 00 & & & 7.58 & k , @xcite + j0956 + 2849 & 09 56 46 & + 28 49 44 & 15.9 & 548261264221011968 & 7.13 & @xcite + leoa & 09 59 26 & + 30 44 47 & 19.0 & & 7.30 & k , @xcite + sextans b & 10 00 00 & + 05 19 56 & 20.5 & & 7.53 & @xcite + sextans a & 10 11 00 & -04 41 34 & & & 7.54 & k , @xcite + kug 1013 + 381@xmath28 & 10 16 24 & + 37 54 44 & 15.9 & 401892408779866112 & 7.58 & k , @xcite + sdss j1025 + 1402 & 10 25 30 & + 14 02 07 & 20.4 & 491964741116231680 & 7.36 & @xcite + ugca 211 & 10 27 02 & + 56 16 14 & 16.2 & & 7.56 & + j1031 + 0434@xmath28 & 10 31 37 & + 04 34 22 & 16.2 & 162635977557278720 & 7.70 & @xcite + hs 1033 + 4757 & 10 36 25 & + 47 41 52 & 17.5 & 271004930616066048 & 7.65 & + j1044 + 0353@xmath28 & 10 44 58 & + 03 53 13 & 17.5 & 162917331083722752 & 7.44 & + hs 1059 + 3934 & 11 02 10 & + 39 18 45 & 17.9 & & 7.62 & + j1105 + 6022 & 11 05 54 & + 60 22 29 & 16.4 & 218086199200317440 & 7.64 & @xcite + j1119 + 5130@xmath28 & 11 19 34 & + 51 30 12 & 16.9 & 247359886311030784 & 7.51 & k , @xcite + j1121 + 0324 & 11 21 53 & + 03 24 21 & 18.1 & 235538034580258816 & 7.64 & @xcite + ugc 6456 & 11 28 00 & + 78 59 39 & & & 7.35 & + sbs1129 + 576 & 11 32 02 & + 57 22 46 & 16.7 & & 7.36 & + j1151 - 0222@xmath28 & 11 51 32 & -02 22 22 & 16.8 & 93111671593107456 & 7.78 & + j1201 + 0211@xmath28 & 12 01 22 & + 02 11 08 & 17.6 & 145464500043120640 & 7.49 & + sbs1159 + 545 & 12 02 02 & + 54 15 50 & 18.7 & & 7.41 & k , @xcite + sbs 1211 + 540@xmath28 & 12 14 02 & + 53 45 17 & 17.4 & 287049377199947776 & 7.64 & + j1215 + 5223 & 12 15 47 & + 52 23 14 & 15.2 & 248767590061572096 & 7.43 & @xcite + tol1214 - 277 & 12 17 17 & -28 02 33 & & & 7.55 & k , + vcc 0428 & 12 20 40 & + 13 53 22 & 17.0 & 497314452257898496 & 7.64 & @xcite + hs 1222 + 3741 & 12 24 37 & + 37 24 37 & 17.9 & 564023911881113600 & 7.64 & * + tol65 & 12 25 47 & -36 14 01 & 17.5 & & 7.54 & k , + j1230 + 1202@xmath28 & 12 30 49 & + 12 02 43 & 16.7 & 454810434122285056 & 7.73 & k , + kissr 85 & 12 37 18 & + 29 14 55 & 19.9 & & 7.61 & @xcite + ugca 292 & 12 38 40 & + 32 46 01 & 18.9 & & 7.28 & + hs 1236 + 3937 & 12 39 20 & + 39 21 05 & 18.5 & & 7.47 & * + j1239 + 1456 & 12 39 45 & + 14 56 13 & 19.8 & & 7.65 & @xcite + sbs 1249 + 493 & 12 51 52 & + 49 03 28 & 18.0 & & 7.64 & @xcite + j1255 - 0213@xmath28 & 12 55 26 & -02 13 34 & 19.1 & 95080395053203456 & 7.83 & + gr8 & 12 58 40 & + 14 13 03 & 17.9 & & 7.65 & k , @xcite + kissr 1490 & 13 13 16 & + 44 02 30 & 19.0 & & 7.56 & @xcite + ddo 167 & 13 13 23 & + 46 19 22 & & & 7.20 & + hs 1319 + 3224 & 13 21 20 & + 32 08 25 & 18.6 & & 7.59 & + j1323 - 0132@xmath28 & 13 23 47 & -01 32 52 & 18.2 & 96204976459612160 & 7.78 & @xcite + j1331 + 4151@xmath28 & 13 31 27 & + 41 51 48 & 17.1 & 412307391565004800 & 7.75 & @xcite + eso577-g27 & 13 42 47 & -19 34 54 & & & 7.57 & k , + j1355 + 4651@xmath28 & 13 55 26 & + 46 51 51 & 19.3 & 361921789577658368 & 7.63 & + j1414 - 0208 & 14 14 54 & -02 08 23 & 18.0 & 258056040799535104 & 7.28 & + sbs1415 + 437 & 14 17 01 & + 43 30 05 & 17.8 & & 7.43 & k , @xcite + j1422 + 5145 & 14 22 51 & + 51 45 16 & 20.2 & & 7.41 & @xcite + j1423 + 2257@xmath28 & 14 23 43 & + 22 57 29 & 17.9 & 600334402043510784 & 7.72 & @xcite + j1441 + 2914 & 14 41 58 & + 29 14 34 & 20.1 & & 7.47 & @xcite + hs1442 + 4250 & 14 44 13 & + 42 37 44 & 15.9 & & 7.54 & + j1509 + 3731@xmath28 & 15 09 34 & + 37 31 46 & 17.3 & 394011865673367552 & 7.85 & @xcite + kissr 666 & 15 15 42 & + 29 01 40 & 19.1 & & 7.53 & + kissr 1013 & 16 16 39 & + 29 03 33 & 18.2 & 396262437692637184 & 7.63 & @xcite + j1644 + 2734 & 16 44 03 & + 27 34 05 & 17.7 & 475922385527111680 & 7.48 & @xcite + j1647 + 2105@xmath28 & 16 47 11 & + 21 05 15 & 17.3 & 442143986740625408 & 7.75 & @xcite + w1702 + 18 & 17 02 33 & + 18 03 06 & 18.4 & & 7.63 & @xcite + hs 1704 + 4332 & 17 05 45 & + 43 28 49 & 18.4 & & 7.55 & + sagdig & 19 29 59 & -17 40 41 & & & 7.44 & k , @xcite + j2053 + 0039 & 20 53 13 & + 00 39 15 & 19.4 & 288175754707992576 & 7.33 & + j2104 - 0035 & 21 04 55 & -00 35 22 & 17.9 & 288457262211530752 & 7.05 & + j2105 + 0032 & 21 05 09 & + 00 32 23 & 19.0 & 288457264749084672 & 7.42 & + j2120 - 0058 & 21 20 26 & -00 58 27 & 18.8 & 289300532823064576 & 7.65 & + hs2134 + 0400 & 21 36 59 & + 04 14 04 & & & 7.44 & @xcite + j2150 + 0033 & 21 50 32 & + 00 33 05 & 19.3 & 414839880762261504 & 7.60 & + eso146-g14 & 22 13 00 & -62 04 03 & & & 7.59 & k , + 2df 171716 & 22 13 26 & -25 26 43 & & & 7.54 & + phl293b & 22 30 37 & -00 06 37 & 17.2 & & 7.62 & + 2df 115901 & 22 37 02 & -28 52 41 & & & 7.57 & + j2238 + 1400@xmath28 & 22 38 31 & + 14 00 30 & 19.0 & 533060792673632256 & 7.45 & + j2250 + 0000 & 22 50 59 & + 00 00 33 & 19.8 & 190501187865280512 & 7.61 & @xcite + j2259 + 1413 & 22 59 01 & + 14 13 43 & 19.1 & & 7.37 & @xcite + j2302 + 0049@xmath28 & 23 02 10 & + 00 49 39 & 18.8 & 190784502518251520 & 7.71 & + j2354 - 0005 & 23 54 37 & -00 05 02 & 18.7 & 307596757485748224 & 7.35 & + [ revision ] in order to put our search into context , we carried out a careful revision of the literature to select all the nearby galaxies with metallicity reported to be one - tenth solar or less . ( @xmath36 ) . those are listed in table [ revision ] . we start off from table 3 in @xcite , who compile all the galaxies found in the literature prior to year 2000 . it comprises 31 targets , from which * 5 * were discarded because their metallicity has been revised upwards . then we cover the last ten years by checking cross - references existing in recent known papers . we begin with the work by , that analyzes * 44 * targets selected in sdss / dr6 . the spectra used in the analysis are not spectra from sdss but obtained elsewhere . we choose from all the galaxies with oxygen abundance below the threshold . then , using ads , we revised all papers cited in and published after @xcite . based on their title and abstract , we examined those papers dealing with galaxy metallicity , separating the appropriate xmp galaxies . high redshift targets were not included ( e.g. , those in * ? ? ? the procedure was repeated with all the papers containing xmp galaxies , until no new reference earlier than @xcite was found . in addition , our recursive searching strategy was repeated with all the papers citing . for example , @xcite is cited by and provides 13 targets from sdss / dr5 . study spectroscopically and morphologically 7 galaxies identified in sdss / dr4 and 6dfgs ( six - degree field galaxy survey , * ? ? ? * three * galaxies are exhaustively analyzed in three separate papers by , and are also included in table [ revision ] . two additional galaxies are from the study by . through this step - by - step search procedure , we revised all the papers that seems to be relevant , and a number of them rendered one or several additional xmp galaxies for table [ revision ] . finally , the galaxies in table [ list1 ] were inspected individually in ned , to assure that they are included in our bibliographic search even when the estimated metallicity exceeds the imposed threshold . table [ revision ] contains all galaxies we found in the literature through this search . it includes only * 129 * galaxies , proving that the xmp targets are really rare objects . the table includes name , coordinates , sdss specobjid ( when available ) , the reference for the oxygen metallicity , as well as a tag to identify galaxies included in the work by @xcite used as reference . the match between tables [ list1 ] and [ revision ] renders 21 objects , i.e. , only 21 of the 32 candidates had been previously identified as xmp galaxies . they are marked in the tables with asterisks . the remaining 11 objects are new . the overlapping between the two tables allows us to confirm the low metallicity of the xmp candidates selected using state - of - the - art metallicity determinations . the oxygen abundance in table [ revision ] are based on self - consistently determined electron temperatures . assuming the twenty one known objects to be representative of the full set , then mean metallicity turns out to be @xmath37 , with the error bar accounting for the standard deviation . figure [ spatial ] shows the spatial distribution of the low metallicity targets , i.e. , the * 129 * galaxies obtained from the literature plus the targets found in this work . note that most works have focused the search in very localized areas on the sky . for example , the targets by are clustered around the equator , even though the search is based on sdss / dr6 which has a much broader scope . our systematic search , however , has selected targets spread throughout the sdss / dr7 field of view , that covers a significant part of the sky ( @xmath38% ) . figure [ diag_xmps ] shows diagnostic plots like fig . [ diag_bcd ] but for the galaxies in the literature with spectra in sdss . assuming this subset to be representative of the full family , it shows several systematic differences with respect to the targets we have selected . the main one has to do with the ratio between @xmath26}\lambda$]6583 and h@xmath0 , which is smaller than those characterizing our targets ( cf . [ diag_bcd]d , and fig . [ diag_xmps]d ) . because of this difference , many of the xmp found in the literature would have escaped our search , which pick out classes with extreme contrast between @xmath26}\lambda$]6583 and h@xmath0 . we do not have a final explanation for the difference , but we can offer two conjectures . our search may yield galaxies in the low metallicity end of the xmp family , i.e. , even more extreme than those existing in the literature and , consequently , even more metal poor than our estimates based on the semi - empirical calibrations by @xcite . alternatively , our search may be missing a fraction of the xmp galaxies , where the difference between the [ nii ] lines and h@xmath0 is not so extreme . the two possibilities explain other significant differences between our set and the targets from the literature . our galaxies have larger surface brightness than those of the xmp galaxies in the literature ( cf . [ diag_bcd]a and fig . [ diag_xmps]a ) . they have larger h@xmath0 equivalent widths as well . the surface brightness difference is attributable to the trend for the most metal poor galaxies to be bcds ( see [ properties ] ) . alternatively , the physical conditions in the hii regions of the bcds may differ systematically from other galaxies so that the same low metallicities render a particularly small ratio between @xmath26}\lambda$]6583 and h@xmath0 in bcds . in order to sort out the two possibilities one would have to determine the metallicities of our targets in a self - consistent way measuring electron temperatures and excitations ( as in , e.g. , * ; * ? ? ? such a detailed analysis clearly goes beyond the scope of the paper , and it is planned for following up work . unfortunately , the twenty one candidates that were already known to be xmp do not clarify the situation . some have particularly low metallicities , but not all of them ( see the names with asterisks in table [ revision ] ) . one of the advantages of carrying out a systematic search in sdss / dr7 is starting from a magnitude limited sample . the relatively simple bias produced by this condition can be corrected for and , thus , the selected xmp galaxies can be used to estimate the volume number density of these unusual objects in the local universe . such estimate is described in the present section . assume the metallicity @xmath39 of all galaxies in sdss / dr7 to be known . then the number of galaxies per unit volume and metallicity is just @xmath40 where the sum over @xmath41 includes all the galaxies in the sample , @xmath42 represents the bin size of the metallicity histogram , the symbol @xmath43 stands for the rectangle function , @xmath44 and @xmath45 represents the maximum volume in which the @xmath41-th galaxy of the sample could be observed . equation ( [ pi_eq ] ) represents the so - called @xmath46 approximation by @xcite used to determine luminosity function of galaxies ( e.g. , * ? ? ? * ) , which has been transformed to derive the number density of any other physical property of the galaxies ( e.g. , * ? ? ? assume the sample to be magnitude limited , so that all galaxies brighter than the apparent magnitude @xmath47 are included . ( this is the way the spectroscopic sample of sdss was defined and therefore it must be good approximation . ) then , @xmath48 with @xmath49 the solid angle covered by the survey and @xmath50 the maximum distance at which the @xmath51th galaxy can be observed , @xmath52 which only depends on its absolute magnitude @xmath53 . a few comments and caveats are in order before applying equation ( [ pi_eq ] ) to the dataset . first , sdss is not truly magnitude limited . some bright galaxies are not observed because of problems to pack the spectrograph fibers in crowded fields ( see * ? ? ? * ) and , as usual , very low surface brightness galaxies tend to be missed even if they are luminous ( e.g. , * ? ? ? none of these two problems seems to be of relevance for the xmp galaxies , which are both isolated and high surface brightness ( [ properties ] ) . sdss / dr7 sets @xmath54 in @xmath55 , but our selection also imposes a cut at redshift @xmath16 which , in principle , modifies the magnitude - limited character of the original sample . this question is of no concern , though . our targets are dwarf galaxies , never reaching a luminosity sufficient to be detected beyond the redshift threshold . finally , computing the galaxy metallicities using detailed chemical abundance analyses ( as in , e.g. , * ? ? ? * ; * ? ? ? * ) goes beyond the scope of the paper . we estimate the metallicities using the strong line calibration of by @xcite described in [ kmeans ] . they are given in table [ list1 ] . specifically , we employ the calibration in equation ( [ pet2 ] ) , which seems to be more appropriate for xmp galaxies where n2@xmath56 ( n2 is the logarithm of ratio between the equivalent widths of the lines , as defined in equation [ [ n2def ] ] ) . the use of this approximation for the estimate of the metallicity implies that our estimates are only indicative . the 32 xmp galaxies selected in [ kmeans ] , together with equations ( [ pet2 ] ) , ( [ pi_eq ] ) , ( [ pi_eq6 ] ) , ( [ pi_eq7 ] ) , and ( [ pi_eq8 ] ) , render the local density of galaxies with ( oxygen ) metallicity less than , approximately , one tenth of the solar value , @xmath57 , @xmath58 the error bar in the previous expression assumes the 32 targets to be drawn from a poisson distribution . in terms of the total number of galaxies , xmp galaxies are just @xmath59 where the total number of galaxies in the local universe ( @xmath60 ) has been taken from the normalization of the luminosity functions by @xcite . these luminosity functions are also based on sdss , and we use a hubble constant of 70 kms@xmath61mpc@xmath61 to revert their normalization . the error bar in equation ( [ per_xmp ] ) considers the dispersion quoted in equation ( [ numberd1 ] ) together with the spread of values among the different luminosity functions corresponding to the five sdss color filter bandpasses . going a step further , we have considered all the galaxies with emission lines in sdss / dr7 to estimate @xmath62 in a broader range of metallicities . the sdss / dr7 data reduction pipeline provides the equivalent widths of [ nii]@xmath56583 and h@xmath0 , which we use to estimate oxygen abundances . the distribution function @xmath62 inferred from all these galaxies is shown in fig . [ histograms]a , the solid line . it has been computed using equation ( [ pi_eq ] ) with @xmath63 , and considering only n2 @xmath64 ( @xmath65 ) , with this upper limit forced by the validity range of the calibration used to estimate abundances ( see * ? ? ? * ) . the horizontal bar in fig . [ histograms]a corresponds to the density in equation ( [ numberd1 ] ) , assuming the 32 xmp galaxies to spread one dex in metallicity . it is consistent with this other independent estimate inferred from the full distribution of sdss / dr7 galaxies if one integrates the histogram @xmath62 in fig . [ histograms]a for galaxies with metallicities smaller than one - tenth solar ( @xmath66 ) , the volume number density of xmp galaxies turns out to be , @xmath67 , which is close to the figure in equation ( [ numberd1 ] ) . the integral of @xmath62 in fig . [ histograms]a for all metallicities turns out to be @xmath68 , which is a factor two smaller than , but consistent with , the number density of local galaxies from @xcite used in equation ( [ per_xmp ] ) for normalization . one have to keep in mind that the distribution in fig . [ histograms]a do not consider neither galaxies without emission lines nor galaxies with super - solar metallicities , which all together can easily account for the factor two difference . xmp galaxies are dwarf and therefore they tend to be underrepresented in surveys . in order to illustrate the importance of such malmquist bias , fig . [ histograms]b contains the actual histogram of observed galaxies @xmath69 from which @xmath62 was derived . in the parlance of equation ( [ pi_eq ] ) , it is given by @xmath70 note how the low metallicity tail of @xmath69 is depressed with respect to @xmath62 . this difference can be better appreciated in fig . [ histograms]a , which also shows a scaled version of @xmath69 forced to agree with @xmath62 at the bin of highest metallicity . the figure shows how xmp galaxies are clearly lacking in the observation ( the dashed line ) as compared to their actual number densities ( the solid line ) . we find than only 0.01% of the observed galaxies with emission lines are xmp , whereas they represent 0.2% considering galaxies in a fixed volume . obviously , the above estimates assume the xmp galaxy set to be complete . however , we can not discard the existence of missing targets ( [ biblio_search ] ) and , if so , the figures we provide have to be scaled up accordingly . although it is difficult to tell how much , we will try to make an educated guess . about sixty known xmp galaxies in the sdss field of view have not been included in our selection ( see fig . [ spatial ] ) . they may have been excluded for several reasons because they are too faint , because noise in the spectra artificially enhances [ nii]@xmath56583 with respect to h@xmath0 , because [ nii]@xmath56583 exceeds the limits of our selection , and possibly others . since some of them are proper reasons for exclusion , the number of sixty undetected galaxies may be taken as an upper limit for those missing . consequently , the number densities we estimate may need to be increased , but only up to a factor of three . $ ] , the solid line ) . the bar represents the average value inferred from the 32 xmp galaxies identified in this study . ( b ) histogram of galaxies with a given metallicity as directly observed in sdss / dr7 ( the dashed line ) . a scaled up version of this histogram is also shown as the dashed line in ( a ) . note how the already scarce xmp galaxies , represented by the solid line , are hindered even further from observation because they are dwarf , and so , only observable in our neighborhood . , title="fig:",scaledwidth=45.0% ] + $ ] , the solid line ) . the bar represents the average value inferred from the 32 xmp galaxies identified in this study . ( b ) histogram of galaxies with a given metallicity as directly observed in sdss / dr7 ( the dashed line ) . a scaled up version of this histogram is also shown as the dashed line in ( a ) . note how the already scarce xmp galaxies , represented by the solid line , are hindered even further from observation because they are dwarf , and so , only observable in our neighborhood . , title="fig:",scaledwidth=45.0% ] we have carried out a systematic search for extremely metal poor ( xmp ) galaxies among the spectroscopic sample of sdss / dr7 . these objects are rare , and have a clear cosmological interest as unevolved galaxies probably tracing physical conditions in an early phase of the universe ( see [ intro ] ) . the search is based on the classification of a narrow spectral region around h@xmath0 , known to be particularly sensitive to the ( hii gas ) metallicity . obviously , the almost one million spectra in the database can not be inspected individually , and we resort to the use of a standard automatic method of classification : k - means . after two nested runs of the procedure , and a subsequent cleaning up for artifacts created by the sdss pipeline , we end up with 32 targets ( [ kmeans ] , table [ list1 ] ) . they represent only 0.01% of the observed galaxies with emission lines . the final metallicity estimate remains pending , however , strong - line empirical calibrations by @xcite imply their oxygen metallicity to be of the order of one - tenth solar . obviously , our candidates must be studied in detail through imaging and spectroscopy in following up work . in order to put the work into context , we carried out a bibliographic search for galaxies with metallicity smaller than one - tenth the solar value ( [ biblio_search ] ) . we find only 129 ( table [ revision ] ) , and only 21 of them overlap with our sample which , consequently , provides 11 new xmp candidates . the oxygen metallicity of the 21 known targets turns out to be @xmath71 . these metallicities are based on electron temperatures and , thus , they are more precise than the empirical calibrations we have been employing . assuming this subset to be representative of the full sample , it confirms the xmp character of our targets . our procedure is systematic , therefore , in principle , it should have identified all the spectra in sdss / dr7 where h@xmath24}\lambda$]6583 . this assumption , together with the fact that the sdss spectroscopic sample is limited in apparent magnitude , allows us to estimate the volume number density of xmp galaxies . using the @xmath72 approximation , we estimate it to be @xmath3 . so far as we are aware of , it provides the first estimate of this number density . the xmp galaxies represent 0.1% of the galaxies in the local volume , or @xmath4 0.2% considering only emission line galaxies . we analyze some of the physical properties of the candidates in [ properties ] . all but four of our xmp candidates turn out to be blue compact dwarfs ( bcds ) . note that this association is by no means trivial . we have selected our sample according to the shape of their spectra in an narrow spectral window around h@xmath0 , and this narrow bit of spectrum turns out to determine many global properties of the galaxy such as color , compactness , and star formation rate . we ignore what causes the association between xmps and bcds . the fact that most metal poor galaxies tend to be bcds is known . @xcite point it out , but warn against a trivial interpretation since it may reflect an observational bias that makes it easier to detect high surface brightness objects such as bcds . actually , there are xmp galaxies whose surface brightness does not suffice to call them bcds ( see [ biblio_search ] ) . however , the fact that our xmp candidates are also bcds is revealing . the surface brightness where sdss starts having problems of completeness is about @xmath73 @xcite , i.e. , one magnitude fainter than the faintest xmp candidate . consequently , if we do not find low surface brightness galaxies with @xmath26}\lambda 6583 \ll { \rm h}\alpha$ ] is because they do not exist in sdss / dr7 . among the 32 xmp candidates , 24 of them have either cometary shape or are formed by chains of knots . already noticed the trend for xmp bcds to reveal a cometary morphology due to the presence of intense star formation at one edge that gradually decreases . this shape is not unique to xmp galaxies , but it comprises only 10% of all bcds @xcite , whereas it is dominant when they are metal poor . the origin of the xmp shapes is also unclear . there are arguments for gravitational triggering due to mergers with low - mass stellar or gaseous companions , or for the self propagation of star formation activity within a pre - existing gas rich galaxy . recently @xcite and @xcite have found that all xmp galaxies have distorted hi morphologies , which may indicate the infall of external unenriched gas feeding the starburst and dropping the metallicity ( e.g. , * ? ? ? . it may also be the signature of gas stripping forced by the interaction with an external medium ( e.g. , * ? ? ? * ; * ? ? ? a concluding remark is in order . the methods for mining massive astronomical databases are still under development . the bases are simply too large for the traditional techniques to be efficient . in this sense , the current paper presents a new approach that may be of interest beyond our particular application . the standard method to find xmp galaxies ( or galaxies with any other property ) would have been to set , beforehand , the observational criteria the targets should fulfill . then those targets complying with the criteria would have been selected . unfortunately , the criteria are often crude and have large uncertainties , which propagate into the selection as false objects sneaking in or true objects leaking out . here the criteria have not been stipulated in advance . we classify the whole database , and only afterward we select those galaxies belonging the classes that have the intended property . the search is comprehensive , and the results are robust against uncertainties in the selection criteria . thanks are due to b. elmegreen for fruitful discussions on the origin of the cometary shapes in xmp galaxies . thanks are also due to an anonymous referee for helping us completing the bibliographic search , as well as for insightful comments that prompted an improvement of both the argumentation and the presentation . this work has been partly funded by the spanish micinn , projects aya 2007 - 67965 - 03 - 01 , aya 2007 - 67752-c03 - 01 , and aya 2010 - 21887-c04 - 04 . jsa , ala and cmt are members of the consolider - ingenio 2010 program , grant micinn csd2006 - 00070 : first science with gtc . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . department of energy , the national aeronautics and space administration , the japanese monbukagakusho , the max planck society , and the higher education funding council for england . the sdss is managed by the astrophysical research consortium for the participating institutions ( for details , see the sdss web site at http://www.sdss.org/ ) . _ facilities : _

we study the effect of galactic outflows on the statistical properties of the ly forest and its correlation with galaxies . the winds are modelled as fully ionised spherical bubbles centered around the haloes in an n - body simulation of acdm model . the observed flux probability distribution and flux power spectrum limit the volume filling factor of bubbles to be less than . we have compared the mean flux as a function of distance from haloes with the adelberger ( assp ) measurement . for a model of bubbles of constant size surrounding the most massive haloes , bubble radii of are necessary to match the high transmissivity at separations but the increase of the transmissivity at small scales is more gradual than observed . the cosmic variance error due to the finite number of galaxies in the sample increases rapidly with decreasing separation . at separations our estimate of the cosmic variance error is , 30% higher than that of assp . the difficulty in matching the rise in the transmissivity at separations smaller than the size of the fully ionised bubbles surrounding the haloes is caused by residual absorption of neutral hydrogen lying physically outside the bubbles but having a redshift position similar to the haloes . the flux level is thus sensitive to the amplitude of the coherent velocity shear near halos and to a smaller extent to the amplitude of thermal motions . we find that the velocity shear increases with halo mass in the simulation . a model where lbgs are starbursts in small mass haloes matches the observations with smaller bubble radii than a model where massive haloes host the lbgs . if we account for the uncertainty in the redshift position of haloes , a starburst model with a bubble radius of and a volume filling factor of 2% is consistent with the assp measurements at the 1 - 1.5 level . if this model is correct the sharp rise of the transmissivity at separations in the assp sample is due to cosmic variance and is expected to become more moderate for a larger sample . # 1_#1 cosmology : theory gravitation dark matter baryons intergalactic medium

astro-ph0311209

stellar feedback , in the form of ionising radiation and mechanical energy in galactic winds / outflows , plays an important role in determining the physical properties of galaxies and the intergalactic medium ( igm ) . the galactic contribution to the hydrogen ionising uv radiation may well exceed that of qsos . galactic superwinds regulate star formation , enrich the igm with metals , and alter the dynamical and observational properties of the igm close to galaxies . galactic winds are driven by the mechanical energy produced by supernova explosions of massive stars . they have been studied extensively in nearby starburst galaxies ( _ e.g. _ heckman 1990 ) . the large velocity shifts between stellar and interstellar lines in the spectra of lyman - break galaxies ( lbgs ) are strong evidence that galactic winds are also present in high - redshift galaxies ( pettini 1998 , 2000 , heckman 2000 ) . the inferred wind velocities are several hundred @xmath9 , far above the sound speed of the igm so that strong shocks are likely to be associated with these winds . these shocks will heat up , collisionally ionise , and sweep up the surrounding igm near galaxies ( heckman 2000 ) . adelberger ( 2003 , hereafter assp ) have analyzed a sample of high resolution qso spectra with lines - of - sight ( los ) passing very close to lbgs . they detected a significant decrease of the ly@xmath0 absorption near galaxies . as discussed by assp and kollmeier et al . ( 2003 ) the increased uv flux close to lbgs falls short of explaining the decreased absorption by a factor of a few . the decreased absorption is thus taken as evidence for the existence of dilute and highly ionised gas bubbles caused by wind - driven shocks . initially a shock propagating in the igm expands adiabatically . during this phase the gas encountered by the shock is heated and collisionally ionised to become part of a bubble of dilute and hot plasma . when radiative losses become important , a thin cool shell forms at the shock front . the shell is driven outwards by the hot low density inner plasma ( e.g. weaver 1977 , ostriker & mckee 1988 , tegmark 1993 ) . eventually the expansion of the plasma comes to a halt when pressure equilibrium with the surrounding igm is reached . to explain the lack of close to the haloes , the bubble has to survive for a sufficiently long time . we defer the discussion of the conditions for the survival of the bubble to another paper . here we assume that the bubbles are long - lived and study the implications of long - lived bubbles devoid of neutral hydrogen on qso absorption spectra . we will investigate how the correlation between the flux in the ly@xmath0 forest and the distance of the los to the closest galaxy depends on physical properties of the ionised bubbles around galaxies . for this purpose we use a n - body simulation of collisionless dark matter ( dm ) . the basic observed properties of the ly@xmath0 forest can be reproduced with such simulations using a simple prescription that relates the distribution of gas and dark matter in the igm ( hui & gnedin 1997 ) . with dm simulations it is possible to probe a wider dynamical range than with hydrodynamical simulations . this allows a better modelling of the statistical properties of the igm - halo relation . measures of the halo - flux correlation , defined for example as the mean flux in the ly@xmath0 forest as a function of the distance to the nearest galaxies , are sensitive to the size of the bubbles . absorption spectra probe the gas distribution in redshift space . thermal motions and coherent peculiar velocities therefore strongly affect the halo - flux correlation at small separations ( e.g. kollmeier 2003 ) . previous modelling of the impact of winds on the ly@xmath0 forest had difficulties to match the large decrease of the absorption measured by assp at comoving separation @xmath10 from the galaxies ( croft 2002 , bruscoli 2002 , theuns 2002 , kollmeier 2003 ) . we will investigate here the role of peculiar velocities , thermal motions and the scaling of bubble size with halo mass in more detail . we will also perform a detailed assessment of the errors to obtain a more robust lower limit on the size and thus the volume filling factor of the bubbles . the paper is organized as follows . in [ lyman ] we describe the model of the ly@xmath0 forest which we have used to calculate synthetic spectra from the dm simulations . in [ nbody ] , we describe the simulation and the corresponding halo catalogs , compute three - dimensional flux maps of the simulation , and discuss the flux probability distribution function and power spectrum in the presence of winds . the main body of the paper are [ xihf ] and [ fluxsim ] where we study in detail the impact of galactic winds on the halo - flux correlation and the corresponding observational errors . we conclude with a discussion of the results in [ discussion ] . the ly@xmath0 forest is now widely believed to originate from an undulating warm ( @xmath11k ) photoionized igm . the ly@xmath0 optical depth in redshift space due to resonant scattering can be expressed as a convolution of the real space density along the line of sight with a voigt profile @xmath12 ( gunn & peterson 1965 , bahcall & salpeter 1965 ) , @xmath13 dx\ ; . \label{depth}\end{aligned}\ ] ] where @xmath14 is the effective cross - section for resonant line scattering and h@xmath15 is the hubble constant at redshift @xmath16 , @xmath17 is the real space coordinate , @xmath18 is the line of sight component of the peculiar velocity field , @xmath12 is the voigt profile , and @xmath19 is the doppler parameter due to thermal / turbulent broadening . for thermal broadening @xmath20 , where t is the temperature of the gas . both @xmath17 and @xmath21 are measured in @xmath9 . the voigt profile is normalized such that @xmath22 and can be approximated by a gaussian for moderate optical depths , @xmath23 ^ 2/b^2\right\}$ ] . the normalized flux obtained by `` continuum fitting '' of the observed spectrum is related to the optical depth along the line of sight as , @xmath24 where @xmath25 is the optical depth , @xmath21 is the redshift space coordinate along the line of sight , @xmath26 is the observed flux and @xmath27 is the flux emitted from the source ( quasar ) that would be observed in the absence of any intervening material . on scales larger than a filtering scale which is related to the jeans scale , @xmath28 , the igm traces the dm distribution very well . for moderate overdensities the balance between adiabatic cooling and photo - heating of the expanding igm establishes power - law relations between temperature , neutral hydrogen density and total gas density ( katz 1996 , hui & gnedin 1997 , theuns 1998 ) , @xmath29 where the parameter @xmath30 is in the range @xmath31 , @xmath32 , and @xmath33 and @xmath34 are the density and temperature of the gas , respectively . on scales smaller than the jeans scale pressure dominates over gravity . the gas pressure smoothes the gas distribution compared to the distribution of collisionless dark matter ( _ e.g. _ theuns , schaye & haehnelt 2000 ) . we thus assume that @xmath35 , where @xmath36 is the dark matter density smoothed in a way that mimics the pressure effects . we adopt the smoothing used in sph simulations ( e.g. springel , yoshida & white 2001 ) to estimate the gas density from the dark matter particle distribution . the density @xmath37 and the smoothed dark matter field @xmath38 are then related as @xmath39^{\alpha}\ ; , \label{neutral}\ ] ] where @xmath40 is the ( volume ) average of @xmath38 , and @xmath41 is the density at @xmath42 . one of our main goals is to assess the error in the observed galaxy flux correlation . this requires a simulation large enough to contain a substantial number of dm haloes which can be identified as hosts of lbgs . at the same time the simulation should still reproduce typical absorption systems . this favours the use of a dm simulation for which a larger dynamical range can be achieved rather than a computationally more expensive hydrodynamical simulation . we use a @xmath1cdm simulation run with the n - body code gadget ( springel , yoshida & white 2001 ) . the cosmological parameters are listed in table [ table1 ] . it is part of a series of simulations of increasingly higher resolution which zoom in on a spherical region of initial comoving radius 26@xmath43 selected from a simulation with box size 479@xmath43 ( cf . stoehr 2002 ) . the inner region of the simulation has thus higher resolution than the outer regions . we only use the particle distribution inside a box of size @xmath44 which belongs to the spherical region of high resolution . the particle mass in this high - resolution region is @xmath45 ( cf . table [ table1 ] ) . the large volume makes the simulation well suited for a direct comparison with the observed galaxy - flux correlation . we use a redshift output @xmath46 to facilitate a comparison with the assp measurements . .the principle parameters of the n - body simulation . the number of particles , @xmath47 , and the particle mass , @xmath48 , in the high resolution region are also listed . [ cols="^,^,^,^,^,^",options="header " , ] [ table2 ] the and the flux distributions are calculated from the dm distribution which was smoothed with a sph kernel containing @xmath49 particles within the smoothing length . the smoothed dm density and velocity fields , @xmath38 and @xmath50 , were interpolated for randomly chosen los through the simulation box . the pixel size of the spectra is @xmath51 and the comoving length @xmath52 . from the density and velocity fields along the los we computed the optical depth as a function of the redshift space coordinate @xmath53 , where @xmath54 is the component of @xmath50 in the line of sight direction . ( [ depth ] ) then takes the following discrete form , @xmath55^{\alpha } { \cal h}_{ij}\ ; , \label{depthsim}\ ] ] where @xmath56 and we assume that @xmath38 is normalized such that its mean is unity and @xmath57 $ ] is a normalized gaussian profile . as we are using a dm simulation we still have to specify a temperature . for this we use the fact that temperature and density are tightly coupled . at moderate densities a simple power - law relation holds , @xmath58 ( cf . eq [ state ] ) . in high - density regions , line cooling is efficient and the gas rapidly cools to @xmath59 ( _ e.g. _ theuns 1998 , dav 1999 ) . we will thus take the following relation between temperature and density @xmath60 in the high - density regions where shock heating is important this is only a very crude approximation of the @xmath61 diagram of hydrodynamical simulation ( theuns 1998 , dav 1999 , croft 2002 ) . however , it is mainly the moderate density regions which are responsible for the ly@xmath0 forest . ( [ statesim ] ) should thus be sufficient for our purposes . the gas temperature @xmath62 at fixed overdensity is expected to evolve with redshift . schaye 2000 and theuns 2002 claim to have detected a peak at @xmath46 with decreasing temperature towards lower and higher redshifts . note that we have ignored a possible redshift dependence and have assumed @xmath63 independent of redshift . we have adopted a very simple model for the effect of wind bubbles . we simply assume that shocks produce long - lived _ fully ionised _ spherical bubbles around galaxies ( see appendix a for a discussion of the physical properties of expanding wind bubbles ) . we assume that the neutral hydrogen density is zero in a spherical region of radius @xmath64 centered on the dm haloes which we have chosen to identify as lbgs . note that this simple model has also been discussed in croft ( 2002 ) , kollmeier et al . 2003 ( 2003a,2003b ) and weinberg et al . ( 2003 ) . in the simulation , most of the haloes lie along filaments , or at the intersection of filaments , and the gas distribution around the haloes is often anisotropic on the scale of galactic winds ( e.g. croft 2002 , springel & hernquist 2003 ) . theuns ( 2002 ) have also shown that winds propagate more easily into the low density igm than in the filaments , giving rise to highly ionised regions which are generally not spherical . it should thus be kept in mind that our assumption of a spherical wind bubble will only be a reasonable approximation for strong isotropic winds . furthermore the bubbles do not necessarily live longer than a hubble time . a discussion of the physical properties of shocks in the igm can be found in appendix [ cooltimescale ] . once we have smoothed the dark matter density @xmath38 , the flux distribution depends only on @xmath65 and @xmath0 and the size distribution of wind bubbles . we have assumed @xmath66 and constrain @xmath65 from a random sample of 500 lines of sight of comoving length @xmath44 by demanding that the mean flux @xmath67 is equal to the observed value @xmath68 ( rauch 1997 , mcdonald 2000 ) . the effect of varying the assumed size distribution of wind bubbles are discussed in the following sections . the spectral resolution is @xmath69 , somewhat smaller than in the assp data , where it is @xmath70 . note that the mean inter - particle distance in the simulation is @xmath71 . this is sufficient to resolve @xmath72 on a scale @xmath73 as in the observations , but is somewhat too low to resolve all the features of the ly@xmath0 absorption lines ( e.g. theuns 1998 , bryan 1999 ) . we also add noise with signal - to - noise s / n=50 per pixel of width @xmath74 . we will first investigate some basic flux statistics of our synthetic spectra to check if they are a reasonable representation of observed spectra and to study the effect of the wind bubbles . we begin with the flux probability distribution function ( pdf ) defined as the fraction of the volume of the igm which has a flux value within a certain range . the right panel of fig . [ fig2 ] compares the flux pdf at @xmath46 for wind bubbles with different radii @xmath64 . the filled circles show the observed pdf from mcdonald 2000 , while the solid , long - dashed , short - dashed and dotted curves correspond to @xmath75 ( i.e. , no wind ) , 0.5 , 1 and @xmath76 , respectively . following mcdonald ( 2000 ) we use 21 bins of width @xmath77 to compute the flux pdf . the pdf of spectra without wind bubbles ( solid curve ) agrees very well with the pdf of the observed spectra ( filled circles ) . including wind bubbles with @xmath78 has no significant effect on the flux pdf , and even when @xmath64 is as large as @xmath76 there is only a slight discrepancy with the flux pdf of the observed spectra in saturated regions with @xmath79 . the small effect of the wind bubbles is due to their small volume filling factor @xmath80 , which is about 5% for the model with @xmath81 . this is consistent with the findings of theuns ( 2002 ) and weinberg et al . ( 2003 ) . we can place an upper bound on the volume filling factor , @xmath82 , of bubbles by demanding a good match between the simulated and observed pdfs . we have computed the flux pdf for several bubbles radii for which @xmath80 is in the range 0.01 - 0.5 . to account for the uncertainty in the mass of lbgs , we performed three computations for each @xmath80 : we selected dm haloes with mass @xmath83 larger than @xmath84 , @xmath85 and @xmath86 , and adjusted @xmath64 which gives the desired @xmath82 . we find that independent of halo mass , a filling factor @xmath87 yields a poor match to the observations , especially in the range @xmath88 , as shown in the small plot inside the right panel of fig . [ fig2 ] . in the left panel of fig . [ fig2 ] we show the dimensionless one - dimensional flux power @xmath89 as a function of @xmath90 ( in @xmath9 ) , defined as @xmath91 where @xmath92 is the fourier transform of the flux contrast @xmath93 . the flux power spectrum was calculated with a fft routine from 500 synthetic spectra . note that our simulation box and therefore the spectra are not periodic which may lead to small artifacts at small scales . the flux power of our synthetic spectra and that of the observed spectra obtained by mcdonald ( 2000 ) also agree very well . bubbles of size as large as @xmath94 have again a very small effect due to their small volume filling factor ( e.g. croft 2002 , weinberg et al . 2003 ) . to demonstrate the role of peculiar velocities , the triangles in fig . [ fig2 ] show the flux pdf and the power spectrum in real space for @xmath75 obtained by setting the peculiar velocities and thermal motions to zero . the real space flux pdf and power spectrum significantly deviate from the observations . there is a variety of statistical measures which can describe the correlation between the flux along a los and nearby galaxies . here we will focus on the conditional flux function , @xmath72 . @xmath72 is the mean flux at real - space distance @xmath95 to the next halo , @xmath96 where @xmath97 is the total number of haloes , @xmath98 is the value of the flux at a distance @xmath99 from the nearest halo . note that we use the position of the haloes which we identify as host haloes of lbgs as a proxy for the galaxy position . results will thus depend on how this identification is done and we will later explore several possibilities . the summation is performed over a representative sample of haloes . the function @xmath72 can be expressed in terms of the unconditional two - point correlation @xmath100 as @xmath101 \;. \label{corr.con}\ ] ] relation ( [ corr.con ] ) also holds in redshift space . in the following @xmath102 and @xmath8 are the coordinates of @xmath103 parallel and perpendicular to the los , respectively . for the numerical simulation we know the full 3-dimensional flux field . we could therefore estimate @xmath104 from the transverse correlation @xmath105 . however , real observation will sample the flux field @xmath106 more sparsely . assp have therefore estimated @xmath104 from the averaged 2d halo flux correlation function @xmath100 using eq . ( [ corr.con ] ) in order to maximize the signal . we proceed as assp do and calculate @xmath107 as follows . we first pick a los direction . we then produce an ensemble of spectra along this direction with random offsets perpendicular to the chosen direction . for each pair of halo / spectrum we determine the flux @xmath108 , the distance @xmath8 at the point of closest approach , and the distance @xmath102 between the halo and a given pixel of the spectrum . this process is repeated for an ensemble of randomly chosen los directions . we then obtain @xmath109 by annular averaging @xmath108 , and binning in bins of comoving size @xmath110 . we further average over all haloes and los . in the left panel of fig . [ fig3 ] we plot the full 2d distribution of the average flux @xmath111 around haloes assumed to host lbgs . the haloes were assumed to have no galactic wind bubbles . the bin size is @xmath112 , twice smaller than that we use for @xmath72 . the shaded area shows bins for which @xmath72 is within 0.05 from the mean flux @xmath113 . contours are for flux levels increasing from 0.2 to 0.6 with decreasing line width . the contours are compressed along the line of sight as a result of the peculiar and thermal motions of the gas . at small separation the transmissivity is much smaller than the mean due to the increased density near haloes . redshift distortions cause @xmath104 ( as computed in assp ) to be different from the transverse correlation @xmath114 . this can be seen in the middle panel of fig . the effect is important for @xmath115 . note that previous work compared also the assp measurements with @xmath104 obtained from numerical simulations ( e.g. croft 2002 , kollmeier 2003 , bruscoli 2003 ) . the sample of assp contains 431 galaxies in 7 fields containing a qso ( one field contains two qsos ) . at large scales ( @xmath116 a few mpc ) the number of galaxy los pairs is reasonably large . however at small scales the number of available galaxies becomes very small . there are _ e.g. _ only 3 galaxies within a transverse separation of @xmath117 to the line - of - sight to a qso . obviously poisson errors will then be large and estimating these errors is thus important . in the following we use the term _ cosmic variance _ to describe the error resulting from the finite number of galaxies and qso spectra . at small scales cosmic variance is the dominant source of error . other sources of errors also affect the calculation of @xmath118 , mainly the uncertainty of the mean flux @xmath113 , and the redshift position of the galaxies . these errors will be discussed in section [ e : role ] . about one los from a bright qso is expected to pass through a region of angular dimensions comparable to that of our simulation box at @xmath46 . however , the correlation length of the flux is around a few @xmath43 . there will thus be about a hundred los through the simulation box which represent independent flux measurements . we therefore estimate the cosmic variance error as follows from the simulation . for each separation bin of width @xmath119 we perform monte - carlo realizations of @xmath104 , where @xmath104 is averaged over the halo / spectrum pairs lying in the given bin . the cosmic variance error is then the @xmath120 scatter around @xmath109 . the error of @xmath104 depends on the total number of galaxies / haloes with separation @xmath99 . a sample of about 40 los through our simulation box gives a cumulative halo function @xmath121 close to the cumulative galaxy function @xmath122 of the assp sample for separation @xmath123 . note also that the number of haloes with @xmath124 in the simulation is only about 110 . this means that we severely oversample the halo catalog when we estimate the cosmic variance on large scales . however , at the small scales in which we are mainly interested this should have little effect . we first explore the shape of @xmath72 assuming that all wind bubbles have the same radius , @xmath64 , independent of the halo mass that they surround . in the right panel of fig . [ fig3 ] we plot @xmath72 for haloes with mass @xmath125 . we show curves for @xmath126 , 1 and @xmath76 as indicated in the figure . the corresponding filling factors are @xmath127% , @xmath128% and @xmath129% respectively , small enough to ensure that the flux pdf , the power spectrum and the line distribution of the ly@xmath0 forest are not affected . the dashed - dotted curve corresponds to @xmath72 in the absence of outflows . the filled symbols show the observed correlation as given in assp . the shaded area shows the cosmic variance error , and is calculated ( cf . section [ errors ] ) from the cumulative halo distribution @xmath121 also shown as a histogram in the right panel of fig . [ fig3 ] ( right axis ) . the cosmic variance error is only shown for @xmath78 . the trend of decreasing mean flux with decreasing distance to the next galaxy in the observations is well reproduced by the numerical simulations . at small separation this trend is inverted if the wind bubbles are included . as kollmeier et al . ( 2003a , 2003b ) we find that the strong increase of the mean flux level at @xmath130 , is only reproduced for bubbles with size as large as @xmath131 . we thus confirm the difficulty of other authors to reproduce the mean flux at small separation unless rather large bubbles are included . bubbles of this size are , however , difficult to explain even with very efficient galactic superwinds ( croft et al . 2002 ) . there are thus two possibilities : the simple model of spherical wind bubbles of constant size around massive haloes does not describe the effect of winds of observed galaxies properly or the observed strong increase of the flux level in the assp is a statistical fluke . we will explore both possibilities in the following sections in more detail . kollmeier ( 2002 ) have pointed out that the peculiar velocities of matter infalling onto the galaxy / dm halo expected from hierarchical structure formation should have a strong influence on @xmath104 at small separations . in fig . [ fig4 ] we investigate the effect of coherent small - scale peculiar velocities and thermal velocities . in the left and middle panel we show @xmath104 with and without peculiar and thermal velocities for wind bubbles of size @xmath132 and @xmath133 , respectively . without peculiar velocity the increase of the mean flux at small separation is much more pronounced . this is easy to understand . in redshift space the velocity shear of the infalling surrounding material fills in the cavity which is caused by the fully ionised wind bubble in real space . for the smaller bubble size such an effect is also visible for the thermal velocities alone . neutral hydrogen lying at the boundary of the ionised bubbles leads to significant ly@xmath0 absorption at the redshift halo position when the ratio @xmath134 is large enough . the peculiar velocity will , however , always dominate over the effect of thermal velocities . this can be seen in the right panel where we have scaled the temperature of the gas up and down . this has very little effect on @xmath104 calculated with peculiar velocities even for the smaller bubble size . we have also computed @xmath104 for various values of the density threshold in eq . ( [ statesim ] ) . we found only small differences for @xmath135 . such overdensities occur typically within @xmath136 ( comoving ) from the halo ( kollmeier 2003 ) which is smaller than the bubble radii considered here . note that the good agreement of the flux pdf of the synthetic spectra with the mcdonald ( 2000 ) data found in section 3 disappeared if peculiar velocities were set to zero . this indicates that our simulations have similar peculiar velocities as the gas responsible for the observed absorption at least in a volume - averaged sense . in the left panel of fig . [ fig5 ] the black dots show the velocity dispersion for haloes of mass @xmath83 . for each dm halo the velocity shear @xmath137 along the line of sight was calculated along three perpendicular los directions for an interval @xmath138 centered on the halo position . the histogram shows the mean velocity dispersion averaged over bins with @xmath139 . the velocity shear clearly increases with halo mass . this trend is present for a range of @xmath140 , @xmath141 . the middle and right panels of fig . [ fig5 ] show the transmitted flux at separation @xmath10 , @xmath142 , as a function of the halo mass @xmath83 assuming that all haloes with @xmath143 are surrounded with bubbles of radius @xmath126 and @xmath94 , respectively . as expected @xmath144 is significantly larger for @xmath94 . the dependence on halo mass appears to be stronger for @xmath94 than for @xmath78 . in the former case , @xmath144 decreases with increasing @xmath83 by about @xmath145 from @xmath146 to @xmath147 . this is consistent with the trend in the left panel which predicts that low mass haloes satisfy the condition @xmath148 more often than massive haloes as a result of the @xmath149 dependence . for smaller bubbles , @xmath150 , this appears to be rarely the case , independent of the mass of the halo . to investigate the effect of the velocity field near haloes further we show the real space density and velocity field of the gas , as well as the absorption spectrum along two lines of sight passing haloes with @xmath151 ( left panel ) and @xmath152 ( right panel ) in fig . solid and dotted line are for the case with and without wind bubbles , respectively . the radius of the bubbles is @xmath78 for both haloes . the difference between the two cases is striking . for the smaller mass halo in the left panel the flux level at the position of the halo is @xmath153 if the bubble is included whereas @xmath72 saturates at the redshift space position of the more massive halo in the right panel . since the density fields near the haloes in both panels show similar features , the difference in the flux values is mainly due to the difference in the velocity fields near the two haloes . the velocity shear in the less massive halo ( left panel ) is quite modest . on the contrary , the velocity field near the massive halo ( right panel ) has a large infall on a physical scale of a few hundred @xmath9 , which moves some of the moderate density peaks close to the redshift space position of the halo . this effect occurs for a large fraction of haloes , and thereby reduces the overall expected @xmath72 at small scales by a few tens of percent . there is a number of other uncertainties which have to be taken into account when comparing the observed @xmath154 with that calculated from the numerical simulation . most important are the uncertain redshifts of the lbgs . there are large systematic shifts of several hundred @xmath9 between the redshift of nebular emission lines , ly@xmath0 emission and interstellar absorption lines which make the assignment of a redshift somewhat ambiguous . assp give @xmath155 or about @xmath156 as a typical error . we know however relatively little about the dynamical state of the responsible gas and there may well be larger systematic errors . the left panel of fig . [ fig7 ] shows the effect of adding a gaussian distributed error to the redshift of the dm haloes . the solid curve shows the same model as the solid curve in the right panel of fig . [ fig3 ] ( `` massive '' haloes , @xmath78 ) . long - dashed and short - dashed curves are for redshift errors of @xmath157 and @xmath158 , respectively . as expected introducing redshift errors smoothes out the large depression of the flux level at distances of up to @xmath159 to the line - of - sight . this substantially improves the agreement with the assp measurements on these scales . in the middle panel of fig . [ fig7 ] the shaded area shows the 1@xmath8 error of the simulated @xmath154 for our ensemble of lbg - size haloes with mass @xmath160 . the errorbars show the errors quoted by assp . to make the comparison easier the errors are plotted for the model with @xmath157 . the solid histogram shows the cumulative halo function @xmath121 ( right axis ) as a function of separation . at @xmath161 the cumulative halo function @xmath121 is close to the cumulative galaxy function @xmath162 of the assp sample , with @xmath163 and @xmath164 ( resp . 3 and 12 in the assp sample ) . our errors are generally larger than that of assp , at @xmath10 by about 30 percent . at larger scales the difference is smaller but note that there our sample is about 20% larger than that of assp . note also that the errors decrease faster than those of the transverse correlation @xmath165 . the mean flux level of qso spectra varies significantly from spectrum to spectrum ( 10 - 20 percent ) . we thus also investigated how @xmath72 depends on the value of the mean flux @xmath113 . in the right panel of fig . [ fig7 ] we show the effect of varying @xmath113 . the long - dashed curve is our fiducial `` massive '' haloes model , which has @xmath78 and @xmath157 . the upper ( solid ) and lower ( short - dashed ) curves show @xmath166 and 0.62 respectively . a larger ( lower ) mean transmitted flux @xmath113 raises ( lowers ) @xmath109 almost linearly by the same amount @xmath167 at all separations . @xmath68 matches the large scale assp measurements best . so far we have considered bubbles with sizes that are independent of mass . unfortunately , it is far from obvious what scaling with mass is appropriate . obviously , on average more stars and a larger total energy input from stellar feedback are expected in the more massive haloes . however , simple binding energy arguments show that winds should escape more easily from low mass haloes ( larson 1974 , dekel & silk 1986 ) . this may be properly taken into account by a constant bubble radius but this is very uncertain . we therefore investigate the effect of different power - law scalings of the bubble size with mass . the left panels of fig . [ fig8 ] compare the case of constant bubble size with those assuming a power - law relation @xmath168 with @xmath169 and @xmath170 , at the minimal halo mass @xmath171 . the top panels are for @xmath126 and the bottom panels are for @xmath172 . for @xmath169 @xmath104 is very sensitive to the value of @xmath64 and the mean flux at @xmath173 rises to values as large as @xmath174 for the larger of the two bubble sizes improving the agreement with the results of assp . however , this is at the cost of the good agreement with the data on larger scales . it appears thus difficult to reproduce the observations of assp at small and large scales simultaneously by changing the mass - radius relation of the bubbles . in the right panels of fig.[fig8 ] we investigate the effect of changing the minimum mass above which dm haloes are assumed to be surrounded by wind bubbles . changing @xmath175 will change the total space density of the halo as given in table [ table2 ] . there is less absorption if @xmath175 decreases , mainly as a result of the increasing bubble filling factor . with a bubble radius @xmath94 , the filling factor is @xmath176 and @xmath177 for @xmath178 and @xmath179 , respectively . so far we have assumed that lbgs are hosted by the most massive haloes available and have introduced a minimum mass to match their space density with that observed for lbgs . this assumes that lbgs are long - lived and have the same brightness over a fair fraction of the hubble time . this picture may be in conflict with the lack of the evolution of the space density of bright lbgs towards very high redshift which is more easily explained if lbgs are starbursts with a duty cycle of high star formation rate of less than 10% . this would mean that there is an about a factor ten or more larger number of dm haloes needed to host these lbgs than in the massive halo picture and the dm haloes hosting lbgs are less massive on average . as should be apparent from the previous section , for a given bubble size surrounding a typical lbg halo , this should reduce the effect of the velocity shear . it is then easier to reproduce the high mean flux levels at small separation . we investigate this alternative `` starburst '' picture here in more detail . kolatt ( 1999 ) developed a simple , _ ad hoc _ procedure to assign lbgs to colliding haloes identified in their n - body simulation and were able to reproduce both the number density and clustering properties of lbgs . here , we will do something simpler and randomly select haloes with mass @xmath143 such that their number density matches the observed lbgs number density , @xmath180 . this corresponds to a duty cycle of ten percent . note that only 10% of the haloes with a mass above @xmath181 are haloes as massive as our `` massive - halo '' picture i.e. have masses @xmath179 ( cf . table [ table2 ] ) . most of the lbg - size haloes have @xmath182 in the `` starburst '' scenario . we have produced 120 different monte - carlo realizations of such catalogs of starbursting haloes . in the left panel of fig . [ fig9 ] we compare @xmath104 for the `` massive - halo '' model ( solid curve ) and the `` starburst '' picture ( dashed curve ) . for both we assume a bubble radius @xmath94 and a redshift error of the halo position of @xmath157 . at @xmath183 @xmath104 is larger in the `` starburst '' model by about 20% . we also plot the 1@xmath8 error for the starburst model and the assp measurements . our starburst model with @xmath94 and a volume filling factor of 2% appears to be fully consistent with the assp measurement . even at @xmath184 , where our estimate of the cosmic variance error is 30% larger than that of assp , @xmath104 in the starburst model is only slightly more than @xmath120 below the assp measurement despite a difference in the mean flux of @xmath185 . note that for the quoted volume filling factor we did only take into account the bubbles around the haloes that host lbgs at a given time . if the bubbles last longer than the lbg phase the volume filling factor would be higher by the same factor . as discussed above our procedure to identify the host haloes of lbgs in the starburst model is rather crude . unlike kolatt et al . ( 1999 ) and wechsler et al . ( 2001 ) we do not identify `` colliding '' haloes but choose a fraction of less massive haloes at random . it is thus not obvious that the host haloes chosen in this way will reproduce the clustering properties of lbgs . in the right panel of fig . [ fig9 ] we compare the two - point correlation function of our `` starburst '' and `` massive - halo '' models . the correlation function in the starburst model was computed from 120 different monte - carlo realizations of catalogs of starbursting haloes . the correlation lengths for the `` starburst '' and `` massive - halo '' models are @xmath186 , and @xmath187 , respectively . the observed correlation length is @xmath188 ( adelberger 1998 , adelberger 2000 , arnouts 1999 , giavalisco & dickinson 2001 ) . the massive halo picture is thus in somewhat better agreement . however , the difference is small and as discussed above , kolatt ( 1999 ) and wechsler ( 2001 ) find that they can reproduce the observed clustering properties if they select `` colliding haloes '' as host galaxies of starbursting lbgs . we thus do not investigate this issue any further here . to understand better why the cosmic variance error becomes so large at small separation we plot the probability distribution @xmath189 in fig . [ fig10 ] . the distribution is very broad . about 40% of the haloes have @xmath144 larger than 0.8 . the probability of finding three ( uncorrelated ) haloes with @xmath190 at separation @xmath191 is thus still @xmath1925 - 10% . the symbols at the bottom of figure 5 show the increase of @xmath144 if the thermal motions and the velocity shear are artificially reduced by a factor of two . the broad probability distribution also explains why the suggestion of croft et . ( 2002 ) that lbgs are preferentially located in haloes with a low - density environment has such a strong effect on the mean transmission . we used mock spectra obtained from numerical simulations to investigate the effect of fully ionised bubbles around the galactic haloes hosting lbgs on the distribution in the igm as probed by the ly@xmath0 forest in qso spectra . by matching the probability distribution and the power spectrum of the flux of mock spectra obtained from a n - body simulation of a @xmath1cdm model with those of observed spectra we derive an upper limit of @xmath2 on the volume filling factor of bubbles . we further considered the mean transmitted flux at a given distance from galactic haloes . to model the observational data we computed the mean transmitted flux @xmath72 from an annular average of the two - dimensional conditional flux @xmath193 , in the same way as done by adelberger et al . ( 2003 ) . as previous investigations we find that the observed decrease of the transmission at separation @xmath194 to a halo is well reproduced by the simulated spectra . it is due to the increased matter density around haloes . we further find that an increase of the transmissivity at the smallest separations @xmath10 requires bubble radii of @xmath195 . this is two to three times larger than the separation at which the increase of the transmissivity is observed . as discussed by kollmeier 2003 this is because in redshift space the velocity shear of the material infalling onto the halo fills in the cavity which is caused by the fully ionised wind bubble in real space . for a model where the haloes are surrounded with bubbles of size @xmath195 there is no effect on either the flux pdf or the flux power spectrum . the flux pdf and the flux power spectrum of our simulated spectra are fully consistent with those of observed spectra . this is because the volume filling factor of the ionised bubbles is still small . we have investigated a variety of scalings of bubble size with halo mass . we also compared a model where lbgs are long - lived and are hosted by the most massive haloes with a model where lbgs are short - lived starburts and are hosted by more numerous less massive haloes . unlike kolatt ( 1999 ) and wechsler ( 2001 ) , we do not identify starbursts as `` colliding haloes '' but choose a fraction of less massive haloes at random . matching the observed transmission at scales @xmath194 is not difficult and there is certainly more than one way to do so . introducing redshift errors at the expected level improves the agreement with the observations . the increase in transmission at small separation in our model spectra spectra is , however , generally more gradual than that of the observed measurements . as other authors we found it challenging to match the observations at the smallest separation . numerical simulations including galactic winds and simple analytical estimates as the one in appendix [ cooltimescale ] agree that the required bubble sizes are difficult to achieve ( croft et al . 2002 , bruscoli et al . we have thus particularly looked into models which are consistent with the observed measurements and are most `` economical '' in terms of bubble size . the velocity shear around haloes increases with increasing mass . haloes of lower mass require thus smaller bubble radii to explain the observed increase of the transmission at @xmath191 . this may favour a model where lbgs are starburst galaxies hosted by more numerous less massive haloes . we have taken special care in estimating the errors on the flux near galaxies in a similar way as done in the observations . for small separations we found that the expected error due to cosmic variance is about 30% larger than estimated by assp . our starburst model is consistent with all measurements for a bubble radius of @xmath196 and a volume filling factor of 2% . a bubble size of @xmath196 appears , however , still difficult to achieve for galactic winds . if for some reason the velocity shear around haloes is smaller than suggested by our ( dm only ) simulation this would further reduce the required bubble size and filling factor . the correlation length for our `` starburst '' model is @xmath186 , slightly lower than the observations . identifying the lbgs with satellites infalling on high mass haloes , as in kolatt ( 1999 ) , would further improve the agreement with the observed clustering properties of lbgs . however , this would be at the expense of matching the observed small - scale flux @xmath72 , since the velocity shear increases with halo mass . therefore we advocate the model in which starbursts occur in small mass haloes . in a @xmath1cdm cosmology of normalisation @xmath197 , this model yields a reasonable match to the observed clustering of lbgs and transmitted flux @xmath72 . the mean transmitted flux of our simulated spectra at the smallest separation is 30 percent smaller than that of the assp sample . we would thus expect the observed mean transmission to decrease substantially for larger samples if the model is correct . a sample with 12 lbgs at separation @xmath10 should reduce the error to @xmath198 . this should help to discriminate against alternative models like the suggestion of croft ( 2002 ) that lbgs are located preferentially in low - density environments . we acknowledge stimulating discussions with j. primack and m. pettini . we thank juna kollmeier , our referee , for constructive comments on the manuscript . this research was supported by the israel science foundation ( grant no . 090929 - 1 ) , the german israeli foundation for scientific research and development , and the ec rtn network `` the physics of the intergalactic medium '' . vd would like to acknowledge the mpa ( garching ) , the institute of astronomy ( cambridge ) and the technion computing center where part of this work was accomplished . adelberger k.l . , steidel c.c . , shapley a.e . , pettini m. , 2003 , apj , 584 , 45 adelberger k.l . , 2000 , in asp conference series , vol . 200 , clustering at high redshift , ed . a. mazure , o. le fvre , v. le brun ( kluwer academic publisher ) adelberger k.l . , steidel c.c . , giavalisco m. , dickinson m. , pettini m. , kellogg m. , 1998 , apj , 505 , 18 arnouts s. , cristiani s. , moscardini l. , matarrese s. , lucchin f. , fontana a. , giallongo e. , 1999 , mnras , 310 , 540 bagla j.s . , 1998 , mnras , 299 , 417 bahcall j.n . , salpeter e.e . , 1965 , apj , 142 , 1677 bruscoli m. , ferrara a. , marri s. , schneider r. , maselli a. , rollinde e. , aracil b. , 2003 , mnras , 343 , 41 bryan g.l . , machacek m. , anninos p. , norman m.l . , 1999 , apj , 517 , 13 croft r.a.c , hernquist l. , springel v. , westover m. , white m. , 2002 , apj , 580 , 634 dav r. , hernquist l. , katz n. , weinberg d.h . , 1999 , apj , 511 , 521 dekel a. , silk j. , 1986 , apj , 303 , 39 gavialisco m. , dickinson m. , 2001 , apj , 550 , 177 gunn j.e . , peterson b.a . , 1965 , apj , 142 , 1633 haehnelt m.g . , natarajan p. , rees m.j . , 1998 , mnras , 300 , 817 heckman t.m . , armus l. , miley g.k . , 1990 , apjs , 74 , 833 heckman t.m . , 2000 , asp conf.ser . , 2000 , 240 , 345 heckman t.m . , lehnert m.d . , strickland d.k . , armus l. , 2000 , apjs , 129 , 493 hui l. , gnedin n.y . , 1997 , mnras , 292 , 27 katz n. , weinberg d.h . , hernquist l. , 1996 , apjs , 105 , 19 kolatt t.s . , bullock j.s . , somerville r.s . , sigad y. , jonsson p. , kravstov a.v . , kyplin a.a , primack j.r . , faber s.m . , dekel a. , 1999 , apj , 345 , 52 kollmeier j.a . , weinberg d.h . , dav r. , katz n. , 2003a , apj , 594 , 75 kollmeier j.a . , weinberg d.h . , romeel d. , katz n. , 2003b , aip conf . proc.666 191 - 194 larson r.b . , 1974 , mnras , 169 , 229 lowenthal j.d . , koo d.c . , guzman r. , gallego j. , phillips a.c . , faber s.m . , vogt n.p . , illingworth g.d . , gronwall c. , 1997 , apj , 481 , 673 mcdonald p. , miralda - escude j. , rauch m. , sargent w.l.w . , barlow t.a . , cen r. , ostriker j.p . , 2000 , apj , 543 , 1 ostriker j.p . , mckee c.f . , 1988 , phys . , 60 , 1 pettini m. , kellogg m. , steidel c.c . , dickinson m. , adelberger k.l . , giavalisco m. , 1998 , apj , 508 , 539 pettini m. , shapley a.e . , steidel c.c . , cuby j .- , dickinson m. , moorwood a.f.m . , adelberger k.l . , giavalisco m. , 2001 , apj , 554 , 981 rauch m. , miralda - escude j. , sargent w.l.w . , barlow t.a . , weinberg d.h . , hernquist l. , katz n. , cen r. , ostriker j.p . , 1997 , apj , 489 , 7 schaye j. , theuns t. , rauch m. , efstathiou g. , sargent w.l.w . , 2000 , mnras , 318 , 817 sedov , l.i . , 1959 , `` similarity and dimensional methods in mechanics '' , ( new - york : academic ) shapley a.e . , steidel c.c . , adelberger k.l . , dickinson m. , giavalisco m. , pettini m. , 2001 , apj , 562 , 95 sheth r.k . , tormen g. , 1999 , mnras , 308 , 119 somerville r.s . , primack j.r . , faber s.m . , 2001 , mnras , 320 , 504 springel v. , hernquist l. , 2003 , mnras , 339 , 289 springel v. , yoshida n. , white s.d.m . , 2001 , new astronomy , 6 , 79 steidel c.c . , giavalisco m. , dickinson m. , adelberger k. , 1996 , aj , 112 , 352 steidel c.c . , kollmeier j.a . , shapley a.e . , churchill c.w . , dickinson m. , pettini m. , 2002 , apj , 570 , 526 stoehr f. , white s.d.m . , tormen g. , springel v. , 2002 , mnras , 335 , l84 sutherland r.s . , dopita m.a . , 1993 , apjs , 88 , 253 tegmark m. , silk j. , evrard a. , 1993 , apj , 417 , 54 theuns t. , leonard a. , efstathiou g. , pearce f.r . , thomas p.a . , 1998 , mnras , 301 , 478 theuns t. , schaye j. , haehnelt m.g . , 2000 , mnras , 315 , 600 theuns t. , schaye j. , zaroubi s. , tae - sun k. , tzanavaris p. , carswell b. , 2002 , apj , 567 , 103 theuns t. , viel m. , kay s. , schaye j. , carswell r.f , tzanavaris p. , 2002 , mnras , 578 , 5 trager s.c . , faber s.m . , dressler a. , oemler a. , 1997 , apj , 485 , 92 weaver r. , mccray r. , castor j. , shapiro p. , moore r. , 1977 , apj , 218 , 377 wechsler r.h . , somerville r.s . , bullock j.s . , kolatt t.s . , primack j.r . , blumenthal g.r . , dekel a. , 2001 , apj , 554 , 85 weinberg d.h . , romeel d. , katz n. , kollmeier j.,2003 , aip conf . proc 666 157 - 169 to explain the lack of close to the haloes , the dilute , highly ionised bubble has to survive for a sufficiently long time . in this appendix we present general order of magnitude estimates of the wind / shock energetics , relate that to the observed speeds , and estimate cooling time scales . we heavily rely on the work of sedov ( 1959 ) , weaver ( 1977 ) , ostriker & mckee ( 1988 ) and tegmark ( 1993 ) . the main characteristics of the wind , such as the temperature of the inner plasma , the size @xmath64 of the bubble swept up by the shock and the wind velocity @xmath199 , depend on the total energy @xmath200 released in the outflow . neglecting energy losses @xmath200 can be estimated as @xmath201 where @xmath202 is the gas mass inside @xmath64 . to get the numerical estimation we have assumed that @xmath203 where @xmath204 is the mean background density and @xmath205 is the density contrast inside @xmath64 . the characteristic radius @xmath206 and velocity @xmath199 of the wind are poorly constrained , but recent observations suggest that @xmath207 and @xmath208 ( heckman 2000 , pettini 2001 , adelberger 2003 ) . on scales @xmath209 the value of @xmath205 is typically about @xmath210 , which gives @xmath211 . this value is consistent with the amount of energy @xmath212 released by supernovae during a burst of duration @xmath213 and of star formation rate @xmath214 . to assess whether the gas will be expelled from the halo one can apply a simple binding argument ( e.g. dekel & silk 1986 ) . on the one hand the potential energy @xmath215 of the gas which lies within the halo is @xmath216 , where @xmath217 is the characteristic radius of the halo , @xmath218 , and @xmath219 is the mass of the gas which lie within a radius @xmath220 . the mass of the gas in the halo is @xmath221 , where @xmath222 is the star formation efficiency and @xmath223 the average fraction of baryons in the halo . the potential energy is then @xmath224 . on the other hand , if one assumes that winds are driven by supernovae , the energy released in the wind is @xmath225 , where @xmath226 is the star formation rate . hence , according to this crude estimate , the ratio of the kinetic to binding energy is @xmath227 . we therefore expect @xmath64 to be a function of @xmath83 , and outflows to escape more easily from low mass haloes ( e.g. from dwarf galaxies ) . on this latter point , the measurements of adelberger ( 2003 ) seem to indicate that , if outflows are the cause of lack of absorption in the observed transmitted flux @xmath72 , they can escape from high mass galaxies as well , and affect the igm properties on comoving scale as large as @xmath117 . a wind ejects energy in the igm , driving a shock which heats up and collisionally ionises the material it encounters . we briefly review now the evolution of the shock front and the ionised bubble it encompasses , before radiative losses become important . we assume that the gas follows the perfect gas law @xmath228 , where @xmath229 , @xmath230 and @xmath231 are , respectively , the pressure , the internal energy and the density of the gas , and @xmath232 is the ratio of specific heats , which is @xmath233 for a mono - atomic gas . if the expansion of the universe can be ignored , the shock is well described by a strong adiabatic blast wave which propagates into the igm . in this regime , accurate self - similar solutions exist . let @xmath234 be the position of the shock front at time @xmath235 after the burst . the jump conditions relate the density @xmath236 of the unperturbed igm to the density @xmath237 , the pressure @xmath238 and the temperature @xmath239 at the shock front . for a strong adiabatic shock the jump conditions for pressure and density can be expressed as @xmath240 hence , @xmath241 for an adiabatic index @xmath233 . assuming that the flow is self - similar , @xmath242 can be expressed as a function of the similarity variable @xmath243 . the position of the shock front corresponds then to a fixed value of @xmath244 . hence , @xmath242 is given by ( sedov 1959 ) @xmath245 where @xmath211 is the energy released in the wind . inserting the value of @xmath200 in the expressions of @xmath242 and @xmath239 , we find @xmath246 since the temperature behind the shock front @xmath247 is a decreasing function of @xmath248 , @xmath249 reaches its minimum at the shock front . this solution holds only in the regime where the expansion is negligible , i.e. @xmath250 where @xmath251 is the hubble time scale . self - similar solutions for @xmath252 including this latter complication exist ( ostriker & mckee 1988 ) , but since energy and momentum conservation break down for @xmath253 , it is difficult to express the asymptotic solution as a function of the initial @xmath254 . the evolution of such an expanding shock depends critically on cooling which will lower the temperature of the shell and its interior , thereby increasing the amount of hydrogen available for ly@xmath0 resonant scattering . the expanding shell cools due to radiative losses . the shock will undergo a phase of shell formation which will occur when the energy dissipated in a volume @xmath255 will be of the order of @xmath200 . cooling is expected to be more efficient at the shock front since , at @xmath256 , the density ( temperature ) is larger ( lower ) than at @xmath257 . thus , shell formation first occurs at the shock front . as long as we are in the regime of low density , @xmath258 , and high temperature , @xmath259 , radiative cooling is dominated by thermal bremsstrahlung . however , for temperature @xmath260 , atomic line cooling might contribute significantly to the cooling rate @xmath1 ( in @xmath261 ) if the wind is enriched with heavy elements . for a wind of metallicity @xmath262 ( e.g. adelberger 2003 ) , the cooling rate is @xmath263 in the range @xmath264 ( e.g. sutherland & dopita 1993 ) . we assume that cooling at the shock front , where the temperature is @xmath260 , is dominated by atomic line transitions . this is true when the temperature of the shell is @xmath265 , i.e. for times @xmath266 . the cooling time scale at the shock front is then @xmath267 , @xmath268 where the cooling rate @xmath269 is in unit of @xmath270 and is assumed to be constant in the temperature range @xmath271 . once the shell starts forming , the temperature @xmath239 of the shell cools to @xmath272 on a very short time scale . however , in the range @xmath273 , the cooling rate @xmath1 drops sharply from @xmath274 down to @xmath275 ( sutherland & dopita 1993 ) and the shell can not cool below temperature @xmath59 . we define now a radius @xmath276 such that the cooling time scale @xmath277 as a function of radius @xmath248 satisfies @xmath278 for @xmath279 . according to this definition , the gas between @xmath276 and @xmath234 has cooled onto the shell by the time @xmath235 . to express @xmath277 as a function of @xmath248 and @xmath235 , it should be noted that the number density @xmath280 behind the shock front ( in @xmath281 ) scales as @xmath282 with @xmath283 typically . this relation is however valid in the asymptotic regime only . therefore , assuming that the plasma has reached pressure equilibrium , the temperature @xmath284 behind the shock front is @xmath285 , and the radiative cooling time scale @xmath277 is @xmath286 on fig . [ figapp2 ] the solid curves are @xmath276 for various cooling time rate @xmath269 , and the dashed curve is @xmath287 . the shell starts forming at the shock front when @xmath276 intersects @xmath234 . for larger time , cooling occurs for @xmath288 ( shaded area ) . since most of the gas swept up by the wind lies at @xmath289 , a large fraction of gas cools down to @xmath59 shortly after the shell formation . for @xmath290 , about 30% of the mass enclosed in the bubble has cooled into the shell over a hubble time @xmath251 , where @xmath291 at @xmath46 . for a high metallicity wind @xmath292 , cooling becomes severe for @xmath293 a few @xmath294 , a time scale shorter than the hubble time scale @xmath251 at @xmath46 . on the contrary , for a low metallicity wind , @xmath295 , the cooling rate is @xmath296 and cooling will not become significant for @xmath297 . hence , most of the bubbles will not survive over a hubble time scale unless the cooling time rate @xmath269 , or the wind metallicity @xmath298 , is very low . for radii @xmath299 , the temperature is higher than @xmath300 , and thermal bremsstrahlung and inverse compton cooling prevail over line cooling , but the corresponding cooling times are still smaller than the hubble time at @xmath46 . to compute the thickness @xmath301 of the shell , we assume that the pressure is constant through the inner plasma and the shell . since the temperature of the shell , @xmath302 , is much lower than the overall temperature @xmath303 of the highly ionised plasma , we expect @xmath304 , i.e. the shell is thin . taking into account this assumption , we find @xmath305\ ; , \label{eq106}\ ] ] a result consistent with a thin shell approximation . assuming @xmath306 and neglecting the expansion of the universe gives @xmath307 . @xmath110 should thus not exceed a few @xmath308 , a scale which can not be resolved in our simulation . the optical depth for resonant ly@xmath0 scattering is given by @xmath309 where @xmath310 is the neutral fraction at temperature @xmath59 . ly@xmath0 absorption is expected for reasonable values of the density contrast @xmath311 , the typical comoving radius of the bubble @xmath312 , and the baryon fraction @xmath313 . however , since we can not resolve the thin shell in our simulation , we have nevertheless ignored its ly@xmath0 absorption . to assess the sensitivity of the transmitted flux to numerical resolution , we perform a calculation with the highest resolution simulation of the series , which has a mass resolution @xmath314 . the halo catalog and the flux in the simulation are computed as in section [ nbody ] with a pixel resolution @xmath51 ( comoving ) . we added a gaussian deviate of variance @xmath315 to the halo positions . in fig . [ figapp3 ] we show the transmitted flux as computed in the `` starburst '' scenario for a bubble radius @xmath94 . the mass resolution is @xmath316 ( dashed curve ) and @xmath317 ( solid curve ) . in the latter the strong absorption lines are better resolved , thereby slightly reducing the normalization of the optical depth @xmath318 . therefore , the optical depth of the absorbing material is lower , and the transmitted flux is slightly larger , by about @xmath319 on scales of @xmath320 .

we revise the conventional nuclear effects of fermi motion , binding and pionic effects in deep inelastic lepton scattering using a relativistic formalism for an interacting fermi sea and the local density approximation to translate results from nuclear matter to finite nuclei . in addition we also consider effects from-meson renormalization in the nucleus . the use of nucleon green s functions in terms of their spectral functions offers a precise way to account for fermi motion and binding . on the other hand the use of many body feynman diagrams in a relativistic framework allows one to avoid using prescriptions given in the past to introduce relativistic corrections in a non relativistic formalism . we show that with realistic nucleon spectral functions and meson nucleus selfenergies one can get a reasonable description of the emc effect for , outside the shadowing region . 1.5 cm -1 in 5.5 in 0.5 in

nucl-th9510060

the emc effect @xcite is probably one of the topics in the interplay of nuclear and particle physics which has attracted more attention . pioneering work on fermi motion and pionic effects @xcite was followed by many different ideas like binding of the nucleons in the nucleus , multiquark cluster effects or @xmath4 rescaling ( see refs . @xcite for reviews on the topic ) . here we shall pay attention only to conventional nuclear degrees of freedom , mesons ( @xmath5 and @xmath0 ) and nucleons . one of the interesting ideas along these lines was the effect of the nuclear binding @xcite , which , with ups and downs , has come to be accepted as , largely or at least partly , responsible for the depletion of @xmath6 in the region of the minimum . criticism on this latter works was raised in @xcite , where it was shown that the introduction of relativistic corrections in the usual nuclear nonrelativistic treatment of the binding effects resulted in a flux factor which reduced the conventional binding effects @xcite . this flux factor led to a different normalization of the spectral function which preserved the baryonic number calculated relativistically @xcite . the argument of the normalization of the baryonic number is an important one and the idea has met with followers @xcite . however , it is a prescription on how to convert the nonrelativistic nuclear wave function into a relativistic spectral function @xcite and the prescription is not shared by others @xcite . furthermore , these are not the only relativistic corrections as shown in @xcite . this issue justifies a work like the present one , where we construct from the beginning a relativistic nucleon spectral function and define everything within a field theoretical formalism which uses the nucleon propagators written in terms of this spectral function . the relativistic formalism is taken from the beginning and the baryonic number is naturally well normalized . since all the nuclear information needed is contained in the nucleon spectral function , one does not need to use ordinary nuclear wave functions , which are anyway static ( no spreading in the energy distribution , which is just concentrated in the single particle energies of the shell model ) and hence one does not need any step to introduce relativistic effects into the nonrelativistic wave functions , as done in @xcite . furthermore , the use of non static spectral functions is important . this was already seen in refs . @xcite which showed that the use of more realistic spectral functions accounting for nuclear correlations resulted in an enhancement of the binding effects . the reason is that , for a given average binding energy , the approach with a realistic spectral function leads to a larger kinetic and potential energy in absolute value than the shell model approach . the other issue that we have revisited here is the pionic contribution . large effects from the pionic cloud associated with the pion excess number in the nucleus were found in @xcite . we have taken up the idea and recalculated these effects within the many body field theoretical approach using input which has been checked in a variety of nuclear reactions testing real and virtual pions : pionic reactions @xcite , muon capture @xcite , inclusive neutrino scattering @xcite and photonuclear reactions @xcite . in addition we have also included corrections from the modification of @xmath0-meson cloud in the nucleus , in complete analogy with the pionic contribution . altogether we find that the use of the spectral function for the nucleons together with the mesonic effects can approximately account for the measured emc effect . the calculations are done using the spectral function for nucleons in nuclear matter , followed by the local density approximation . this was shown to be an excellent tool to deal with photonuclear reactions in the absence of shadowing effects @xcite . hence , a natural limit of our results is the region of shadowing , @xmath7 , where indeed there are discrepancies with the data , and other ingredients should be considered that we do not want to tackle @xcite . let us recall first the nonrelativistic nucleon propagator for a noninteracting fermi sea , which is given in momentum space by @xmath8 where @xmath9 is the fermi occupation number @xmath10 for @xmath11 for @xmath12 and @xmath13 is the nonrelativistic nucleon energy . eq . ( 1 ) can be recast as @xmath14 which separates the propagator into the free propagator and the medium correction . for an interacting fermi sea the nucleon propagator can be written in terms of its nonstatic selfenergy @xmath15 @xmath16 which can be rewritten in terms of the spectral functions for holes and particles as @xcite @xmath17 with the following relationships @xmath18 ^ 2 + [ i m \sigma ( p^0 , p)]^2}\\[2ex ] & \hbox{for } \ ; p^0 \leq \mu \end{array}\ ] ] @xmath19 ^ 2 + [ i m \sigma ( p^0 , p)]^2}\\[2ex ] & \hbox{for } \ ; p^0 > \mu \end{array}\ ] ] let us also recall that the momentum distribution of the nucleon in this interacting fermi sea is given by @xmath20 @xmath21 with the automatic sum rule @xmath22 in passing we also note that in physical reactions @xmath23 does not factorize out in the physical cross sections because other factors dependent on @xmath24 and @xmath25 appear simultaneously in the formulae , and restrictions due to energy and momentum conservation do not allow the infinite ranges in the @xmath24 integration required in eq . failure to realize that , and the naive substitution of @xmath9 in eq . ( 1 ) by @xmath23 of eq . ( 6 ) , as sometimes done , leads to erroneous results which can be off by three orders of magnitude in some cases @xcite . this gives us a warning that we should express all our magnitudes in terms of the spectral functions , not the momentum distributions . the free relativistic nucleon propagator is given by @xmath26 where we have separated in the second member the contribution from positive and negative energy states @xcite . @xmath27 in eq . ( 8) are the nucleon mass and the relativistic nucleon energy @xmath28 , and @xmath29 are the ordinary spinors which we take normalized as @xmath30 . we recall that @xmath31 are functions of three momentum and they will be the only spinors which will appear in our framework . the relativistic nucleon propagator for a noninterating fermi sea is easily derived and , by analogy to eq . ( 2 ) , can be written as @xmath32 which by means of the identity of eq . ( 8) can be recast as @xmath33 \right.\ ] ] @xmath34 apart from the negative energy contribution , which will play no role in our problem , the only difference between eq . ( 10 ) and the nonrelativistic propagator of eq . ( 1 ) is the presence of the factor @xmath35 and the projector over the space of positive energies @xmath36 , which are both unity in the nonrelativistic approximation . now we proceed to construct the relativistic propagator in the interacting fermi sea . we wish to sum the dyson series for the diagrams shown in fig . 1 , where although not shown , one would also have other sources of nucleon selfenergies . we will write them in terms of the operator @xmath15 . it will become clear later on that we only need the imaginary part of the nucleon propagator for the positive energy states , in which case we neglect from the beginning the negative energy states ( their weight becomes negligible compared to the singular part of the positive energy propagator ) . hence , for the purpose of the present problem , the nucleon propagator needed will be @xmath37 @xmath38 @xmath39 where we have used the fact that @xmath40 should be diagonal in spin for spin saturated matter which we only consider . comparison of eqs . ( 11 ) and ( 3 ) shows again the differences between the relativistic and nonrelativistic propagators . the structure of eq . ( 11 ) allows one to define a spectral representation of the nucleon propagator by means of @xmath41 and @xmath42 as @xmath43 @xmath44\ ] ] with the relationships @xmath45 ^2 + \left [ \frac{m}{e ( \vec{p } ) } i m \sigma ( p^0 , p ) \right ] ^2}\ ] ] for @xmath46 @xmath47 ^2 + \left [ \frac{m}{e ( \vec{p } ) } i m \sigma ( p^0 , p ) \right ] ^2}\ ] ] for @xmath48 where for simplicity @xmath40 is now @xmath49 which is independent of the spin . eqs . ( 13 ) are now the generalizations of eqs . ( 5 ) using relativistic kinematics . note that @xmath50 and @xmath51 defined in eq . ( 12 ) are not the nonrelativistic spectral functions normally used . hence one should not expect the same normalization as in @xcite . the normalization of @xmath51 , which we will need , is easily obtained by imposing baryon number conservation , as done in @xcite . for this purpose we evaluate the electromagnetic form factor at @xmath52 . for the case of the nucleon ( fig . 2a ) ) we have ( assume all baryons have charge unity for normalization purposes ) , @xmath53 for the case of nucleons in the medium we must evaluate the many body diagram of fig . @xmath54 e^{i p^{0 } \eta}\ ] ] where exp@xmath55 , with @xmath56 , is the convergence factor for loops appearing at the same time @xcite and v the volume of our normalization box . by means of eq . ( 12 ) we can see that the convergence factor limits the contribution to the hole spectral function and we get @xmath57\ ] ] @xmath58 @xmath59 \int^{\mu}_{- \infty } s_h ( \omega , p ) \ ; d \omega\ ] ] @xmath60 where in the last step we have imposed that this matrix element gives the right current with b baryons , in analogy to eq . ( 14 ) , and @xmath61 is the momentum of the nucleus . note that @xmath62 appears in eq . ( 16 ) because the operator @xmath63 comes from @xmath64 which depends only on @xmath25 ( it corresponds to free particles with @xmath65 . obviously eq . ( 16 ) must be evaluated in the rest frame of our fermi sea where all magnitudes are defined . only @xmath66 is then relevant and we obtain the desired normalization @xmath67 ( note that the factors @xmath68 have cancelled in eq . the factor 2 is a spin factor . by a simple inspection of eq . ( 17 ) we can see that with the definition of the spectral functions in eq . ( 12 ) one can use eq . ( 6 ) to determine momentum distributions in both the nonrelativistic and relativistic cases . in our formalism we do not have a box of constant density , but elements of volume @xmath69 with local density @xmath70 , the nuclear proton and neutron densities at the point @xmath71 . hence our spectral functions for protons and neutrons are functions of the local fermi momentum @xmath72^{1/3 } \ ; ; \ ; k_{f , n } ( \vec{r } ) = [ 3 \pi^2 \rho_n ( \vec{r } ) ] ^{1/3}\ ] ] and then the equivalent normalization to eq . ( 17 ) is @xmath73 in practice we shall work with symmetric nuclear matter of density @xmath74 . hence we have a unique fermi momentum defined as @xmath75^{1/3}$ ] and then one has @xmath76 or equivalently @xmath77 with @xmath78 the mass number of each nucleus . the density @xmath74 of each nucleus is taken from experiment in our case , and expressed in terms of a two fermi parameter distribution for medium and heavy nuclei @xcite and mod harmonic oscillator for light nuclei @xcite . ( 21 ) is fulfilled at the level of 2 - 3 @xmath79 in our case , in spite of the non trivial structure of the spectral function and the integrals involved . the small numerical deviation from the right normalization is taken care by dividing by the integral of eq . ( 21 ) instead of by @xmath78 in the evaluation of @xmath80 , since a similar integration weighed by the structure functions appears in the numerator , as we shall see . although one could separate the contribution of protons and neutrons in the calculation , we have only applied the results to nuclei with @xmath81 , or very close , like @xmath82 and hence , we work with the symmetric nuclear matter version . let us recall the basic ideas in deep inelastic scattering . consider the @xmath83 process on a nucleon of fig . the invariant @xmath84 matrix for the process is @xmath85 where @xmath86 is the invariant matrix element of the hadronic current . the cross section for the process @xmath87 is given in mandl and shaw normalization @xcite by @xmath88 @xmath89 @xmath90 where @xmath91 is the electron mass , @xmath92 stands for fermions and @xmath93 for bosons in the final state x. the factor @xmath94 becomes @xmath95 if we study the cross section on a pion . the index @xmath96 is split in @xmath97 for fermions and bosons respectively . in the nucleon rest frame one can then write the differential cross section , with @xmath98 referring to the outgoing electron , as @xmath99 with @xmath100 and @xmath101 the leptonic tensor @xmath102 and @xmath103 the hadronic tensor defined as @xmath104 with @xmath105 @xmath106 where @xmath107 is the momentum of the virtual photon , @xmath108 the spin of the nucleon and @xmath109 the spin of the fermions in @xmath110 . lorentz covariance and gauge invariance allow one to write @xmath103 as @xcite @xmath111 where @xmath112 are the two structure functions of the nucleon and are functions of @xmath113 . now we evaluate the cross section for @xmath83 on the nucleus . in order not to miss flux factors and be able to write everything in terms of propagators , we evaluate the electron selfenergy corresponding to the diagram in fig . we obtain @xmath114 @xmath115 which for unpolarized electrons can be written as @xmath116 with @xmath117 the photon selfenergy . in order to evaluate the cross section we need only @xmath118 which can be easily evaluated by means of cutkosky rules @xcite . @xmath119 with the result @xmath120 with @xmath121 . the cross section is readily evaluated from there . inspection of eq . ( 11 ) for the relativistic fermion propagator tells us that the electron width is given by @xmath122 by means of which the contribution to the cross section from an element of volume @xmath69 in the rest frame of the nucleus is @xmath123 @xmath124 hence we immediately write the @xmath83 cross section in the nucleus as @xmath125 with @xmath126 , which is always positive in this experiment , so we drop the @xmath127 function . comparison of eq . ( 35 ) with eq . ( 24 ) used for nuclear targets tells as that @xmath128 next we evaluate @xmath117 corresponding to the right hand side of the diagram of fig . 3b ) using again the feynman rules in terms of propagators . we have @xmath129 @xmath130 which by means of cutkosky rules ( 31 ) and the use of free propagators , eq . ( 8) , for the final states and the medium propagator , eq . ( 12 ) , for @xmath131 , plus eq . ( 36 ) , leads immediately to @xmath132 @xmath133 @xmath134 which by means of eq . ( 27 ) can be rewritten for an isospin symmetric nucleus as @xmath135 @xmath136 with @xmath137 note that @xmath138 appears with the off shell arguments of @xmath139 , the bound nucleon . in the steps from eq.(37 ) to ( 38 ) the spinors @xmath140 are included in the matrix elements of the currents and we have considered that there is necessarily a fermion loop ( hence the first minus sign in eq . ( 37 ) ) with a free particle in the final state and the nucleon in the medium in the initial state . the corresponding energy integration in the loop ( if we had used a wick rotation explicitly instead of cutkosky rules ) necessarily picks up the hole part of the propagator of eq . ( 12 ) , and at the same time relaxes the condition @xmath141 of cutkosky rules which does not appear for the hole part . in eq . ( 39 ) there is an apparent lack of normalization , since assuming @xmath103 constant ( which actually can not be in practice ) we would expect @xmath142 . however given the normalization of the spectral function in eq . ( 21 ) , this is not the case . in eq . ( 39 ) we get the extra factor @xmath143 which does not appear in eq . ( 21 ) . it is interesting to see the meaning of this factor in eq . ( if we look at the formula of the @xmath144 cross section in eq . ( 23 ) , and by means of eqs . ( 25 ) and ( 27 ) we find @xmath145 where @xmath146 is a lorentz invariant and the content of the integral in eq . ( 40 ) also . for collinear frames of reference we also have @xmath147 where @xmath148 is the electron momentum in the frame where the nucleon is at rest and hence @xmath149 now if we have a system of moving nucleons the cross section for scattering of the electron with the nucleus can not be obtained as a sum of individual cross sections , because the relative @xmath144 flux is different for each nucleon . instead , one has to sum the probabilities of collision per unit time for each nucleon and divide by a unique flux , the one relative to the cm of the nucleus . by taking for @xmath150 in eq . ( 40 ) the velocity of the electron with respect to the cm of the nucleus and suming over all the nucleons in eq . ( 40 ) we will be calculating the electron nucleus cross section . hence we obtain in the rest frame of the nucleus @xmath151 @xmath152 where @xmath148 is the electron momentum in the nucleus rest frame . since our nuclear cross section is given in terms of @xmath153 by ( see eqs . ( 35 ) , ( 36 ) ) @xmath154 then eq . ( 39 ) follows immediately . the previous discussion has shown that the factor @xmath143 is a factor appearing in the probability of reaction per unit time for each nucleon , and remains in the integral when we divide by a unique flux in order to obtain the nuclear cross section . it is thus a lorentz contraction factor . in the limit of small densities , when @xmath155 , eq . ( 39 ) with the consideration of eq . ( 21 ) would give @xmath156 as it should be . ( 39 ) accounts for fermi motion and binding and includes the relativistic lorentz contraction factor @xmath157 and the change of the arguments in @xmath158 . note in passing that the relativistic factor @xmath159 of eq . ( 33 ) , which we extracted from the relativistic propagator of eq . ( 11 ) , has been essential to provide the right normalization . let us first see the free pion structure function . the same formula eq . ( 24 ) is used for pions and this defines @xmath160 . given the normalization of the fields which we follow @xcite , the cross section of eq . ( 23 ) contains the factor @xmath161 instead of @xmath162 . hence , this means that the definition @xmath160 is given by eq . ( 27 ) dividing the right hand side of the equation by @xmath163 ( and obviously the average over the spin of the nucleon does not appear now for the pion case ) . in order to derive the contribution from the virtual pions in the medium we evaluate again the electron selfenergy related to the diagram of fig . 4 . we can save all the steps given before simply by noting the differences in the two cases : \i ) the bound nucleon propagator is substituted by a pion propagator . from the use of cutkosky rules we must change @xmath164 by @xmath165 with @xmath166 the pion propagator ( in the medium ) . \ii ) one must take into account that @xmath160 is given by eq . ( 27 ) divided by @xmath163 . \iii ) the sum over spins of the bound nucleon in eq . ( 37 ) does not appear now for the case of the pion . \iv ) there are three charged states of pions . with only these four rules we can already write @xmath167 now there are two obvious subtractions to eq . first one should subtract the contribution from a free pion , which has nothing to do with medium effects . however , this is zero because one electron can not decay into another electron , one pion and x. then assuming the pion is dressed in the medium by exciting @xmath168 and @xmath169 as we shall do , one is left with the contributions shown in fig . now there is no problem to get a contribution since @xmath170 gets strength from @xmath171 excitation . the physical channel would correspond to @xmath172 , or equivalently @xmath173 , which is now allowed . the physical channels are easily visualized by cutting the intermediate states in the diagramas with a horizontal line and placing on shell the particles cut by the line . this is actually the essence of cutkosky rules to obtain the imaginary part of the selfenergy of a diagram . the former discussion also tells us that part of what we are calculating is already contained in the nucleon structure function . this is because we are also calculating the contribution from the pions contained in a free nucleon this has to be subtracted . this is easily done by substituting in eq . ( 46 ) @xmath174 since we substract @xmath78 times the contribution from the pion cloud to the structure function of the free nucleon . in technical words , we can say that we are only considering terms with at least two @xmath171 or @xmath175 ( the most important terms ) etc . in fig . 5 ( up to pauli blocking corrections in the @xmath171 lindhard function , which are automatically included by the procedure of eq . ( 47 ) ) . hence the genuine pionic contribution is given by @xmath176 eqn . ( 39 ) and ( 48 ) are the basic equations which provide the nucleonic and pionic contributions . in passing we can mention that the distribution of the excess number of pions , per unit volume in the nucleus , @xmath177 , which contains the averages of @xmath178 , @xmath179 and @xmath180 , is given by @xcite @xmath181 such that in the case of a structure function @xmath182 independent of @xmath183 , eq . ( 48 ) could be considered as a weighed integral of the pion structure function over the pion excess distribution in the nucleus . however , the strong dependence of @xmath184 on @xmath183 ( imposed by energy and momentum conservation ) does not allow that factorization , and hence a relationship of the pion excess number with the pionic contribution to the structure function can not be established . the apparent extra factor 2 which we obtain in this counting ( apart from the lorentz contraction factor , @xmath185 is obtained because one is automatically accounting for the imaginary part of the compton @xmath186 amplitude which is crossing symmetric and contains the two diagrams of fig . 6 , while the pion structure function for on shell pions contains only the imaginary part of the diagram 6a ) . ( see refs . @xcite and @xcite for an elaborate discussion of these issues in the problem of the pion cloud contribution to @xmath187 nucleus scattering ) . it is worth noting that there is no overlap between the feynman diagrams accounted for in the pionic contribution , fig 5 , and those of the nucleonic contribution with fermi motion and binding , which come from selfenergy insertions in the nucleon line of fig . 3b ) . hence , these contributions to the nuclear structure function are independent . we have evaluated the contribution of nucleons and pions to the hadron structure function of the nucleus . we now proceed to write these expressions in terms of the bjorken structure functions @xcite . for nucleons ( and similarly for pions or the nucleus ) one introduces the bjorken variables @xmath188 and for large values of @xmath189 and @xmath4 simultaneously and fixed @xmath190 one has the bjorken scaling @xmath191 @xmath192 and the callan - gross relation @xmath193 up to some , @xmath194 corrections in @xmath195 . since the same @xmath4 will be chosen for the nucleus and the nucleon and we perfom ratios of structure functions we shall not worry about this dependence here . in view of the relations ( 51 ) , ( 52 ) the most practical way to proceed is to work with transverse components of @xmath103 . for this purpose assume @xmath196 along the @xmath197 direction , as usually done in the study of the @xmath198 reaction , and evaluate @xmath199 . we find from eq . ( 28 ) @xmath200 @xmath201 this component has the virtue that the coefficient of @xmath202 is independent of @xmath139 and hence is the same for on shell or off shell nucleons , or pions , or the nucleus . hence we can write @xmath203 @xmath204 @xmath205 @xmath206 where the extra minus sign in @xmath207 is because of the direction of @xmath139 in fig . 4 . here @xmath208 as implicit in eq . ( 39 ) . we still have to exert some caution since @xmath209 . on the other hand @xmath210 but @xmath211 because in our scheme the emerging particle from the coupling of the pion to a nucleon of the nucleus is on shell when we excite @xmath212 with the pion and take the imaginary part of @xmath166 @xcite . since usually one compares ratios of the @xmath213 structure functions , this is easily accomplished by making use of the callan - gross relation ( 52 ) and we find @xmath214 @xmath215 where we have again @xmath216 as implicit in eq . ( 39 ) . in the bjorken limit the evaluation of eq . ( 55 ) does not require the knowledge of the variable @xmath107 since @xmath217 but it is implicit in the structure functions which are taken at a certain @xmath4 . when reaching this point it is worth considering also the contribution of the @xmath0 meson cloud . both the pion and the @xmath0 meson couple to nucleons and delta with derivative couplings , which give rise to relatively large meson selfenergies in the range of momenta which contributes to the structure functions . furthermore , as found already in ref . @xcite , the pion cloud contribution comes mostly from the combined @xmath171 and @xmath169 excitation , with the @xmath171 on shell ( in @xmath218 , and one finds a negligible contribution of two @xmath171 excitations . only @xmath5 and @xmath0 can excite the @xmath169 components and this makes these two mesons special when looking at the mesonic contribution to the nuclear structure function . in addition there is an interplay between @xmath5 and @xmath0 exchange . indeed , the @xmath0 meson , through nuclear correlations , contributes both to the longitudinal and transverse parts of the spin - isospin @xmath171 and @xmath169 interaction and it is an important element contributing to the landau - migdal @xmath219 parameter in a microscopic derivation of this interaction @xcite . the value of this parameter governs to some extend the pionic ( and @xmath0-meson ) contribution to the structure function @xcite . the contribution of the @xmath0-meson cloud to the structure function is given , by analogy to eq . ( 56 ) by @xmath220 @xmath221 where @xmath222 is now the @xmath0-meson propagator and @xmath223is the @xmath0-meson structure function , which we take equal to the one of the pion following refs . in addition @xmath224 is also given , in analogy to eq . ( 57 ) , by @xmath225 eq . ( 58 ) contains an extra factor of two compared to the pionic contribution of eq . this is because of the two transverse polarizations of the @xmath0-meson and the fact that the coupling of the @xmath0 to nucleons and deltas which we consider , following ref . @xcite is only of transverse nature @xmath226 for nucleons and @xmath227 for deltas , with @xmath228 the polarization vector of the @xmath0 meson and @xmath229 the spin transition operator from spin @xmath230 to @xmath231 ) . the expression of @xmath232 in eq . ( 54 ) shows the same lack of normalization discussed in connection with eq . ( 39 ) , since assuming @xmath233 constant ( which is not the case ) @xmath234 is not @xmath78 times @xmath235 , due to the lorentz contraction factor @xmath236 . the same can be said about eq . ( 55 ) in general . however , eq . ( 55 ) , shows a particular normalization property . indeed , if we take an ensemble of nucleons on shell @xmath237 and @xmath238 ( and hence @xmath239 , by virtue of eq . ( 57 ) ) , then the lorentz contraction factor @xmath240 cancells in average the dynamical factor @xmath241 and we get @xmath242 for on shell nucleons . this cancellation , however , will not show up when the nucleons are off shell since in our formalism we still obtain the factor @xmath143 but the dynamical factor @xmath241 will now be different . in this discussion we are implicitly assuming that eqs . ( 54 ) and ( 55 ) stand as they are for the case of off shell nucleons simply by taking for @xmath243 the expression of eq . ( 57 ) using the nucleon off shell variables . this is certainly the easiest form of the analytical continuation in the off shell regime , although there have been other prescriptions in the literature @xcite . we would like to justify our assumption . indeed , in our frawework we use nucleon propagators which are based on the free spinors ( eq . hence in eq . ( 38 ) the matrix elements would be defined in terms of free spinors , the final particles are free particles and the off shell dependence appears only in the @xmath244 function . obviously one can not look in detail at all channels implicit in eq . however , one can use the same philosophy in the parton model which is used to find out the scaling of the structure functions . we follow here the steps of ref . @xcite and assume that partons carry a fraction @xmath245 of the nucleon momentum @xmath139 and its mass @xmath246 , that the electron parton amplitude is given by the on shell expression and that the outgoing parton is a free one , but the @xmath244 function appears with the off shell variables . one finds then @xmath247 @xmath248 where @xmath249 appears whith argument @xmath243 which is defined in terms of the off shell variables . similarly one obtains @xmath250 whith @xmath251 , and @xmath243 defined in terms of the off shell variables . we would like to note here that our finding @xmath252 , or equivalently @xmath253 , where @xmath254 , is not so unconventional . indeed , in ref . @xcite the authors also find this property and they literally quote note that a feature of the results is that @xmath253 ... it does not reflect any violation of baryon conservation , which is ensured by the normalization condition " , in our case eq . ( 21 ) . the issue of the normalization still stirs much controversy . we have devoted many thoughts to it throughout this paper looking at it from some points of view not discussed before . we see that for an ensemble of uncorrelated nucleons @xmath255 but as soon as interactions are accounted for this normalization is lost . a similar thing would happen should we evaluate the structure function @xmath256 , which appears in neutrino scattering . this structure function when integrated over @xmath190 is normalized to 3 for the nucleon if one assumes models with only three valence quarks , or to @xmath257 for the nucleus . once again this normalization would be lost if qcd interaction corrections are accounted for @xcite and equivalently if nn interactions are included in the nuclear case , and experimentally this is the case @xcite . also , experimentally @xmath258 . clear as the question looks to us , we are aware that this point of view is not universally accepted . further thoughts and discussions on the issue should be welcome which would help settle the question in a way acceptable to all . ( 55 , 56 , 58 ) are the final equations which we use in the analysis . for the nucleon and pion structure functions @xmath259 we take the experimental values of refs . the pion propagator in the medium is given by @xmath260^{- 1}\ ] ] with @xmath261 the pion selfenergy . we consider the contribution of the @xmath262 and @xmath169 excitations to the pion selfenergy in connection with the landau - migdal correction in the pionic channel , as well as off shell pion nucleon form factors . since for @xmath263 we need @xmath264 , it is practical to write @xmath265 where @xmath266 is the @xmath267 form factor , which we take of the monopole type @xmath268 with @xmath269 @xcite and @xmath270 . @xmath271 is the longitudinal part of the spin - isospin interaction and @xmath272 is the irreducible pion selfenergy , which contains all selfenergy diagrams which are not connected by @xmath271 . for the @xmath0-meson we can write @xmath273 with @xmath274 the transverse part of the spin - isospin interaction , @xmath275 @xcite and @xmath276 the @xmath277 form factor given by @xmath278 and @xmath279 @xcite . for @xmath280 we take the expressions which are derived from a model with @xmath5 and @xmath0 exchange in the presence of short range nuclear correlations @xcite , which are given by @xmath281 @xmath282 @xmath283 @xmath284 here @xmath285 is the inverse of a typical correlation distance and @xmath286 are the corresponding propagators and form factors substituting @xmath287 by @xmath288 . the irreducible selfenergy @xmath272 is in our case the sum of the lindhard functions @xmath289 for @xmath171 and @xmath169 excitation with the normalization and analytical expressions of the appendix of ref . it is interesting to note that for the values of @xmath290 which contribute most to the structure function , both @xmath271 and @xmath274 are negative and this leads to positive values of @xmath291 and @xmath292 . it was also found in @xcite that in order to evaluate the contribution of the pion cloud by using the pion propagator , as done here , it is important that it satisfies the sum rule @xmath293 which expresses the equal time commutation relation of the pion fields . our model satisfies this equation at the level of one per thousand . section 2.2 has established the framework for the relativistic nucleon propagator which we need here . the only input needed is the nucleon selfenergy . we take it from the work of ref . this is a semiphenomenological , quite succesful approach , which uses as input the @xmath294 cross section and the spin - isospin effective interaction . this allows one to evaluate @xmath295 , which is remarkably close to @xmath295 of the elaborate many body calculations of ref . the real part is evaluated by means of a dispersion relation , and the fock term from the pionic contribution is also included . only pieces of the hartree type , which should be independent of the momentum , are missing for which one needs more information . hence , up to an unknown momentum independent piece in the selfenergy the rest of the nucleon properties in the medium can be calculated , like effective masses , spectral functions , etc , which are also in good agreement with sofisticated many body calculations @xcite . actually , what might appear a drawback is now a welcome feature because since the proper binding energy is an important ingredient in the @xmath296 effect , we also include phenomenologically a function @xmath297 in the nucleon selfenergy and demand that the binding energy per nucleon be the experimental one for each nucleus . then the model is complete , realistic and technically much simpler to handle than the sofisticated many body calculations @xcite . with this improvement , momentum distributions and average binding energies are also in good agreement with other infinite nuclear matter @xcite and finite nuclei calculations @xcite . a small inconvenience appears because the selfenergy of ref . @xcite is evaluated non relativistically . however we have checked that a proper calculation including relativistic factors of the type @xmath35 in the nucleon propagators in the integrals over the loops which appear in the evaluation of @xmath298 in @xcite , would only introduce corrections in @xmath298 below the level of @xmath299 . second , we have changed @xmath298 by @xmath299 and found that the ratio @xmath80 changes only at the level of @xmath300 . thus we take the values for @xmath40 from ref . @xcite and use them in the relativistic propagators of section 2.2 . for the average kinetic and total nucleon energy we have @xmath301 @xmath302 and the binding energy per nucleon is then given by the sum rule @xcite @xmath303 which is also used in @xcite in connection with the study of the @xmath296 effect . we take experimental numbers for each nucleus for @xmath304 and adjust the function @xmath297 to fit @xmath304 . we take @xmath297 linear in the density , @xmath305 . this quantity provides around @xmath306 repulsion at @xmath307 in most of the nuclei . detailed values for @xmath308 and @xmath309 can be seen in table i. as noted in refs . @xcite , the use of nucleon propagators in terms of non static spectral functions leads to bigger values of the average kinetic energy and @xmath310 than the shell model of the nucleus and as a consequence to reduced values of @xmath80 ( for @xmath311 ) . we can see this here also by taking the uncorrelated fermi sea and adding a function @xmath312 to the ordinary thomas - fermi potential @xmath313 such as to get the same binding energy via eq . we can make use of the same formalism by simply considering that @xmath314 @xmath315 eq . ( 72 ) associates one energy to a given momentum , ( the essence of the shell model in infinite nuclear matter ) while the spectral function has a peak around the quasiparticle energy and then spreads out at larger values of @xmath25 for a given energy . this results in a larger value of the average kinetic energy . in table i we show the results for different nuclei and compare them with those of ref . @xcite . as one can see , the results that we obtain with the uncorrelated fermi sea and the spectral functions are remarkably close respectively to those of the hartree fock and spectral function used in @xcite . we shall evaluate @xmath80 using both the spectral function approach and the uncorrelated fermi sea . we have integrated over the momentum up to four times the fermi momentum for each energy . this gives the normalization of @xmath78 at the level of @xmath316 which is sufficient for our purposes , but can lead to higher uncertainties in the kinetic energy , which weighs the integral with a higher power of @xmath25 . even a conservative error of @xmath317 in the kinetic energy has repercussions in the @xmath318 effect only at the level of @xmath319 . in fig . 7 we show the results for @xmath80 from the nucleonic contribution calculated as @xmath320 we do not divide by @xmath321 as experimentally done . one reason for it is that the techniques used here with the local density approximation can not be used for deuterium and hence we can not calculate the nuclei and deuterium with the same model . however the price we pay is small . we rely upon the results for @xmath321 calculated for deuterium in @xcite . we can see there that for @xmath322 ranges between @xmath323 . hence , our results should be increased by @xmath324 and we will not worry about this amount . more serious is the region for @xmath325 , where due to fermi motion the former ratio increases rapidly . hence , in that region we should expect to overcount the experiment as it is indeed the case . in fig . 7 we plot the results obtained for @xmath82 . this nucleus has @xmath326 but by a little amount . furthermore , the experimental results are corrected by the isoscalarity factor to convert them into an equivalent isoscalar nucleus @xcite , hence our calculations done for symmetric nuclear matter are appropriate . we see a minimum around @xmath327 as in the experiment and a steep rise at @xmath328 as it correponds to fermi motion @xcite . the region around @xmath327 agrees well with experiment . however at @xmath329 we are below the data by about @xmath330 . this region will be filled up latter by the mesonic contribution . in the same figure we show the results obtained with the uncorrelated fermi sea ( local step function distribution ) of eq . we observe that @xmath331 takes values closer to unity than the results with the spectral function for @xmath332 the reduction of @xmath331 with the use of the spectral function with respect to a static picture of the nucleus , like hartree fock or the equivalent uncorrelated fermi sea in our case , was already shown and explained in ref . the results here are qualitatively similar to those in @xcite and the explanation lies in the increased binding provided by the spectral functions . both the factor @xmath333 of eq . ( 57 ) , as well as the restrictions of phase space @xmath334 , are responsible for the decrease of @xmath331 in this region . on the other hand there is a novelty in these results with respect to those in @xcite . @xmath331 does not go to 1 at @xmath238 as in @xcite and are systematically lower in all the range of @xmath190 . the reduction at @xmath238 is easy to see from the formulae . in eq . ( 55 ) @xmath335 will take a constant value , and with respect to the normalization integral of eq . ( 21 ) the novelties in eq . ( 55 ) are the extra factors @xmath157 and @xmath333 which both go into reducing the contribution of the integrand for off shell nucleons . the overshooting of the results in the region of @xmath336 was already announced as a result of dividing @xmath337 by @xmath338 and not @xmath339 . but the qualitative features due to fermi motion are reproduced . a very detailed discussion of these effects is given in ref . @xcite , but qualitatively we can see that @xmath340 @xmath341 and one can pick up values of @xmath342 in the integrations such that @xmath343 . hence @xmath344 will be different from zero while @xmath345 , and @xmath331 from eq . ( 73 ) necessarily goes to @xmath346 . it is interesting to call on the attention to the crossing of the two lines in fig . the binding effects reduce @xmath347 . on the other hand fermi motion increases @xmath347 close to 1 , as we noted . fermi motion is more important in the interacting fermi sea because now one has larger momentum components . on the other hand the interacting fermi sea has also more binding . as a consequence we see that at @xmath332 , where the binding effects dominate over the fermi motion , @xmath347 for the interacting fermi sea is smaller than with the non interacting fermi sea , while for @xmath325 , where the fermi motion effect dominates , the situation is just opposite . in figs . 8 , 9 , 10 , 11 we show the mesonic contribution to @xmath80 , together with the nucleonic one discussed above , for different nuclei , @xmath348 , @xmath349 , @xmath350 and @xmath351 . the general features are the same in all nuclei , but , of course , the mesonic contribution is smaller in lighter nuclei . the mesonic contribution is calculated with the structure function of ref . we have also calculated it with the older structure functions of ref . we find that around @xmath352 , the results for @xmath80 decrease in about @xmath353 if one uses the pion structure function of ref . @xcite and increase in about @xmath354 if one uses the one of ref . this should give us an idea of the uncertainties of this contribution . in fig . 11 , for @xmath351 , we split the mesonic contribution into the pion and @xmath0-meson ones . we observe , that although the pionic contribution is bigger , the one from the @xmath0-meson cloud is also important , and both of them are positive in all the range of @xmath190 . similarly , as obtained in other calculations @xcite the mesonic contribution vanishes around @xmath327 and increases as @xmath190 decreases . we can see that thanks to the mesonic contribution the agreement with the data becomes much better . the slope of @xmath355 from @xmath356 to @xmath327 is not reproduced by the contribution of the nucleons alone , a feature which is shared by the results of ref . the mesonic contribution comes to produce the right slope . as we said , the qualitative features of the pionic contribution are similar to those in @xcite . there are also some differences . in ref . @xcite the pionic contribution is evaluated assuming different amounts of pion excess in the nucleus , but no evaluation of this excess is made . furthermore one should recall our warnings in section 4 not to use the excess number in the evaluation . in ref . @xcite an actual evaluation is done of the pionic contribution . even if the formalisms here and there might look quite different , they are actually quite similar and one can see that @xmath357 of eq . ( 48 ) appears in ref . @xcite as @xmath358 , with @xmath359 changed to @xmath360 , which is not very problematic when the pions are off shell . furthermore , the @xmath169 contribution is obtained from an extrapolation of the pionic atom data , which is more problematic when one goes to the off shell situations which one finds here . furthermore , this @xmath169 is taken real in @xcite , while here the @xmath169 lindhard function is explicitly evaluated as a function of @xmath361 by keeping the @xmath362 width . an accurate evaluation of the imaginary parts of the diagrams is necessary in order to fulfill the sum rule of eq . ( 69 ) . on the other hand what one evaluates in ref . @xcite is the fractional increase of @xmath355 with respect to the one in the sullivan process ( deep inelastic scattering with the pion cloud of a free nucleon ) @xcite and not the absolute value of the contribution of the pionic cloud . in ref . @xcite the absolute contribution of the pion cloud is obtained by subtracting the `` free '' parts contained in the response function of a free nucleon , as done here . the formalism is similar to ours but some approximations are done which are improved here . for instance the @xmath169 contribution is again taken real and other approximations are done to relate some magnitudes to the pion excess number ( recall our warnings about this ) . the input which we have used for the meson nucleus interaction , @xmath363 , etc , is the one of section 6 . it has the virtue of having been tested in a large variety of reactions and we have not changed it here . this gives us much confidence about the strength of the mesonic contribution obtained here . the fact that it fills up the part of @xmath80 missed by the nucleonic contribution is certainly a welcome feature which reinforces our confidence on this mesonic model . the agreement of the results that we obtain with the experimental data can be considered rather good by comparison with results obtained with other theoretical approaches . the trend of the data is well reproduced and the remaining discrepancies are not incompatible with the intrinsic theoretical uncertainties of our model , particularly of the mesonic contribution and more specially of the @xmath0-meson cloud which is somewhat tied to the form factors and the nuclear correlation function , the fact that we divide by @xmath364 instead of @xmath365 , etc . our results for @xmath355 in @xmath350 and @xmath351 look already very similar and we have checked that @xmath355 does not change much for heavier nuclei . obviously one should take into account that for heavier nuclei @xmath326 and our approach with symmetric nuclear matter should be less accurate . actual calculations keeping different neutron and proton densities lead to a slight decrease of the minimum of @xmath355 as @xmath366 increases @xcite . we have evaluated the nucleonic and mesonic contributions to the ratio @xmath80 of the @xmath296 effect , with particular emphasis on an accurate treatment of effects shown in the past to be relevant , like binding effects , fermi motion and a dynamical ( non static ) treatment of the nucleons and the mesons in the medium . in order to avoid having to take some prescription on how to include relativistic effects in the approach , which has led to many discussions in the past , we started with a relativistic approach from the beginning and have made a covariant treatment which allows us to write all magnitudes in terms of nucleon and meson propagators in the medium . the approach was made practical by evaluating the relevant magnitudes in an infinite medium and calculating the structure functions in finite nuclei by means of the local density approximation . this procedure is fine for @xmath367 but certainly breaks down for @xmath368 where there is nuclear shadowing . we could see that the use of the spectral functions to construct the nucleon selfenergy was relevant in reducing somewhat the ratio @xmath80 with respect to a static picture of the nucleus , like a shell model , or in our case an uncorrelated fermi sea . this reconfirmed qualitatively earlier findings in the same direction . although our results for the nucleonic contribution differ somewhat from other results in the literature , we share their conclusions that the nucleonic contribution alone does not explain the data , particularly the slope from @xmath238 to @xmath327 . on the other hand , we evaluated the contribution from the pion and @xmath0-meson clouds rather accurately . recent work done before on the contribution of the pion cloud to the @xmath369 nucleus selfenergy had taught us some important lessons , particularly the importance of using an input which satisfies a sum rule , not trivial to satisfy unless the analytical properties of the pion selfenergy are strictly fulfilled , and the need to avoid any relationship to the `` pion excess number '' . in addition , experience gained in dealing with reactions which involve real and virtual pions allowed us to use information on the pion nucleus selfenergy which is realistic enough and has been tested in many such reactions . hence , we consider the present calculation of the pionic effects as an improvement over work done in the past and we think these results are rather reliable . however , there are still small uncertainties in the pionic contribution stemming from different results for the pion structure function obtained in different analyses of the drell - yan process . the strength of the pionic effects is moderate . so is the one from the @xmath0-meson cloud , but , when they are added to the nucleonic contribution one obtains a good description of the data in the region of @xmath370 . in summary we could conclude that the main features of the @xmath296 effect can be described in terms of conventional degrees of freedom , nucleons and mesons . it does not exclude explanations in terms of more elementary degrees of freedom like quarks and gluons . it is simply a question of which degrees of freedom are more economical and transparent , as stressed by jaffe in ref . the fact that we could deal with these degrees of freedom with a certain accuracy , and establish the relationship of the effects found to familiar concepts in conventional nuclear physics , makes these degrees of freedom rather appropriate to look at the @xmath296 and related effects . e. m. collab . , j. j. aubert et al . , phys . lett . b 123 ( 1983 ) 275 c. h. llewellyn smith , phys . lett . 128 b ( 1983 ) 107 m. ericson and a. w. thomas , phys . 128 b ( 1983 ) 112 v. sanjos , v. vento and s. noguera , nucl . a470 ( 1987 ) 509 ; p. gonzlez and v. vento , mod . a8 ( 1993 ) 1563 l. l. frankfurt and m. i. strikman , phys . 160 ( 1988 ) 236 r. p. bickerstaff and a. w. thomas , j. phys . g15 ( 1989 ) 1523 a. b. migdal , e. e. saperstein , m. a. troitskii and d. n. voskresenskii , phys . ( 1990 ) 179 s. v. akulinichev , s. a. kulagin and g. m. vagradov , pis ma zh . eksp . fiz 42 ( 1985 ) 105 ; jetp lett . 42 ( 1985 ) 127 ; phys . 158 b ( 1985 ) 485 g. v. dunne and a. w. thomas , phys . d 33 ( 1986 ) 2061 s. v. akulinichev and s. shlomo , phys . rev . c 33 ( 1986 ) 1551 l. l. frankfurt and m. i. strikman , phys . b 183 ( 1987 ) 254 g. l. li , k. f. liu and g. e. brown , phys . lett . b 213 ( 1988 ) 531 c. ciofi degli atti and s. liuti , phys . rev . c 41 ( 1990 ) 1100 b. l. birbrair , e. m. levin and a. g. shuvaev , nucl . phys . a 491 ( 1989 ) 618 r. l. jaffe , nucl . phys . a 478 ( 1988 ) 3c f. gross and s. liuti , phys . c 45 ( 1992 ) 1374 s. liuti and f. gross , phys . b 356 ( 1995 ) 157 j. nieves , e. oset and c. garcia - recio , nucl . phys . a 554 ( 1993 ) 509 ; ibid , pag . 554 h. c. chiang , e. oset and p. fernndez de crdoba , nucl . a 510 ( 1990 ) 591 s. k. singh and e. oset , nucl . phys . a 542 ( 1992 ) 587 r. c. carrasco and e. oset , nucl . phys . a 536 ( 1992 ) 445 ; r. c. carrasco , e. oset and l. l. salcedo , nucl . phys . a 541 ( 1992 ) 585 s. a. kulagin , g. piller and w. weise , phys . c 50 ( 1994 ) 1154 n. n. nikolaev and b. g. zakharov , phys . b 327 ( 1994 ) 157 a. l. fetter and j. d. walecka , quantum field theory of many particle systems , ( mcgraw hill , ny , 1971 ) p. fernndez de crdoba and e. oset , nucl . phys . a 528 ( 1991 ) 736 c. itzykson and j. b. zuber , quantum field theory ( mcgraw hill , ny , 1980 ) h. de vries , c. w. de jager and c. de vries , at . dat . nucl . dat . tables 36 ( 1987 ) 495 g. c. li , i. sick , r. r. whitney and m. r. yearian , nucl . phys . a 162 ( 1971 ) 583 f. mandl and g. shaw , quantum field theory , john wiley , 1984 e. amaldi , s. fubini and g. furlan , pion electroproduction , springer tracts in modern physics , vol 83 ( springer , berlin , 1979 ) c. garcia - recio , j. nieves and e. oset , phys . c 51 ( 1995 ) 237 m. f. jiang and d. s. koltun , phys . c 46 ( 1992 ) 2462 i. j. r. aitchison and a. j. g. hey , gauge theories in particle physics , edit adam hilger g. baym and g. e. brown , nucl . phys . a 247 ( 1975 ) 395 w. y. p. hwang , j. speth and g. e. brown , z. phys . a 339 ( 1991 ) 383 a. szczurek and j. speth , nucl . phys . a 555 ( 1993 ) 249 l. heller and a. w. thomas , phys . rev . c 41 ( 1990 ) 2756 u. oelfke , p. u. sauer and f. coester , nucl . phys a 518 ( 1990 ) 593 w. melnitchouk , a. w. schreiber and a. w. thomas , phys . rev . d 49 ( 1994 ) 1183 p. berge et al . z. phys . c49 ( 1991 ) 187 emc , j. j. aubert et al . lett b 114 ( 1982 ) 291 m. glck , e. reya and a. vogt , z. phys . c 53 ( 1992 ) 651 r. machleidt , k. holinde and ch . elster , phys . reports 149 ( 1987 ) 1 e. oset and w. weise , nucl . phys . a 319 ( 1979 ) 477 e. oset , p. fernndez de crdoba , l. l. salcedo and r. brockmann , phys . reports 188 ( 1990 ) 79 p. fernndez de crdoba and e. oset , phys . c 46 ( 1992 ) 1697 s. fantoni , b. l. friman and v. r. pandharipande , nucl . phys . a 399 ( 1983 ) 51 s. fantoni and v. r. pandharipande , nucl . a 427 ( 1984 ) 473 c. mahaux , p. f. bortignon , r. a. broglia and c. h. dasso , phys . 120 ( 1985 ) 1 a. ramos , a. polls and w. h. dickhoff , nucl . a 503 ( 1989 ) 1 h. mther , g. knehr and a. polls , phys . c 52 ( 1995 ) 2955 h. mther , a. polls and w. h. dickhoff , phys . c 51 ( 1995 ) 3040 d. s. koltun , phys . c 9 ( 1974 ) 484 c. ciofi degli atti and s. liuti , phys . b 225 ( 1989 ) 215 c. ciofi degli atti and s. liuti , phys . rev . c 44 ( 1991 ) r 1269 r. g. arnold et al . 52 ( 1984 ) 727 a. bodek and j. l. ritchie , phys . d 23 ( 1981 ) 1070 j. badier et al . , z. phys . c 18 ( 1983 ) 281 p. castorina and a. donnachie , z. phys . c 45 ( 1990 ) 497 j. d. sullivan , phys . d 5 ( 1972 ) 1732 ch . h. chung and w. y. p. hwang , phys . d 49 ( 1994 ) 2221 j. gmez et al . d 49 ( 1994 ) 4348 a. c. benvenuti et al . b 189 ( 1987 ) 483 nmc collaboration , m. arneodo et al . , b 441 ( 1995 ) 12 nmc collaboration , p. amaudruz et al . , nucl . b 441 ( 1995 ) 3 \a ) free nucleon , b ) fermi sea with b baryons . + @xmath371 figure 3 : ( a ) feynman diagram for deep inelastic electron - nucleon scattering and ( b ) , electron selfenergy diagram associated . + @xmath371 figure 4 : electron selfenergy diagram accounting for electron - pion deep inelastic scattering . + @xmath371 figure 5 : diagrams of the electron selfenergy including @xmath372 , etc .. + @xmath371 figure 6 : two diagrams ( a ) direct and ( b ) crossed , which contribute to compton @xmath186 scattering . + @xmath371 figure 7 : results of @xmath331 for @xmath82 . solid line : using the spectral function ; dashed line : using the uncorrelated fermi sea . experimental points from ref . @xcite ( solids dots ) , ref . @xcite ( open squares ) . + @xmath371 figure 8 : results of @xmath355 for @xmath373 . solid lines : whole calculation including the nucleons and the mesons ; dashed line : contribution of the nucleons . experimental points from ref . @xcite ( solid dots ) . density for @xmath373 from ref . @xcite . @xmath371 figure 9 : same as fig . 8 for @xmath374 . experimental points from ref . @xcite ( solid dots ) , ref . @xcite ( open squares ) . + @xmath371 figure 10 : same as fig . 8 for @xmath375 . experimental points from ref . @xcite ( solid dots ) , ref . @xcite ( open squares ) . + @xmath371 figure 11 : results for @xmath355 for @xmath351 . solid line : whole calculation including the nucleons and the mesons ; dashed line : contribution of the nucleons ; dot - dashed line : contribution of nucleons plus pions . experimental points from ref . @xcite ( solid dots ) , ref . @xcite ( open squares ) . llllc & & @xmath376t@xmath377 [ mev ] & @xmath378e - m@xmath379 [ mev ] & @xmath380 [ mev ] + + @xmath348 & ufs & 9.3 & 22.1 & 5.2 + & sf & 18.8 & 33.8 & 5.2 + & & & & + @xmath349 & ufs & 13.7 @xmath381 ( 17.0 ) & 31.0 @xmath381 ( 23.0 ) & 8.0 + & sf & 31.7 @xmath381 ( 37.0 ) & 50.4 @xmath381 ( 49.0 ) & 7.8 + & & & & + @xmath350 & ufs & 16.0 @xmath381 ( 16.5 ) & 33.5 @xmath381 ( 26.6 ) & 8.5 + & sf & 40.4 @xmath381 ( 36.0 ) & 58.6 @xmath381 ( 52.1 ) & 8.6 + & & & & + @xmath351 & ufs & 16.1 @xmath381 ( 17.0 ) & 34.1 @xmath381 ( 25.0 ) & 8.9 + & sf & 40.4 @xmath381 ( 33.0 ) & 58.6 @xmath381 ( 49.8 ) & 8.7 + ufs : uncorrelated fermi sea . sf : spectral function . the numbers in brackets correspond to those obtained in ref . @xcite respectively for the hartree - fock and spectral functions ( the latter called there src , from short range correlations ) .

the two - dimensional xy - model with random phase - shifts on bonds is studied . the analysis is based on a renormalization group for the replicated system . the model is shown to have an ordered phase with quasi long - range order . this ordered phase consists of a glass - like region at lower temperatures and of a non - glassy region at higher temperatures . the transition from the disordered phase into the ordered phase is not reentrant and is of a new universality class at zero temperature . in contrast to previous approaches the disorder strength is found to be renormalized to larger values . several correlation functions are calculated for the ordered phase . they allow to identify not only the transition into the glassy phase but also an additional crossover line , where the disconnected vortex correlation changes its behavior on large scales non - analytically . the renormalization group approach yields the glassy features without a breaking of replica symmetry . 2

cond-mat9601131

the two - dimensional xy - model with random phase - shift , defined in eq . ( [ h ] ) below , captures the physics of a variety of physical systems . among them are xy - magnets with non - magnetic impurities , which are coupled to the xy - spins via the dzyaloshinskii - moriya interaction,@xcite josephson - junction arrays with geometric disorder,@xcite and crystals with quenched impurities.@xcite disordered xy - models are related to even more physical systems , like impure superconductors . it is in particular the interest in vortex - glass phases in type - ii superconductors , which motivates an analysis of possible glassy features in the paradigmatic two - dimensional bond - disordered system . the most fundamental question is , whether this model has an ordered phase . in the absence of disorder , there exists at low temperatures a phase with quasi long - range order . the famous kosterlitz - thouless transition separates it from the high - temperature phase with short - ranged order.@xcite the analysis by cardy and ostlund@xcite and more explicitly by rubinstein , shraiman and nelson@xcite predicted that the phase transition should be reentrant in the presence of disorder . later korshunov suggested,@xcite that the ordered phase might be completely destroyed by disorder . ozeki and nishimori@xcite showed for models with gauge - invariant disorder distributions , that the phase transition can not be reentrant , leaving open whether the ordered phase does exist or not . experiments@xcite on josephson junction arrays and simulations@xcite were in favor of the existence of an ordered phase without reentrance . in this controversial situation nattermann , scheidl , korshunov , and li@xcite as well as cha and fertig@xcite reconsidered only recently the problem and found that the ordered phase _ does _ exist and its boundary _ does not _ have reentrant shape . the earlier observation of reentrance was attributed to an overestimation of vortex fluctuations at low temperatures . the present study extends refs . and , which focused on the absence of reentrance . the purpose of this article is to examine in more detail the ordered phase , which is shown to be composed of different _ regions_. they are distinguished by the behavior of vortex correlation functions on large length scales , which are calculated _ quantitatively_. a low - temperature regime is found to exhibit glassy behavior in its correlations . in addition to the previous studies we find that the strength of disorder is increased under renormalization . this effect does not destroy the ordered phase but it is crucial for the universality type of the transition at low temperatures . in the following a generalized self - consistent screening approach is applied to the replicated system . the main improvement compared to earlier approaches@xcite on the same basis is achieved by including all relevant fluctuations and by taking special care when the number of replicas is sent to zero . we use the replica formalism mainly for two reasons : i ) the calculation of correlations becomes very convenient : one may use an expansion technique ( small fugacity and density of vortices ) which fails@xcite in the unreplicated system . ii ) we show that in the present model there is _ no _ need for replica symmetry breaking , even at lowest temperatures . in particular , the reentrance disappears and the glassiness occurs without a breaking of replica symmetry . this might astonish , since for various disordered systems reentrance has been explained as an artifact of a replica - symmetric ( exact mean - field or approximate variational ) approach , which has been overcome by breaking the replica symmetry.@xcite the paper is organized as follows : in sec . ii we set up our model and map it onto an effective coulomb gas . we recapitulate the breakdown of fugacity expansion in the presence of disorder . in sec . iii we set up the renormalization group within a self - consistent formulation of screening . the consistency of this approach in the original and replicated system is demonstrated . iv is devoted to the evaluation of the self - consistency and the proper performance of the replica limit . the results of the renormalization group treatment are worked out in sec . v , which are discussed and compared to the results of previous work in sec . the xy - model with random phase - shifts is given by the reduced hamiltonian @xmath0 it refers to a square lattice with unit spacing in two dimensions . an xy - spin with angle @xmath1 is attached to every lattice site ( position @xmath2 ) . the reduced spin coupling between nearest neighbors @xmath3 reads @xmath4 . the interaction involves quenched random phase - shifts with variance @xmath5 , which are uncorrelated on different bonds . thermal fluctuations are weighted according to the partition sum @xmath6 this xy - model can be mapped approximately onto an effective coulomb gas . since this procedure is well known in the absence of disorder@xcite and can be performed similarly in the presence of disorder , we recall only briefly the main manipulations leading to the effective model . in order to simplify the analysis of model ( [ z ] ) , one can replace approximately the cosine - interaction by the villain - interaction.@xcite thereby an additional variable , the vortex density @xmath7 is introduced . the decisive advantage of this approximation is , that the hamiltonian becomes bilinear in the angles . all configurations of the original angles can be expressed in terms of spin - waves and vortices . just as in the pure case , spin - waves and vortices decouple energetically . since spin - waves are simply harmonic , all non - trivial physics arises from vortices with an effective hamiltonian [ h.eff ] @xmath8 in fourier and real space , and the partition sum @xmath9 here @xmath10 is the integer - valued vortex density , which is subject to thermal fluctuations . on the square lattice @xmath10 is associated with plaquettes ( dual sites ) , which are labeled by @xmath11 in contrast to sites @xmath2 . thermal _ vortices @xmath7 couple to a background of _ quenched _ vortices @xmath12 . the last expression means , that the plaquette variable @xmath13 is given by the rotation of variable @xmath14 on the surrounding bonds . from this relation one immediately derives , that the quenched vortices are gaussian random variables with variance @xmath15 . a single phase - shift @xmath16 creates a dipole of quenched vortices situated in the plaquettes which have the bond @xmath17 as side . therefore quenched vortices are anti - correlated and their total vorticity vanishes . we recognize from eq . ( [ hv.k ] ) , that only neutral configurations of vortices will have finite energy . hence the partition sum can be restricted to neutral configurations ( i.e. @xmath18 ) like in the pure case . every such neutral configuration can be imagined as a superposition of dipoles of a vortex ( @xmath19 ) and an antivortex ( @xmath20 ) . in analogy to electrostatics , vortices of the same sign repel each other and vortices of different sign attract each other . since the interaction is logarithmic , vortices build a coulomb gas . the interaction is identical among the thermal and quenched component . the fourier representation ( [ hv.k ] ) of the vortex hamiltonian makes evident , that thermal vortices try to compensate quenched vortices as well as possible . a perfect compensation is prevented by the fact , that @xmath10 is restricted to integer values , whereas @xmath13 is a continuous random variable . in eq . ( [ hv.r ] ) we introduced the vortex core energy @xmath21 , where @xmath22 takes the value @xmath23 on a square lattice . it emerges from the lattice greens function [ cf . ( 4.13a ) of ref . ] . although @xmath22 has a unique value for the lattice model , it will occasionally be considered as an independent parameter which allows to control the density of vortices irrespectively of @xmath24 . in a dilute vortex system with @xmath25 only , @xmath26 is comparable to a chemical potential . in a region of parameters , where vortices are negligible , fluctuations of the original angle variables are given only by spin waves . they lead to a decay of the spin correlation function@xcite @xmath27 \rangle } \sim r^{- \eta}\ ] ] with exponent @xmath28 . if the correlation decays algebraically even in the presence of vortices , the system has quasi long - range order . as vortices are excited thermally , they give rise to additional fluctuations in the angles , the correlation will decay faster . this effect can be expressed by a renormalized @xmath29 , which becomes scale dependent . if vortices lead to a divergence of the renormalized @xmath29 on large scales , quasi long - range order is destroyed . however , vortex fluctuations are not only responsible for such a quantitative effect , they also drive the phase transition into the high - temperature phase , where the spin - spin correlation decays exponentially.@xcite let us estimate , to what extent disorder can favor the excitation of vortices . in the pure case @xmath30 , the ground state of ( [ hv.k ] ) is vortex - free . the density of vortices vanishes for vanishing temperature , since the creation of quantized vortices always costs finite energy . therefore the pure system has a phase with quasi long - range order ( [ def.gamma ] ) at low temperatures . a simple argument shows , that the _ disordered _ system has even at _ zero _ temperature a _ finite density _ of vortices:@xcite we consider two positions at distance @xmath31 and determine the probability of finding there a vortex - antivortex pair . the energy of such a dipole is given by @xmath32 , where we neglect the core energy for large @xmath31 . disorder gives and additional energy contribution @xmath33 . at @xmath34 , the dipole will be present , if its total energy @xmath35 is negative . the original gaussian distribution of the random phase - shifts results in a gaussian distribution of @xmath36 with variance @xmath37 . thus the probability @xmath38 of finding a pair at @xmath34 is finite . in other words , this shows that in the presence of disorder the ground state is no longer vortex - free and that renormalization effects can be important even at arbitrarily small temperatures . the goal of the present work is therefore a calculation of such renormalization effects in the presence of random phase - shifts . we use the conceptual most simple approach , a self - consistent linear response approach . for the pure coulomb gas , this approach is equivalent@xcite to the real - space renormalization group of kosterlitz.@xcite we have checked that this equivalence still holds in the presence of disorder . we now define vortex renormalization effects by macroscopic properties of our system . as a probe we introduce additional test vortices into the system . they experience a screened interaction and the screened background potential acting on them . let us denote the unscreened interaction between vortices @xmath39 . due to quenched disorder vortices @xmath40 , thermal vortices @xmath7 are subject to a gaussian background potential @xmath41 with variance @xmath42 . the screened background potential and interaction are identified from the contributions to the free energy of test vortices , which are of first and second order in their vorticity@xcite ( see also appendix [ app.lr ] for some intermediate steps ) [ screen.v ] @xmath43 it is convenient to introduce the ordinary , disconnected , and connected correlation , [ def.corr ] @xmath44 for a specific realization of disorder , the screened interaction in principle depends explicitly on the coordinates of both interacting vortices , it is not translation invariant . although we should in principle consider screening in the particular disorder environment , we approximately evaluate only the average screening effect by taking the disorder average , which restores the translation symmetry of the interaction . from the large - scale behavior of the screened interaction and the variance of the screened potential one can then identify screened parameters @xmath45 and @xmath46 as ( for the precise definition see appendix [ app.lr ] ) [ total.screen ] @xmath47 although we do not yet know the correlation functions , we may establish from these expressions criteria for the existence of an ordered phase . in this phase , we expect vortices to give rise only to finite screening effects , i.e. the screened parameters must have finite values . this requires @xmath48 and a similar condition for the disconnected correlation . in the limit of large core energy , one can calculate the correlations to leading order by considering only a single vortex dipole . in th absence of disorder , one finds * * ? ? ? * * @xmath49 and @xmath50 . the condition for the ordered phase then simply reads @xmath51 , in agreement with the condition that the free energy a single vortex has to be positive.@xcite a similar argument can be constructed for the disordered system at @xmath34 , where one can estimate @xmath52 using eq . ( [ prob.dipole ] ) . then the condition for order reads @xmath53 . this argument already disproves previous predictions,@xcite that infinitesimal @xmath54 would destroy order at @xmath34 . for finite core energy , screening effects have to be taken into account quantitatively for the calculation of the correlations . for this purpose we develop a self - consistent scheme in analogy to the pure case.@xcite we introduce scale - dependent variables @xmath55 and @xmath56 , which include screening by dipoles of radius @xmath57 only . ( [ total.screen ] ) can be cast in differential form : [ partial.screen ] @xmath58 both equations combine to a simple flow equation of the spin - spin correlation exponent , @xmath59 . ( [ partial.screen ] ) suggests a simple `` two - component picture '' illustrating the screening effects of vortices . this picture virtually separates dipoles of thermal vortices @xmath7 into a frozen and a polarizable component . frozen _ component gives rise to the @xmath60 . it is not polarizable and does not contribute to screening of the interaction . however , since it is frozen , it behaves like the quenched disorder background and leads to an effective disorder strength represented by the screened @xmath61 . polarizable _ component is the source of @xmath62 and leads to the screening of @xmath24 . both components contribute equally to a renormalization of the spin exponent . this virtual separation into two components must not be taken literally . it is impossible to assign a particular dipole uniquely to one of the two components . the two - component picture is analogous to the two - fluid model of superfluidity , which also must not be taken literally in the sense that a particular atom would be either superfluid or normalfluid . in order to take advantage from the differential screening equations , we will calculate the correlations at @xmath63 in a self - consistent way which includes screening effects by smaller dipoles . at each differential step we will also perform a simultaneous rescaling @xmath64 . then we can interpret the differential equations as renormalization group flow equations . for the further investigation of the model , we apply the standard replica technique@xcite to the partition function ( [ z.vc ] ) . we are going to rewrite screening in the replicated system in order to expose the consistency between screening in the unreplicated and replicated system . the coulomb gas ( [ hv.r ] ) is replicated @xmath65 times and after disorder averaging one obtains the effective hamiltonian @xmath66 we introduce the couplings @xmath67 , @xmath68 and the core energy @xmath69 , where @xmath70 and @xmath71 . on the partition sum the neutrality condition @xmath72 is imposed for every replica . in the replicated system , screening effects can be calculated in the same linear response scheme as before . the leading term of the free energy of test replica - vortices is of second order in the infinitesimal test vorticity . from this order , we derive a differential screening @xmath73 in analogy to eq . ( [ partial.screen ] ) . we introduced the replica correlation @xmath74 , where @xmath75 denotes a thermal average in the replicated system . since the initial interactions are replica - symmetric , i.e. the coupling has the form @xmath76 , we use the replica symmetric ansatz @xmath77 for correlations . the relation to the correlations in the unreplicated system is provided by @xmath78 for all three types @xmath79 ( ordinary , disconnected , and connected ) of correlations , if we identify @xmath80 consistently . exploiting the replica symmetry of correlations , eq . ( [ partial.screen.rep ] ) decays into flow equations for @xmath24 and of @xmath81 , which read in terms of @xmath24 and @xmath82 : [ partial.screen.ren ] @xmath83 in order to approach the usual notation of renormalization groups , we define _ effective _ fugacities @xmath84 again for all three types of rescaled correlations . they are related by @xmath85 . in these definitions we took already account of rescaling of lengths , which generates an additional flow of the core energies , [ flow.e ] @xmath86 eqs . ( [ partial.screen.ren ] ) together with ( [ flow.e ] ) constitute the main renormalization group flow equations . from relations ( [ rel.corr ] ) we recognize , that the flow equations of @xmath24 and @xmath61 of the unreplicated system coincide with those of the replicated system in the limit @xmath87 . in the next section these equations will be completed by a flow equation of the free energy , which however does not feed into the other flow equations . the flow equations ( [ partial.screen.ren ] ) together with ( [ flow.e ] ) can not yet be evaluated since they are not closed . in this section we are going to achieve this closure by expressing the effective fugacities in terms of @xmath24 , @xmath61 , @xmath26 and @xmath88 . in the replica language , the effect of disorder is encoded in additional interactions between vortices and we first have to discuss their nature . an inspection of eq . ( [ h.rep ] ) shows , that vortices of opposite sign in the same replica interact with a potential @xmath89 . for @xmath90 they always attract each other . these dipoles are stable against dissociation due to thermal fluctuations only for @xmath91.@xcite in previous work @xcite the stability of precisely such dipoles was used as criterion to determine the boundary of the ordered phase . however , for @xmath92 and low temperature or strong disorder , @xmath93 becomes negative ( repulsion ! ) and this criterion becomes questionable . the increasing instability of these dipoles gives rise to a reentrant phase boundary.@xcite however , it is not sufficient to consider interactions only within replicas . independently on temperature and replica number @xmath65 , vortices of the same sign but in different replicas attract each other with a potential @xmath94 . in the limit @xmath92 the coupling between different replicas reads @xmath95 . for low temperature or strong disorder , the interaction between different replicas becomes stronger to the same extent as the intra - replica coupling becomes weaker and correlations have to be calculated taking into account the competition between intra- and inter - replica interactions . the importance of inter - replica interactions has been recognized by korshunov.@xcite the essence of the replica problem is a suitable treatment of the inter - replica interaction for integer @xmath90 and to construct an analytic continuation which allows to take then the limit @xmath96 in a proper way . in the following this is done in a way different from korshunov s way , leading to opposite conclusions . on the technical level , this step is the essential progress of the present work . as necessary prerequisite for the analytic continuation @xmath87 , we must discuss vortex fluctuations in the replicated system for integer @xmath90 . this happens in some detail since our final conclusions contradict refs . . we first check , that the _ ground state _ of the _ replicated _ hamiltonian is the state without vortices . for this purpose we examine the hamiltonian in fourier space , @xmath98 . there we easily recognize , that every creation of vortices costs energy , because @xmath99 is positive definite : @xmath99 has one eigenvalue @xmath100 and @xmath101 eigenvalues @xmath102 . since the ground state has no vortices , we are allowed to use for low temperatures a low density expansion . this is in contrast to the unreplicated system , where the ground state has a finite density of vortices , as discussed above . for the following it is convenient to introduce the notion of a `` replica - vortex '' ( compare fig . [ fig_dipole ] ) : in a state with vorticity @xmath103 at position @xmath11 , we say that a `` replica - vortex '' of _ type _ @xmath104 is at position @xmath105 . ( upper indices at @xmath7 are not exponents but replica indices ! ) since the core energy contains a term @xmath106 , which suppresses large @xmath107 , we may restrict our consideration for large @xmath26 or @xmath22 to states with @xmath108 : one vortex - free state and @xmath109 different types of replica - vortices , antivortices included . because of the neutrality condition ( [ neutrality ] ) the elementary excitations in the replicated system are again vortex dipoles . in the dilute limit ( large @xmath22 ) , some type of such replica - vortices will form with their antivortices those dipoles , which are most instable against thermal fluctuation and thus destroy order . to find out this type , we have to discuss the energy of such dipoles . the core energy @xmath110 of replica - vortex @xmath111 depends only on the total vortex number @xmath112 and the number @xmath113 of vortices with @xmath114 . the same holds for the coupling @xmath115 , which gives the strength of the logarithmic interaction between a replica - vortex of type @xmath116 and its antivortex : [ def.e_m ] @xmath117 due to replica symmetry , there is a certain degeneracy between different types @xmath116 , which we include when we speak of the _ class _ @xmath118 of replica vortices . the existence of order requires stability of _ all _ types of dipoles against thermal dissociation . for the dilute system this is guaranteed by @xmath119 according to the simple stability criterion of kosterlitz and thouless.@xcite the least stable class with the weakest coupling can be determined explicitly : it is given by replica - vortices of class @xmath120 with @xmath121 in a high temperature region @xmath122 , and by replica - vortices of classes @xmath123 with @xmath124 in a low temperature region @xmath125 . the resulting phase diagram for @xmath126 is shown in fig . [ fig_pdgn ] . the fact , that the class of the least stable replica - vortices for @xmath127 depends on the physical parameters , can be interpreted as an indication for a phase transition even for @xmath92 . we take it as a warning , that we must not focus on a _ single _ class of vortices . the origin for the failure of the early replica approaches@xcite is just the focus on classes @xmath120 only . we even do not restrict ourselves to the two classes which are dominant for @xmath128 , since it turns out _ a posteriori _ , that also other classes contribute to the replica limit @xmath87 . now we calculate the correlations and the effective fugacities taking into account contributions by all types of replica - vortices . for a low density of replica - dipoles ( realized for large @xmath22 or at low temperature ) we can neglect the interaction between different replica - dipoles ( `` independent dipole approximation '' ) . this means that we can calculate approximate correlations in the partition sum containing only up to one dipole of replica - vortices , @xmath129 the fugacity of replica - vortex @xmath116 is defined by @xmath130 . the factor @xmath131 is present , since @xmath116 runs over vortices and their antivortices and we should count every realization of a dipole only once . in eq . ( [ z.lo ] ) integration is restricted to @xmath132 for all @xmath133 , since lengths are rescaled . before we turn to the calculation of correlations , we wish to make sure that calculations in this ensemble make sense in the limit @xmath134 . for this purpose we have to convince ourselves , that @xmath135 . this is equivalent to the condition , that the free energy per replica has a finite limit . then also the renormalization group flow of the free energy per replica , which follows from the contribution of dipoles of radius @xmath136 to @xmath137 , is finite . the flow reads @xmath138 where @xmath139 denotes the free energy per unit volume , per replica , and divided by temperature . the first term on the right - hand side originates from rescaling . from eq . ( [ flow.f ] ) one can derive ( see appendix [ app.extreme ] ) , that a restriction to a single class of replica - vortices leads to an inconsistency with the replica trick , namely the divergence of @xmath139 for @xmath87 . therefore we retain _ all _ classes of vortices . before we turn to the correlations , we evaluate the simpler sum , which is the source for the flow of free energy : @xmath140 where we introduced a `` dummy '' gaussian random variable @xmath141 with @xmath142 and @xmath143 and a `` shell partition sum '' @xmath144 with weights @xmath145 physically , the `` dummy '' gaussian random variable @xmath141 represents nothing but the disorder component which couples to the dipoles with radius in the shell being integrated out . since disorder is uncorrelated on different length scales , we can average over disorder on a given scale right when we consider screening by dipoles with a radius of just the same scale . for this reason we can _ undo the replica trick _ in the flow equations in the very same way as we introduced it initially in the full unrenormalized system . in eq . ( [ sum.y ] ) the replica number @xmath65 became an explicit variable and we may now perform the analytic continuation @xmath146 , @xmath147 as shown in appendix [ app.extreme ] , this yields a well defined free energy per replica only because we have retained _ all _ classes of replica - vortices . this means , that types of fluctuations , which seem to be _ irrelevant _ for @xmath90 ( in our case : energetically expensive replica - dipoles ) can become _ important _ in the replica limit . in a similar way we can proceed to determine the effective fugacities . in the independent dipole approximation ( [ z.lo ] ) the rescaled correlation functions read ( for @xmath148 ) @xmath149 we ignored normalization by the factor @xmath150 , since this factor becomes unity for @xmath92 . with the help of a generating function @xmath151 we find ( @xmath152 ) [ undo.corr ] @xmath153 the @xmath31-dependence of correlations can be restored by the substitutions @xmath154 and @xmath155 . as before , @xmath65 became an explicit variable such that we can send @xmath87 . the effective fugacities then read [ undo.yeff ] @xmath156 now we have achieved our goal of expressing the effective fugacities of the disordered system ( @xmath92 ) as functions of @xmath26 and @xmath88 . in principle these fugacities could also depend on @xmath24 and @xmath61 . such a dependence is absent in the independent dipole approximation used above : correlations are determined neglecting interactions between dipoles , and the energy of single dipoles of radius unity does not depend on @xmath24 and @xmath61 . equations ( [ partial.screen.ren ] ) , ( [ flow.e ] ) , and ( [ undo.yeff ] ) form a closed set of equations . the flow of free energy ( [ flow.f.n=0 ] ) does not feed back into the other flow equations . appealing to the two - component picture , we may call @xmath157 `` density of dipoles '' , @xmath158 `` reduced free energy per dipole '' , @xmath159 `` fraction of polarizable dipoles '' , and @xmath160 `` fraction of frozen dipoles '' . these quantities are given by eq . ( [ undo.yeff ] ) as functions of @xmath26 and @xmath88 . for the convenience of the reader we summarize the flow equations with the two - component parameters : [ flow.compact ] @xmath161 as initial values for @xmath24 and @xmath61 one has to use the unrenormalized values , which also enter @xmath70 , @xmath162 and thereby @xmath163 , @xmath164 , @xmath165 , and @xmath166 . initially @xmath167 since no fluctuations are included . in the present form the flow equations are not very convenient since the dependence of the effective fugacities and the two - component parameters on @xmath26 and @xmath88 is still quite intricate . therefore we perform an additional approximation which is valid on large length scales . first the integrals over @xmath141 ( which are to be performed as disorder averages in eqs . ( [ undo.yeff ] ) ) are split into intervals where one of the contributions to the shell partition sum dominates . then the normalizing denominators in eqs . ( [ undo.yeff ] ) can be expanded with respect to the smaller contributions . the resulting infinite series can be integrated over @xmath141 analytically term by term . from these series one then can extract easily the leading terms for large @xmath168 after approximating @xmath169 and @xmath170 . since @xmath171 and @xmath172 , these approximations are asymptotically correct provided @xmath24 and @xmath61 converge to finite values . for convenience we introduce an effective temperature variable @xmath173 the asymptotic approximation yields for the flow of the reduced free energy @xmath174 for the fugacities we find analogously [ y.asy ] @xmath175 although we were starting from _ unique _ expressions , the _ asymptotic _ behavior differs for high temperature or weak disorder and for low temperature or strong disorder . there are _ three regimes _ of parameters , namely @xmath176 ( called regime ia ) , @xmath177 ( called regime ib ) , and @xmath178 ( called regime ii ) . their physical properties will be analyzed in the next section . approaching the separatrices @xmath179 or @xmath180 using the expressions for the high- and low - temperature regime , the expressions should coincide . this is true for their exponents . the comparison of the prefactors is not meaningful , since in the low - temperature expression the diverging factor @xmath181 is multiplied with the asymptotically vanishing factor @xmath182 . remarkably , for zero temperature ( @xmath183 ) we find @xmath184 , which means that the probability @xmath185 , eq . ( [ prob.dipole ] ) , coincides with the ordinary and connected correlation . for the analysis of the flow equations , we wish to use the quantities of the two - component picture , as introduced in eqs . ( [ flow.compact ] ) . the reduced free energy per dipole , fraction of polarizable dipoles , and fraction of frozen dipoles converge asymptotically to [ def.fpq ] @xmath186 here we have neglected contributions which vanish on large scales . for example @xmath166 is not strictly zero but vanishes exponentially with @xmath187 for @xmath188 and @xmath189 . remarkably , the asymptotic two - component parameters depend on @xmath81 and @xmath61 only through the effective temperature @xmath190 . in order to eliminate @xmath26 and @xmath88 completely from the asymptotic flow equations , we determine the flow equation for the fugacity : @xmath191 to close this subsection of the paper , we wish to point out , that the flow equations ( [ flow.compact ] ) for @xmath24 and ( [ flow.y.asy ] ) for @xmath163 with expression ( [ def.p ] ) for the polarizability coincide with those of nattermann _ et al_@xcite apart from factors smaller than 2 . this small deviation is due to a rougher asymptotic approximation in ref . . in the previous sections the technical part of deriving the flow equations has been achieved . the most important flow equations are eqs . ( [ flow.compact]a-c ) . originally , the two - component parameters @xmath164 , @xmath165 , and @xmath166 have been functions of @xmath26 and @xmath88 . in the asymptotic approximation @xmath26 and @xmath88 have been eliminated , the two - component parameters are given by eq . ( [ def.fpq ] ) as simple functions of @xmath192 , and the fugacity flows according to eq . ( [ flow.y.asy ] ) . these flow equations can be summarized by [ flow.sum ] @xmath193 thus we are in position to discuss the physical implications of these flow equations . to start with , we address qualitative aspects and show that the flow equations satisfy some fundamental physical criteria . according to eqs . ( [ def.y.eff ] ) and ( [ undo.yeff ] ) all correlations @xmath194 are negative , since all fugacities are positive @xmath195 . this is consistent with a screening that weakens the attractive interaction between vortices of the same sign and weakens the repulsive interaction between vortices of the same sign . as a consequence , the two - component parameters are also non - negative . our asymptotic expressions show , that the fraction of polarizable dipoles @xmath196 vanishes linearly at small temperature , whereas the fraction of frozen dipoles @xmath197 becomes unity . screening reduces the coupling @xmath198 since @xmath199 , and it increases the strength of disorder , @xmath200 . in comparison to the pure case the screening of @xmath201 is reduced by the factor @xmath202 , the fraction of polarizable dipoles . whereas previous theoretical approaches did not yield a renormalization of @xmath61 , eq . ( [ flow.sum]c ) manifests that our renormalization scheme leads to an _ increase _ of @xmath61 . the free energy density is reduced by screening , see eq . ( [ flow.sum]d ) . due to disorder the contribution of dipoles is modified by a factor @xmath203 . the increase of @xmath61 seems to contradict the tendency of thermal vortices @xmath7 to _ neutralize _ quenched vortices @xmath40 . this neutralization is expected from eq . ( [ h.eff ] ) . however , the screened @xmath61 is _ not _ defined by the fluctuations of @xmath204 . rather , it was introduced via the fluctuation of the screened disorder potential @xmath205 . the fluctuations of @xmath205 are proportional to @xmath206 . the expected neutralization shows up as @xmath207 . this holds even though @xmath208 since the reduction of @xmath24 dominates the increase of @xmath61 , which follows from @xmath209 that is easily proven with the asymptotic expressions . the most important quantity to analyze is the flow of entropy . from spin - glass models one knows , that a replica - symmetric theory might lead to a negative entropy . this unphysical feature then indicates that the symmetric approach is insufficient and one has to allow for a breaking of replica symmetry . we now show that the present replica - symmetric theory does not suffer from this problem . therefore we believe that a _ breaking of replica symmetry is not necessary_. when the entropy @xmath210 is derived from the reduced free energy , the unrenormalized values of @xmath201 and @xmath61 have to be kept fixed . according to the flow equation for @xmath139 , a non - negative entropy is guaranteed by @xmath211 . the demonstration of this inequality is complicated by the temperature dependence , which is implicit to the screened @xmath201 and @xmath61 for @xmath212 . if one ignores this implicit dependence , one easily verifies @xmath213 , since @xmath214 and @xmath215 hold for @xmath189 and @xmath216 separately . the implicit temperature dependencies can be taken into account by showing @xmath217 . this is verified in the leading order , where one has @xmath218 for @xmath219 . if one would use the same expression for @xmath216 , one would obtain a negative entropy flow . however , for @xmath216 we find @xmath220 . therefore the entropy flow is non - negative even at very low temperatures . a numerical integration of the flow equations ( even without asymptotic approximations ) shows , that below some transition line @xmath221 in the @xmath222-plane the renormalization flow indeed does converge for @xmath223 to fixed points with @xmath224 , @xmath225 , and @xmath226 . this is true not only in the limit of large @xmath22 ( small fugacity ) , but also for arbitrarily small @xmath22 . only the extent of the ordered phase shrinks with increasing @xmath22 . this implies in particular , that the original xy - model described by @xmath23 has an ordered phase . above @xmath227 all quantities diverge under renormalization and the present renormalization group leaves its range of validity , which is limited to small fugacities already due to the independent dipole approximation . the properties of the ordered phase are evident in terms of the renormalized parameters @xmath228 and @xmath229 . due to screening they are related to the bare parameters by @xmath230 and @xmath231 , where the inequalities become equalities for vanishing fugacity within the ordered phase . in the ordered phase , the irrelevance of vortices requires that the fugacities vanish on large scales . therefore the flow equation ( [ flow.sum]a ) of fugacity must have a negative eigenvalue . this is true at temperatures and disorder strength below @xmath221 , which is located at @xmath232 in terms of the renormalized parameters . it is depicted in fig . [ fig_pdg0 ] as function of the renormalized parameters and of the unrenormalized parameters for @xmath233 . the most striking feature of the phase boundary is the absence of reentrance . reentrance was predicted by earlier approaches.@xcite absence of reentrance without a loss of the ordered phase was found only recently in improved approaches.@xcite as a function of the renormalized parameters , the phase boundary @xmath234 has a horizontal part at large @xmath228 , i.e. at low temperatures . in terms of bare parameters , the line @xmath221 can be parametrized by a function @xmath235 or @xmath236 . due to the renormalization of disorder strength , @xmath237 and due to the reduction of the coupling @xmath238 the end points are found at @xmath239 and @xmath240 and @xmath241 has no longer a horizontal part for finite fugacity . concerning the asymptotic behavior of the vortex correlation functions , one can distinguish three regions within the ordered phase . they are separated by a line @xmath242 , fixed by @xmath243 ( equivalent to @xmath244 ) , and a line @xmath245 , fixed by @xmath246 ( equivalent to @xmath247 ) , see fig . [ fig_pdg0 ] . in the @xmath248-diagram these lines emerge from the origin . they are straight lines in the limit of zero fugacity and bent upwards for finite fugacity . the correlations are retrieved from the fugacities , eq . ( [ y.asy ] ) , by omitting rescaling and inserting @xmath249 . thereby we find ( @xmath250 ) [ cs ] @xmath251 for @xmath252 and @xmath253 for @xmath254 . the limiting values for the fraction of polarizable and frozen dipoles are related to the correlations by [ pqf.inf ] @xmath255 naturally the question arises , whether these regimes are separated by a true thermodynamic transition at @xmath242 and @xmath245 . as we have observed , that asymptotic form of the flow equations and of the correlations changes its dependence on @xmath61 and @xmath24 non - analytically at @xmath179 and @xmath180 . however , a phase transition in the strict sense requires a non - analytic dependence of the integrated free energy on the unrenormalized parameters @xmath61 and @xmath24 . we were not able to prove such a singularity at @xmath242 and @xmath245 , which we therefore consider as crossover lines only . however , a true transition is expected at @xmath221 just as in the pure case . region ii can be distinguished _ qualitatively _ from the other regions . one can consider @xmath256 as order parameter for the low - temperature regime ii . it measures the relative strength of the disconnected vortex density correlation to the usual correlation on large scales . in the definition of this order parameter the normalization of the disconnected correlation by the usual correlation is necessary , since true long - range order is absent . the glassy order reflected by the edwards - anderson like correlation can therefore be only quasi long - ranged , too . nevertheless one may compare @xmath256 to the edwards - anderson order parameter in spin - glasses and we interpret it as order parameter for the _ glassiness _ of in the vortex system . the regions ia and ib differ only by the analytic dependence of the connected correlation on @xmath228 and @xmath229 . the correlation under consideration do not provide an order parameter comparabe to @xmath166 for the line @xmath245 . the universality class of the pure xy - model is defined by the behavior of the critical exponents @xmath257 and @xmath258 at the transition and the exponential divergence of the correlation length when the transition is approached from above.@xcite in the presence of disorder , the spin correlation decays with exponent@xcite @xmath259 along the line @xmath221 the exponent takes the value @xmath260 for @xmath261,@xcite which decreases with temperature until it reaches the value @xmath262 for @xmath34.@xcite in the disordered phase , where correlations decay exponentially , @xmath263 formally . thus the value of @xmath264 at the transition becomes non - universal , i.e. it is disorder dependent . only for @xmath265 or @xmath34 the jump of @xmath264 is universal ( from 1/4 or 1/16 to @xmath266 respectively ) in the sense that it does not depend on the bare coupling @xmath201 or bare fugacity @xmath163 . for @xmath188 or @xmath267 the jump lies inbetween and will depend on microscopic details . in the present work we have not included a magnetic field . therefore we can not determine @xmath268 in the presence of disorder . we now attempt to analyze how the correlation length @xmath269 diverges when the transition line is approached from above . even though the flow equations become invalid for large fugacity , kosterlitz@xcite has argued that @xmath269 is given by the length scale where the fugacity becomes of order unity . in the pure case , where the flow equations reduce to @xmath270 and @xmath271 he found @xmath272 with a non - universal constant @xmath273 . in the zero - temperature limit the flow equations read @xmath274 and @xmath275 . by a replacement @xmath276 they can be mapped onto the flow equations of the pure case . therefore we conclude , that the zero temperature transition has @xmath277 with some other non - universal @xmath273 . for finite temperature _ and _ finite disorder , our flow equations are too complicated for being integrated up analytically . however , we naturally expect that the functional form of the divergence is preserved and that only the argument of the square - root has to be replaced by the distance to the transition line , say @xmath278^{1/2}$ ] . the present theory is based on a selfconsistent linear screening approach , which is solved in a differential form . this differential form has been presented in the renormalization group language . in fact , the very same differential equations can be directly derived by a real space renormalization group , generalizing the scheme of kosterlitz@xcite to the replicated system . in this scheme the flow equations obtain contributions from integrating out _ all _ classes of replica dipoles with the smallest ( cut - off ) diameter . the controversy among earlier theoretical treatments of the disordered xy - model requires to put the present work into perspective with these earlier approaches . when cardy and ostlund ( co ) considered the more complicated problem of the xy - model in the presence of random fields,@xcite they used a replica - symmetric approach to derive renormalization group flow equations , which generated also random bond disorder . rubinstein , shraiman , and nelson@xcite ( rsn ) focused on the random bond problem , which they analyzed without using the replica formalism . however , the resulting flow equations were a subset of those of co. their starting point was essentially eq . ( [ partial.screen.k ] ) , which they evaluated only to the leading order in the fugacity of vortices in the unreplicated system . consequently , they neglected the disconnected correlation and could not find a flow of @xmath61 . their flow equations for @xmath24 and @xmath157 coincide with our asymptotic flow equations for @xmath219 . however , they used them also for @xmath178 , which gave rise to reentrance and a negative flow of entropy . the present approach is in principle equivalent to co and rsn . the essential difference lies in the calculation of the correlations , or in the contributions which are taken into account when the dipoles of smallest distance are integrated out . co and rsn thereby made too crude approximations , which treat the disorder effectively as annealed . this yields correct results only for high temperatures @xmath279 but fails for @xmath178 . technically , they considered only contributions of replica - vortices of class @xmath120 . we pointed out , that already for replica numbers @xmath280 this restriction is insufficient at low temperatures . therefore we included _ all classes _ of replica - vortices . the additional vortex classes turned out to be important for @xmath178 and remove reentrance and the negative flow of entropy . from the disappearence of reentrance after the inclusion of additional vortex classes we have to conclude , that these _ additional _ fluctuations in the replicated system ( @xmath128 ) _ reduce _ the effect of thermal fluctuations in the original unreplicated system ( @xmath92 ) . this strange feature reminds of the fact , that in variational replica approaches one has to maximize and not to minimize the free energy for @xmath92 in order to find the physical state.@xcite when korshunov@xcite considered the problem in the replica representation , he attempted to identify the transition in the limit of zero fugacity with the dissociation of the least stable class of dipoles . the @xmath7-complexes in his notation are identical to dipoles formed by replica vortices of class @xmath281 , where one has to identify @xmath282 . they indeed drive the transition for @xmath283 . from my point of view , there are problems in his way of taking the limit @xmath87 , which cause the conclusion , that the ordered phase should not exist at all . one problem is , that it does not make sense to consider replica vortices which occupies @xmath284 replicas even for a number of replicas @xmath285 , since @xmath286 should be preserved during the analytic continuation . in addition , the consideration of such restricted classes of dipoles is not sufficient for an analytic continuation @xmath87 . even then , if one takes into account all classes of replica - vortices , it is the _ most stable _ and not the least stable class , which drives the transition in the limit @xmath87 . this is again a strange feature of replica theory . the arguments for these statements are detailed in appendix [ app.extreme ] . in appendix [ app.sin.vor ] we further show , that @xmath221 can be derived already from considering a _ single _ vortex in a finite system . after replication its free energy can be calculated by taking into account certain types of fluctuations , which are again parametrized by some @xmath287 in the range @xmath288 . the transition @xmath221 follows from a _ maximization _ of the free energy with respect to this parameter . for several disordered systems the occurence of reentrance and negative contributions to the flow of entropy have been shown to be artefacts of a replica - symmetric theoretical approach , which could be removed by breaking replica symmetry.@xcite for the present model these artefacts have emerged in the early replica - symmetric approaches , indeed . here we have shown that these problems can be cured by including the relevant fluctuations within a symmetric theory . hence a breaking of replica symmetry is not necessary . the recent approaches of nattermann , scheidl , korshunov , and li@xcite and also of cha and fertig@xcite found the existence of the ordered phase and the absence of reentrance without using the replica formalism . their results agree with the present theory at least in the limit of zero fugacity , for finite fugacity there are quantitative deviations . in a very recent real - space renormalization group approach in the unreplicated system , tang@xcite has derived flow equations identical to eqs . ( [ flow.sum ] ) and achieves even quantitative agreement with the present results . a qualitative progress of the present theory is the inclusion of a renormalization of @xmath61 . this renormalization is driven by the disconnected vortex density correlation , which in earlier approaches was neglected from the very beginning or found to be zero due to later approximations . although @xmath61 is increased under renormalization , this effect does not destroy the ordered phase . the flow of @xmath61 is essential for the universality type of the zero - temperature transition , which deviates from the kosterlitz - thouless transition in the pure system only by the value of the critical exponents but not in the exponential divergence of the correlation length . due to the renormlaization of @xmath61 one can exclude a horizontal part of @xmath221 at low temperatures even for finite fugacity . such a horizontal part was shown to exist by ozeki and nishimori@xcite for a wide class of disordered systems with a certain gauge symmetry . but the xy - model under consideration does not possess this symmetry . to sum up , is has been shown that the two - dimensional xy - model with random phase - shifts on bonds has a phase with quasi long - range order . it has exhibits a zero temperature transition at a critical strength of disorder . near the transition the correlation length diverges exponentially as in the pure case , but the jump of the spin correlation exponent is of different magnitude . within the ordered phase there is a glassy low - temperature region . in the absence of true long - range order , we identified the ratio of the edwards - anderson correlation to the usual vortex - density correlation on large length scales as order parameter for glassiness . we used a replica - symmetric renormalization group approach . the reentrance and negative entropy flow of earlier approaches of that kind could be cured by taking into account additional fluctuations . no need for a breaking of replica symmetry was found . several generalizations of the present work are of interest . in a forthcoming publication,@xcite the present theory will be extended to include ( random ) magnetic fields on the ordered phase of the xy - model . the properties of the system in the disordered phase are beyond the range of our perturbative renormalization approach . in particular the question about possible glassy properties in this phase deserves further research . in the limit @xmath289 , the present model becomes the so - called gauge glass model , which also exhibits a glass transition.@xcite the author gratefully acknowledges stimulating discussions with s.e . korshunov , t. nattermann , and l .- h . tang . this work was supported financially by the deutsche forschungsgemeinschaft sfb 341 . the author also enjoyed hospitality of i.s.i . foundation funded by eu contract erbchrx - ct920020 . since eqs . ( [ partial.screen ] ) are the basis of the present analysis , it seems worthwhile to present some intermediate steps of their derivation . we start by including a test vorticity @xmath166 into the effective vortex hamiltonian ( [ hv.k ] ) , @xmath290 the free energy of the system with test vortices reads @xmath291 = -\ln \sum_{\ { n\ } } e^{-{\cal h}_{\rm v}}.\ ] ] the screened potential and interaction are defined by [ app.uv.scr ] @xmath292 } { \delta q_{-\bf k } } \right|_{q=0 } , \\ u_{\bf k}^{\rm scr } \delta({\bf k}+{\bf k}')&:= & \left . \overline{\frac { \delta^2 { \cal f } [ \ { q \ } ] } { \delta q_{-\bf k}\delta q_{-\bf k ' } } } \right|_{q=0 } , \end{aligned}\ ] ] and result in eqs . ( [ screen.v ] ) . the screened parameters are identified via [ app.ks.scr ] @xmath293 fourier transformation then leads to eqs . ( [ total.screen ] ) . on the way one needs the probability distribution of quenched vortices , @xmath294 \propto \exp \left\{- \frac { 2 \pi^2}{\sigma } \int_{\bf k } \frac{1}{k^2 } |q_{\bf k}|^2 \right\},\ ] ] and one has to integrate partially @xmath295 q_{\bf k } \langle n_{-\bf k } \rangle \nonumber \\ & = & - \frac { \sigma k^2}{4 \pi^2 } \int { \cal d}\{q\ } \frac{\delta p[\ { q\}]}{\delta q_{-\bf k } } \langle n_{-\bf k } \rangle \nonumber \\ & = & \frac { \sigma k^2}{4 \pi^2 } \int { \cal d}\{q\ } p[\ { q\ } ] \frac{\delta \langle n_{-\bf k } \rangle}{\delta q_{-\bf k } } \nonumber \\ & = & \sigma k c_{\rm con}({\bf k } ) . \end{aligned}\ ] ] the resulting screening formulae are also valid in the case , where the interaction and disorder correlation on small scales have additional contributions due to @xmath26 and @xmath88 , which flow under renormalization independently of @xmath24 and @xmath61 . it is important to realize , that @xmath46 could alternatively be defined by @xmath296 for small @xmath297 , which also might be plausible at first sight but gives different results as soon as @xmath298 . then , however , it would be necessary to distinguish the screened interactions among thermal vortices @xmath7 from the screened interactions between thermal vortices @xmath7 and quenched vortices @xmath40 . this complication is avoided by the definition actually used , which is based on the correlation of @xmath299 . only in this way screening effects can be captured by a renormalization of the parameters @xmath24 and @xmath61 , which were present already in the unrenormalized model . in this appendix we address the question , whether the correct vortex flow equations can be found , if one considers only the contributions of a _ single _ class of replica vortices to the partition sum and the correlations . for @xmath128 one naturally expects , that fluctuations of the vortices with the lowest energy cost give the most important contribution to renormalization . since the interaction between different replicas favors vortices of the same sign , optimal replica vortices will be of class @xmath300 or @xmath301 , which we call simply @xmath302 vortices . according to eq . ( [ def.e_m ] ) , their core energy is @xmath303 since @xmath26 and @xmath88 are positive , the dipoles with minimum energy are always given by @xmath304 for @xmath305 and by @xmath306 for @xmath307 . they determine the phase diagram of fig . [ fig_pdgn ] . the _ most stable _ pairs are given by @xmath308 ( or , for @xmath97 , by one of the two nearest integers to @xmath309 ) . now we consider screening effects by @xmath287-vortices only . we would like to point out , that the restriction to this class of vortices does not break replica symmetry , since we include all @xmath310 possibilities to place such vortices in @xmath65 replicas . then the flow equations read [ total.screen.m ] @xmath311 with fugacity @xmath312 and effective fugacities @xmath313 , @xmath314 , and @xmath315 . they can be related by the identities @xmath316 , @xmath317 , and @xmath318 . thereby we find the two - component parameters and flow of @xmath163 [ flow.m ] @xmath319 where @xmath320 . to simplify the discussion of the replica - limit @xmath92 , we use the asymptotic approximation @xmath169 and @xmath321 . one should consider @xmath287 in the range @xmath322 , which follows from an analytic continuation of the condition @xmath323 . one easily convinces oneself in the case _ without _ disorder ( @xmath324 ) , that one obtains unphysical results , if one looks for @xmath287 which _ minimizes _ @xmath325 : this would be @xmath326 and , the systen would be unstable to the creation and dissociation of dipoles ( @xmath327 ) . however , if we look for the _ maximum _ , which lies in the pure case always at @xmath304 , we find a positive core energy @xmath328 and the correct phase boundary at @xmath329 . therefore let us try to look for the most stable type of dipoles also in the presence of disorder . it is given by @xmath330 . recalling the constraint @xmath331 in the limit of @xmath92 , we have to distinguish a high- and low - temperature regime with [ m^ * ] @xmath332 if the core energy is large and renormalization is negligible , the criterion @xmath333 for the dissociation of the most stable dipoles then yields the correct transition line @xmath221 as in eq . ( [ sc ] ) . surprisingly , even the two - component parameters @xmath334 , @xmath335 , and @xmath336 for @xmath178 according to eq . ( [ flow.m ] ) are in perfect agreement with ( [ def.fpq ] ) . so , was it essentially unnecessary to consider _ all _ classes of vortices ? even though the flow equations are correct , we must know the initial values to integrate them up , in particular we must know the unrenormalized value of @xmath163 . the problem is , that for @xmath134 we have @xmath337 , which _ diverges _ @xmath338 in the low temperature regime with @xmath339 . this pretends erroneously the relevance of vortices . in particular , there is an inconsistency with the replica trick , since according to eq . ( [ total.screen.m]a ) the free energy per replica diverges like @xmath340 . these problems are removed by including the contributions of all classes of vortices . it is instructive to show , that the phase boundary @xmath221 in the limit of zero fugacity can be derived by considering a single vortex . similar to the reasoning of kosterlitz and thouless,@xcite we examine the possibility of finding single free vortices in a system of finite linear dimension @xmath341 by determining its free energy . the energy of a vortex is composed of its self - energy @xmath342 and of a potential energy @xmath205 due to disorder . the disorder - averaged free energy reads @xmath343 we use @xmath344 and find the leading power of @xmath345 in @xmath341 by considering all possible partitionings @xmath346 of replicas into groups:@xcite @xmath347 } .\ ] ] here @xmath348 denotes the number of groups , and @xmath349 or @xmath350 if replicas @xmath351 and @xmath273 are in the same or in different groups . if we consider only the possibility , where @xmath352 groups of @xmath287 replicas are formed , we find @xmath353 } .\ ] ] the restriction to these partitionings looks at first sight like a one - step replica symmetry breaking , but it is not . if one partitioning is included , also all other partitionings , wich are related by a permutation of replicas , are also included . the phase boundary @xmath221 is there , where the free energy of the single vortex changes sign . ( [ f.ht ] ) gives the high - temperature part of @xmath221 , whereas eq . ( [ f.lt ] ) gives the low - temperature part of @xmath221 in agreement with eq . ( [ sc ] ) . note , that the boundary @xmath361 separating the validity of eqs . ( [ f.ht ] ) and ( [ f.lt ] ) does _ not _ coincide with @xmath242 of @xmath245 . this , however , can not be expected since within the ordered phase fluctuations are dominated by vortex dipoles and not by single vortices . the appearence of the separatrix @xmath361 is specific for the physics of single vortices . g. parisi , j. phys . a * 13 * , l115 ( 1980 ) ; j. phys . a * 13 * , 1101 ( 1980 ) ; j. phys . a * 13 * , 1887 ( 1980 ) ; phys . lett * 50 * , 1946 ( 1983 ) ; j .- bouchaud and a. georges , phys . * 68 * , 3908 ( 1992 ) ; j .- naef and a. canning , j. phys . i france * 2 * , 247 ( 1992 ) ; h. steffan and r. khn , z. phys . b * 95 * , 249 ( 1994 ) . for screening in ordinary coulomb gases nelson , in _ fundamental problems in statistical mechanics v _ , ed . cohen ( north - holland , amsterdam , 1979 ) ; p. minnhagen and g.g . warren , phys . b. * 24 * , 2526 ( 1981 ) ; j.m . thijssen and h.j.f . knops , phys . b * 38 * , 9080 ( 1988 ) . see , e.g , g. parisi , phys . lett * 43 * , 1754 ( 1979 ) ; or the random energy model as the simplest case of glassy system : b. derrida , phys . lett * 45 * , 79 ( 1980 ) ; b. derrida , phys . b * 24 * , 2613 ( 1981 ) ; d.j . gross and m. mzard , nucl . phys . * b240 * , 431 ( 1984 ) . d.a . huse and s. seung , phys . rev . b * 42 * , 1059 ( 1990 ) ; m.a . moore and s. murphy , phys . b * 42 * , 2587 ( 1990 ) ; m.p.a . fisher , t.a . tokuyasu , and a.p . young , phys . lett . * 66 * , 2931 ( 1991 ) ; m. cieplak , j.r . banavar and a. khurana , j. phys . a * 24 * , l145 ( 1991 ) ; m.j.p . gingras , phys . b * 45 * , 7547 ( 1992 ) .

we study via monte carlo simulation a generalisation of the so - called vertex interacting self - avoiding walk ( visaw ) model on the square lattice . the configurations are actually not self - avoiding walks but rather restricted self - avoiding trails ( bond avoiding paths ) which may visit a site of the lattice twice provided the path does not cross itself : to distinguish this subset of trails we shall call these configurations _ grooves_. three distinct interactions are added to the configurations : firstly the visaw interaction , which is associated with doubly visited sites , secondly a nearest neighbour interaction in the same fashion as the canonical interacting self - avoiding walk ( isaw ) and thirdly , a stiffness energy to enhance or decrease the probability of bends in the configuration . in addition to the normal high temperature phase we find three low temperature phases : ( i ) the usual amorphous liquid drop - like globular " phase , ( ii ) an anisotropic ``-sheet '' phase with dominant configurations consisting of aligned long straight segments , which has been found in semi - flexible nearest neighbour isaw models , and ( iii ) a maximally dense phase , where the all sites of the path are associated with doubly visited sites ( except those of the boundary of the configuration ) , previously observed in interacting self - avoiding trails . we construct a phase diagram using the fluctuations of the energy parameters and three order parameters . the-sheet and maximally dense phases do not seem to meet in the phase space and are always separated by either the extended or globular phases . we focus attention on the transition between the extended and maximally dense phases , as that is the transition in the original visaw model . we find that for the path lengths considered there is a range of parameters where the transition is first order and it is otherwise continuous .

1510.06686

there are many lattice models of a single polymer in solution that take account of different physical scenarios and capture different aspects of such a system , for example by modelling the inclusion of hydrogen bonding or stiffness . the first ingredient in these models is the type of lattice path that is considered . the simplest example is the unrestricted random walk , which is a lattice path that may visit sites and bonds of the lattice multiple times . on the other hand the classical model of polymers is based on the _ self - avoiding walk _ , which is a lattice path that can not visit either bonds or sites of the lattice more than once . there are however several configuration systems used that lie between these two extremes . one commonly used for modelling polymers is the self - avoiding trail , which is a lattice path that is bond avoiding but may visit sites multiple times . importantly , this path may both touch at a site , or may cross . consider the subset , @xmath1 , of @xmath2-step trails @xmath3 on the square lattice with the added restriction that no crossings are allowed . this is an important type of configuration since they appear in the high temperature expansion of the @xmath4 model on the square lattice ( see @xcite for a recent example ) . let us call these configurations _ grooves_. the second ingredient in a lattice model of polymers in solution is the set of interactions that are associated with various combinatorial features of the configurations . given a groove @xmath5 , we associate the following set of boltzmann weights : a weight @xmath6 for _ site interaction _ , i.e. for a site that the groove visits more than once , such as occuring in the interacting self - avoiding trail model ( isat ) , a weight @xmath7 for every straight segment , which moderates stiffness in the polymer , and finally a weight @xmath8 for each _ nearest - neighbour interaction _ , i.e. for a pair of visited sites that are nearest - neighbours on the lattice but non sequential in the groove . this nearest - neighbour interaction occurs in the canonical interacting self - avoiding walk ( isaw ) model of polymer collapse . figure [ fig : configuration1 ] shows an example of a configuration belonging to @xmath9 and its associated weights . steps with one ( @xmath10 ) collision , five ( @xmath11 ) straight segments , and four ( @xmath12 ) nearest neighbours . there are @xmath13 singly visited sites and @xmath14 doubly visited sites . the total weight of this configuration is @xmath15 . ] there has been long term interest in special cases of this model over and above the canonical isaw model , which has a weak second order phase transition between a high temperature expanded phase and a low - temperature phase . one can distinguish these phases for example by the different scaling of the size of the polymer as given by its radius of gyration . the associated exponent takes the value @xmath16 in the high - temperature phase and @xmath17 in the low - temperature phase . in the low temperature phase the polymer is disordered and dense , but not maximally dense : it has been described as a liquid - like disordered drop . the phase transition known as the @xmath18-point is conjectured in two dimensions to have a cusp singularity in a convergent specific heat with the associate finite length exponent @xmath19 . the size exponent at the transition is @xmath20 . the isaw model is realised by letting @xmath21 and @xmath22 . on the other hand setting @xmath23 and @xmath22 gives the so - called vertex interacting self - avoiding walk " ( visaw ) , which is a misnomer since the configurations are actually grooves ( as described above ) and not self - avoiding walks . this model first arose as a specialisation of an integrable lattice model introduced by blte and nienhuis in 1989 @xcite related to the izergin - korepin vertex model . in fact , that model included more generally a stiffness parameter @xmath7 . the model was studied numerically @xcite for various values of @xmath7 including @xmath22 . it has also been studied as part of a generalisation of the interacting self - avoiding trail model ( asymmetric isat ) that distinguishes collisions from crossings , using transfer matrix calculations @xcite and monte carlo simulations @xcite . the monte carlo work @xcite concluded that the transition in the visaw @xmath24 would apparently seem to be a strong second order transition . the exponent describing the finite length singularity in the specific heat has been estimated @xcite as @xmath25 , which is in accord with the divergence seen in the isat model of @xmath26 @xcite , estimated from much longer configurations . the size exponent of visaw has not been measured using monte carlo , though the isat model has @xmath17 with a scaling form containing multiplicative logarithmic corrections : this clearly differs from the @xmath18-point isaw value of @xmath27 . in work on the asymmetric isat @xcite the low temperature phase of visaw ( with @xmath22 ) was considered ; it was found to have a similar structure to the low temperature phase in the isat model . this low temperature phase is not the same as the globular , liquid - like phase in the low temperature region of isaw : for visaw and isat the low temperature phase is maximally dense , i.e. every site of the path , apart from those on the outside of a dense ball , is associated with a doubly visited site of the lattice . in contrast , setting @xmath28 excludes the possibility of straight segments and by also setting @xmath29 the model is known as interacting self - avoiding trails on the l - lattice ( _ lsat _ ) . note that by setting @xmath30 the trails do not cross themselves and so are actually _ grooves_. the model can also be mapped onto isaw on the manhattan lattice @xcite . a kinetic growth algorithm that produces long l - lattice trails maps to configurations with boltzmann weights @xmath31 : an analysis of simulations of very long configurations @xcite has shown that the transition is @xmath18-like with a weak second order transition with an exponent conjectured to be the same as the @xmath18-point value of @xmath19 . the scaling of the size of the polymer given by the radius of gyration or similar is described by the exponent @xmath20 . interestingly , the low temperature phase of the lsat model has not been previously considered . of course it has been implicitly assumed that it would be like the isaw model , i.e. disordered and dense but not maximally dense . the parameter space of the semi - flexible visaw model ( @xmath29 ) also includes the special multi - critical point of the @xmath4 model found in @xcite , known as the blte - nienhuis point or bn - point . the location of this point is given by @xmath32 with @xmath33 , and @xmath34 . recently there have been various investigations of the bn - point itself @xcite . in the monte carlo work @xcite the size exponent was estimated as close to @xmath20 in agreement with earlier work @xcite . the specific heat exponent was not estimated and since the study looked at just that point the low temperature phase was not considered . very recently , a theoretical framework for the bn - point has been elucidated where it has been proposed that this point is related to a particularly unusual conformal field theory with continuous exponents @xcite . so while the semi - flexible visaw model has been studied previously by monte carlo at several values of @xmath7 , there are clearly some outstanding gaps in our knowledge . to gain a wider perspective , the generalised model we study here allows us to interpolate between the semiflexible visaw model and the canonical interacting self - avoiding walk ( isaw ) model directly . the set of @xmath2-step self - avoiding walks @xmath35 is a subset of grooves , @xmath36 . hence , if we set @xmath37 we suppress all multiple visits to sites and recover self - avoiding walks as configurations . we are now left with two boltzmann weights @xmath8 and @xmath7 which gives us the semi - flexible interacting self - avoiding walk model or semi - flexible isaw model . this model was studied on the square lattice in @xcite . to present a phase diagram that can be readily compared to the data we describe below , we present the schematic phase diagram for semi - flexible isaw . this has been obtained from our own data and is in accord with the phase diagram elucidated in @xcite previously . in figure [ fig : sfisawpd ] we see a density plot of a generalised specific heat and a schematic phase diagram as per @xcite . ) and a schematic phase diagram inferred from this that is in accord with previous work . darker shades indicate larger values of the fluctuations . later we shall see that appropriate order parameters / indicator variables reinforce these conclusions . there are three phases : extended , collapsed ( globular ) and @xmath0-sheet - like . solid lines represent the approximate location of a second - order phase transitions . the dashed line indicates first order behaviour.,scaledwidth=80.0% ] the data in figure [ fig : sfisawpd ] is a slice obtained from the dataset 3p by fixing @xmath37 ( see table [ appendix : table ] ) . there are three phases : two that occur in the canonical isaw model , that is the extended and globular phases , and a third which is an anisotropic collapsed phase that can be described as @xmath0-sheets . these two - dimensional @xmath0-sheets are actually parallel lines of polymer in the @xmath38 or @xmath39 directions . the extended and globular collapsed phases are separated by a line of weak second - order transitions . the @xmath18-point lies on this line and the conjecture is that the entire line belongs to the @xmath18-universality class . on the other hand the extended phase and @xmath0-sheet - like phase are separated by a line of first order transition points . according to @xcite , the @xmath0-sheet - like phase and the collapsed phase are separated by a line of second - order phase transitions associated with a positive @xmath40-exponent ( around @xmath41 ) and therefore a diverging specific - heat . as can be seen from this summary of the models that are contained within our model there are very different transitions and different collapsed low temperature phases . here we investigate our generalised model filling in the gaps in our knowledge and providing a coherent picture of how the transitions and where the different low temperature phases exist . given an @xmath2-step groove @xmath42 with @xmath43 collisions , @xmath44 straight segments and @xmath45 nearest neighbours , we assign to this groove a boltzmann weight @xmath46 . summing over all @xmath2-step grooves , we arrive at the partition function @xmath47 if we denote the number of singly visited sites of the path as @xmath48 and the number of doubly visited sites by @xmath49 we can infer their values from @xmath2 and @xmath43 by observing that @xmath50 . we find that @xmath51 and @xmath52 . the average of any quantity @xmath53 over the ensemble set of grooves @xmath9 is given generically by @xmath54 in this paper we are interested in the following quantities . we calculate a measure of the size of the polymer , @xmath55 , by using the mean end - to - end distance @xmath56 defined as follows : letting any @xmath2-step path @xmath57 on a lattice by a sequence @xmath58 of vector positions of the vertices of that path the average - square end - to - end distance is given by @xmath59 in the above formulae we use @xmath60 . we are actually more interested in analysing the polymer density which is given by @xmath61 as noted in the introduction one of the phases we have is anistropic so to assist in identifying such a phase we define an anisotropy by @xmath62 the number of singly visited sites per unit length is given by @xmath63 the size of the polymer defined above is expected to scale as @xmath64 where the amplitude @xmath65 is non - universal and temperature dependent , while @xmath66 is expected to be universal , depending only on the phase . for high temperature it is always expected that @xmath16 which is the value for non - interacting self - avoiding walks and self - avoiding trails so @xmath67 which goes to zero in the extended phase . for each of the low temperature phases we expect @xmath17 so that @xmath68 will attain a non - zero positive value in the thermodynamic limit . the scaling of the peak value of the specific heat @xmath69 or indeed any similar measure such as the largest eigenvalue of the fluctuations matrix at the collapse transition is expected to behave as @xmath70 assuming the transition is second order . the need to consider the background constant @xmath71 depends on whether @xmath40 is positive or negative . for any isotropic phase one expects that @xmath72 so that @xmath73 will converge to a constant less than one . if it is fully isotropic @xmath74 and , moreover , @xmath75 so that @xmath73 converges to zero in the thermodynamic limit . on the other hand for a fully anisotropic phase one expects that @xmath76 so that @xmath77 and @xmath73 will converge to one in such a phase . for a maximally dense phase one may expect that @xmath78 we studied our generalised model for various slices of the three - dimensional parameter space @xmath79 using monte carlo simulations . our simulations were based on the flatperm algorithm @xcite which is a flat histogram version of the pruned and enriched rosenbluth method ( perm ) developed in @xcite . the perm algorithm generates a polymer configuration kinetically , which is to say that each growth step is selected at random from all possible growth steps . during this process the algorithm keeps track of a weight factor to correct the sample bias . at each growth step , configurations with very high weight relative to other configurations of the same size are duplicated ( or _ enriched _ ) while configurations with low weight or that can not be grown any further ( i.e. because they are trapped " or because they have reached a limit length ) are discarded ( or _ pruned _ ) . a single iteration is then concluded when all configurations have been pruned . the total number of samples generated during each iteration depends on the specifics of the problem at hand and on the details of the enriching / pruning strategy . this simple mechanism produces valid configurations with many steps very efficiently but introduces correlation between samples which are grown from a same smaller configuration . we keep account of these correlations in two ways : first , we count `` effective '' samples by the fraction of independent steps , and second , we run around 10 completely independent runs for each simulation to obtain an a posteriori measure of the statistical error . flatperm extends this method by cleverly choosing the enrichment and pruning steps to generate for each polymer size @xmath2 a quasi - flat historgram in some choosen micro - canonical quantities @xmath80 and producing an estimate @xmath81 of the total weight of the walks of length @xmath2 at fixed values of @xmath82 . from the total weight one can access physical quantities over a broad range of temperatures through a simple weighted average , e.g. @xmath83 the quantities @xmath84 may be any subset of the physical parameters of the model . for the results presented in this paper , we have run different simulations with different choices for the micro - canonical quantites @xmath82 depending on region of the parameter space of interest . we point out that the number of micro - canonical quantities has a dramatic impact on the memory requirement of algorithm which might need to keep in memory many multi - dimensional histograms like @xmath81 and @xmath85 . the complete list of our simulation runs is provided in table [ appendix : table ] . .summary of flatperm simulations . the simulations range from uniform sampling across all four parameters @xmath86 with the length of configurations restricted by a maximal length of @xmath87 ( data set 3p ) to uniform sampling across only two parameters , with boltzmann weights for the other parameters at fixed values . for example in the data set nn-2 , two boltzmann weights are fixed at @xmath88 and @xmath22 , and uniform sampling is performed across @xmath2 and @xmath45 . this enables simulations for longer lengths , allowing to sample up to @xmath89 . [ appendix : table ] [ cols= " < , < , < , < , < , < , < " , ] from a simulation with micro - canonical quantities , it is possible to obtain a mixed distribution that is canonical in one of the quantities by fixing its associated weight . for example : from @xmath90 we can obtain `` a slice '' at fixed value of @xmath91 by computing @xmath92 the normalisation of the distribution is irrelevant . to obtain a landscape of possible phase transitions , we compute the largest eigenvalue of the matrix of second derivatives of the free energy with respect to @xmath93 effectively measuring the strength of the fluctuations and covariance in @xmath82 . e.g. given the flatperm weights @xmath81 , we first compute the partition function : @xmath94 and then we find the maximum eigenvalue of the hessian matrix @xmath95 where @xmath96 is the free energy . more explicitedly , in the diagrams where @xmath7 is fixed the remaining parameters are @xmath97 and we plot the eigenvalue of the matrix defined by @xmath98 \frac{\partial^2 f}{\partial \tau\ , \partial \omega } & \frac{\partial^2 f}{\partial \omega^2 } \end{pmatrix}.\ ] ] in the diagrams where @xmath8 is fixed the remaining parameters are @xmath99 and we plot the eigenvalue of the matrix defined by @xmath100 \frac{\partial^2 f}{\partial \tau\ , \partial p } & \frac{\partial^2 f}{\partial p^2 } \end{pmatrix}.\ ] ] finally , when @xmath6 is fixed , the remaining parameters are @xmath101 and we plot the eigenvalue of the matrix defined by @xmath102 \frac{\partial^2 f}{\partial p\ , \partial \omega } & \frac{\partial^2 f}{\partial \omega^2 } \end{pmatrix}.\ ] ] we started our new investigation by examining the semi - flexible visaw model for all values of stiffness @xmath7 . this would allow us to see how the bn point belongs to a critical line . to do so , we simulated a visaw model with additional stiffness . ) and a schematic phase diagram inferred from this and the order parameters / indicator variables of polymer density and singly visited sites . darker shades indicate larger values of the fluctuations . there seems to be only two phases : extended and maximally dense . the solid line ( red ) represents the estimated location of second order phase transitions . data is obtained from the dataset sf.,scaledwidth=70.0% ] a density plot of the largest eigenvalue of fluctuations of a semi - flexible visaw model is shown in figure [ fig : sfvisaw_phase_diagram ] . we found that the bn point seems to lie on a line of second - order phase transitions that extends from the l - lattice trails point to high value of @xmath6 and @xmath7 , connecting the bn point and the visaw critical point . for some high value of @xmath6 the transition seems to turn first order . the l - lattice trails point is known to be in the @xmath18-point universality class , while the critical visaw has a strongly divergent specific - heat , as it was reported in @xcite . therefore along the line the transition starts weak and becomes stronger and stronger until it turns first order . for @xmath30 and @xmath29 and for several values of @xmath6 . for @xmath103 and @xmath104 , this quantity goes to zero as @xmath2 diverges . this indicates the existence of the maximally dense phase where all but surface sites are involved in collisions and so are doubly visited by the polymer . data is obtained from the dataset vi l - lattice.,scaledwidth=70.0% ] as done in @xcite we considered the fraction of sites visited only once as an order parameter . in figure [ fig : visaw_low_temperature_omega_eq_1_p_eq_0 ] we show that number of singly visited sites per unit length in the maximally dense phase goes to zero as length diverges . this indicates the existence of the maximally dense phase where all but surface sites are involved in collisions and so are doubly visited by the polymer . in figure [ fig : visaw_density ] we show a density plot of the fraction of singly visited sites indicating that the maximally dense region extends all the way to the line of second order phase transitions . . the solid line ( red ) shows the approximate location of second order phase transitions . data is obtained from the dataset sf.,scaledwidth=70.0% ] inspecting typical configurations at various parameter values further confirms this scenario . in figure [ fig : visaw_configs ] we show three configurations for @xmath29 and @xmath30 at @xmath105 ( swollen ) , @xmath106 ( critical ) and @xmath103 . ( note that the choice of @xmath30 forces a right - angle turn after every single step , so these are configurations on the l - lattice . ) one clearly sees the typical characteristics of a swollen chain for @xmath105 , and observes a clear difference between this configuration and the configuration at @xmath106 , which is critical . in contrast , the configuration for @xmath103 is close to maximally dense with only a few small internal defects . and @xmath30 at @xmath105 , @xmath106 ( at the location of the transition ) and @xmath103 . we see that the low temperature ( @xmath103 ) configuration appears to be almost maximally dense . data is obtained from the dataset vi l - lattice . ] our model has three boltzmann weights and so three parameters that can be varied . the models of previous focus have been two parameter slices of the space , namely the semi - flexible isaw that is defined by @xmath105 and the semi - flexible visaw that is defined by @xmath29 . because simulations of longer lengths can be made when we fix certain boltzmann weights we have analysed our model in slices . to analyse the phase diagram in a slice we have used three main pieces of information . firstly we have used the appropriate hessian and considered the landscape of the largest eigenvalue of that matrix : this gives us an indication of divergences in appropriate fluctuations , equivalently specific heats . we have looked at scaling of those possible divergences . when we suspect a first order transition inferred from linear or super linear scaling of the divergence we consider the distribution of the appropriate micro - canonical quantity so look for emerging bimodality . we have also considered the order parameters associated with the known phases in the semi - flexible isaw and visaw models to indicate which phases occur in which regions of the parameter space . finally we have looked at typical " ( random ) configurations to re - enforce our conjectures . in the following sections we consider two - dimensional slices of the parameter space , first by fixing @xmath6 , then by fixing @xmath8 , and finally by fixing the stiffness @xmath7 . the first key observation to make is that the order parameters clearly indicate that the three low temperature phases from the semi - flexible isaw and visaw models occur in different parts of the three dimensional phase space and so that there are at least three low temperature phases . before looking at the range of @xmath6 slices we consider @xmath107 , first to see how the phase diagram changes from @xmath105 when we fix at a value of @xmath6 that should induce the maximally dense phase in some part of the phase diagram but also to demonstrate the method we are using to conjecture phase diagrams . in figure [ fig : visaw_phase_diagram_tau_eq_5 ] a plot of the maximum eigenvalue of the matrix of fluctuations is given for the slice @xmath107 . darker shades indicate larger values of the fluctuations . there is clearly a strong peak in these fluctuations for large @xmath7 separating regions of small and large @xmath8 . we then identify this transition to be between an extended phase and a @xmath0-sheet phase given schematically by the dashed phase boundary drawn for @xmath108 : the two phases are identified using the order parameter plots seen in figure [ fig : order_parameters_tau_eq_5 ] . . peaks in this plot can be used to infer the phase diagram . darker shades indicate larger values of the fluctuations . solid lines show the approximate location of second order phase transitions and dashed lines indicate first order behaviour . as indicated by the arrow , the phase transition turns first - order as it approaches the vertical axis . the data for this plot is obtained from the dataset sf - nn.,scaledwidth=70.0% ] we plot three order parameter / indicator variables that can be used to re - enforce the conclusions regarding the phase diagram made from the peaks in the fluctuations . in the top diagram the anisotropy parameter is plotted : it is clearly large for large @xmath7 and @xmath8 this is a good indicator for the @xmath0-sheet phase . in the middle diagram the density is plotted : it is clearly small ( tending to zero ) in the extended phase for small @xmath8 excluding very small @xmath7 . in the bottom diagram the fraction of singly visited sites is plotted : in the maximally dense phase is elected to go to zero . these plots are all obtained from the same dataset sf - nn.,title="fig:",scaledwidth=55.0% ] we plot three order parameter / indicator variables that can be used to re - enforce the conclusions regarding the phase diagram made from the peaks in the fluctuations . in the top diagram the anisotropy parameter is plotted : it is clearly large for large @xmath7 and @xmath8 this is a good indicator for the @xmath0-sheet phase . in the middle diagram the density is plotted : it is clearly small ( tending to zero ) in the extended phase for small @xmath8 excluding very small @xmath7 . in the bottom diagram the fraction of singly visited sites is plotted : in the maximally dense phase is elected to go to zero . these plots are all obtained from the same dataset sf - nn.,title="fig:",scaledwidth=55.0% ] we plot three order parameter / indicator variables that can be used to re - enforce the conclusions regarding the phase diagram made from the peaks in the fluctuations . in the top diagram the anisotropy parameter is plotted : it is clearly large for large @xmath7 and @xmath8 this is a good indicator for the @xmath0-sheet phase . in the middle diagram the density is plotted : it is clearly small ( tending to zero ) in the extended phase for small @xmath8 excluding very small @xmath7 . in the bottom diagram the fraction of singly visited sites is plotted : in the maximally dense phase is elected to go to zero . these plots are all obtained from the same dataset sf - nn.,title="fig:",scaledwidth=55.0% ] we have then verified that the divergence of the fluctuations is at least linear in length and that at least at some point along that line the distribution of nearest neighbour contacts @xmath45 is bimodel . the other lines are all seen to be second order transitions . we find that a maximally dense phase does exist for very small @xmath7 and moderately small @xmath8 in the bottom left corner of the diagram . not all the conjectured phase boundaries are equally well delineated , with the extended - collapsed boundary the least defined : this is not to be unexpected as the transition should be in the @xmath18-universality class which has a convergent specific heat singularity . in general this phase boundary is the most difficult to estimate . at @xmath109 . of particular interest is the @xmath0-sheet type configuration . these configurations come from the dataset nn . ] finally to re - enforce our identification of phases we have considered typical configurations . in figure [ fig : tau - eq-5 ] we show a typical configuration in the part of the phase diagram we label as @xmath0-sheets ( actually for @xmath109 in this instance ) : it clearly shows the anisotropic structure we expect . ranging from @xmath21 ( top left ) to @xmath110 ( bottom ) a density plot of the larger eigenvalue is displayed . information from this has been combined with the three order parameter indicator variables mentioned above to infer a phase diagram schematically drawn in each diagram . we could not determine the phase boundary between the extended and collapsed phases for @xmath111 with much accuracy : it could move to smaller values of @xmath8 especially for small @xmath7 . as in figure [ fig : visaw_phase_diagram_tau_eq_5 ] the phase transition turns first - order as it approaches the vertical axis . these plots come from an appropriate slice of the dataset 3p , except the phase diagram at @xmath112 which comes from dataset sf - nn.,title="fig:",scaledwidth=45.0% ] ranging from @xmath21 ( top left ) to @xmath110 ( bottom ) a density plot of the larger eigenvalue is displayed . information from this has been combined with the three order parameter indicator variables mentioned above to infer a phase diagram schematically drawn in each diagram . we could not determine the phase boundary between the extended and collapsed phases for @xmath111 with much accuracy : it could move to smaller values of @xmath8 especially for small @xmath7 . as in figure [ fig : visaw_phase_diagram_tau_eq_5 ] the phase transition turns first - order as it approaches the vertical axis . these plots come from an appropriate slice of the dataset 3p , except the phase diagram at @xmath112 which comes from dataset sf - nn.,title="fig:",scaledwidth=45.0% ] ranging from @xmath21 ( top left ) to @xmath110 ( bottom ) a density plot of the larger eigenvalue is displayed . information from this has been combined with the three order parameter indicator variables mentioned above to infer a phase diagram schematically drawn in each diagram . we could not determine the phase boundary between the extended and collapsed phases for @xmath111 with much accuracy : it could move to smaller values of @xmath8 especially for small @xmath7 . as in figure [ fig : visaw_phase_diagram_tau_eq_5 ] the phase transition turns first - order as it approaches the vertical axis . these plots come from an appropriate slice of the dataset 3p , except the phase diagram at @xmath112 which comes from dataset sf - nn.,title="fig:",scaledwidth=45.0% ] ranging from @xmath21 ( top left ) to @xmath110 ( bottom ) a density plot of the larger eigenvalue is displayed . information from this has been combined with the three order parameter indicator variables mentioned above to infer a phase diagram schematically drawn in each diagram . we could not determine the phase boundary between the extended and collapsed phases for @xmath111 with much accuracy : it could move to smaller values of @xmath8 especially for small @xmath7 . as in figure [ fig : visaw_phase_diagram_tau_eq_5 ] the phase transition turns first - order as it approaches the vertical axis . these plots come from an appropriate slice of the dataset 3p , except the phase diagram at @xmath112 which comes from dataset sf - nn.,title="fig:",scaledwidth=45.0% ] ranging from @xmath21 ( top left ) to @xmath110 ( bottom ) a density plot of the larger eigenvalue is displayed . information from this has been combined with the three order parameter indicator variables mentioned above to infer a phase diagram schematically drawn in each diagram . we could not determine the phase boundary between the extended and collapsed phases for @xmath111 with much accuracy : it could move to smaller values of @xmath8 especially for small @xmath7 . as in figure [ fig : visaw_phase_diagram_tau_eq_5 ] the phase transition turns first - order as it approaches the vertical axis . these plots come from an appropriate slice of the dataset 3p , except the phase diagram at @xmath112 which comes from dataset sf - nn.,title="fig:",scaledwidth=45.0% ] ranging from @xmath21 ( top left ) to @xmath110 ( bottom ) a density plot of the larger eigenvalue is displayed . information from this has been combined with the three order parameter indicator variables mentioned above to infer a phase diagram schematically drawn in each diagram . we could not determine the phase boundary between the extended and collapsed phases for @xmath111 with much accuracy : it could move to smaller values of @xmath8 especially for small @xmath7 . as in figure [ fig : visaw_phase_diagram_tau_eq_5 ] the phase transition turns first - order as it approaches the vertical axis . these plots come from an appropriate slice of the dataset 3p , except the phase diagram at @xmath112 which comes from dataset sf - nn.,title="fig:",scaledwidth=45.0% ] ranging from @xmath21 ( top left ) to @xmath110 ( bottom ) a density plot of the larger eigenvalue is displayed . information from this has been combined with the three order parameter indicator variables mentioned above to infer a phase diagram schematically drawn in each diagram . we could not determine the phase boundary between the extended and collapsed phases for @xmath111 with much accuracy : it could move to smaller values of @xmath8 especially for small @xmath7 . as in figure [ fig : visaw_phase_diagram_tau_eq_5 ] the phase transition turns first - order as it approaches the vertical axis . these plots come from an appropriate slice of the dataset 3p , except the phase diagram at @xmath112 which comes from dataset sf - nn.,scaledwidth=45.0% ] one key observation to make is the maximally dense phase and the @xmath0-sheet phase never meet and that there are only ever three phases meeting at a point in the phase diagram . starting with the semi - flexible isaw model again at @xmath21 we now compare how the phase diagram changes with changing @xmath6 . we specifically consider @xmath113 and @xmath114 . the associated density plots of the largest eigenvalue are shown in figure [ fig : blah ] . the extended , @xmath0-sheet and collapse ( globular ) phases exist at each value of @xmath6 and exist for roughy in the same regions of parameter space : the @xmath0-sheet phase always exists when @xmath7 and @xmath8 are both large enough . for large enough @xmath6 a region of the maximally dense phase appears for small @xmath7 and small @xmath8 . this region increases in size with increasing @xmath6 . we now compare how the phase diagram changes from the one shown above for the semi - flexible visaw model ( @xmath29 ) by considering different slices of fixed @xmath8 : see figure [ fig : omega_slices ] . the extended to maximally dense transition is always first order except possibly at @xmath30 . for other values of @xmath8 there seems to be a special value of @xmath115 that signals a change from second to first order in that transition : second order for small @xmath7 and first order for large @xmath7 . the extended to @xmath0 sheet transition is expected to be first order while the collapse to @xmath0 sheet transition is expected to be second order in line with previous work . the extended to collapsed phase is the canonical @xmath18-transition that is weakly second order . these plots come from an appropriate slice of the dataset 3p , except the phase diagram at @xmath23 which comes from dataset sf.,title="fig:",scaledwidth=45.0% ] the extended to maximally dense transition is always first order except possibly at @xmath30 . for other values of @xmath8 there seems to be a special value of @xmath115 that signals a change from second to first order in that transition : second order for small @xmath7 and first order for large @xmath7 . the extended to @xmath0 sheet transition is expected to be first order while the collapse to @xmath0 sheet transition is expected to be second order in line with previous work . the extended to collapsed phase is the canonical @xmath18-transition that is weakly second order . these plots come from an appropriate slice of the dataset 3p , except the phase diagram at @xmath23 which comes from dataset sf.,title="fig:",scaledwidth=45.0% ] the extended to maximally dense transition is always first order except possibly at @xmath30 . for other values of @xmath8 there seems to be a special value of @xmath115 that signals a change from second to first order in that transition : second order for small @xmath7 and first order for large @xmath7 . the extended to @xmath0 sheet transition is expected to be first order while the collapse to @xmath0 sheet transition is expected to be second order in line with previous work . the extended to collapsed phase is the canonical @xmath18-transition that is weakly second order . these plots come from an appropriate slice of the dataset 3p , except the phase diagram at @xmath23 which comes from dataset sf.,title="fig:",scaledwidth=45.0% ] the extended to maximally dense transition is always first order except possibly at @xmath30 . for other values of @xmath8 there seems to be a special value of @xmath115 that signals a change from second to first order in that transition : second order for small @xmath7 and first order for large @xmath7 . the extended to @xmath0 sheet transition is expected to be first order while the collapse to @xmath0 sheet transition is expected to be second order in line with previous work . the extended to collapsed phase is the canonical @xmath18-transition that is weakly second order . these plots come from an appropriate slice of the dataset 3p , except the phase diagram at @xmath23 which comes from dataset sf.,title="fig:",scaledwidth=45.0% ] the extended to maximally dense transition is always first order except possibly at @xmath30 . for other values of @xmath8 there seems to be a special value of @xmath115 that signals a change from second to first order in that transition : second order for small @xmath7 and first order for large @xmath7 . the extended to @xmath0 sheet transition is expected to be first order while the collapse to @xmath0 sheet transition is expected to be second order in line with previous work . the extended to collapsed phase is the canonical @xmath18-transition that is weakly second order . these plots come from an appropriate slice of the dataset 3p , except the phase diagram at @xmath23 which comes from dataset sf.,title="fig:",scaledwidth=45.0% ] the extended to maximally dense transition is always first order except possibly at @xmath30 . for other values of @xmath8 there seems to be a special value of @xmath115 that signals a change from second to first order in that transition : second order for small @xmath7 and first order for large @xmath7 . the extended to @xmath0 sheet transition is expected to be first order while the collapse to @xmath0 sheet transition is expected to be second order in line with previous work . the extended to collapsed phase is the canonical @xmath18-transition that is weakly second order . these plots come from an appropriate slice of the dataset 3p , except the phase diagram at @xmath23 which comes from dataset sf.,title="fig:",scaledwidth=45.0% ] recalling that for @xmath29 we conjecture just two phases being extended and maximally dense we can now see how the other phases emerge . for @xmath116 there are only these two phases . it is worth noting that for @xmath117 the extended to maximally dense transition appears to be first order for all @xmath7 but as @xmath8 is increased a region of second order transition appears for small @xmath7 . for @xmath118 we can first see the emergence @xmath119 of the @xmath0-sheet phase for large @xmath7 from the extended phase this is in accord with the semi - flexible isaw diagram and other @xmath6 slices discussed above . again in accord with these @xmath6-slices we see that for @xmath120 and @xmath121 the extended phase has been completely replaced with the collapsed and @xmath0-sheet phases . also we note that as @xmath8 is increased the maximally dense phase retreats to large values of @xmath6 and smaller values of @xmath7 . this highlights the competition of the doubly - visited site and nearest - neighbour interactions : they favour different and competing low temperature phases . we also note that for @xmath122 the extended to maximally dense phase transition is completely second order for the range of @xmath7 and @xmath6 considered in these diagrams . finally we consider fixing the stiffness @xmath7 . there are important values of the stiffness , namely @xmath123 that are worth considering first . when @xmath30 all the configurations bend at each step and so effectively exist on the l - lattice . in particular , this slice contains the lsat model for @xmath29 . see figure [ fig : phase - diagram - p - eq-0 ] for the density plot of the largest eigenvalue of the fluctuations and a conjecture phase diagram . drawn on top of the density plot of the maximum eigenvalue of the fluctuations . for @xmath29 one obtains the lsat model . there are three phases of maximally dense , extended and collapsed . the phase transition between the extended and maximally dense becomes first order as it approaches the vertical axis for small @xmath8 . otherwise the transitions are second order . the data for this plot comes from dataset vi - nn l - lattice.,scaledwidth=70.0% ] . the left diagram shows the frequency of singly - visited sites going to zero for large @xmath6 and small @xmath8 , and the right diagram shows a clearly demarcated region of extended configurations where the density is small . taken together , these diagrams indicate the presence of three distinct phases . the data for this plot comes from dataset vi - nn l - lattice.,title="fig:",scaledwidth=48.0% ] . the left diagram shows the frequency of singly - visited sites going to zero for large @xmath6 and small @xmath8 , and the right diagram shows a clearly demarcated region of extended configurations where the density is small . taken together , these diagrams indicate the presence of three distinct phases . the data for this plot comes from dataset vi - nn l - lattice.,title="fig:",scaledwidth=48.0% ] . according to the phase diagram in figure [ fig : phase - diagram - p - eq-0 ] , on the vertical line @xmath124 and for values of @xmath6 above 2 , we are in the maximally dense phase . this is confirmed in the above plot which shows that the ratio of singly visited sites goes to zero . the plot on the left is obtained from dataset vi - nn l - lattice.,scaledwidth=70.0% ] the phase diagram is seen to contain three phases : extended , collapsed ( globular ) and maximally dense , see figure [ fig : order_parameters_p - eq-0 ] . this is as expected since the @xmath0-sheet phase exists only for sufficiently large stiffness . the demarcation of the maximally dense phase is fairly well defined but the boundary between the extended and globular phases is very difficult to ascertain : the fluctuation data and the order parameter data given quite different phase boundaries . we have conservatively conjectured a boundary that does not change much as @xmath6 changes but we have weak evidence that this is correct . the key observation to be made here is that the low temperature phase of the lsat model is maximally dense ! this is completely unexpected and we do not have a explanation for this . the lsat model has a collapse transition that is expected to be in the @xmath18 universality class and not the isat / visaw one and yet its low temperature collapsed phase seems to coincide with those models : recall that collapse transition has a strongly divergent specific heat in those models . we confirm that as @xmath8 is varied that the low temperature phase is indeed maximally dense : see figure [ fig : visaw_low_temperature_omega_eq_0_5_p_eq_0 ] . for non - zero @xmath8 we see that the extended to maximally dense transition is second order and only at @xmath117 might it turn to first order . this is in accord with the lsat model at @xmath29 of course . , as obtained from dataset vi - nn . first- and second - order transitions are shown respectively as dashed and solid lines . there are three phases , as at @xmath30 , with extended , collapsed and maximally dense phases existing.,scaledwidth=70.0% ] we now consider fixing the stiffness to the bn - point value @xmath126 so as to consider how the bn - point sits in the wider phase space varying @xmath8 and @xmath6 : see figure [ fig : phase_diagram_bn ] . the semi - flexible visaw model along the line @xmath127 is shown in figure [ fig : phase_diagram_bn ] as the line @xmath29 . the analysis here is done analogously to the one above for the lsat model ( @xmath30 ) . one can see that in common with the lsat model the low temperature phase is maximally dense , as demonstrated in figure [ fig : order - parameters - bn ] , which shows diagrams for the relevant order parameters . from the scaling of the number of singly visited sites ( figure [ fig : single - visited - bn ] ) and typical configurations ( figure [ fig : configs_bn ] ) it follows that the low temperature phase is indeed maximally dense . . the left diagram shows the frequency of singly - visited sites going to zero for large @xmath6 and small @xmath8 , and the right diagram shows a clearly demarcated region of extended configurations where the density is going to zero . taken together , these diagrams indicate the presence of three distinct phases . the data for this plot comes from dataset vi - nn.,title="fig:",scaledwidth=48.0% ] . the left diagram shows the frequency of singly - visited sites going to zero for large @xmath6 and small @xmath8 , and the right diagram shows a clearly demarcated region of extended configurations where the density is going to zero . taken together , these diagrams indicate the presence of three distinct phases . the data for this plot comes from dataset vi - nn.,title="fig:",scaledwidth=48.0% ] , suggesting that increasing @xmath6 leads to a collapsed phase that is maximally dense . this data comes from dataset vi.,scaledwidth=70.0% ] along the line @xmath128 at @xmath129 , @xmath130 ( at the approximate location of the transition ) and @xmath103 . the low - temperature phase is clearly compact . these configurations are obtained from dataset vi . ] it is interesting to consider the fully - flexible case of our model ( @xmath22 ) . this model contains explicitly the fully flexible visaw for @xmath29 and the standard isaw model for @xmath105 . see figure [ fig : phase - diagram_p_1 ] . again the three phases of extended , globular and maximally dense occur . the region of the conjectured first order transition between the extended and maximally dense phase has increased significantly over @xmath127 . it may be that in the thermodynamic limit that it ends at @xmath29 a question that deserves further consideration . . first- and second - order transitions are shown respectively as dashed and solid lines . this data comes from dataset vi - nn-2 . the canonical isaw model lies on the horizontal line @xmath105.,scaledwidth=70.0% ] it is worth considering the typical configurations along a line parallel in @xmath6 to the isaw model . in figure [ fig : configs - p_1 ] we have illustrated some typical configurations for the line @xmath88 . at low temperature ( large @xmath8 ) we see a configuration from the globular collapsed phase which can be compared to the maximally dense and @xmath0-sheet phases . and @xmath131 , at @xmath132 , @xmath133 and @xmath134 . the low - temperature phase is globular : dense but with amorphous structure . these configurations come from the dataset nn-2.,scaledwidth=100.0% ] we now compare a fully range of @xmath7 values in figure [ fig : phase - diagrams - variable - p ] . we see that as one increases @xmath7 the region of the maximally dense phase retreats to larger values of @xmath6 and smaller values of @xmath8 eventually disappearing from our diagrams range of @xmath6 and @xmath8 once @xmath7 reaches @xmath135 . the other significant feature to note is the eventual appearance of the @xmath0-sheet phase for @xmath136 at large values of @xmath8 . the extended to maximally dense transition is always second order except possibly at @xmath117 . in line with the previously discussed @xmath8-slices there exists a special value of @xmath115 that signals a change from second to first order in that transition at other values of @xmath7 . the first three plots come from datasets vi - nn l - lattice , vi - nn and vi - nn-2 respectively . the remaining plots come from an appropriate slice of the dataset 3p.,title="fig:",scaledwidth=45.0% ] the extended to maximally dense transition is always second order except possibly at @xmath117 . in line with the previously discussed @xmath8-slices there exists a special value of @xmath115 that signals a change from second to first order in that transition at other values of @xmath7 . the first three plots come from datasets vi - nn l - lattice , vi - nn and vi - nn-2 respectively . the remaining plots come from an appropriate slice of the dataset 3p.,title="fig:",scaledwidth=45.0% ] the extended to maximally dense transition is always second order except possibly at @xmath117 . in line with the previously discussed @xmath8-slices there exists a special value of @xmath115 that signals a change from second to first order in that transition at other values of @xmath7 . the first three plots come from datasets vi - nn l - lattice , vi - nn and vi - nn-2 respectively . the remaining plots come from an appropriate slice of the dataset 3p.,title="fig:",scaledwidth=45.0% ] the extended to maximally dense transition is always second order except possibly at @xmath117 . in line with the previously discussed @xmath8-slices there exists a special value of @xmath115 that signals a change from second to first order in that transition at other values of @xmath7 . the first three plots come from datasets vi - nn l - lattice , vi - nn and vi - nn-2 respectively . the remaining plots come from an appropriate slice of the dataset 3p.,title="fig:",scaledwidth=45.0% ] the extended to maximally dense transition is always second order except possibly at @xmath117 . in line with the previously discussed @xmath8-slices there exists a special value of @xmath115 that signals a change from second to first order in that transition at other values of @xmath7 . the first three plots come from datasets vi - nn l - lattice , vi - nn and vi - nn-2 respectively . the remaining plots come from an appropriate slice of the dataset 3p.,title="fig:",scaledwidth=45.0% ] the extended to maximally dense transition is always second order except possibly at @xmath117 . in line with the previously discussed @xmath8-slices there exists a special value of @xmath115 that signals a change from second to first order in that transition at other values of @xmath7 . the first three plots come from datasets vi - nn l - lattice , vi - nn and vi - nn-2 respectively . the remaining plots come from an appropriate slice of the dataset 3p.,title="fig:",scaledwidth=45.0% ] the extended to maximally dense transition is always second order except possibly at @xmath117 . in line with the previously discussed @xmath8-slices there exists a special value of @xmath115 that signals a change from second to first order in that transition at other values of @xmath7 . the first three plots come from datasets vi - nn l - lattice , vi - nn and vi - nn-2 respectively . the remaining plots come from an appropriate slice of the dataset 3p.,title="fig:",scaledwidth=45.0% ] while there are several different transitions in our model we have concentrated our investigation to the extended to maximally dense transition as the least all understood but also assessable using our data . first let us consider the other transitions . the extended to globular transition has been well studied previously , see @xcite and reference therein , as this is the canonical polymer collapse transition . we find no evidence that it changes with parameters of our model and we conjecture that it always lies in the @xmath18-point universality class of the isaw model . the extended to @xmath0-sheet transition has been studied previously @xcite and a first order transition has been found . we confirm this and see it also is unchanged by the parameter variations in our model . the globular to @xmath0-sheet transition is a low temperature transition that is difficult to study . nevertheless it has been studied in @xcite and they found @xmath137 : a moderately strong divergence specific heat . we are unable to provide any further precision to the previous study . the transition between the collapsed phase and the maximally dense phase has not been studied previously and unfortunately our simulations are not long enough to elucidate this low temperature transition . the only comment we can make is that we do not see evidence that it is first order . the transition in the visaw model , and as we have seen it is apparently the one in the lsat model , is the transition between the _ extended to maximally dense _ phases . since analysis of the visaw model has previously found a diverging specific hear with @xmath25 ( we confirm this earlier estimate in figure [ fig : specific - heat - expon - visaw ] ) while lsat has a conjectured negative exponent value of @xmath19 in common with the @xmath18-point there is no consistent value in the literature we can attribute to this transition . , @xmath23 as a function of @xmath6 . on the bottom is the scaling of the peak heights in length , which shows an exponent close to @xmath138 . these plots come from dataset vi-2.,title="fig:",scaledwidth=70.0% ] , @xmath23 as a function of @xmath6 . on the bottom is the scaling of the peak heights in length , which shows an exponent close to @xmath138 . these plots come from dataset vi-2.,title="fig:",scaledwidth=70.0% ] the analysis by blte - nienhuis @xcite of the semiflexible visaw model and the subsequent numerical work by foster and pinettes @xcite find a variety of phase transition including first order behaviour , which indicates this transition has a richer structure that depends on model parameters . in fact we have found that there is a region where it is first order and a region where it is second order . importantly our second order transition does not have exponents that are consistent . while it is possible that the scenario is the same as in the triangular isat model where it has been conjectured that a first order line ends in an isat universality class point followed by a line of @xmath18-like transitions , our data is not precise enough to confirm this . however , that scenario would be the simplest : it begs the question of what happens at the point where the maximally dense globular and extended phases meet . it is also confused by at least one set of data , which is shown in figure [ fig : nu_visaw ] . if the fully flexible visaw model is in the isat universality class , we should find a size exponent of @xmath17 with logarithmic corrections . however , the plot of the effective exponent seems to indicate that @xmath20 would be a compatible value . this confuses any possible simple conjecture . when @xmath139 and @xmath22 , that for the visaw model , against @xmath6 for different values of length . the different lines cross at a point close to the expected transition temperature indicating the critical value of @xmath66 to be close to the @xmath18-value of @xmath27 . this plot comes from dataset vi-2.,scaledwidth=70.0% ] it is worthwhile noting that the estimation of the exponents for the convergent @xmath18-like transitions can be deceptive . in figure [ fig : omega_eq_1_p_eq_0 ] we show an effective exponent estimate for @xmath40 directly from the fluctuations which gives a positive value of around @xmath140 . however , knowing that a negative exponent is likely , an estimate of the exponent using @xmath141 gives a value of around @xmath142 at the lengths considered , which are up to @xmath143 . we are aware from earlier work that good estimates of @xmath40 in these problems require polymer lengths of many thousands @xcite . on the vertical line @xmath30 , @xmath144 . bottom : the scaling of the specific heat peak with respect to the size of the walk . these plots come from dataset vi l - lattice . we believe this critical point belongs to a line of phase transitions in the @xmath145 universality class , which is know to feature a convergent specific - heat . therefore we interpret the low effective exponent 0.22 as the result of strong finite - size corrections to a converging specific - heat.,title="fig:",scaledwidth=70.0% ] on the vertical line @xmath30 , @xmath144 . bottom : the scaling of the specific heat peak with respect to the size of the walk . these plots come from dataset vi l - lattice . we believe this critical point belongs to a line of phase transitions in the @xmath145 universality class , which is know to feature a convergent specific - heat . therefore we interpret the low effective exponent 0.22 as the result of strong finite - size corrections to a converging specific - heat.,title="fig:",scaledwidth=70.0% ] a very similar picture is produced at @xmath146 suggesting the transition to the maximally dense phase looks @xmath18-like when @xmath30 regardless of @xmath8 . of interest has been the bn - point value of the stiffness @xmath126 , and in figure [ fig : specific - heat - expon - bn ] we plot the specific heat at @xmath147 where we obtain @xmath148 . , @xmath144 . the scaling of the peak heights in length shows an exponent close to @xmath149 . these plots come from dataset vi-3.,title="fig:",scaledwidth=70.0% ] , @xmath144 . the scaling of the peak heights in length shows an exponent close to @xmath149 . these plots come from dataset vi-3.,title="fig:",scaledwidth=70.0% ] however , at @xmath146 we see in figure [ fig : omega_eq_0.5_p_eq_0 ] that an estimate of @xmath150 is obtained , compatible with the isat universality class . it could be that at @xmath146 there are just very strong finite size effects and a negative value will eventually be reached for sufficiently long lengths . 0.6 on the vertical line @xmath127 , @xmath124 . ( ) the scaling of the specific heat peak with respect to the size of the walk . ( ) the distribution of site interactions near the specific heat peak has a single mode . all plots come from dataset vi.,title="fig : " ] 0.6 on the vertical line @xmath127 , @xmath124 . ( ) the scaling of the specific heat peak with respect to the size of the walk . ( ) the distribution of site interactions near the specific heat peak has a single mode . all plots come from dataset vi.,title="fig : " ] 0.6 on the vertical line @xmath127 , @xmath124 . ( ) the scaling of the specific heat peak with respect to the size of the walk . ( ) the distribution of site interactions near the specific heat peak has a single mode . all plots come from dataset vi.,title="fig : " ] to demonstrate that the extended to maximally dense phase transition can also display first order characteristics we consider now @xmath146 with @xmath22 . in figure [ fig : omega_eq_0_5_p_eq_1 ] we plot the specific heat and the distribution of doubly visited sites which is clearly bimodal . the specific heat looks like it is diverging super - linearly ( impossible in the large length limit ) which is the indication of the finite size early build up of a first - order transition . 0.6 on the vertical line @xmath22 , @xmath124 . ( ) the scaling of the specific heat peak with respect to the size of the walk . the scaling is super - linear , which is a clear indication of a first - order phase transition . ( ) the distribution of site interactions near the specific heat peak presents two clear peaks . these plots come from dataset vi-4.,title="fig : " ] 0.6 on the vertical line @xmath22 , @xmath124 . ( ) the scaling of the specific heat peak with respect to the size of the walk . the scaling is super - linear , which is a clear indication of a first - order phase transition . ( ) the distribution of site interactions near the specific heat peak presents two clear peaks . these plots come from dataset vi-4.,title="fig : " ] 0.6 on the vertical line @xmath22 , @xmath124 . ( ) the scaling of the specific heat peak with respect to the size of the walk . the scaling is super - linear , which is a clear indication of a first - order phase transition . ( ) the distribution of site interactions near the specific heat peak presents two clear peaks . these plots come from dataset vi-4.,title="fig : " ] finally , we estimate , using from simulations of short walks ( up to length 100 ) , the extent of the first order region for different values of @xmath7 : this is given in figure [ fig : first - order - region ] . and @xmath7 , the lowest value of @xmath6 for which we detected a double peak in the energy distribution . these lines have been obtained from the very short walks ( @xmath87 ) in the dataset 3p and therefore these numbers are to be considered upper bounds ( e.g. in the thermodinamic limit the lines would extend further on the right ) . , scaledwidth=50.0% ] we have considered a generalised model of polymer collapse in two dimensions that contains several well know models as subcases . we have found three low temperature phases as well as the normal high temperature extended phase that is dominated by the excluded volume effect . the three low temperature phases can be described as globular , which occurs in the canonical polymer collapse models , maximally dense , and as an anisotropic crystal phase , that has typical configurations analogous to a three - dimensional @xmath0-sheet . the @xmath0-sheet phase and maximally dense phase do not meet . the phase transition between the extended and maximally dense phase is the most complicated and displays first order or second order characteristics depending on the parameters . more work needs to be done to fully elucidate the transitions but we now have an overall picture in a larger parameter space of the how these phases arise . financial support from the australian research council via its support for the centre of excellence for mathematics and statistics of complex systems and through its discovery program is gratefully acknowledged by the authors . a l owczarek thanks the school of mathematical sciences , queen mary , university of london for hospitality . 10 url # 1#1urlprefix[2][]#2

we analyze the scenario of molecular cloud formation by large - scale supersonic compressions in the diffuse warm neutral medium ( wnm ) . during the early stages of this process , a shocked layer forms , and within it , a thin cold layer . an analytical model and high - resolution 1d simulations predict the thermodynamic conditions in the cold layer . after myr of evolution , the layer has column density , thickness pc , temperature k and pressure k . these conditions are strongly reminiscent of those recently reported by heiles and coworkers for cold neutral medium sheets . in the 1d simulations , the inflows into the sheets produce line profiles with a central line of width and broad wings of width . being the result of the inflows rather than of turbulence , these linewidths do not imply excessively short lifetimes for the sheets . at later times , 3d numerical simulations show that the cold layer undergoes a dynamical instability that causes it to develop turbulent motions and to increase its thickness , until it becomes a fully three - dimensional turbulent cloud . the destabilization mechanism appears to be the nonlinear thin shell instability , triggered by the thermal instability - induced density contrast . in the simulations , fully developed turbulence arises on times ranging from myr for inflow mach number to myr for . due to the limitations of the simulations , these numbers should be considered upper limits . in the turbulent regime , the highest - density gas ( hdg , ) is always overpressured with respect to the mean wnm pressure by factors 1.54 , even though we do not include self - gravity . the intermediate - density gas ( idg , < 100 $ ] ) has a significant pressure scatter that increases with , so that at , a significant fraction of the idg is at a higher pressure than the hdg . the ratio of internal to kinetic energy density varies from the inflow to the idg and the hdg , and increases with density in the most turbulent runs . our results suggest that the turbulence and at least part of the excess pressure in molecular clouds can be generated by the compressive process that forms the clouds themselves , and that thin cnm sheets may be formed transiently by this mechanism , when the compressions are only weakly supersonic .

astro-ph0509127

molecular cloud complexes ( or `` giant molecular clouds '' , gmcs ) are some of the most studied objects in the interstellar medium ( ism ) of the galaxy . yet , their formation mechanism and the origin of their physical conditions remain uncertain ( see , e.g. , * ? ? ? * ; * ? ? ? in particular , gmcs are known to have masses much larger than their thermal jeans masses @xcite ( i.e. , thermal energies much smaller than their absolute gravitational energy ) , but comparable gravitational , magnetic , and turbulent kinetic energies @xcite . this similarity has traditionally been interpreted as indicative of approximate virial equilibrium , and of rough stability and longevity of the clouds ( see , e.g. , * ? ? ? * ; * ? ? ? * ) . in this picture , the fact that molecular clouds have thermal pressures exceeding that of the general ism by roughly one order of magnitude @xcite is interpreted as a consequence of the fact that they are strongly self - gravitating . because of their self - gravitating and overpressured nature , molecular clouds were left off global ism models based on thermal pressure equilibrium , such as those by @xcite and @xcite . recent work suggests instead that molecular clouds and their substructure may actually be transient features in their respective environments . in this picture , the clouds and their substructure are undergoing secular dynamical evolution , being assembled by large - scale supersonic compressions in the atomic medium on a crossing time and becoming self - gravitating in the process @xcite . specifically , @xcite suggested that the accumulation process of atomic gas may last a few tens of megayears until the gas finally becomes molecular , self - gravitating and supercritical at roughly the same time ( see also * ? ? ? * ) . observations of atomic inflows surrounding molecular gas support this scenario @xcite . in this scenario , the excess pressure in molecular clouds may arise from the fact that they have been assembled by motions whose ram pressures are comparable or larger than the local thermal pressure , so that both the excess pressure and the self - gravitating nature of the clouds may originate from the compression that forms the cloud , and need not be an indication of any sort of equilibrium @xcite . additionally , it has been suggested by @xcite that a substantial fraction of the internal turbulence of the clouds may originate from thin - shell instabilities in the compressed gas , rather than from local energy injection from the clouds own stellar products ( see also * ? ? ? this would explain the fact that even clouds with no apparent stellar content ( see , e.g. , * ? ? ? * ) have levels of turbulence comparable to clouds with healthy star - forming activity . since the typical velocity dispersion in the warm medium ( which we will denote generically as wnm , as we will not distinguish between neutral and ionized warm components ) is @xmath15 km s@xmath16 @xcite , the flow is transonic ( i.e. , with rms mach number @xmath0 ) . under these conditions , transonic compressions in the wnm may render the gas unstable to two main types of instabilities . on one hand , there is the thermal instability ( ti ) , originally studied by @xcite ( see also the reviews by * ? ? ? * ; * ? ? ? * ) , and by @xcite in the context of radiative post - shock cooling regions , in which it causes the formation of thin cold sheets parallel to the shock front . the formation of cold neutral medium ( cnm ) structures induced by compressions in the wnm has been more recently studied analytically and numerically by @xcite and @xcite . on the other hand , there are bending mode instabilities , which cause rippling of the compressed layer , and , eventually , fully developed turbulence . the nonlinear theory was laid down by @xcite in the isothermal case , who referred to the instability as nonlinear thin - shell instability ( ntsi ) , a name which we use throughout this paper . more recently , @xcite have studied the stability of cooling shocks and its dependence on the mach number and the final post - shock temperature , finding that lower final temperatures cause greater overstability . numerically , several groups have studied the problem in a variety of contexts , such as the interiors of molecular clouds at sub - pc scales @xcite , stellar wind interactions @xcite , cloud collisions @xcite , and in a general fashion @xcite . in particular , the simulations of the latter authors show that turbulence is generated and maintained for the entire duration of the inflowing motions . in the context of the thermally bistable atomic medium , a two - dimensional study of converging flows in the diffuse atomic medium has been recently presented by @xcite , but focusing on the competition between externally imposed turbulence and the tendency of ti to produce two - phase structure , rather than in the production of turbulence by the process . in a related context , @xcite and @xcite have studied the production of supersonic turbulence behind a propagating shock wave in the warm neutral ism , showing that cold , dense cloudlets are formed , with bulk velocities that are supersonic with respect to their internal temperatures , proposing this mechanism as the origin of molecular cloud turbulence . however , these authors attributed the development of turbulence to ti alone without indicating the precise mechanism at play , and focused on the structure behind a single traveling shock wave , rather than on the structure in shocked compressed layers between converging flows . instead , large molecular masses typical of gmcs probably require focused , large - scale compressive motions that may last several tens of megayears , as in the scenario proposed by @xcite . in this case , the turbulence generation might last for as long as the accumulation ( compression ) lasts , and the turbulence in molecular clouds could be considered as driven , rather than decaying . a preliminary study of the development of the turbulence in these circumstances has been recently presented by @xcite , who also investigated the masses and possible self - gravitating nature of the clumps formed by the induced turbulence . the advanced stages of molecular cloud evolution are also poorly understood . traditionally , the greatest concern about molecular clouds has been how to support them against self - gravity , and it was speculated that turbulence could be prevented from decaying if it consisted primarily of mhd waves @xcite , although numerical work strongly suggests that mhd turbulence decays just as fast as hydrodynamic turbulence @xcite . clouds were assumed to be finally dispersed by the kinetic energy injection from their stellar products ( e.g. , * ? ? ? * ) instead , in the secular - evolution scenario cloud support may not be a concern at all . clouds start as atomic entities with negligible self - gravity that increase their mean density as they are compressed , so that their self - gravity also increases in the process . when they finally become strongly self - gravitating , collapse _ does _ occur , albeit in a localized fashion that involves a small fraction of the total cloud mass , due to the action of their internal turbulence , which forms nonlinear density fluctuations that may collapse themselves on times much shorter than the parent cloud s free - fall time @xcite . the mass that does not collapse sees its density reduced , and may possibly be dispersed as soon as the external compression ends . indeed , even in simulations of decaying , non - magnetic self - gravitating turbulence , the star formation efficiency is smaller than unity ( e.g. , * ? ? ? * ) , and some of the cloud s mass is dispersed away from the simulation box @xcite . the present paper is the first in a series that analyzes quantitatively the turbulent scenario of molecular cloud evolution , assuming that they are formed by large - scale compressions in the warm diffuse medium . here we focus on the formation stages , investigating the physical conditions that may be produced in the compressed layer as a function of the mach number of the converging streams , and so we neglect the chemistry and self - gravity of the gas and use moderate - resolution simulations to test the order - of - magnitude estimates that can be made from simple considerations , the triggering of thin - shell instabilities , and the statistics of the physical conditions in the turbulent gas . we also report on the formation of transient , thin sheets of cold gas during the initial stages of the cloud formation process . these sheets turn out to closely match the properties of cold neutral medium ( cnm ) sheets recently observed by heiles and collaborators @xcite . these authors have reported on the existence of extremely thin cnm sheets , with thicknesses @xmath17 pc , column densities , @xmath18 , temperatures @xmath19 k and linewidths @xmath20 , and have argued that these properties are difficult to understand in terms of turbulent clouds . we will argue that objects with these properties are naturally formed as part of the cloud formation process , in cases of low - mach number compressions . the plan of the paper is as follows : in [ sec : basic_phys ] we describe the basic physics of the pre - turbulent stages of the system , and then in [ sec : num_sim ] we present numerical simulations that confirm the estimates and follow the nonlinear stages in which the compressed and cooled layers become turbulent . we describe the numerical method in [ sec : num_meth ] , relevant resolution considerations in [ sec : num_res ] , and the results in [ sec : results ] . in [ sec : discussion ] we discuss our results , comparing them with observational data and previous work . finally , in [ sec : concl ] we give a summary and some conclusions . we consider the atomic medium in the ism , whose flow and thermal conditions are governed by the equations @xmath21 = \rho \left(\gamma - \rho \lambda(t)\right ) , \label{eq : int_en}\end{aligned}\ ] ] where @xmath22 is the gas density , * v * is the fluid velocity , @xmath23 is the total energy per unit volume , @xmath24 is the thermal pressure , @xmath25 $ ] is the internal energy per unit mass @xmath26 is the temperature , @xmath27^{-1}$ ] is the specific heat at constant volume , @xmath28 is the heat capacity ratio of the gas , @xmath29 is the heating rate , and @xmath30 is the radiative cooling rate . we use a piecewise power - law fit to the cooling function , based on a fit to the standard thermal - equilibrium ( te ) pressure versus density curve of @xcite , as decribed in @xcite and @xcite . the fit is given by @xmath31 the background heating is taken as a constant @xmath32 , where `` h@xmath16 '' means `` per hydrogen atom '' . this value is roughly within half an order of magnitude of the value of the dominant heating mechanism ( photo - electric heating ) reported by @xcite throughout the range @xmath33 . note that we have for now neglected magnetic fields , self - gravity , and heating and cooling processes adequate for the molecular regime . the condition of thermal equilibrium between heating and cooling at a given density defines te - values of the temperature and thermal pressure , which we denote by @xmath34 and @xmath35 . figure [ fig : peq_vs_rho ] shows @xmath36 versus number density @xmath37 for the cooling and heating functions defined above . as is well known @xcite , at the mean midplane thermal pressure of the ism ( @xmath38 k @xmath39 or sightly higher , * ? ? ? * ) , the atomic medium is thermally bistable . for our chosen fits to the cooling and heating functions , and a mean pressure of @xmath40 k@xmath5 , a warm diffuse phase ( @xmath41 @xmath39 , @xmath42 k ) is able to coexist in pressure equilibrium with a cold dense one ( @xmath43 @xmath39 , @xmath44 k ) . a third , unstable phase with ( @xmath45 @xmath39 , @xmath46 k ) also corresponds to the same equilibrium pressure , but is not expected to exist under equilibrium conditions because it is unstable . the equilibrium values of the density are also shown in fig . [ fig : peq_vs_rho ] . due to our neglect of molecular - phase cooling and heating , our @xmath36-@xmath37 curve does not correctly represent the approximately isothermal behavior of molecular gas above densities of a few hundred @xmath39 , and so we introduce a slight error in the thermodynamic conditions of the densest gas . we expect to address this shortcoming in subsequent papers , which will also take into account the chemistry in the gas and the transition to the molecular phase . within the warm medium described above , we consider the collision of two oppositely - directed gas streams with speeds comparable to the sound speed , since the velocity dispersion in the warm ism is known to be roughly sonic ( e.g. * ? ? ? * ; * ? ? ? this compressive motion may have a variety of origins ( the passage of a spiral density wave , or a large - scale gravitational instability , or general transonic turbulence in the diffuse medium ) , whose details are not of our concern here . the discussion below on the early evolution of the system is inspired by that of @xcite , except that in our case we assume that the inflows are maintained in time rather than being impulsive , and that our simulations are three - dimensional . the colliding streams have the density and thermal pressure conditions of the warm diffuse phase , forming a shock - bounded slab at the collision site , as illustrated in fig . [ fig : schematic ] . during the initial stages the evolution is adiabatic , the bounding shocks move outwards from the collision site , and the gas inside the shocked slab is heated and driven away from thermal equilibrium by the shocks . the gas in the slab then has conditions dictated by the adiabatic jump relations ( see , e.g. , * ? ? ? * ) . for later use , we transcribe here the jump conditions for the velocity and the pressure : @xmath47 where the subindex `` 1 '' denotes the pre - shock ( or upstream ) quantities , subindex `` 2 '' denotes quantities _ immediately _ downstream of the shock , and the symbol @xmath48 denotes velocities in the frame of reference of the shock . velocities in the rest frame are denoted by the symbol @xmath49 . denoting by @xmath50 the shock speed ( in the rest frame ) one has @xmath51 and @xmath52 at the time of the shock formation , the velocity @xmath53 is simply zero and it is thus possible to calculate the initial shock speed by combining eqs . ( [ eq : u_jump ] ) , ( [ eq : u2_vs ] ) and ( [ eq : u1_v1_vs ] ) . one obtains the following quadratic equation for @xmath50 as a function of @xmath54 @xmath55 = 0,\ ] ] where @xmath56 is the sound speed in the ( warm diffuse ) pre - shock medium . the solution of this equation , expressed in terms of the inflow mach number in the rest frame @xmath57 , is @xmath58 + 16}}{4}. \label{eq : vs_m1r}\ ] ] as time progresses , the gas in the slab begins to cool significantly , so that most of the internal energy increase caused by the shock is radiated away after roughly a cooling time , which we take here as @xmath59 where @xmath60 is the cooling function evaluated at the immediate post - shock conditions . equation ( [ eq : tau_cool ] ) is well defined in the shock - heated , out - of - te post - shock gas , in which @xmath61 . at @xmath62 , the flow returns close to its local equilibrium temperature . if the adiabatic - jump value of the immediate post - shock pressure is higher than the maximum te pressure allowed for the warm gas , then the cooling brings the gas to the cold branch of the te curve , creating a thin , dense , cold layer in the middle of the compressed slab ( * ? ? ? * see also hennebelle & prault 1999 ) . we interchangeably refer to this dense layer as the `` cold '' or the `` condensed '' layer , and to its boundary as the condensation front . the strong cooling undergone by the gas as it flows into the central regions causes the thermal pressure to drop to values comparable to that of the inflow , the density to increase , and the outer shock to decelerate . ti exacerbates this process by creating such a large density contrast ( @xmath63 ) between the inflow and the central layers that a quasi - stationary situation is reached in which shocked layer thickness becomes almost constant and the outer shock is almost at rest . for @xmath64 , the cold slab has a half - thickness of roughly one cooling length @xmath65 , given by @xmath66 , and the velocity field across the shocked layer takes values between @xmath53 ( immediately behind the shock ) and zero at the center of the cold slab . the cooling time ( @xmath67 ) and length ( @xmath65 ) as functions of the inflow mach number are plotted in fig . [ fig : cool_time_length ] . as we will discuss in [ sec : num_meth ] , the cooling length constitutes the minimum physical size that the numerical box must have . the pressure in the central cold layer can be estimated assuming that the total pressure , sum of the kinetic and ram pressures , is constant in the shocked region . the immediate post - shock pressures and velocities are given by the jump conditions , eqs.([eq : u_jump ] ) and ( [ eq : p_jump ] ) , using an upstream velocity equal to the inflow speed , since at this time the shock is not moving in the rest frame . the central pressure @xmath68 ( in what follows , we denote quantities in the condensed layer by the subindex `` 3 '' ) is thus estimated as @xmath69 since the velocity is zero at the center of the condensed layer . the density in the condensed layer can then be estimated from the fact that its constituent gas has undergone a phase transition from the warm to the cold phase , in which the approach to te is extremely fast . this means that the density in the cold layer is given by the thermal - equilibrium value at @xmath68 . for a power - law cooling function @xmath70 and a constant heating function , it can be shown that the equilibrium pressure satisfies @xmath71 , with @xmath72 being an effective polytropic exponent ( see , e.g. , * ? ? ? thus , the dense branch of our @xmath36 vs. @xmath22 curve , in which @xmath73 , is described by the equation @xmath74 setting @xmath75 in the above equation and inverting to solve for the density , we obtain the dependence of the cold layer density @xmath76 on the inflow mach number @xmath14 , shown as the solid line in fig.[fig : ncold_minfl ] . note that for molecular gas the density is not expected to increase as rapidly with @xmath14 , because in this case the flow is approximately isothermal so that @xmath77 @xcite . if the problem were strictly one - dimensional , then the condensed layer would simply thicken in time , with its bounding front moving at a speed @xmath78 , an estimate based on mass conservation from the immediate post - shock region to the cold slab . this velocity is shown as the dotted line in fig . [ fig : ncold_minfl ] . we see that , somewhat counterintuitively , the front speed decreases with increasing @xmath14 . this is because the density contrast between the condensed layer and its surroundings increases rapidly with @xmath14 , and thus the mass entering the condensed layer is compacted very efficiently , causing a very mild increase in the slab thickness . finally , the dash - dotted line in fig . [ fig : ncold_minfl ] gives the pressure in the condensed layer . the column density through the slab increases with time and is given by @xmath79 , where @xmath80 is the time since the beginning of the condensation process , and the factor of 2 comes from the fact that @xmath81 is the half - thickness of the cold layer . the dashed line in fig . [ fig : ncold_minfl ] shows the column density through the slab after 1 myr in units of @xmath82 @xmath83 , as a function of @xmath14 . we see that the column density in the cold layer varies relatively slowly with @xmath14 , with values @xmath84@xmath85 @xmath83 at @xmath86 myr . the simple model given in this section then predicts the values of the physical variables in the cold layer . in [ sec : mod_sim_comp ] we compare its predictions with the results of 1d high - resolution numerical simulations of the process . the description of the late stages of the evolution requires numerical simulations , which we present in [ sec : num_sim ] . here we just give some general discussion and expectations . shock compressed layers are known to be _ nonlinearly _ unstable in the isothermal case @xcite , meaning that large enough inflow mach numbers are needed to trigger the instability . specifically , displacements of the compressed layer comparable to its thickness are necessary to trigger the instability . this would seem to create a difficulty for destabilizing the compressed layers in the wnm , given the transonic nature of the flow . however , cooling apparently helps in bringing down the threshold for instability , as suggested by the recent results of @xcite . these authors found that there exists a critical value @xmath87 of the cooling exponent for the appearance of the well - known global overstability in radiative shocks @xcite , and that this exponent depends on both the upstream mach number and the ratio @xmath88 of the cold layer temperature to the pre - shock temperature . the relevant result for our purposes is that low - mach number , low-@xmath88 shocks have values of @xmath87 that are comparable to those of high - mach number shocks with high @xmath88 . that is , lower cold - layer temperatures bring down the required mach number for overstability . since the temperature ratio between the cold and warm phases in the atomic medium is @xmath89 , the compressed layers formed by transonic compressions in this medium can probably become overstable , producing oscillations that can bend the layer strongly enough to trigger a nonlinear thin shell - like ( nts - like ) instability even at relatively low mach numbers . indeed , in the simulations reported in [ sec : num_sim ] , we have always found that the compressed layer eventually becomes unstable , given enough time and resolution . we have found no threshold for suppression of the nts - like instability in the range of inflow mach numbers we have explored . in this section we present moderate - resolution simulations that are intended mostly as an exploratory tool of parameter space and of the phenomena described in the previous section . detailed , high - resolution simulations using adaptive - mesh and smoothed - particle hydrodynamics techniques will be presented in future papers to investigate the details of the small - scale gasesous structures as well as the star formation that result in this scenario . we solve the hydrodynamic equations together with the energy conservation equations using eulerian hydrodynamics code based on the total variation diminishing ( tvd ) scheme @xcite . it is a second - order accurate upwind scheme , which conserves mass , momentum and energy . we include the heating and cooling functions described in [ sec : eq ] as source terms after the hydrodynamic step . tests have shown that this procedure is accurate enough , so that we did not have to implement an``operator splitting '' algorithm ( where the execution of the hydrodynamic and source stages is alternated ) which formally preserves the second order accuracy . we solve the equations both in one ( 1d ) and three ( 3d ) dimensions . the 1d runs are used for testing the analytical model of [ sec : early ] . the 3d runs , in which the inflows enter along the @xmath90 direction , use periodic boundary conditions along the @xmath91 and @xmath92 directions and the same resolution in all three directions ( 200 grid cells ) , except for two runs that use a lower resolution ( 50 grid cells ) in the @xmath92 direction and open boundary conditions in @xmath91 and @xmath92 , that we used at early stages of this work for studying the dependence of the time for turbulence development on the @xmath90 and @xmath91 resolution . in the 3d runs , a random component in time and space , of amplitude @xmath93 , is added to the inflow speed at each cell of the boundary as a generic way of triggering the dynamical instabilities of the compressed layer without the biasing that might be introduced if we used , for example , fluctuations with a pre - defined power spectrum . being so fast and small - scale , most of this component is erased by diffusivity , so that the velocity inside the box fluctuates only by a few percent . no explicit term for heat conduction is included . in the ideal , absolutely non - conducting case , the fastest - growing mode of ti has a vanishing length scale @xcite . numerically , this would imply instability at the scale of the grid size , producing numerical instabilities . in the presence of thermal conductivity , the smallest unstable length scale ( the `` field length '' ) is finite @xcite , and sensitively dependent on the local temperature . if the field length is not resolved , numerical diffusion still stabilizes the smallest available scales , producing a `` numerical field length '' larger than the real one @xcite , even if explicit heat conduction is not included , and in this case the scale of the fastest - growing mode is resolution - dependent @xcite . in practice , this means that if the resolution is insufficient , the size of the structures _ formed by ti _ is not adequately resolved and determined . however , in the presence of turbulence , in which large - scale supersonic motions are the main drivers of density fluctuations rather than the development of ti , the _ statistics _ of the dense gas ( pressure and density distributions ) , which will be our main focus in this paper , are rather insensitive to the resolution even in simulations without thermal conduction @xcite . we conclude that , in spite of not including thermal conduction explicitly , the moderate resolution we have used should cause an overestimate of the sizes of the structures ( `` clumps '' ) formed , but should have no serious effect on the density and pressure statistics of the dense gas we discuss below . aware of the above limitations , we adopt the following convenient set of units for the simulations : @xmath94 @xmath39 , @xmath95 k , @xmath96 km s@xmath16 , where @xmath97 is the adiabatic sound speed at @xmath98 . all simulations start with inflows at @xmath99 k and @xmath100 @xmath39 , which correspond to the equilibrium warm phase at a pressure @xmath101 k @xmath39 . the inflow s velocity , given in terms of its mach number with respect to its sound speed of 9.89 km s@xmath16 , is varied between @xmath102 to @xmath103 . the physical size of the box deserves some special discussion . in [ sec : early ] we have noted that during the very early stages of the system s evolution , the shocks move outwards , stopping after roughly one cooling time , and having traversed one cooling length . in order to capture the full dynamics of the problem , we must make sure that the box has a large enough physical size that the shocks do not leave the simulation before stopping . we have found that actual cooling lengths are larger by factors of @xmath104 than indicated by fig.[fig : cool_time_length ] . moreover , at late stages of the evolution , the slab becomes heavily distorted and thickens considerably ( see also , e.g. , * ? ? ? so we use box sizes significantly larger than the cooling length given in fig.[fig : cool_time_length ] . as seen in this figure , the cooling length is a rapidly decreasing function of the inflow mach number @xmath14 in the range we have considered , and so simulations with smaller @xmath14 require larger box lengths . finally , the physical box size implies a physical time unit given by @xmath105 , we take as a fiducial simulation one with @xmath9 and a box length @xmath106 pc . the coefficients of the cooling function for the fiducial case in code units are 561.641 for 15 k @xmath107 k , 4.6178 for 141 k @xmath108 k , 1.0244 for 313 k @xmath109 k , and 4.7432 for @xmath110 k. other box sizes ( and derived time units ) are obtained by simply multiplying the cooling and heating coefficients by the same factor as that for the box size , because the only simulation parameter on which the cooling rates depend is the time unit . table [ tab : runs ] summarizes the main parameters of the runs we consider below , in which the runs are denoted mnemonically by their inflow mach number @xmath14 and their box size @xmath111 as mn.nlnn , where n is an integer . we denote 1d runs by the suffix `` 1d '' , high - resolution versions of another run by the suffix `` hr '' , and simulations with lower resolution in the @xmath92 direction , by the suffix `` lz '' , for `` low-@xmath92 '' . before we discuss the results of the numerical simulations , it is important to discuss the resolution limitations of the simulations we present here and their effects , in order to be able to usefully interpret their results . the cold layer is estimated to be very thin , @xmath112 pc for @xmath113 , for which @xmath114 , when taking @xmath115 myr . moreover , we have seen ( fig . [ fig : ncold_minfl ] ) that the front speed @xmath116 is a decreasing function of @xmath14 , and therefore the cold layer is thinner for higher @xmath14 . this means that only run m1.03l64 - 1dhr should barely resolve ( with @xmath117 grid cells ) the cold layer at this time . in practice the layer is not quite resolved even at this time , since apparently at least 89 zones are necessary to resolve the layer ( see below ) . this implies that the thickness of the cold layer during the early , non - turbulent stages depends on the resolution , and is exaggerated , in some cases grossly , by our simulations . figure [ fig : mach0.96 ] shows profiles along the @xmath90-axis at half the @xmath91 and @xmath92 extensions of the @xmath90-velocity ( _ solid _ line ) , density ( _ dash - dotted _ line ) , temperature ( _ dahed _ line ) and pressure ( _ dotted _ line ) for runs m1.03l64 - 1d ( _ left panel _ ) and m1.03l64 - 1d - hr ( _ right panel _ ) at @xmath118 myr . these runs differ only in the resolution used . it is clear that the cold layer is thinner in the high - resolution case . the fact that the cold layer thickness is initially resolution - limited also implies that the density within it is artificially reduced , since the layer is thicker than it should , and the total mass in the layer is determined only by the accretion rate , which is independent of the resolution . in turn , this implies that the pressure is below its physical value , as clearly seen in fig . [ fig : mach0.96 ] . the pressure in the cold layer can only reach the equilibrium value with its surroundings at later times when the slab thickness is enough for the density to reach its physical value . this happens at times @xmath119 16.0 and 8.0 myr for runs m1.03l64 - 1d and m1.03l64 - 1d - hr , respectively , at which the slab can be considered to start being well resolved . this condition is shown in fig.[fig : mach0.96_interm ] , in which the peak density is seen to have converged in the two runs at different times . however , since the total mass in the layer is independent of the resolution , we expect a similar situation for the layer s column density . indeed , the thickness of the slab in run m1.03l64 - 1d at @xmath118 myr ( fig . [ fig : mach0.96 ] ) is @xmath120 grid zones , or @xmath121 pc , with a maximum density of @xmath122 @xmath39 . the column density through the layer is measured at @xmath123 @xmath83 . this can be compared with the estimates of [ sec : early ] . for @xmath124 , the model gives @xmath125 km s@xmath16 and @xmath126 @xmath39 , so that after @xmath127 myr , the layer thickness should be @xmath128 pc , and the column density @xmath129 . thus , the column density in the simulation is within a @xmath130% error of the predicted value , even though the layer is @xmath131 times thicker than predicted at this time . for run m1.03l64 - 1d - hr , the slab thickness is @xmath132 zones , or @xmath133 pc , and the maximum density is @xmath134 @xmath39 , so the layer is close to being well resolved , but not quite yet . the measured column density is @xmath135 , within a @xmath136% error of the predicted value . the column densities measured in the simulations at this early time are lower than the predicted value at this time because the layers in the simulations have actually bell - shaped profiles , rather than the flat ones that develop at later times , at which the agreement with the model predictions is much better ( see [ sec : mod_sim_comp ] and fig.[fig : m0.96_mod_sim_compar ] ) . from the above discussion we conclude that the resolution of the simulations in this paper is generally insufficient for correctly resolving the structure inside the cold layer during the initial stages of the evolution . nevertheless , the column density of the cold layer is reasonably recovered even from the early stages of the simulations . with respect to the late stages of the evolution , we have found empirically that as the resolution is increased , the instability in the compressed layer develops earlier . this is shown in fig . [ fig : mach0.96_fin ] , which depicts constant-@xmath92 slices of the density ( _ upper panels _ ) and of the pressure ( _ lower panels _ ) for runs m1.03l64-lz ( _ left _ ) and m1.03l64-hr - lz ( _ right _ ) at the last computed step ( @xmath137 myr ) . it is seen that the instability has already developed in the high - resolution run , but not quite yet in the low - resolution one , although in this run the cold layer is already significantly bent , and presumably turbulence will be generated soon after this time . since our 3d simulations are still far from resolving the true field length in the various ism regimes , the effect of numerical diffusion roughly approaches that of the correct heat conduction as the resolution is increased . therefore , the observed trend of decreasing turbulence development time with increasing resolution implies that the timescales reported in this paper should be considered upper limits only . concerning the structures formed by the turbulence in the compressed layer , once it has developed fully , the main limitation introduced by the limited resolution is that the size of the cold structures will be artificially bounded from below by the numerical cell size , but we expect no other serious limitation . the structures are mainly formed by the turbulent flow , not by ti , and therefore they can in principle form with a variety of sizes , rather than at the characteristic scales of the ti , which indeed are very small . finally , we note that , as seen from figs . [ fig : m1.2_movie ] and [ fig : m2.4_movie ] , in runs m1.2l32 and m2.4l16 , as the turbulent layer thickens , its bounding shock eventually touches the inflow boundary . this happens at times @xmath138 myr for run m1.2l32 ( frame 32 in the corresponding animation , fig . [ fig : m1.2_movie ] , with the frame - count starting at frame zero ) , and @xmath139 myr for run m2.4l16 ( frame 11 in fig . [ fig : m2.4_movie ] ) . at this point , the simulations are in principle not valid anymore , as the interaction of the material in the turbulent layer with the inflowing gas ceases to be followed in full . in practice , however , we have found that the statistical properties of the simulation are not affected by the collision of the shock with the boundary . this fact can be understood because the inflow boundary conditions effectively act as outflow conditions for the material reaching the boundary from the inside of the box , and so effectively this gas just leaves the box as through standard outflow conditions , while fresh gas continues to flow in through these boundaries , and to interact with the gas remaining in the simulation . thus , in [ sec : turb_gas ] we discuss various physical and statistical properties both at the time the shocks leave the simulations , and at the time when the statistics become stationary . the predictions of the analytical model of [ sec : early ] can be compared with the results of the 1d numerical simulations , which resolve the layer at not - too - late times , and in which the slab does not become unstable . to be consistent with the hypotheses of the model , we consider times late enough that the outer shock has essentially stopped . as illustrations , we discuss the cases with @xmath11 and 1.2 . the left panel of fig . [ fig : m0.96_mod_sim_compar ] shows the density field in the central 6 pc of run m1.03l64 - 1dhr at times 26.6 myr and 79.8 myr . we can see that the right - hand side of the dense cold layer has moved 0.8 pc , from @xmath140 pc to @xmath141 pc . this gives a velocity @xmath142 . we also read the density inside the layer as 255@xmath5 . the right panel of fig . [ fig : m0.96_mod_sim_compar ] shows the pressure throughout the entire simulation at @xmath143 myr . the maximum value of the pressure , occurring at the center of the dense layer , is @xmath144 k@xmath5 . this can be compared with the model predictions for @xmath145 , which are @xmath146 k@xmath5 , @xmath147 and @xmath114 , in excellent agreement with the measurements . figure [ fig : m1.2_mod_sim_compar ] shows the corresponding comparison in the case @xmath148 . in this case , the left panel shows that the front moves from @xmath149 pc at @xmath150 myr to @xmath151 pc at @xmath152 myr , implying @xmath153 . we also read a cold layer density of @xmath154 . in turn , the right panel shows the pressure , whose maximum value is @xmath155 . for a mach number @xmath156 , our model gives @xmath157 k@xmath5 , @xmath158 and @xmath159 ( cf . [ fig : ncold_minfl ] ) . this is again in good agreement with the results of the simulation . we conclude that the analytical model and the 1d numerical simulations are consistent with each other , giving us confidence in both , and confirming the physical scenario ( an outer shock front and an inner condensation front ) , in which thin sheets form during the initial stages of the compression . in this section we now discuss the evolution of the compressed and the cold layers in runs with various inflow mach numbers . in the animations of figures [ fig : m0.96_movie ] , [ fig : m1.2_movie ] and [ fig : m2.4_movie ] we respectively show the evolution of runs m1.03l64 , m1.2l32 and m2.4l16 . the spacing between frames in the animations is @xmath160 code time units ( cf . table [ tab : runs ] for the time unit in each run ) , which corresponds to @xmath161 2.66 , 1.33 and 0.66 myr , respectively . the animations in figures [ fig : m1.2_movie ] and [ fig : m2.4_movie ] show constant-@xmath92 cross - sections of the simulation at different , equally - spaced @xmath92 values , with @xmath162 grid cells , to illustrate the evolution of the thickness and structure of the dense `` layer '' ( which , in the late stages , becomes a turbulent cloud ) . the animation in fig.[fig : m0.96_movie ] shows instead a transluscent projection of the density evolution in run m1.03l64 , to illustrate the fragmentation process of the thin dense layer . the still figures of the printed version show selected panels from the animations to also illustrate these features . from the animations and figures , several points are noticed . in general , we see that , after the formation of the thin cold layer , the latter begins to fragment into a filamentary honeycomb pattern , with denser clumps at the sites where the filaments intersect . subsequently , the density structures begin to move on the plane of the dense layer , merging and forming larger clumps . however , apparently the random fluctuations in the inflow velocity cause the merging clumps to collide with slight offsets , which therefore cause the layer to thicken and to develop vorticity . ultimately , the motion in the cold thin layer appears to completely destabilize the entire thick shocked slab , and fully developed turbulence ensues . in fig . [ fig : m1.2_movie ] it is interesting to note that the density peaks ( `` clouds '' ) appear surrounded by a low - pressure interface in the pressure images . this region probably corresponds in our simulations to a numerical effect , with the pressure gradient being compensated there by numerical diffusion , although in real clouds this may correspond to a conducting interface . this interface , however , is not apparent in the pressure images for run m2.4l16 ( fig . [ fig : m2.4_movie ] ) . these results suggest that sharp phase transitions between the warm and cold gas still exist at @xmath156 @xcite , but tend to be erased at @xmath163 , as also suggested by the pressure histograms discussed in [ sec : rho_p_dists ] . nevertheless , even in run m1.2l32 , the pressure in the `` clouds '' is seen to fluctuate significantly , because of their dynamical origin , and in general they are not all at the same pressure nor at uniform pressure inside . an important datum of the simulations is the time they require for attaining a saturated turbulent state . this can be defined in practice as the timescale for reaching a stationary shape of the statistical indicators , such as the density and pressure histograms . we find that this occurs at @xmath164 myr for run m1.2l32 ( frame 36 in the corresponding animation , fig . [ fig : m1.2_movie ] ) , and at @xmath165 myr for run m2.4l16 ( frame 11 in fig.[fig : m2.4_movie ] ) . run m1.03l64 does not seem to have reached a stationary state by the end of the integration time we have considered , @xmath166 myr . as mentioned in [ sec : num_res ] , these times are larger than those at which the shocks reach the boundary , and therefore in the next section we discuss both . [ sec : rho_p_dists ] in this section we concentrate on the cases with @xmath148 and @xmath167 , as they are the ones in which significant amounts of turbulence can develop within realistic timescales in the shocked layer . figure [ fig : rho_hists]a shows the density histograms of run m1.2l32 at @xmath168 myr ( _ solid line _ , frame 32 in the animation of fig . [ fig : m1.2_movie ] ) , when the shock touches the boundary , and at @xmath169 myr , when the statistics become stationary ( _ dotted line _ , frame 36 in the animation ) . the histograms at the two times are very similar , although the former one contains slightly lower numbers of grid cells with intermediate- and high - densities , indicative of the not yet completely stationary turbulent regime at that time . similarly , fig . [ fig : rho_hists]b shows the density histograms at the corresponding times for run m2.4l16 ( @xmath170 myr , frame 11 , _ solid line _ , and @xmath171 myr , frame 16 , _ dotted line _ ) . the same trends are observed for this run . the histograms of both runs have narrow and tall peaks at the density of the unperturbed inflowing streams ( the density of the warm phase , @xmath172 ) . the histogram of the mildly supersonic run m1.2l32 is significantly bimodal , although it extends to densities @xmath173 @xmath39 , well into what is normally associated with typical molecular cloud densities . the peak of the high - density maximum of the distribution is between @xmath174 and @xmath175 . in contrast , the histogram for the strongly supersonic run m2.4l16 has a less pronounced bimodal character , and extends at roughly constant height to densities typical of the cold phase , to then start decreasing at higher densities , to reach values close to 1000 @xmath39 . thus , a higher inflow mach number tends to erase the signature of bistability of the flow by increasing its level of turbulence , in agreement with the studies by @xcite , @xcite and @xcite . of particular interest is the pressure distribution in these simulations , as a means to understanding the overpressured nature of molecular clouds . figures [ fig : p_hists]a and b show the pressure histograms of runs m1.2l32 ( at @xmath138 myr ) and run m2.4l16 ( at @xmath176 myr ) . shown are the histograms for the entire simulation ( _ dotted _ lines ) , for gas with densities 10 @xmath177 @xmath39 ( _ dashed _ lines ) , which we will refer to as the `` intermediate - density '' gas ( idg ) , and for gas with densities @xmath178 @xmath39 ( _ solid _ lines ) , which we will refer to as `` high - density '' gas ( hdg ) . the histograms for all components are normalized to the total number of points . we will generally identify the idg with the cnm , although we do not use this nomenclature because in some cases the idg is highly pressurized , and thus warm rather than cold . on the other hand , the hdg can be identified with the `` molecular '' component , although we will maintain the notation hdg because we do not follow the chemistry nor have cooling appropriate for the molecular gas . the times shown for figs.[fig : p_hists]a and b are the same as the eralier times in figures [ fig : rho_hists]a and b. figures [ fig : p_vs_rho]a and b in turn show the distribution of points in the simulations in the @xmath179 plane , overlayed on the @xmath36 vs.@xmath22 curve , at the same times as fig . [ fig : p_hists ] . several points are worth noting in these figures . first , again the pressure of the unperturbed inflow gas is noticeable as the sharp peak at @xmath180 ( @xmath101 k @xmath39 ) in the pressure histograms of the entire simulations . however , the global shape of the two histograms is very different . the total histogram for run m1.2l32 ( fig.[fig : p_hists]a ) is quite narrow , with a total width of slightly over one order of magnitude , and moreover has a second , wide maximum centered at @xmath181 k @xmath39 . instead , the total pressure histogram of run m2.4l16 ( fig . [ fig : p_hists]b ) extends over two orders of magnitude and has a nearly lognormal shape ( except for the sharp peak noted above ) , rather than the bimodal shape of the @xmath148 run . all of this indicates a more developed state of the turbulence in run m2.4l16 . focusing on the pressure distributions for the idg and hdg , we note that , in run m1.2l32 , the pressure in the idg is mostly confined to lower values than those of the hdg , in the range 10004000 k @xmath39 , with a very low tail extending to @xmath182 k@xmath5 . the most probable value of the pressure of this gas practically coincides with that of the unperturbed inflowing wnm , with @xmath183 k @xmath39 . the hdg , on the other hand , is systematically overpressured with respect to the idg and the unperturbed wnm , with @xmath184 10,000 . it is interesting that the broad high - pressure maximum in the total histogram overlaps with the range of pressure values of the `` molecular '' gas . this maximum corresponds to the shocked , low - density gas that is in transit from the warm to the cold phases , crossing the unstable density range , @xmath185 , as can be seen in fig.[fig : p_vs_rho]a . _ the pressure coincidence between the hdg and the shocked unstable gas strongly suggests that the hdg is in pressure balance with the shocked , compressed gas , rather than with the ambient wnm , explaining its higher - than - average pressure . _ in the case of run m2.4l16 , some new features arise . most notably , the pressure distribution of the wnm and the idg now extend beyond that of the hdg ( see also fig . [ fig : p_vs_rho]b ) . this is somewhat surprising , and probably indicates that a substantial fraction of the pressure in the shocked gas is converted into kinetic energy of the hdg by the dynamical instabilities , rather than into internal energy . this picture is supported by the fact discussed in the next subsection , that the ratio of turbulent kinetic - to - internal energy density is highest in the hdg in this high-@xmath14 run . it is also interesting that the pressure distribution of the hdg extends over a very similar range ( @xmath186 ) as that of the mildly supersonic case m1.2l32 , in spite of the much higher pressures present in the idg and wnm distributions . table [ tab : energies ] summarizes the energy densities of the inflowing wnm , the idg and the hdg for the two runs under consideration . for the idg and the hdg , the table additionally gives the rms speed , rms mach number , and mean temperature . the data for each run are given at two times : the time at which the shocks first touch the @xmath90-boundaries , and the ( later ) time at which the density and pressure histograms become stationary ( cf . [ sec : global_evol ] ) . we see that the statistics for the idg indiceate that indeed it is slightly more turbulent at the later times for both runs , with larger velocity dispersions , rms mach numbers and lower mean temperatures ( indicating larger densities due to stronger compressions ) . the statistics for the hdg , however , are nearly indistinguishable at the two times . note that the velocity statistics reported in table [ tab : energies ] refer to the total velocity dispersion of these components , and thus includes the bulk motions of the moving dense gas parcels . although we have not measured it here , it has consistently been reported by various groups @xcite that the _ internal _ velocity dispersion of the dense gas regions is subsonic . we do not attempt these measurments here , and defer the task for future papers using higher resolution simulations . it can be seen from table [ tab : energies ] that in run m1.2l32 both the idg and the hdg have higher internal than kinetic energy densities . the opposite is true for run m2.4l16 , in which the kinetic energy density in these two components is 2.53.5 times larger than the internal energy density . note in fact that the kinetic energy density in the turbulent idg and hdg is larger than that in the inflowing streams . this does not constitute any violation of energy conservation , since the volume occupied by the idg and the hdg is small compared with that of the diffuse wnm . finally , from the rms speed and mach number data for the idg and hdg in the two runs we see that their motions are in general transonic ( in the idg ) or supersonic ( in the hdg ) with respect to their own sound speeds . the rms mach number in the hdg is lower than typical values for molecular clouds , but this can be attributed mainly to the relatively low resolution of the simulations , causing the velocities and the density fluctuations to be somewhat damped at the scales of the dense gas by numerical diffusion . this causes lower velocities and higher temperatures , thus lowering the mach number . moreover , we do not model the transition from atomic to molecular hydrogen , so that the simulations do not account for the reduction of the sound speed upon the formation of molecules . but we see that the velocity dispersion , @xmath187 would correspond to mach numbers @xmath188 in molecular gas at @xmath189 k. our results from both the qualitative analysis and the numerical simulations show that during the early stages of evolution , the collision of wnm streams may form thin sheets of cnm , whose properties are described in [ sec : early ] and [ sec : mod_sim_comp ] and fig . [ fig : ncold_minfl ] . since the sheets last the longest at low inflow mach numbers , we consider the results of simulation m1.03l64 - 1dhr , reported in [ sec : mod_sim_comp ] . according to this simulation , the outward velocity of the cold layer boundary is @xmath190 km s@xmath16 , implying that the layer thickness is @xmath191 { \rm ~pc},\ ] ] and , with a number density @xmath192 , its column density is @xmath193 { { \rm ~cm}^{-2}}.\ ] ] the pressure in the cold layer is @xmath194 k @xmath5 , and therefore its temperature is @xmath195 k. these values are interestingly similar to those derived by ( * ? ? ? * hereafter ht03 ) ( see also * ? ? ? * ) for cloud `` a '' of @xcite , of @xmath196 , @xmath197 k , @xmath198 and thickness @xmath17 pc . in general , the column densities in fig . [ fig : ncold_minfl ] are within the range of the values reported in table 5 of ht03 . these similarities suggest that sheets such as those reported by ht03 can be formed by transonic compressions in the wnm , as modeled by our simulations . two important remarks are in order . first , @xcite assigns to this cloud a characteristic time scale of @xmath199 yr , on the basis of an observed line - of - sight turbulent velocity component @xmath0 km s@xmath16 and the thickness of 0.05 pc , while our estimates above are for a time of 1 myr after the compression started . this apparent discrepancy may be resolved as follows . from our fig . [ fig : mach0.96_interm ] we see that , although the velocity at the center of the slab is very close to zero , it rapidly increases in the transition front , since the velocity of the gas right outside the cold layer is close to @xmath200 . thus , sampling the gas out to sufficiently distant positions from the collision center should pick up higher velocities . we can investigate this effect in the 1d simulations , which by construction are not turbulent , to determine the linewidths that can be produced by the inflow alone . our simulations do not yet resolve well the cold slab at 1 myr , although run m1.03l64 - 1dhr begins to resolve it at @xmath201 myr ( cf . [ sec : num_res ] ) . the line profile of the slab at this time can be approximated by the mass - weighted velocity histogram for gas with @xmath202 k , using a velocity resolution of @xmath203 in order to crudely mimic blurring by thermal broadening . the resulting profile is shown in fig.[fig : m1.03_rho_v_hist ] . a central line with fwhm @xmath6 is seen , with broad wings of width @xmath204 , in reasonable agreement with the observations . moreover , we have found that the linewidth is nearly invariant over time , as a consequence of the fact that the greatest contribution comes from the material in the interface between the dense slab and its surrounding medium . thus , we expect the line profile at @xmath205 myr to be a good estimate of that at @xmath206 myr . these results suggest that the observed linewidths of the thin cnm sheets can be almost entirely accounted for by the accretion onto the sheet . moreover , since none of our 3d simulations develop turbulence by times as early as @xmath0 myr , these results suggest that the observed linewidths are not representative of internal motions , but of the accretion onto the sheets . the second remark is that ht03 estimated the number density in the sheets they observed from the spin temperature , assuming a pressure @xmath207 k@xmath5 . our model and simulations both suggest that the sheets are actually at a significantly higher pressure than the mean interstellar value , due first to the outer shock and then to the deceleration towards the center . this would raise their density estimates by factors @xmath208 , in agreement with the fact that our densities are in general larger , although making the observed sheets even thinner . if our identification of the cold layers in our simulations with the thin cnm sheets reported by ht03 is correct , then this implies that they can be the `` little sisters '' of molecular clouds , produced by compressions that are not strong enough to rapidly develop turbulence nor produce very dense gas , with the only difference in interpretation with respect to @xcite being that the 1-km s@xmath16 linewidth does not imply rapid destruction of the sheet , but instead just represents the velocity of the gas entering the sheet . the appropriate destruction time is that required for the _ development _ of turbulence in the cold layer which , as we have seen , is a rapidly varying function of the inflow mach number , but is in general greater than 5 myr in the cases we have investigated . the results of [ sec : late ] are complementary to those presented by @xcite and @xcite . in particular , the latter authors presented a physical setup very similar to ours , at higher resolutions in two - dimensions , and recognized three instabilities that may be at play in the problem , namely ti , ntsi and also the kelvin - helmholz instability , with the former one working to create the dense cold layer and the latter two working to produce disordered motions . in their study , these authors focused on the competition between these instabilities , the mass distribution of the cold clumps , and the generation of vorticity in the cold gas . in our study , we have focused primarily on the pressure of the cold gas , as a step in understanding the overpressured nature of molecular clouds , the fractions of thermal and kinetic energies in the cold gas , and the rapidity of turbulence development as a function of the inflow mach number . in summary , our work , taken together with those previous studies strongly suggests that various physical properties of molecular clouds , such as rms velocities of a few @xmath209 , densities of several hundred @xmath5 , and thermal pressures several times larger than the mean interstellar values , can be produced during the formation stages of the clouds , without the need for external energy sources , other than the ones that produced the large - scale compression . in this paper we have studied the process of cloud formation by large - scale stream collisions in the wnm , presenting a simple analytical study of the initial stages , and numerical simulations of the whole process . the analytical model and high - resolution 1d simulations show that thin sheets of cold neutral medium can be formed within the shock - bounded layer by transonic compressions ( @xmath210 ) on timescales @xmath0 myr . these sheets are reminiscent of those reported by @xcite , with column densities @xmath211 , thicknesses @xmath2 pc , temperatures @xmath3 k and pressures @xmath212 k@xmath5 . in our simulations the sheets have linewidths @xmath204 , again comparable to the value reported by @xcite , although these linewidths do not correspond to turbulent motions in the layer , but rather to the inflowing speed of the gas . also , our sheets are at higher pressures than those assumed by @xcite , implying that their number densities are higher than those authors estimated . at later times , the simulations show that the boundary of the cold layer becomes dynamically unstable , through an nts - like instability that occurs even though the flow is always subsonic inside the shocked layer . eventually , fully developed turbulence arises , on times that can be as short as @xmath213 myr for inflow mach numbers @xmath167 , and as long as over 80 myr for @xmath145 . in this turbulent regime , the highest - density gas ( hdg , with @xmath214 ) is always overpressured with respect to the mean wnm pressure by factors 1.55 . since our simulations do not include self - gravity , this result shows that dense , overpressured gas can be readily formed by dynamical compressions in the wnm , possibly explaining at least part of the excess pressure in molecular clouds . the intermediate - density gas ( idg , with @xmath215 < 100 $ ] ) has a significant pressure scatter at a given value of the density , which increases with inflow mach number , so that at @xmath167 a significant fraction of the idg has pressures larger than those of the hdg . in general , the ratio of internal to kinetic energy density of the inflowing gas changes as the gas is incorporated into the idg and the hdg , with a tendency to increase as one considers higher - density gas in the fully turbulent regime . finally , the density probability distribution tends to lose the bimodal signature of thermal bistability as the inflow mach number is increased . our calculations are not free of caveats , with the most notable ones being our neglect of molecular cooling , thermal conduction , magnetic fields , and self - gravity . the relatively low resolutions we have used imply that the structure within the dense gas is not resolved . finally , due to somewhat small box sizes used as a compromise between acceptable resolution at the early times and sufficient spatial coverage at the late , turbulent times , the simulations cease to be valid in a strict sense ( because the bounding shocks leave the box ) before the turbulence becomes stationary , although their statistical properties at later times do not seem to be affected by this fact . we plan to address these shortcomings in future papers . our results , together with those of previous groups @xcite suggest that the turbulence and at least part of the excess pressure in molecular clouds are generated during the compression that forms the clouds themselves , and that the cnm sheets reported by @xcite may be formed by the same mechanism , in cases where the compressions are only mildly supersonic . we are glad to acknowledge useful discussions with carl heiles and ethan vishniac , and a useful and thorough report by an anonymous referee . this work has received partial support from conacyt grant 36571-e to e.v .- s . , korea research foundation grant krf-2004 - 015-c00213 to d. r. , and from the french national program pcmi to t. p. the numerical simulations have been performed on the linux cluster at centro de radioastronoma y astrofsica of unam , funded by the above conacyt grant . we have made extensive use of nasa s ads and lanl s astro - ph abstract services . heiles , c. 2004 , in milky way surveys : the structure and evolution of our galaxy , proceedings of asp conference # 317 , ed . d. clemens , r. shah & t. brainerd ( san francisco : astronomical society of the pacific ) , 323 vzquez - semadeni , e. , gazol , a. , snchez - salcedo , passot , t. & f. j .. 2003b , in turbulence and magnetic fields in astrophysics , lecture notes in physics , vol . e. falgarone , and t. passot , ( berlin : springer ) , 213 cccccccccccc m2.4l16 & 3d & 200 & 200 & 2.4 & 23.7 & 16 & @xmath216 & 0.08 & 0.67 + m1.2l32 & 3d & 200 & 200 & 1.2 & 11.9 & 32 & @xmath217 & 0.16 & 1.33 + m1.2l32 - 1d & 1d & 1000 & & 1.2 & 11.9 & 32 & @xmath217 & 0.032 & + m1.03l64 & 3d & 200 & 200 & 1.03 & 10.2 & 64 & @xmath218 & 0.32 & 2.67 + m1.03l64-lz & 3d & 200 & 50 & 1.03 & 10.2 & 64 & @xmath218 & 0.32 & + m1.03l64-hr - lz & 3d & 400 & 50 & 1.03 & 10.2 & 64 & @xmath218 & 0.16 & + m1.03l64 - 1d & 1d & 1000 & & 1.03 & 10.2 & 64 & @xmath218 & 0.064 & + m1.03l64 - 1d - hr & 1d & 4000 & & 1.03 & 10.2 & 64 & @xmath218 & 0.016 & + cccccccccccc m1.2l32 & inflow & @xmath219 & @xmath220 + ( @ 42.6 myr ) & idg & @xmath221 & @xmath222 & 1.3 & 1.25 & 45 . + & hdg & @xmath223 & @xmath224 & 0.72 & 1.0 & 21 . + + ( @ 47.9 myr ) & idg & @xmath225 & @xmath226 & 1.4 & 1.3 & 45 . + & hdg & @xmath223 & @xmath227 & 0.73 & 1.1 & 21 . + + + m2.4l16 & inflow & @xmath219 & @xmath228 + ( @ 7.37 myr ) & idg & @xmath229 & @xmath230 & 2.8 & 2.3 & 92 . + & hdg & @xmath231 & @xmath232 & 1.7 & 2.4 & 21 . + + ( @ 10.7 myr ) & idg & @xmath233 & @xmath234 & 3.1 & 2.6 & 78 . + & hdg & @xmath231 & @xmath232 & 1.7 & 2.4 & 21 .

typically percent of the total baryonic mass in luminous elliptical galaxies is in the form of cooled interstellar gas . although the mass contributed by cooled gas is small relative to the mass of the old stellar system in these galaxies , it is almost certainly concentrated within the optical effective radius where it can influence the local dynamical mass . however , the mass of cooled gas can not be confined to very small galactic radii ( ) since its mass would greatly exceed that of known central mass concentrations in giant ellipticals , normally attributed to massive black holes . we explore the proposition that a population of very low mass , optically dark stars is created from the cooled gas . for a wide variety of assumed radial distributions for the interstellar cooling , we find that the mass of cooled gas contributes significantly ( percent ) to stellar dynamical mass to light ratios which , as a result , are expected to vary with galactic radius . however , if the stars formed from cooled interstellar gas are optically luminous , their pertubation on the the mass to light ratio of the old stellar population may be reduced . cooling mass dropout also perturbs the local apparent x - ray surface brightness distribution , often in a positive sense for centrally concentrated cooling . in general the computed x - ray surface brightness exceeds observed values within , suggesting the presence of additional support by magnetic stresses or non - thermal pressure . the mass of cooled gas inside is sensitive to rate that old stars lose mass , but this rate is nearly independent of the initial mass function of the old stellar population . .2 in

astro-ph0002385

perhaps the most perplexing and long - standing problem associated with galactic and cluster cooling flows is the uncertain physical nature and spatial distribution of the gas that cools . the apparent absence of large masses of cooled gas in elliptical galaxies has led some to argue that little or no cooling actually occurs and to postulate some source of heating that offsets the radiative losses in x - ray emission . but the energy required to balance radiative losses is prohibitively large and appropriate heating sources may not be universally available . if the expected radiative cooling actually occurs , two questions arise : ( 1 ) what is the nature of the objects that condense from the hot gas ? and ( 2 ) where is most of the cooled mass located in the galaxy ? regarding the first question , a variety of physical arguments support the hypothesis , or even the inevitability , of low mass star formation ( fabian , nulsen & canizares 1982 ; thomas 1986 ; cowie & binney 1988 ; vedder , trester , & canizares 1988 ; sarazin & ashe 1989 ; ferland , fabian & johnstone 1994 ; mathews & brighenti 1999 ) . here we shall address the second question in the context of cooling flows in elliptical galaxies where the known stellar mass and light profiles strongly constrain the spatial distribution of cooled gas . we adopt the generally accepted hypothesis that only stars of very low mass ( e.g. @xmath5 @xmath6 ) form in cooling flows ( e.g. ferland , fabian & johnstone 1994 ) , so that the mass to light ratio of the young stellar population formed from the cooling gas is essentially infinite . in view of the difficulties we encounter with this hypothesis , described below , it seems more likely that the stellar population formed from cooled gas extends to somewhat more massive stars that are optically luminous . our gas dynamical models for the evolution of hot interstellar gas in giant ellipticals indicate that the origin of the gas varies with galactic radius . most of the gas in the inner , optically luminous regions originates from the ejected envelopes of evolving stars ; gas in the outer halo is supplied by cosmological secondary infall or tidal acquisitions from neighboring galaxies ( mathews & brighenti 1998b ) . circumgalactic gas around massive ellipticals is enriched by type ii supernovae that accompanied early star formation . the variability of circumgalactic gas among luminous ellipticals is responsible for some of the enormous dispersion in x - ray luminosity @xmath7 among ellipticals of similar optical luminosity @xmath8 ( mathews & brighenti 1998a ) . since the hot interstellar gas in a bright elliptical emits observable x - rays , it is clearly losing energy . however , as the gas loses energy it is compressed toward the galactic center by gravitational forces and @xmath9 work maintains the high temperatures observed , @xmath10 k , producing a galactic cooling flow . the positive interstellar temperature gradients typically observed within a few effective radii are often cited as evidence of radiative cooling in a cooling flow , but this cooling is due instead to the mixing of somewhat cooler , locally virialized gas ejected from stars with hotter gas arriving from larger galactic radii ( mathews & brighenti 1998b ; brighenti & mathews 1998 , 1999a ) . if large entropy fluctuations are present in the hot gas , catastrophic cooling can occur at any radius in the flow . regions of low entropy ( low temperature , high density ) radiate more and cool sooner . the amplitude distribution of entropy fluctuations in the interstellar gas determines the radius where cooling mass dropout occurs in the cooling flow . for example , if the entropy in some region in the flow is only slightly less than in the ambient flow , the differential radiative cooling will be small and the region will cool out of the flow at small galactic radii ; conversely , localized regions with entropy much lower than the ambient flow cool rapidly and deposit their mass at large radii . some possible sources of interstellar entropy variations are stellar winds , explosions of type ia supernovae , non - uniform snii heating at early times , and mergers with small , gas - rich galaxies . the total rate that mass cools and drops out of the flow is closely related to the x - ray luminosity @xmath7 . the x - ray luminosity can be approximately expressed as the product of the total cooling rate @xmath11 and the enthalpy per gram in the hot gas , or @xmath12 here we have used data from the giant virgo elliptical ngc 4472 : @xmath13 k ; @xmath14 ; @xmath15 . if @xmath7 and @xmath16 are reasonably constant over the hubble time , a mass @xmath17 @xmath6 of cold gas is expected to condense from the hot ism somewhere within ngc 4472 . although this mass is very large , it is only about 4 percent of the total stellar mass in ngc 4472 today . the mass that cools can therefore be ignored if it is widely distributed throughout the galactic volume . however , the central concentration of h@xmath18 emission in ellipticals ( e.g. macchetto et al . 1996 ) suggests that the cooling is concentrated toward the galactic center where the interstellar density is highest and the bulk of the x - ray energy is emitted . the motivation of this paper is to explore a variety of options for the mass dropout profile of cooled gas in bright ellipticals appropriately constrained by the known radial distributions of total stellar and non - baryonic mass . the radial mass dropout profile of cooled gas can not be determined from first principles because the distribution and amplitude of the entropy and magnetic fluctuations in the hot gas are unknown and difficult to evaluate from simple physical arguments . nevertheless , the total mass of cooled gas inside an effective ( half - light ) radius @xmath3 must be consistent with the mass to light ratio determined from stellar velocities and with the total mass inferred from x - ray observations within @xmath3 . assuming that the stellar mass to light ratio is uniform with radius , the stellar mass @xmath19 and the x - ray mass @xmath20 appear to be in nearly perfect agreement for two bright virgo ellipticals in the range @xmath21 ( brighenti & mathews 1997a ) . because of constraints on the mass distribution of cooled gas provided by x - ray and stellar dynamical observations , galactic cooling flows provide a critical venue for testing the physics of mass deposition in cooling flows . the mass of cold ( @xmath22 k ) gas @xmath23 actually observed in ellipticals is many orders of magnitude less than the total cooled mass estimated above . for example , neither hi nor h@xmath24 gas has been observed in ngc 4472 , only upper limits , @xmath25 @xmath6 ( bregman , roberts & giovanelli 1988 ; braine , henkel & wiklind 1988 ) . if stars form they must either be indistinguishable from the old stellar population or non - luminous . we explore here the possibility that most of the cooled gas forms dense baryonic clouds or stars that are dark at optical and radio frequencies . the very low mass stars advanced by ferland , fabian & johnstone ( 1994 ) satisfy this invisibility criterion , while the star formation models of mathews & brighenti ( 1999 ) indicate that ( luminous ) stars of mass @xmath26 @xmath6 can form in galactic cooling flows . x - ray studies indicate significant masses of cold , absorbing gas in cluster and galactic cooling flows ( e.g. white et al . 1991 ; allen et al . 1993 ; fabian et al . 1994 ; allen & fabian 1997 ; buote 1999 ) , but these results are inconsistent with the absence of radio frequency emission from the cold gas ( braine & dupraz 1994 ; donahue & vogt 1997 ) and should be regarded as controversial until this inconsistency is resolved . even taken at face value , the total mass of cooled gas implied by the x - ray absorption in cluster cooling flows is typically only a small fraction of the total mass that should have cooled in a hubble time ( allen & fabian 1997 ; wise & sarazin 1999 ) , implying that most of the cooled gas may have formed stars . if cooled gas forms into small stars , these stars will have apogalatica near their point of origin where they will spend most of their orbital time . we shall assume that the gas mass that cools and drops out of the flow contributes optically dark ( stellar ) mass at the radius where the cooling occurred . in the following we describe a series of gas dynamical calculations for the evolution of x - ray emitting interstellar gas over the hubble time and investigate a variety of assumptions about the radial distribution of optically dark cooled gas . to be specific , we compare our models with the well - observed massive elliptical ngc 4472 . we find that the mass of cooled gas contributes significantly to dynamical mass to light determinations within @xmath3 based on stellar velocities . the e2 elliptical ngc 4472 is a luminous , slowly rotating [ @xmath27 galaxy in the virgo cluster . with an adopted distance @xmath28 mpc , its optical luminosity is @xmath29 @xmath30 and its half light or effective radius @xmath31 is @xmath32 kpc ( faber et al . 1989 ) . the total stellar mass @xmath33 @xmath6 is found from the mass to light ratio @xmath34 determined by van der marel ( 1991 ) with a two - integral stellar distribution function . this mass to light ratio is appropriate to the galactic region within about 0.4@xmath3 where stellar velocities are well determined , although the mass determined from x - ray observations suggests that @xmath35 remains constant to at least @xmath3 ( brighenti & mathews 1997a ) . if @xmath35 is spatially constant , the stellar mass also has a de vaucouleurs profile . within a central core or break radius @xmath36 ( @xmath37 pc ) the stellar density profile flattens ( faber et al . 1997 ) , but we shall not consider this small feature here . ngc 4472 contains a central black hole of mass @xmath38 @xmath6 ( magorrian et al . 1998 ) . the total mass distribution in luminous ellipticals can most easily be determined from the radial variation of density and temperature in the hot interstellar gas , assuming hydrostatic equilibrium . figure 1 illustrates the interstellar density and temperature profiles in ngc 4472 . the filled circles in figure 1 are _ einstein _ hri observations ( trinchieri , fabbiano & canizares 1986 ) and open circles are rosat hri and pspc data from irwin & sarazin ( 1996 ) ( also see forman et al . 1993 ) . the @xmath39 and @xmath40 profiles have been fit with analytic curves as described by brighenti & mathews ( 1997a ) . hydrostatic equilibrium in the hot interstellar gas is an excellent approximation since the cooling flow velocity is very subsonic . the total mass interior to radius @xmath41 determined from x - ray observations is @xmath42 where @xmath43 is the proton mass and @xmath44 is the mean molecular weight . the last term , representing the possibility of magnetic pressure @xmath45 , is negative if @xmath46 as seems likely . if the magnetic term is important but not included , the total mass will be underestimated . assuming @xmath47 , the total mass @xmath20 in ngc 4472 is shown with a solid line in figure 1b . in the outer halo , @xmath48 , the total mass is dominated by the dark halo . the dark halo mass distribution in ngc 4472 can be approximated with an nfw halo profile ( navarro , frenk & white 1996 ) of virial mass @xmath49 @xmath6 although the observed halo is somewhat less centrally peaked than nfw ( brighenti & mathews 1999a ) . within @xmath3 the contribution of the dark nfw halo mass in the model is small ; for example at @xmath50 the total mass to light ratio is @xmath51 , only 10 percent greater than the dynamic value 9.2 determined in @xmath52 . it is remarkable that the total mass found from the x - ray data @xmath20 is nearly identical to the expected dynamical mass @xmath19 ( based on stellar velocities and @xmath53 ) in the range @xmath21 ( figure 1 ) . an almost identical agreement in this radius range is indicated by x - ray observations of another bright virgo elliptical , ngc 4649 ( brighenti & mathews 1997a ) . in this important region the hot gas is confined by the _ stellar _ potential . the excellent agreement of the stellar and x - ray masses supports the consistency of two radically different mass determinations : from stellar velocities and from the radial equilibrium of hot interstellar gas . the apparent agreement of the x - ray and stellar masses in this range of galactic radii also indicates that the hydrostatic support of the hot gas is not strongly influenced by local magnetic fields and rotation . however , it is not obvious why @xmath35 would be constant with galactic radius , particularly when the cooling dropout mass is considered , and why the agreement between stellar dynamic and x - ray masses no longer obtains in the central regions @xmath54 . in this central region the total mass indicated by the x - ray observations in figure 1 is considerably less than the expected mass based on an assumed de vaucouleurs profile and constant mass to light ratio . this type of deviation could be due to magnetic or other non - thermal pressure in this region , to rotation , or to local cooling dropout in the hot interstellar gas . the lower mean temperature in cooling regions lowers the total apparent gas temperature and results in an underestimate of the total internal mass ( equation 1 ) . we assume that currently available _ einstein _ and rosat observations are accurate in the central region of ngc 4472 , @xmath54 . these observations have been reduced assuming no ( non - galactic ) photoelectric absorption by low temperature gas in the central regions . if x - ray absorption is present , the true hot gas density would be more centrally peaked than shown in figure 1 and the total mass indicated by the x - ray observations would increase . buote ( 1999 ) finds that two - temperature models fit the x - ray spectrum for ngc 4472 quite well . the two temperatures do not necessarily need to be spatially mixed ; they could also approximately represent the range of the radial temperature variation observed in ngc 4472 . in buote s two temperature model , only the cooler component ( @xmath55 kev located in @xmath56 ) requires an absorption column @xmath57 @xmath58 in excess of the galactic value . however , the influence of cold gas having this column density on the hot gas density plotted in figure 1 is small . for a worst case example , suppose that absorbing material with column density @xmath57 @xmath58 is in a disk oriented perpendicular to the line of sight and that this disk absorbs _ all _ x - rays from the back side of the galaxy . the radius of this disk would be @xmath59 pc if it contained the maximum mass @xmath60 @xmath6 allowed by co and hi observations ( bregman , roberts & giovanelli 1988 ; braine , henkel & wiklind 1997 ) or 20 kpc if it contained all of the gas that has cooled , @xmath61 @xmath6 . the x - ray surface brightness within the opaque disk would be reduced by 2 , but the corresponding gas density would be lowered by only @xmath62 since the volume emissivity @xmath63 . such a small correction in the density ( gradient ) could not account for the large mass discrepancy between the x - ray mass and the stellar dynamical mass shown in figure 1 within 0.1@xmath3 . the densities and temperatures in figure 1 were determined from x - ray data assuming the abundance of the hot gas is uniformly solar . since the gas is likely to be more metal rich at smaller galactic radii ( matsushita 1997 ; brighenti & mathews 1999b ) , an allowance for this gradient would tend to lower the derived density gradient and the internal mass , _ increasing _ the discrepancy in @xmath64 in figure 1 by a small amount . in the following discussion we shall ignore the relatively small possible influence of absorption or metallicity gradients on the results shown in figure 1 the hydrodynamical models we use in this paper are similar to those in our recent papers ( e.g. brighenti & mathews 1999a ) so we provide only a brief review here . hot interstellar gas in ellipticals has a dual origin : ( i ) mass loss from an evolving old stellar population and ( ii ) secondary infall into the overdensity perturbation that formed the galaxy group within which the elliptical formed by early merging events . for a given set of cosmological parameters , dark and baryonic matter flow toward an overdensity region . the dark matter forms an nfw halo , growing in size with time . spherical geometry is assumed . within the accretion shock at time @xmath65 gyr , when enough baryons have accumulated , we form the current de vaucouleurs stellar profile and release the energy of all type ii supernovae according to a salpeter imf ( slope : @xmath66 , mass limits : @xmath67 and @xmath68 @xmath6 ) . all stars greater than 8 @xmath6 produce type ii supernovae each of energy @xmath69 ergs . we assume that a fraction @xmath70 of this energy is delivered to the internal energy of the gas . we have shown ( brighenti & mathews 1999b ) that such a galaxy formation scheme can work in a variety of cosmologies : flat or low density , with or without a lambda term . the evolution of gas within the optical effective radius @xmath3 , of most interest here , is insensitive to these cosmological parameters . for simplicity therefore we assume a simple flat universe with @xmath71 , @xmath72 km s@xmath73 mpc@xmath73 and @xmath74 . we characterize the dark halo with an nfw profile ( navarro , frenk & white 1996 ) having a virial mass @xmath49 @xmath6 at the current time @xmath75 gyrs . the models we discuss here are identical to the standard model of ( brighenti & mathews 1999a ) except we now use a finer central spatial zoning ( 65 pc for the innermost zone ) , a snii efficiency @xmath70 , and a mass `` dropout '' function @xmath76 with more adjustable parameters ( see below ) . our objective is to seek solutions of the gasdynamical equations including mass dropout that jointly satisfy several observational constraints at time @xmath77 : ( i ) the observed hot gas density , temperature and x - ray surface brightness profiles , ( ii ) the known dynamical mass in the galactic center usually attributed to a massive black hole , ( iii ) the apparent dynamical mass to light ratio @xmath34 determined in @xmath52 , and ( iv ) an apparent internal mass @xmath20 in @xmath78 based on equation ( 1 ) that agrees with the constant-@xmath79 de vaucouleurs profile as shown in figure 1 . the baryonic component in our models has a complex evolution . much of the initial baryonic mass is consumed in creating the stellar system . when the type ii supernova energy is released , a significant mass of gas is expelled as a galactic wind . after these early events , the interstellar gas is re - established and sustained by stellar mass loss and by inflow of circumgalactic gas ( secondary infall ) , most of which was previously enriched and expelled by snii . we assume that the stars form during a short epoch that can be described by a single burst salpeter imf as discussed above . the stellar mass loss rate for this imf varies as @xmath80 where @xmath81^{-1.3}$ ] sec@xmath73 . although galactic stars form at @xmath82 gyr , their mass loss contribution to the ambient interstellar gas is assumed to begin at a later time , @xmath65 gyrs . since galactic stars have been enriched by supernova ejecta , the single burst model can not be strictly correct , but our approximation @xmath83 is consistent with several early starbursts closely spaced in time and allows for metal enrichment of old galactic stars that are not in the first single - burst population . if the de vaucouleurs profile is a result of largely dissipationless merging , some or most of the star formation must have occurred at a time @xmath84 before the important merging events at @xmath65 gyrs . by taking @xmath85 we reduce by @xmath86 @xmath6 the total amount of gas ejected by stars within the galactic potential . we recognize the inconsistencies in these approximations of complex stellar formation and dynamical processes that are poorly understood . however , once the galaxy is formed , we follow the interstellar gas dynamical evolution in full detail , conserving mass and energy . continued heating by type ia supernova is assumed to vary inversely with time , snu@xmath87 snu@xmath88@xmath89 , where the current rate , snu@xmath90 snia per 100 yrs per @xmath91 @xmath30 , is near the lower limit of observed values , snu@xmath92 ( cappellaro et al . 1997 ) , as required to maintain the low interstellar iron abundance . for the models discussed here the equation of continuity includes a `` mass dropout '' term : @xmath93 where @xmath94 is the time for gas to cool locally by radiative losses at constant pressure ( see , e. g. sarazin & ashe 1989 ) . the cooling is assumed to be instantaneous without advection in the cooling flow , i.e. @xmath95 , although in practice this inequality may not always be satisfied . while this type of cooling dropout has been widely used in past models , there is no adequate physical model for mass dropout . clearly , the gas must cool somewhere the emission of x - rays indicates a large net energy ( and mass ) loss from the interstellar medium . when small regions of low entropy cool , the pressure remains constant since the sound crossing time is much less than the flow time . following fabian , nulsen & canizares ( 1982 ) and ferland , fabian , & johnstone ( 1994 ) , we assume that cooled gas converts to a second population of optically dark , low mass stars . regarding h@xmath18 emission as a tracer for the cooling gas , mathews & brighenti ( 1999 ) have shown that the cooling occurs at a multitude ( @xmath96 ) of cooling sites distributed throughout the inner galaxy and that only stars with mass @xmath97 can form at each cooling site . this is supported by the observed absence of young massive stars and snii in elliptical galaxies . nevertheless , in the following discussion we assume that the maximum stellar mass in the dropout population is sufficiently low that the optical light from these stars is unobservable . we expect cooling and low mass star formation to be concentrated within at least 2 kpc , which is the observed extent of h@xmath18 emission in ngc 4472 ( macchetto et al . while the region containing optical emission lines provides a natural guideline for selecting the dropout profile , we consider a wider range of constant or variable dropout coefficients @xmath76 parameterized by @xmath98 which concentrates the cooling within radius @xmath99 . note that even when @xmath100 is constant , the mass dropout term is spatially concentrated , @xmath101 . in a series of recent papers in this journal we have presented evolutionary cooling flow models for ngc 4472 that agree quite well with the observed distributions of interstellar gas density , temperature and metallicity ( brighenti & mathews 1999a ; 1999b ) , particularly at intermediate and large radii . since the total mass of cooled gas in these calculations was only a few percent of the stellar mass , the gravitational contribution of cooled gas was ignored , even in models including mass dropout . for a better understanding of the inner galaxy , we now include the gravity of all gas , both hot and cold , except when specifically noted . in this section we begin with models having @xmath102 everywhere , so that cooling to low temperatures occurs in a small central region , then we investigate models in which @xmath100 is constant throughout the cooling flow . we begin by considering a perfectly homogeneous interstellar flow in which all the gas reaches the central computational zone ( @xmath104 pc ) or its neighboring zones where it then cools to @xmath105 k. for comparison we discuss two cases : in model 1 we ignore the gravity of this cooled gas ; in model 2 ( and subsequent models ) we include its gravitational influence on the flow . the mass of gas that cools into the central gridzone is dynamically equivalent to a massive black hole , but we do not necessarily regard our calculation as a realistic model for black hole formation . in figure 2a we illustrate the radial variation of gas density and temperature for models 1 and 2 after the interstellar gas has evolved to @xmath106 gyrs . several global parameters for these and subsequent models are listed in table 1 . when the gravity of the cooled gas is considered ( model 2 ) , the interstellar gas within a few kpc of the galactic center is compressed and sharply heated in the local potential . such hot central thermal cores are not generally observed ( but see colbert & mushotzky 1998 ) . the central mass in both models , @xmath107 @xmath6 is about 16 times larger than the black hole mass found in ngc 4472 , @xmath38 @xmath6 ( magorrian et al . for these reasons neither model 1 nor 2 provides a realistic interstellar mass distribution for this galaxy . rosat band x - ray surface brightness profiles @xmath108 for models 1 and 2 are shown in figure 2b and the total rosat band luminosities are listed in table 1 . for model 1 the x - ray brightness peaks strongly in the galactic center , diverging from the observations within about 3 kpc ; it was just this sort of disagreement that initially led to the assumption of distributed mass dropout in cooling flows ( thomas 1986 ; vedder , trester , & canizares 1988 ; sarazin & ashe 1989 ) . if the efficiency of heating by early snii were lowered , models 1 and 2 could be made to agree better with @xmath109 observations in @xmath110 kpc , but the computed @xmath109 would rise even further above the observations within a few kpc . the x - ray luminosities @xmath7 for models 1 and 2 listed in table 1 are unreliable in part because of numerical inaccuracies caused by the extremely steep variation of gas temperature and density in the central 2 or 3 computational zones . this numerical difficulty is common to all cooling flows that proceed to the very center of the galaxy before cooling , for any reasonable central grid spacing . since more @xmath9 work is done on the flow in model 2 ( where the gravity of cooled gas is included ) , we expect that @xmath7 should also be larger than for model 1 . the opposite sense of the change in @xmath7 shown in table 1 evidently results from numerical inaccuracies near the central singularity where zone to zone variations are no longer linear . if the gas has not cooled before flowing into the core ( @xmath111 pc ) , as in models 1 and 2 , a significant fraction of the total x - ray emission should come from this very central region , again in disagreement with observations . in models 1 and 2 a sphere of cold ( @xmath112 k ) , dense gas accumulates in the central gridzones and grows in mass and size over the hubble time . this sphere is an unrealistic artifact of our computational assumptions . therefore , to explore further the central numerical difficulty with @xmath7 encountered in models 1 and 2 , we considered two additional models using higher spatial resolution ( radius of central zone is only 15 pc ) , both of which include the gravitational influence of cooled gas . the first model ( model 2.1 ) is similar to model 2 with @xmath102 in all zones . in the second model ( model 2.2 ) we set @xmath102 in all zones except the central zone where @xmath113 . model 2.2 is the appropriate limit of the series of models discussed below in which the cooling dropout is more and more centrally concentrated . while the flow in @xmath114 kpc is very similar in models 2.1 and 2.2 , the behaviour at smaller radii is quite different . for example the flow parameters for model 2.1 at 100 pc and @xmath75 gyrs are : @xmath115 k , @xmath116 @xmath117 , and @xmath118 km s@xmath73 . at the same radius and time for model 2.2 the flow is quite different : @xmath119 k , @xmath120 @xmath117 , and @xmath121 km s@xmath73 . at projected radius @xmath122 pc , the rosat x - ray surface brightness in model 2.2 is 20 times larger than model 2.1 and the total rosat x - ray luminosity integrated over the entire cooling flow in model 2.2 is larger by a factor 32 . the differences between these two models result from upstream propagation of information about the central boundary conditions made possible by subsonic flow near the origin . fortunately , these numerical and physical difficulties near the origin do not arise in more realistic cooling flows discussed below in which @xmath123 at larger galactic radii . while solutions of the gasdynamical equations provide the temperature as a function of physical radius @xmath39 , the observed temperature @xmath124 is an emission - weighted mean temperature along the line of sight at projected radius @xmath125 . for symmetric galaxies and at small galactic radii , these two temperatures are nearly identical because of the steep radial variation in x - ray emissivity . the variation of temperature with physical radius @xmath126 in the background flow is shown with light solid lines in figure 2a . in the presence of spatially distributed cooling dropout , the local temperature is an emission - weighted mean of the background ( uncooled ) gas and the cooling regions . cooling is assumed to occur at a large number of cooling sites where the gas remains in pressure equilibrium as it cools , apparently unrestricted by magnetic stresses ( see mathews & brighenti 1999 for details ) . the heavy solid lines in figure 2a show the mean apparent temperature @xmath127 including contributions from locally cooling regions , @xmath128 where @xmath16 is the background flow temperature and the slowly - varying functions @xmath129 are plotted in brighenti & mathews ( 1998 ) . note that the effective temperature is independent of the local gas density . the temperature that is actually observed is an average of @xmath130 along the line of sight ; this temperature @xmath131 is shown with dotted lines in figure 2a . lacking an acceptable physical model for spatially distributed mass dropout in galactic cooling flows , we are at liberty to choose any variation for the dropout coefficient @xmath76 , appropriately constrained _ a posteriori _ by the known dynamical mass from stellar velocities , the x - ray mass , the central black hole mass and the observed radial variation of hot gas density , temperature and x - ray surface brightness . it is natural to begin with constant @xmath100 solutions similar to those considered by sarazin & ashe ( 1989 ) in their steady state cooling flow solutions . for simplicity we assume that the mass of cooled gas remains at the cooling site where it contributes to the gravitational potential . in the central panels of figure 2a we illustrate the hot interstellar gas density and temperature that results after @xmath75 gyrs assuming uniform @xmath137 and 4 ; these are listed in table 1 as models 3 and 4 respectively . when @xmath100 is a constant independent of radius , the ratios of true to apparent values @xmath138 , @xmath139 and @xmath140 are also approximately uniform with galactic radius . here @xmath141 is the true total mass within @xmath41 and @xmath20 is the value that would be determined from x - ray observations of the models by assuming hydrostatic equilibrium ( as in fig . the influence of constant-@xmath100 mass dropout is similar at all galactic radii since the factors that convert background temperature and density to effective values in equations ( 3 ) and ( 4 ) depend only on @xmath39 which is slowly varying , not on @xmath40 . because of the enormous volume and time available to gas dropping out at large @xmath41 , the total mass of cooled gas ( listed in table 1 ) becomes very large in the outer galaxy . in figure 3 we compare the radial distribution of stellar and dark halo mass with the distributed dropout mass that results after 13 gyrs with @xmath137 and 4 . the de vaucouleurs `` stellar '' mass profile in figure 3 is constructed assuming uniform @xmath34 . as @xmath100 increases from 1 to 4 , the background hot gas density ( light solid lines in fig . 2a ) decreases and its radial gradient flattens , causing a rise in temperature to provide enough pressure to support the cooling flow atmosphere . note that the total dropped - out mass within @xmath142 decreases with larger @xmath100 ( figure 3 and table 1 ) . although more overall cooling dropout occurs as @xmath100 increases , most of this cooling occurs at very large @xmath41 and , somewhat paradoxically , less gas remains for dropout closer to the galactic center , @xmath56 . however , the effective ( i.e. apparent ) density ( heavy solid lines in fig . 2a ) is less sensitive to @xmath100 . the @xmath137 solution ( model 3 ) is preferred for its fit to observed temperatures while the @xmath113 solution ( model 4 ) agrees better with the observed density in @xmath143 kpc and almost exactly with the x - ray surface brightness ( fig . in addition , the centrally concentrated dropped out mass in the @xmath137 solution ( model 3 ) contributes a larger fraction of the dynamical mass in @xmath144 and the total mass within the central gridzone ( @xmath104 pc ) is almost equal to the known mass of the black hole in ngc 4472 ( table 1 ) . evidently , cooled masses in models with @xmath145 would exceed the central dark mass observed . for both values of @xmath100 considered , the dropped out mass listed in table 1 contributes appreciably to the total mass within @xmath146 . when both old stellar and dropout mass are included , the @xmath35 values at @xmath142 shown in table 1 12.19 and 11.21 exceed both the dynamical value @xmath34 found by van der marel ( 1991 ) within @xmath147 and the value @xmath148 in our models which includes a small additional contribution of non - baryonic dark matter . to quantify the influence of the dark baryonic dropout mass on the mass to light ratio we include in table 1 an entry for @xmath149 @xmath150 where @xmath151 @xmath6 is the combined mass of luminous stars and dark halo matter at @xmath142 and @xmath152 also includes the dropout mass within this radius . since our computed total mass @xmath153 exceeds the observed dynamical value 9.2 ( and also 10.18 ) , to be fully consistent we should have chosen @xmath154 for the old stellar population ( see below ) , provided @xmath35 for stars produced from the cooled gas is infinite as we have assumed . if these constant @xmath100 models are physically appropriate , the agreement between @xmath20 and @xmath19 in @xmath78 , as shown in figure 1 ( and for ngc 4649 ) , is surprising since @xmath155 and @xmath156 differ by 10 - 40 percent ( table 1 ) for the constant @xmath100 models . this suggests that the mass to light ratio of the dropout stellar population is not infinite as we have assumed , but comparable with that of the old stellar population . we now seek evolutionary gasdynamical solutions with variable dropout coefficients @xmath76 that strongly concentrate the mass dropout in the inner galaxy , @xmath56 . the limited spatial extent of optical emission lines in the cores of bright ellipticals suggests that cooling dropout occurs well within @xmath3 . for example , the h@xmath18 + [ nii ] image of ngc 4472 is observed out to approximately @xmath157 or 2 kpc ( macchetto et al . as hot interstellar gas cools in this region , its temperature pauses at @xmath158 k where the gas is heated and ionized by stellar uv radiation . therefore , we consider parameters @xmath99 and @xmath159 in equation ( 2 ) that concentrate the mass dropout in the central region , but not at the very center as in model 2 . the dropout parameters @xmath160 , @xmath99 and @xmath159 , are listed in table 1 for models 5 , 6 and 7 . the mass dropout in these three models is progressively more concentrated toward the galactic center . the results for model 5 , for which the dropout scale length is very large , @xmath161 kpc , are similar in most respects to those of model 4 except the massive dropout in the outer galaxy is no longer present . in particular , the mass to light ratio at @xmath142 in table 1 is very similar for models 4 and 5 . the current interstellar density and temperature variations for model 6 with @xmath162 kpc are shown in figure 2a and @xmath108 is plotted in figure 2b . the apparent density and @xmath109 ( heavy solid lines ) are considerably greater than observed values in @xmath163 kpc , although the projected apparent temperature is a reasonable fit to the ngc 4472 data . the radial distribution of dropout mass for model 6 is shown in figure 3 . from table 1 the total ( old stars , halo and dropout ) mass to light ratio at @xmath142 is @xmath164 , 29 percent higher than van der marel s value and 22 percent greater than the corresponding value for our background model galaxy . also shown in figures 2a and 2b are similar results for model 7 in which the mass dropout , now approximated with a gaussian , is concentrated within @xmath165 pc . model 7 is constructed so that most of the mass dropout occurs in @xmath166 , corresponding to the region of apparent disagreement between @xmath167 and @xmath168 in figure 1 . it is of interest that the gas density and @xmath109 within 1 kpc for model 7 , where the dropout is greatest , exceeds those of model 2 in which there is no dropout at all . this can be understood from the flow velocity distribution . for model 2 without distributed dropout the inward moving gas velocity increases through the entire region illustrated and shocks at @xmath169 kpc . in the presence of dropout , the flow velocity at corresponding radii in model 7 is much slower and reaches a maximum ( negative ) value near @xmath170 kpc then approaches zero subsonically at the origin . the density and @xmath109 enhancements in the background flow for model 7 at @xmath163 kpc are due to a local compression as the gas flows into slowly moving gas in the core . within about 3 kpc , both the background and apparent densities exceed the observations by a larger factor than those of model 6 and @xmath109 also peaks unrealistically in this same region ( fig . the dropout mass for model 7 shown in figure 3 equals that of the old population stellar mass at @xmath171 kpc ( @xmath172 ) . for these reasons model 7 seems less satisfactory than model 6 , but neither is as generally successful as model 3 ( @xmath137 ) . although increased mass dropout can decrease the x - ray surface brightness @xmath109 , as when uniform @xmath100 increased from 1 to 4 ( models 3 and 4 in fig . 2b ) , this is not always the case . when the dropout is concentrated more toward the galactic center , @xmath109 actually increases , as in the transition from model 6 to 7 . it is interesting to determine the influence of distributed dropout on the total apparent mass @xmath20 found from the model by assuming hydrostatic equilibrium ( equation 1 ) . due to the contribution of low temperature cooling regions to the total x - ray emission , @xmath39 and therefore @xmath20 is always lower than the true mass @xmath173 in cooling dropout regions . a difference in this sense is apparent in table 1 at @xmath146 for models 3 - 7 ; this is similar to the mass discrepancy in figure 1 at @xmath54 . the high central apparent gas density and surface brightness for model 6 shown in figures 2a and 2b are obvious problems for this model . however , the gas density can be reduced if a strong magnetic field or other non - thermal energy density is present in @xmath174 kpc . dynamically important magnetic fields may also be required to fit the x - ray data of ngc 4636 ( brighenti & mathews 1997a ) and may be generally expected in luminous ellipticals ( mathews & brighenti 1997 ; godon , soker & white 1998 ) . additional non - thermal pressure support is also implied for model 6 since the effect of central cooling dropout fails to lower the apparent mass @xmath20 below the actual mass @xmath175 as much as the observed deviation shown in figure 1 . like many bright ellipticals , ngc 4472 has a weak non - thermal radio source within the central @xmath176 kpc , indicating @xmath177 @xmath178 g ( ekers & kolanyi 1978 ) . for all calculated models in table 1 with distributed dropout , the total mass @xmath179 significantly exceeds the mass of the old stellar population plus dark halo @xmath180 throughout the inner galaxy , indicating that dropout material makes an important additional contribution to the total mass . however , the apparent mass @xmath167 determined from equation ( 1 ) is less than @xmath175 in the inner galaxy and , for model 6 , can also be less than @xmath181 . of particular interest is the region @xmath182 ( i.e. , @xmath183 ) in figure 1 . although the agreement in figure 1 is excellent in this region , for models 6 and 7 the total mass is larger than @xmath173 and values of @xmath184 in table 1 suggest that the dropped out mass contributes 25 - 35 percent of the total mass in this region . therefore , if the dropout mass is optically dark , the true stellar mass to light ratio of the old stellar population must be @xmath185 rather than the value @xmath34 found by van der marel which includes the dropout mass . to investigate such a possibly more self - consistent old stellar component , we consider model 8 based on the same @xmath76 used in model 6 , but with @xmath186 for the old stellar population . for additional consistency in model 8 , @xmath187 is increased by the ratio of assumed stellar mass to light ratios 9.2/6 = 1.53 as described below ; the total stellar mass ejected is identical to that in model 6 . figure 3 illustrates the dropout mass profile for model 8 . the apparent density , temperature and @xmath109 profiles for model 8 shown in figures 4a and 4b are very similar to those of model 6 , so this adjustment of the stellar @xmath35 has had little effect . the radial mass profiles of models 6 and 8 are compared in figure 5 . in this plot the open circles show the observed x - ray mass @xmath20 for ngc 4472 using equation ( 1 ) and the solid lines shows @xmath20 based on equation ( 1 ) using @xmath40 and @xmath39 from the models . the dashed lines show @xmath141 and the dotted lines are the stellar mass @xmath19 based on a de vaucouleurs profile with @xmath34 in the upper panel ( model 6 ) and @xmath186 in the lower panel ( model 8) . the superiority of model 8 is evident from the closer agreement between the x - ray mass data points for ngc 4472 and the solid line for that model . this agreement for model 8 would be even closer in the range @xmath188 if we had used a dark halo mass profile less centrally peaked than nfw . model 8 may provide the most self - consistent overall fit to the mass constraints for ngc 4472 ; if so , the old stars in ngc 4472 have a mass to light ratio @xmath185 , about @xmath189 percent lower than the mass to light ratio determined from stellar dynamics . in summary , none of our models is fully satisfactory in every respect . the total mass to light ratio ( including dropout mass ) in @xmath144 is 10 - 35 percent higher than the value for the underlying galaxy . however , in model 8 in which @xmath186 for the old stars , the difference between the x - ray mass determined from the models and ngc 4472 is appreciably reduced . nevertheless , the central apparent gas density and x - ray surface brightness in model 8 are still larger than observed , requiring additional non - thermal support . there is no independent theoretical justification for the dropout profile @xmath76 assumed in models 6 and 8 ; as explained earlier , the dropout distribution depends on unknown interstellar entropy fluctuations . nevertheless , for all models considered here the mass of cooled interstellar gas contributes significantly to the total mass and the dynamically determined mass to light ratio within the inner galaxy . in this series of calculations we have taken a census of all baryons involved in the evolution of a large elliptical galaxy : the original stellar component , the interstellar medium , and of most interest the small but troublesome mass of hot gas that cools over cosmic time . we have shown that the radial distribution of cooled interstellar gas influences dynamical and x - ray determinations of the total interior mass and the radial profiles of apparent density , temperature and x - ray brightness of the hot gas . in our models , cooled gas is slowly deposited in the central galaxy @xmath56 ( see figure 3 ) as indicated by the extent of observed h@xmath18 emission . since there is little or no direct observational evidence for the mass that has dropped out in ellipticals like ngc 4472 , the cooled mass must either be dark at optical and radio frequencies or indistinguishable from the old stars . low mass stars are an obvious and physically reasonable endstate for the cooled gas ( ferland , fabian & johnstone 1994 ; mathews & brighenti 1999 ) . we also suppose that the cooled gas remains at the dropout site where it contributes to the galactic potential . in an attempt to reduce the influence of cooling and cooled gas on the models , we have altered many of the model parameters . the chosen cosmology ( @xmath71 ; @xmath190 and @xmath191 , etc . ) or the baryon fraction @xmath192 have little influence on the total dropout mass . changing the times when the stars and the galactic potential form ( @xmath84 and @xmath193 ) or the spatial scale of the release of snii energy within reasonable limits have only a modest influence on the dropout mass that accumulates by time @xmath75 gyrs . increasing the interval @xmath194 between star and galaxy formation does reduce the total dropout mass , but this interval can not be too large since luminous ellipticals are observed at large redshifts . the cooling dropout would not be dramatically reduced if massive ellipticals were all only a few gyrs old since most of the stellar mass loss occurs just after star formation ; however , many or most luminous ellipticals are thought to be very old . perhaps the most effective way to preserve the excellent agreement between @xmath181 and @xmath167 in figure 1 in @xmath195 , without constraining the mass dropout profile @xmath76 , would be to reduce the total mass that has cooled over cosmic time . the most sensitive parameter influencing the cooled mass is the specific rate of stellar mass loss , @xmath187 . although the total mass of hot gas increases with time due to the continued influx of secondary infalling gas , the mass of hot gas within the optical galaxy originates mostly from stellar mass loss and the x - ray luminosity there scales as @xmath196 ( appendix b of tsai & mathews 1995 ) . computed models similar to those described here but with arbitrarily reduced @xmath187 fit the @xmath40 and @xmath108 data rather well at time @xmath77 and produce much less cooling dropout however , @xmath187 can not be lowered without also increasing the stellar mass . for all reasonable power law initial mass functions , @xmath187 varies inversely with the stellar mass to light ratio : @xmath197 { 1 \over ( m_*/l_b ) } \propto { 1 \over ( m_*/l_b)}.\ ] ] the physical explanation for the constancy of @xmath198 is that both @xmath8 ( dominated by post - main sequence stars ) and @xmath199 depend on the instantaneous rate that stars leave the main sequence so this imf - dependent factor cancels out ( e.g. renzini & buzzoni 1986 ) . to illustrate this result , we use the renzini - buzzoni procedure and plot in figure 6 the relationship between @xmath200 and @xmath35 at time @xmath75 gyrs for 84 power law imfs with slopes @xmath2010.6 , 0.8 , 1.0 , 1.2 , 1.4 , 1.6 , and 1.8 , each with lower and upper mass limits of @xmath2020.01 , 0.032 , 0.1 , 0.32 and @xmath20310 , 32 , 100 . the remarkable linearity in figure 6 shows that the quantity in square brackets in the equation above is almost invariant to large changes in the slope or mass limits for power law imfs . therefore , if @xmath187 is reduced by 2 or 3 in an attempt to reduce the total dropout mass , the stellar mass ( and @xmath35 ) must be increased by the same factor ; as a result the total mass ejected , @xmath204 , and the total dropout mass do not change . although massive ellipticals are not rotationally flattened , they do rotate significantly , e.g. @xmath205 for ngc 4472 ( faber et al . if the hot interstellar gas rotates in the same sense as the bulk of mass - losing stars , stars formed from the cooled gas should form into a disk of scale @xmath206 ( brighenti & mathews 1997b ) , although the development of such disks is likely to be suppressed by the mass dropout process . nevertheless , to the extent that the cooled gas has a disk - like distribution , its global influence on the stellar dynamics in @xmath52 would be less than if the same dark mass were distributed spherically . remarkably , there is no observational evidence at present for rotational flattening in the x - ray images of giant ellipticals like ngc 4472 . dynamical determinations of the mass to light ratio from stellar velocities reflect the entire mass within the stellar orbits , including non - baryonic mass . the nfw halo we use in our models for ngc 4472 agrees with x - ray observations in the extended halo but is slightly too massive near @xmath207 relative to the x - ray mass @xmath20 observationally determined for ngc 4472 . this may indicate that dark halos are less centrally peaked than nfw ( see kravtsov et al . the mass of our nfw halo model contributes about 10 percent to the dynamical mass to light ratio measured within @xmath142 , but the nfw profile is probably disturbed in this region . when the dominant baryonic mass in @xmath56 compressed to form the de vaucouleurs profile , we expect that the nfw halo was dragged inward and distorted . however , the dark halo cores of the earlier galactic condensations that merged to form the elliptical may have expanded due to starburst driven galactic winds . because of these various counteracting effects , the small non - baryonic contribution to dynamical mass to light determinations is uncertain . for all models studied here based on a wide variety of mass dropout profiles @xmath76 the mass of cooled interstellar gas contributes substantially to the total mass within @xmath147 where the stellar mass to light ratio is determined from stellar velocities . if optically dark low mass stars form from the cooled gas , the `` stellar '' mass to light ratios in the literature refer to two distinct stellar populations having radically different initial mass functions and spatial distributions . the stellar mass to light ratio of the original , optically luminous stellar population is lower than published values indicate . a complete solution of this problem requires a better understanding of the physics of star formation and the processes that control the stellar imf . we have assumed here that a salpeter imf provides a satisfactory approximation to the original single - burst star formation at early times . yet we argue that the younger dropout imf is strongly skewed toward low mass stars . the imf has evolved over time . it is possible therefore that the early , more intense mass dropout resulted in a more nearly salpeter - like imf , producing a fraction of currently observed luminous stars in ellipticals . if so , this early dropout would not contribute to the excess dropout mass that we find in our models , but would have already been included in the original de vaucouleurs population . high density , metal enriched stellar cores in elliptical galaxies may have derived from normal - imf star formation from early , more intense interstellar cooling . this is similar to assumptions made for dissipative galactic core formation from the convergence of gas following major mergers ( mihos & hernquist 1996 ) . while such notions can not be entirely dismissed , the approximate universality of the de vaucouleurs light profile among ellipticals may argue against a dual formation process for the radial distribution of luminous stars : violent relaxation and cooling flow dropout . throughout this discussion we have assumed that the stellar population formed from cooling flow dropout is optically dark . although there is little or no evidence that normal massive ob stars ( or snii ) are present in elliptical galaxies , it is possible that younger stars having masses up to @xmath26 @xmath6 are present and that such intermediate mass stars could form from the cooled gas ( mathews & brighenti 1999 ) . this type of dropout stellar population could contribute to the total optical light . if the mass to light ratio of dropout and old stellar populations are similar , the dropout component could be difficult to detect by the means we have discussed here and its perturbation on the observed @xmath35 would be greatly lessened . in this case the dropout population would introduce an additional radial light profile that would differ slightly from that of the old stellar population . the dropout mass profiles in figure 3 indicate that models 4 and 6 would be rather difficult to detect against the background stellar light . intermediate mass dropout stars could therefore provide a satisfactory resolution to the problems we have discussed here . the ensemble of elliptical galaxies is known to have global parameters that deviate slightly from the assumptions of virial equilibrium and homologous structure . the deviation of this fundamental plane relationship is in the sense that the dynamical mass to light ratio increases with galactic mass , @xmath208 ( dressler et al . 1987 ; djorgovski & davis 1987 ) . such a non - homologous deviation could in principle be produced by the small amount of cooled interstellar gas @xmath23 provided it increases appropriately with @xmath168 . to test this idea , we performed an identical hydrodynamic calculation for an elliptical galaxy having a mass one fourth that of ngc 4472 . the dark halo mass was also reduced by the same factor but the cosmological environment and mass dropout distribution were identical to those used for ngc 4472 , scaled to a smaller @xmath3 . we found that the amount of mass dropout @xmath23 is _ higher _ in smaller ellipticals relative to the total baryonic mass @xmath168 . this is opposite to the trend observed in the fundamental plane . however , if hot gas and dark matter in the outer halos of smaller ellipticals is tidally stripped in group environments , as suggested by mathews & brighenti ( 1998b ) , then the mass of cooled gas would be reduced and its effect on the fundamental plane would be reduced . nevertheless , explanations of the deviations of the fundamental plane from virial scaling must recognize the possible additional influence of dropout mass , regardless of the trend of @xmath209 with @xmath168 . \(1 ) if the hot interstellar gas cools only in the very center of ngc 4472 for @xmath210 gyrs , the total accumulated mass there would be @xmath211 times larger than the mass of the central black hole observed . if such large concentrated masses were generally present in bright ellipticals , interstellar gas in @xmath166 would be compressed and heated to temperatures @xmath212 kev . such hot thermal cores are not generally observed . we conclude that the cooling dropout in massive ellipticals must occur before the gas reaches the galactic center . the hypotheses of distributed mass dropout and low mass star formation were proposed many years ago ( fabian , nulsen & canizares 1982 ; thomas 1986 ; cowie & binney 1988 ; vedder , trester , & canizares 1988 ; sarazin & ashe 1989 ; ferland , fabian , & johnstone 1994 ) . however , these historical arguments were generally based on the notion that mass dropout would help reduce computed x - ray surface brightness profiles at small projected radii , as required by the observations , but we have shown here that in some cases enhanced dropout at small galactic radii can cause @xmath109 to increase , not decrease . ( the total bolometric x - ray luminosity @xmath7 should always be lower in distributed cooling models since the hot gas experiences only a fraction of the galactic potential . ) the best arguments for distributed cooling dropout are ( i ) limits on the central black hole mass and ( ii ) the absence of rotational flattening in x - ray images . \(2 ) we have considered a wide variety of possible mass profiles for the radial deposition of cooled interstellar gas in ngc 4472 . the dropout mass is assumed to be optically dark , consistent with the formation of very low mass stars . in every case the ( stellar plus dropout ) mass to light ratio at @xmath213 significantly exceeds the mass to light ratio determined from stellar velocities . if the dropout mass is optically dark , dynamical mass to light ratios in luminous ellipticals should be substantially enhanced by dark baryonic matter . in this case the true @xmath35 for luminous stars may be @xmath189 percent smaller than published values . \(3 ) the excellent agreement between the x - ray and `` stellar '' mass in ngc 4472 shown in figure 1 ( and also ngc 4649 ) in the range @xmath195 may be a coincidence if our estimates of the cooling flow dropout mass are correct and if this mass is non - luminous . \(4 ) dynamical mass to light determinations within @xmath3 refer to a superposition of two stellar populations : an old luminous population with a de vaucouleurs galactic profile and a younger population having a bottom - heavy imf and a different galactic mass profile . if the younger population is optically dark , the mass to light ratio is not likely to be constant with galactic radius within @xmath3 . \(5 ) it is not possible to reduce the total amount of mass deposited from the cooling flow simply by lowering the specific rate of stellar mass loss @xmath214 . for all reasonable power law initial mass functions we show that @xmath215 . for given @xmath8 , larger stellar masses @xmath19 must accompany lower values of @xmath200 so the total amount of mass ejected from stars @xmath216 is nearly independent of the imf . \(6 ) among the models we consider , those with centrally concentrated mass dropout perform best in minimizing the overall disagreement with the central @xmath35 and the x - ray determined mass in @xmath195 . constant @xmath100 models in which the dropout is proportional to the local gas emissivity at every radius deposit less mass in @xmath195 , but may have gas temperatures that are too low ( @xmath217 ) or central masses that are too large ( @xmath218 ) . \(7 ) even in the presence of mass dropout , the computed central interstellar gas densities and x - ray surface brightnesses @xmath108 are generally too large . such deviations would be reduced if the hot gas is partially supported in @xmath166 by magnetic or other non - thermal pressure associated with the extended radio source . \(8 ) if the stellar population formed from cooled interstellar gas extends to intermediate masses , @xmath26 @xmath6 , its mass to light ratio may blend with that of the older population . in this case the dropout mass would already be represented in the de vaucouleurs profile representing the stellar mass distribution in our models . dynamical determinations of @xmath35 would be a weighted mean of the two populations . if the dropout stellar population is luminous , some of the difficulties we have discussed here would be alleviated , but not those regarding @xmath108 . \(9 ) if the influence of rotation and non - thermal pressure can be understood , high resolution images of the central regions of elliptical galaxies using the _ chandra _ ( axaf ) satellite may detect the presence of dark baryonic dropout material and an accurate determination of the mass to light ratio of the old stellar population . thanks to karl gebhardt for providing useful information . studies of the evolution of hot gas in elliptical galaxies at uc santa cruz are supported by nasa grant nag 5 - 3060 and nsf grant ast-9802994 for which we are very grateful . fb is supported in part by grant murst - cofin 98 . allen , s. w. , fabian , a. c. , johnstone , r. m. , white , d. a. , daines , s. j. , edge , a. c. & stewart , g. c. 1993 , mnras 262 , 901 allen , s. w. & fabian , a. c. 1997 , mnras 286 , 583 braine , j. & dupraz , c. 1994 , a&a 283 , 407 braine , j. , henkel , c. & wiklind , t. 1997 , a&a 321 , 765 bregman , j. n. , roberts , m. s. & giovanelli , r. 1988 apj 330 , l93 brighenti , f. & mathews , w. g. 1997a , apj 486 , l83 brighenti , f. & mathews , w. g. 1997b , apj 490 , 592 brighenti , f. & mathews , w. g. 1998 apj 495 , 239 brighenti , f. & mathews , w. g. 1999a , apj 512 , 65 brighenti , f. & mathews , w. g. 1999b , apj ( in press ) buote , d. a. 1999 , mnras ( submitted ) cappellaro , e. , turatto , m. , tsvetkov , d. yu . , bartunov , o. s. , pollas , c. , evans , r. , & hamuy , m. 1997 , astron . & ap . 322 , 431 colbert , e. & mushotzky , r. f. 1998 , ( astro - ph/9809150 ) cowie , l. l. & binney , j. 1988 , apj 215 , 723 djorgovski , s. g. & davis , m. 1987 , apj 313 , 59 donahue , m. & vogt , m. 1997 , in _ galactic and cluster cooling flows _ , ed . n. soker , asp conf . series 115 , 48 dressler , a. , lynden - bell , d. , burstein , d. , davies , r. l. , faber , s. m. , terlevich , r. j. , & wegner , g. 1987 , apj 313 , 42 ekers , r. & kolanyi , c. 1978 , a&a 67 , 47 faber , s. m. , wegner , g. , burstein , d. , davies , r. l. , dressler , a. , lynden - bell , d. & terlevich , r. j. 1989 , apjs 69 , 763 faber , s. m. , tremaine , s. , ajhar , e. a. , byun , y - i , dressler , a. , gebhardt , k. , grillmair , c. , kormendy , j. , lauer , t. r. , & richstone , d. 1997 , aj 114 , 1771 fabian , a. c. , nulsen , p. e. j. & canizares , c. r. 1982 , mnras 201 , 933 fabian , a. c. , arnaud , k. a. , bautz , m. w. & tawara , y. 1994 , apj 436 , l63 ferland , g. j. , ferland , a. c. & johnstone , r. m. 1994 , mnras 266 , 399 forman , w. , jones , c. , david , l. , franx , m. , makishima , k. , & ohashi , t. 1993 , apj 418 , l55 godon , p. , soker , n. & white iii , r. e. 1998 , aj 116 , 37 irwin , j. a. & sarazin , c. l. 1996 , apj , 471 , 683 kravtsov , a. v. , klypin a. a. , bullock , j. s. & primack , j. r. 1998 , apj 502 , 48 macchetto , f. , pastoriza , m. , caon , n. , sparks , w. b. , giavalisco , m. , bender , r. & capaccioli , m. 1996 , a&as 120 , 463 magorrian , j. , tremaine , s. , richstone , d. , bender , r. , bower , g. , dressler , a. , faber , s. m. , gebhardt , k. , green , r. , grillmair , c. , kormendy , j. , & lauer , t. 1998 , aj 115 , 2285 mathews , w. g. & brighenti , f. 1997 , apj 488 , 595 mathews , w. g. & brighenti , f. 1998a , apj 493 , l9 mathews , w. g. & brighenti , f. 1998b , apj 503 , l15 mathews , w. g. & brighenti , f. 1999 , apj ( submitted ) matsushita , k. 1997 , ph.d . thesis , university of tokyo mihos , j. c. & hernquist , l. 1996 , apj 464 , 641 navarro , j. f. , frenk , c. s. , & white , s. d. m. 1996 , apj , 462 , 563 renzini , a. & buzzoni , a. 1986 , in _ the spectral evolution of galaxies _ eds . c. chiosi & a. renzini ( reidel : dordrecht ) , p. 195 sarazin , c. l. & ashe , g. a. 1989 , apj 345 , 22 thomas , p. a. 1986 , mnras 220 , 949 trinchieri , g. , fabbiano , g. , & canizares , c. r. 1986 , apj , 310 , 637 trinchieri , g. , fabbiano , g. , & kim , d .- w . 1997 , a&a 318 , 361 tsai , j. c. & mathews , w. g. 1995 , apj 448 , 84 van der marel , r. p. 1991 , mnras , 253 , 710 vedder , p. w. , trester , j. j. , & canizares , c. r. 1988 , apj 332 , 725 white , d. a. , fabian , a. c. , johnstone , r. m. , mushotzky , r. f. , & arnaud , k. a. 1991 , mnras 252 , 72 wise , m. w. & sarazin , c. l. 1999 , apj ( submitted ) crccccccccccc 1 & 0 & ... & ... & @xmath219 & 4.64 & @xmath220 & @xmath221 & @xmath222 & 0.00 & @xmath223 & 28.6 & 28.6 2 & 0 & ... & ... & @xmath224 & 4.66 & @xmath225 & @xmath226 & @xmath227 & 0.34 & @xmath228 & 21.7 & 21.7 3 & 1 & ... & ... & @xmath229 & @xmath230 & @xmath231 & @xmath232 & @xmath233 & 0.19 & @xmath234 & 18.7 & 36.4 4 & 4 & ... & ... & @xmath235 & @xmath236 & @xmath237 & @xmath238 & @xmath239 & 0.10 & @xmath240 & 8.8 & 42.6 5 & 4 & 15 & 1 & @xmath241 & @xmath242 & @xmath243 & @xmath244 & @xmath245 & 0.11 & @xmath240 & 13.9 & 27.7 6 & 4 & 2 & 1 & @xmath241 & @xmath246 & @xmath247 & @xmath248 & @xmath249 & 0.27 & @xmath250 & 22.1 & 33.5 7 & 4 & 0.8 & 2 & @xmath241 & @xmath251 & @xmath252 & @xmath253 & @xmath254 & 0.34 & @xmath255 & 29.1 & 43.3 8 & 4 & 2 & 1 & @xmath256 & @xmath257 & @xmath258 & @xmath259 & @xmath260 & 0.40 & @xmath261 & 20.0 & 28.1

color ( or categorical ) range reporting is a variant of the orthogonal range reporting problem in which every point in the input is assigned a _ color_. while the answer to an orthogonal point reporting query contains all points in the query range , the answer to a color reporting query contains only distinct colors of points in . in this paper we describe an-space data structure that answers one - dimensional color reporting queries in optimal time , where is the number of colors in the answer and is the number of points in the data structure . our result can be also dynamized and extended to the external memory model .

1306.5029

in the orthogonal range reporting problem , we store a set of points @xmath5 in a data structure so that for an arbitrary range @xmath6\times \ldots\times [ a_d , b_d]$ ] all points from @xmath7 can be reported . due to its importance , one- and multi - dimensional range reporting was extensively studied in computational geometry and database communities . the following situation frequently arises in different areas of computer science : a set of @xmath8-dimensional objects @xmath9 must be preprocessed so that we can enumerate all objects satisfying @xmath10 for arbitrary @xmath11 , @xmath12 . this scenario can be modeled by the orthogonal range reporting problem . the objects in the input set can be distributed into _ categories_. instead of enumerating all objects , we may want to report distinct categories of objects in the given range . this situation can be modeled by the color ( or categorical ) range reporting problem : every point in a set @xmath5 is assigned a color ( category ) ; we pre - process @xmath5 , so that for any @xmath6\times \ldots\times [ a_d , b_d]$ ] the distinct colors of points in @xmath7 can be reported . color range reporting is usually considered to be a more complex problem than point reporting . for one thing , we do not want to report the same color multiple times . in this paper we show that complexity gap can be closed for one - dimensional color range reporting . we describe color reporting data structures with the same space usage and query time as the best known corresponding structures for point reporting . moreover we extend our result to the external memory model . [ [ previous - work . ] ] previous work . + + + + + + + + + + + + + + we can easily report points in a one - dimensional range @xmath13 $ ] by searching for the successor of @xmath14 in @xmath5 , @xmath15 . if @xmath16 is known , we can traverse the sorted list of points in @xmath5 starting at @xmath17 and report all elements in @xmath18 $ ] . we can find the successor of @xmath14 in @xmath5 in @xmath19 time @xcite ; if the universe size is @xmath20 , i.e. , if all points are positive integers that do not exceed @xmath20 , then the successor can be found in @xmath21 time @xcite . thus we can report all points in @xmath18 $ ] in @xmath22 time for @xmath23 . henceforth @xmath3 denotes the number of elements ( points or colors ) in the query answer . it is not possible to find the successor in @xmath24 time unless the universe size @xmath20 is very small or the space usage of the data structure is very high ; see e.g. , @xcite . however , reporting points in a one - dimensional range takes less time than searching for a successor . in their fundamental paper @xcite , miltersen et al . showed that one - dimensional point reporting queries can be answered in @xmath25 time using an @xmath26 space data structure . alstrup et al . @xcite obtained another surprising result : they presented an @xmath1-space data structure that answers point reporting queries in @xmath25 time and thus achieved both optimal query time and optimal space usage for this problem . the data structure for one - dimensional point reporting can be dynamized so that queries are supported in @xmath25 time and updates are supported in @xmath27 time @xcite ; henceforth @xmath28 denotes an arbitrarily small positive constant . we refer to @xcite for further update - query time trade - offs . solutions of the one - dimensional point reporting problem are based on finding an arbitrary element @xmath29 in a query range @xmath30 $ ] ; once such @xmath29 is found , we can traverse the sorted list of points until all points in @xmath30 $ ] are reported . therefore it is straightforward to extend point reporting results to the external memory model . janardan and lopez @xcite and gupta et al . @xcite showed that one - dimensional color reporting queries can be answered in @xmath31 time , both in the static and the dynamic scenarios . muthukrishnan @xcite described a static @xmath1 space data structure that answers queries in @xmath25 time if all point coordinates are bounded by @xmath4 . we can obtain data structures that use @xmath1 space and answer queries in @xmath32 or @xmath33 time using the reduction - to - rank - space technique . no data structure that answers one - dimensional color reporting queries in @xmath34 time was previously known . a dynamic data structure of mortensen @xcite supports queries and updates in @xmath35 and @xmath36 time respectively if the values of all elements are bounded by @xmath4 . recently , the one- and two - dimensional color range reporting problems in the external memory model were studied in several papers @xcite . larsen and pagh @xcite described a data structure that uses linear space and answers one - dimensional color reporting queries in @xmath37 i / os if values of all elements are bounded by @xmath1 . in the case when values of elements are unbounded the best previously known data structure needs @xmath38 i / os to answer a query ; this result can be obtained by combining the data structure from @xcite and reduction of one - dimensional color reporting to three - sided\times [ 0,c]$ ] and @xmath30\times [ c,+\infty]$ ] are three - sdied queries . ] point reporting @xcite . in another recent paper @xcite , chan et al . described a data structure that supports the following queries on a set of points whose values are bounded by @xmath1 : for any query point @xmath39 and any integer @xmath3 , we can report the first @xmath3 colors that occur after @xmath39 . this data structure can be combined with the result from @xcite to answer queries in @xmath2 time . unfortunately , the solution in @xcite is based on the hive graph data structure @xcite . therefore it can not be used to solve the problem in external memory or to obtain a dynamic solution . [ [ our - results . ] ] our results . + + + + + + + + + + + + as can be seen from the above discussion and table [ tab : res ] , there are significant complexity gaps between color reporting and point reporting data structures in one dimension . we show in this paper that it is possible to close these gaps . in this paper we show that one - dimensional color reporting queries can be answered in constant time per reported color for an arbitrarily large size of the universe . our data structure uses @xmath1 space and supports color reporting queries in @xmath2 time . this data structure can be dynamized so that query time and space usage remain unchanged ; the updates are supported in @xmath27 time where @xmath20 is the size of the universe . the new results are listed at the bottom of table [ tab : res ] . our internal memory results are valid in the word ram model of computation , the same model that was used in e.g. @xcite . in this model , we assume that any standard arithmetic operation and the basic bit operations can be performed in constant time . we also assume that each word of memory consists of @xmath40 bits , where @xmath20 is the size of the universe . that is , we make a reasonable and realistic assumption that the value of any element fits into one word of memory . furthermore , we also extend our data structures to the external memory model . our static data structure uses linear space and answers color reporting queries in @xmath41 i / os . our dynamic external data structure also has optimal space usage and query cost ; updates are supported in @xmath42 i / os . [ cols="<,<,^,^,^,^ " , ] we start by describing a static data structure that uses @xmath1 space and answers color reporting queries in @xmath2 time . all elements of a set @xmath5 are stored in a balanced binary tree @xmath43 . every leaf of @xmath43 , except for the last one , contains @xmath44 elements , the last leaf contains at most @xmath44 elements , and every internal node has two children . for any node @xmath45 , @xmath46 denotes the set of all elements stored in the leaf descendants of @xmath47 . for every color @xmath48 that occurs in @xmath46 , the set @xmath49 ( @xmath50 ) contains the minimal ( maximal ) element @xmath51 of color @xmath48 . the list @xmath52 contains the @xmath44 smallest elements of @xmath49 in increasing order . the list @xmath53 contains the @xmath44 largest elements of @xmath50 in decreasing order . for every internal non - root node @xmath47 we store the list @xmath52 if @xmath47 is the right child of its parent ; if @xmath47 is the left child of its parent , we store the list @xmath53 for @xmath47 . all lists @xmath52 and @xmath53 , @xmath45 , contain @xmath1 elements in total since the tree has @xmath54 internal nodes . we define the middle value @xmath55 for an internal node @xmath47 as the minimal value stored in the right child of @xmath47 , @xmath56 where @xmath57 is the right child of @xmath47 . the following _ highest range ancestor _ query plays a crucial role in the data structures of this and the following sections . the answer to the highest range ancestor query @xmath58 for a leaf @xmath59 and values @xmath60 is the highest ancestor @xmath47 of @xmath59 , such that @xmath61 ; if @xmath18=\emptyset$ ] , the answer is undefined . the following fact elucidates the meaning of the highest range ancestor . [ fact : eluc ] let @xmath62 be the leaf that holds the smallest @xmath63 , such that @xmath64 ; let @xmath65 be the leaf that holds the largest @xmath63 , such that @xmath66 . suppose that @xmath67\not=\emptyset$ ] for some leaf @xmath59 and @xmath47 is the answer to the highest range ancestor query @xmath58 . then @xmath47 is the lowest common ancestor of @xmath62 and @xmath65 . let @xmath68 denote the lowest common ancestor of @xmath62 and @xmath65 . then @xmath62 and @xmath65 are in @xmath68 s left and right subtrees respectively . hence , @xmath69 and @xmath68 is not an ancestor of @xmath47 . if @xmath68 is a descendant of @xmath47 and @xmath68 is in the right subtree of @xmath47 , then @xmath70 . if @xmath68 is in the left subtree of @xmath47 , then @xmath71 . we will show that we can find @xmath47 without searching for @xmath62 and @xmath65 and answer highest range ancestor queries on a balanced tree in constant time . for every leaf @xmath59 , we store two auxiliary data structures . all elements of @xmath72 are stored in a data structure @xmath73 that uses @xmath74 space and answers color reporting queries on @xmath72 in @xmath2 time . we also store a data structure @xmath75 that uses @xmath76 space ; for any @xmath60 , such that @xmath67\not=\emptyset$ ] , @xmath75 answers the highest range ancestor query @xmath58 in @xmath77 time . data structures @xmath73 and @xmath75 will be described later in this section . moreover , we store all elements of @xmath5 in the data structure described in @xcite that supports one - reporting queries : for any @xmath60 , some element @xmath78 $ ] can be found in @xmath77 time ; if @xmath18=\emptyset$ ] , the data structure returns a dummy element @xmath79 . finally , all elements of @xmath5 are stored in a slow data structure that uses @xmath1 space and answers color reporting queries in @xmath80 time . we can use e.g. the data structure from @xcite for this purpose . [ [ answering - queries . ] ] answering queries . + + + + + + + + + + + + + + + + + + all colors in a query range @xmath30 $ ] can be reported with the following procedure . using the one - reporting data structure from @xcite , we search for some @xmath81 $ ] if at least one such @xmath29 exists . if no element @xmath29 satisfying @xmath82 is found , then @xmath18=\emptyset$ ] and the query is answered . otherwise , let @xmath83 denote the leaf that contains @xmath29 . using @xmath84 , we search for the highest ancestor @xmath47 of @xmath83 such that @xmath85 . if no such @xmath47 is found , then all @xmath29 , @xmath82 , are in @xmath86 . we can report all colors in @xmath87 $ ] using @xmath88 . if @xmath84 returned some node @xmath47 , we proceed as follows . let @xmath89 and @xmath57 denote the left and the right children of @xmath47 . we traverse the list @xmath90 until an element @xmath91 is found or the end of @xmath90 is reached . we also traverse @xmath92 until an element @xmath93 is found or the end of @xmath92 is reached . if we reach neither the end of @xmath90 nor the end of @xmath92 , then the color of every encountered element @xmath94 , @xmath66 , and @xmath95 , @xmath64 , is reported . otherwise the range @xmath30 $ ] contains at least @xmath44 different colors . in the latter case we can use any data structure for one - dimensional color range reporting @xcite to identify all colors from @xmath96 $ ] in @xmath97 time . [ [ leaf - data - structures . ] ] leaf data structures . + + + + + + + + + + + + + + + + + + + + + a data structure @xmath73 answers color reporting queries on @xmath72 as follows . in @xcite , the authors show how a one - dimensional color reporting query on a set of @xmath98 one - dimensional elements can be answered by answering a query @xmath30\times [ 0,a]$ ] on a set of @xmath98 uncolored two - dimensional points . a standard priority search tree @xcite enables us to answer queries of the form @xmath30\times [ 0,a]$ ] on @xmath98 points in @xmath99 time . using a combination of fusion trees and priority search trees , described by willard @xcite , we can answer queries in @xmath100 time . the data structure of willard @xcite uses @xmath101 space and a universal look - up table of size @xmath102 for an arbitrarily small @xmath28 . updates are also supported in @xmath100 time points , but extension to the case of any @xmath103 is straightforward . ] . since @xmath72 contains @xmath104 elements , we can answer colored queries on @xmath72 in @xmath105 time . updates are also supported in @xmath77 time ; this fact will be used in section [ sec : dyncol ] . now we describe how @xmath75 is implemented . suppose that @xmath67\not=\emptyset$ ] for some leaf @xmath59 . let @xmath106 be the path from @xmath59 to the root of @xmath43 . we say that a node @xmath107 is a _ left parent _ if @xmath108 for the left child @xmath89 of @xmath47 ; a node @xmath107 is a _ right parent _ if @xmath109 for the right child @xmath57 of @xmath47 . if @xmath72 contains at least one @xmath110 $ ] , then the following is true . [ fact : val ] if @xmath107 is a left parent , then @xmath111 . if @xmath107 is a right parent , then @xmath112 . if @xmath113 is a left parent , then @xmath72 is in its left subtree . hence , @xmath55 is greater than any @xmath114 and @xmath111 . if @xmath47 is the right parent , than @xmath72 is in its right subtree . hence , @xmath55 is smaller than or equal to any @xmath114 and @xmath115 . [ fact : mono ] if @xmath116 is a left parent and @xmath117 is an ancestor of @xmath118 , then @xmath119 . if @xmath116 is a right parent and @xmath117 is an ancestor of @xmath118 , then @xmath120 . if @xmath117 is a left parent , then @xmath121 is in its left subtree . hence , @xmath122 by definition of @xmath55 . if @xmath121 is a right parent , then @xmath117 is in its right subtree . hence , @xmath123 by definition of @xmath55 . suppose that we want to find the highest range ancestor of @xmath59 for a range @xmath30 $ ] such that @xmath67\not=\emptyset$ ] . let @xmath124 be the set of middle values @xmath55 for left parents @xmath113 sorted by height ; let @xmath125 be the set of @xmath55 for right parents @xmath113 sorted by height . by fact [ fact : mono ] , elements of @xmath126 ( @xmath127 ) increase ( decrease ) monotonously . by fact [ fact : val ] , @xmath111 for any @xmath128 and @xmath129 for any @xmath130 . using fusion trees @xcite , we can search in @xmath126 and find the highest node @xmath131 such that @xmath117 is a left parent and @xmath132 . we can also search in @xmath127 and find the highest node @xmath118 such that @xmath121 is a right parent and @xmath133 . let @xmath47 denote the higher node among @xmath117 , @xmath121 . then @xmath47 is the highest ancestor of @xmath59 such that @xmath134 $ ] . = 1000 [ [ removing - duplicates . ] ] removing duplicates . + + + + + + + + + + + + + + + + + + + + when a query is answered , our procedure returns a color @xmath48 two times if @xmath48 occurs in both @xmath135 $ ] and @xmath136 $ ] . we can easily produce a list without sorting in which each color occurs exactly once . let @xmath137 denote an array with one entry for every color that occurs in a data structure . initially @xmath138=0 $ ] for all @xmath139 . we traverse the list of colors @xmath140 produced by the above described procedure . every time when we encounter a color @xmath48 in @xmath140 such that @xmath141=0 $ ] , we set @xmath141=1 $ ] ; when we encounter a color @xmath48 such that @xmath141=1 $ ] , we remove the corresponding entry from @xmath140 . when the query is answered , we traverse @xmath140 once again and set @xmath142=0 $ ] for all @xmath143 . [ theor : col1d ] there exists an @xmath1-space data structure that supports one - dimensional color range reporting queries in @xmath2 time . the static data structure of section [ sec : statcol ] can be used for answering queries in external memory . we only need to increase the sizes of @xmath72 , @xmath53 , and @xmath52 to @xmath144 , and use an external memory variant of the slow data structure for color reporting @xcite . this approach enables us to achieve @xmath41 query cost , but one important issue should be addressed . as explained in section [ sec : statcol ] , the same color can be reported twice when a query is answered . however , we can not get rid of duplicates in @xmath41 i / os using the method of section [ sec : statcol ] because of its random access to the list of reported colors . therefore we need to make further changes in our internal memory solution . for an element @xmath63 , let @xmath145 denote the largest element @xmath146 of the same color . for every element @xmath29 in @xmath52 and any @xmath147 , we also store the value of @xmath145 . we define each set @xmath72 for a leaf @xmath59 to contain @xmath148 points . lists @xmath149 and @xmath150 for an internal node @xmath151 contain @xmath144 leftmost points from @xmath152 ( respectively , @xmath144 rightmost points from @xmath153 ) . data structures @xmath75 are implemented as in section [ sec : statcol ] . a data structure @xmath73 supports color reporting queries on @xmath72 and is implemented as follows . we can answer a one - dimensional color reporting query by answering a three - sided point reporting query on a set @xmath154 of @xmath155 two - dimensional points ; see e.g. , @xcite . if @xmath156 , @xmath72 and @xmath154 contain @xmath157 points . in this case we can use the data structure from @xcite that uses linear space and answers three - sided queries in @xmath158 i / os . if @xmath159 , @xmath72 and @xmath154 contain @xmath160 points . using the data structure from @xcite , we can find the predecessor of any value @xmath151 in a set of @xmath160 points in @xmath77 i / os . therefore we can apply the rank - space technique @xcite and reduce three - sided point reporting queries on @xmath154 to three - sided point reporting queries on a grid of size @xmath161 ( i.e. , to the case when coordinates of all points are integers bounded by @xmath161 ) using a constant number of additional i / os . larsen and pagh @xcite described a linear space data structure that answers three - sided point reporting queries for @xmath98 points on an @xmath162 grid in @xmath41 i / os . summing up , we can answer a three - sided query on a set of @xmath144 points in @xmath41 i / os . hence , we can also answer a color reporting query on @xmath72 in @xmath41 i / os using linear space . a query @xmath13 $ ] is answered as follows . we find the highest range ancestor @xmath47 for any @xmath78 $ ] exactly as in section [ sec : statcol ] . if @xmath47 is a leaf , we answer the query using @xmath163 . otherwise the reporting procedure proceeds as follows . we traverse the list @xmath92 for the left child @xmath89 of @xmath47 until some point @xmath164 is found . if @xmath64 for all @xmath95 , then there are at least @xmath144 different colors in @xmath30 $ ] and we can use a slow data structure to answer a query in @xmath165 i / os . otherwise we traverse @xmath90 and report all elements @xmath29 such that @xmath166 . if @xmath167 for @xmath168 , then an element of the same color was reported when @xmath92 was traversed . traversal of @xmath90 stops when an element @xmath169 is encountered or the end of @xmath90 is reached . in the former case , we reported all colors in @xmath30 $ ] . in the latter case the number of colors in @xmath30 $ ] is at least @xmath144 . this is because every element in @xmath90 corresponds to a distinct color that occurs at least once in @xmath30 $ ] . hence , we can use the slow data structure and answer the query in @xmath170 i / os . [ theor : col1dext ] there exists a linear - space data structure that supports one - dimensional color range reporting queries in @xmath171 i / os . in this section we show how the base tree and auxiliary data structures of the static solution can be modified for usage in the dynamic scenario . to dynamize the data structure of section [ sec : statcol ] , we slightly change the balanced tree @xmath43 and secondary data structures : every leaf of @xmath43 now contains @xmath172 elements of @xmath5 and each internal node has @xmath173 children . we store the lists @xmath52 and @xmath53 in each internal non - root node of @xmath47 . we associate several values @xmath174 to each node @xmath47 : for every child @xmath175 of @xmath47 , except the leftmost child @xmath117 , @xmath176 . the highest range ancestor of a leaf @xmath59 is the highest ancestor @xmath47 of @xmath59 such that @xmath177 for at least one @xmath178 . data structures @xmath73 and @xmath75 are defined as in section [ sec : statcol ] . we also maintain a data structure of @xcite that reports an arbitrary element @xmath78 $ ] if the range @xmath30 $ ] is not empty . we implement the base tree @xmath43 as the weight - balanced b - tree @xcite with the leaf parameter @xmath44 and the branching parameter @xmath179 . this means that every internal node has between @xmath180 and @xmath181 children and each leaf contains between @xmath182 and @xmath183 elements . each internal non - root node on level @xmath184 of @xmath43 has between @xmath185 and @xmath186 elements in its subtree . if the number of elements in some node @xmath47 exceeds @xmath185 , we split @xmath47 into two new nodes , @xmath187 and @xmath188 . in this case we insert a new value @xmath189 for the parent @xmath68 of @xmath47 . hence , we may have to update the data structures @xmath75 for all leaf descendants of @xmath68 . a weight - balanced b - tree is engineered in such a way that a split occurs at most once in a sequence of @xmath190 insertions ( for our choice of parameters ) . since @xmath75 can be updated in @xmath77 time , the total amortized cost incurred by splitting nodes is @xmath77 . when an element @xmath29 is deleted , we delete it from the set @xmath72 . if @xmath191 for a deleted element @xmath29 and some node @xmath47 , we do not change the value of @xmath174 . we also do not start re - balancing if some node contains too few elements in its subtree . but we re - build the entire tree @xmath43 if the total number of deleted elements equals @xmath192 , where @xmath193 is the number of elements that were stored in @xmath43 when it was built the last time . updates can be de - amortized without increasing the cost of update operations by scheduling the procedure of re - building nodes ( respectively , re - building the tree ) @xcite [ [ auxiliary - data - structures . ] ] auxiliary data structures . + + + + + + + + + + + + + + + + + + + + + + + + + + we implement @xmath73 in the same way as in section [ sec : statcol ] . hence color queries on @xmath72 are answered in @xmath194 time and updates are also supported in @xmath77 time @xcite . we need to modify data structures @xmath75 , however , because @xmath43 is not a binary tree in the dynamic case . let @xmath106 denote a path from @xmath59 to the root for some leaf @xmath59 . we say that a node @xmath47 is an @xmath139-node if @xmath195 for the @xmath139-th child @xmath175 of @xmath47 . [ fact : val2 ] suppose that @xmath67\not=\emptyset$ ] and @xmath106 is the path from @xmath59 to the root . if @xmath113 is an @xmath139-node , then @xmath196 for @xmath197 and @xmath198 for @xmath199 . we say that a value @xmath200 for @xmath113 is a _ left value _ if @xmath201 and @xmath47 is an @xmath139-node . a value @xmath200 for @xmath113 is a _ right value _ if @xmath202 and @xmath47 is an @xmath139-node . [ fact : mono2 ] if @xmath203 is a left value and @xmath131 is an ancestor of @xmath118 , then @xmath204 for any @xmath205 . if @xmath203 is a right value and @xmath131 is an ancestor of @xmath118 , then @xmath206 for any @xmath205 . it is easy to check facts [ fact : val2 ] and [ fact : mono2 ] using the same arguments as in section [ sec : statcol ] . we store all left values @xmath200 , @xmath113 , in a set @xmath126 ; @xmath200 in @xmath126 are sorted by the height of @xmath47 . we store all right values @xmath200 , @xmath113 , in a set @xmath127 ; @xmath200 in @xmath127 are also sorted by the height of @xmath47 . using fusion trees on @xmath126 , we can find the highest node @xmath117 , such that at least one left value @xmath207 . we can also find the highest @xmath121 such that at least one right value @xmath208 . since @xmath126 and @xmath127 contain @xmath76 elements , we can support searching and updates in @xmath77 time ; see @xcite . by fact [ fact : val2 ] , @xmath209 and @xmath210 . if @xmath47 is the higher node among @xmath117 , @xmath121 , then @xmath47 is an answer to the highest range ancestor query @xmath30 $ ] for a node @xmath59 . in this section we describe a dynamic data structure with optimal query time . our improvement combines an idea from @xcite with the highest range ancestor approach . we also use a new solution for a special case of two - dimensional point reporting problem presented in section [ sec : fast3sid ] . let @xmath211 denote the height of a node @xmath47 . for an element @xmath63 let @xmath212 , where @xmath187 is the highest ancestor of the leaf containing @xmath29 , such that @xmath213 . we define @xmath214 in the same way with respect to @xmath50 . all colors in a range @xmath30 $ ] can be reported as follows . we identify an arbitrary @xmath78 $ ] . using the highest range ancestor data structure , we can find the lowest common ancestor @xmath47 of the leaves that contain @xmath14 and @xmath215 . let @xmath216 and @xmath217 be the children of @xmath47 that contain the successor of @xmath14 and the predecessor of @xmath215 . let @xmath218 , @xmath219 ; let @xmath220 for @xmath221 and @xmath222 for @xmath223 . we can identify unique colors of relevant points stored in each node @xmath224 , @xmath225 , by finding all @xmath226 $ ] such that @xmath227 . this condition is equivalent to reporting all @xmath226 $ ] such that @xmath228 . we can identify all colors of relevant points in @xmath216 by reporting all @xmath229 $ ] such that @xmath230 . queries of the form @xmath110 $ ] , @xmath231 , ( respectively @xmath110 $ ] , @xmath232 ) can be supported using lemma [ lemma : fast ] . while the same color can be reported several times , we can get rid of duplicates as explained in section [ sec : statcol ] . when a new point is inserted into @xmath5 or when a point is deleted from @xmath5 , we can update the values of @xmath233 and @xmath214 in @xmath234 time . we refer to @xcite for details . while updates of data structures of lemma [ lemma : fast ] are fast , re - balancing the base tree can be a problem . as described in section [ sec : dyncol ] , when the number of points in a node @xmath47 on level @xmath184 exceeds @xmath235 , we split it into two nodes , @xmath187 and @xmath188 . as a result , the values @xmath233 for @xmath29 stored in the leaves of @xmath188 can be incremented . hence , we would have to examine the leaf descendants of @xmath188 and recompute their values for some of them . since the height of @xmath43 is logarithmic , the total cost incurred by re - computing the values @xmath233 and @xmath214 is @xmath76 . the problem of reducing the cost of re - building the tree will be solved in the following sections . in appendix [ sec : slow ] we describe another data structure that supports fast updates but answering queries takes polynomial time in the worst case . in section [ sec : fullydyn ] we show how the cost of splitting can be reduced by modifying the definition of @xmath233 , @xmath214 and using the slow data structure from appendix [ sec : slow ] when the number of reported colors is sufficiently large . let @xmath236 denote the number of leaves in the subtree of a node @xmath47 . let @xmath237 denote the set of @xmath238 smallest elements in @xmath49 ; let @xmath239 denote the set of @xmath238 largest elements in @xmath50 . we maintain the values @xmath240 and @xmath241 for @xmath63 , such that for any @xmath45 we have : @xmath242 if @xmath243 and @xmath244 if @xmath245 ; @xmath246 if @xmath247 and @xmath248 if @xmath249 . we keep @xmath240 and @xmath241 in data structures of lemma [ lemma : fast ] . we maintain the data structure described in section [ sec : slow ] . this data structure is used to answer queries when the number of colors in the query range is large . it is also used to update the values of @xmath240 and @xmath241 when a node is split . to answer a query @xmath30 $ ] , we proceed in the same way as in section [ sec : del ] . let @xmath47 , @xmath216 , @xmath217 , and @xmath250 , @xmath251 , @xmath252 be defined as in section [ sec : del ] . distinct colors in each @xmath253 $ ] , @xmath252 , can be reported using the data structure of lemma [ lemma : fast ] . if the answer to at least one of the queries contains at least @xmath254 elements , then there are at least @xmath254 different colors in @xmath30 $ ] . the total number of elements in @xmath30\cap s$ ] does not exceed @xmath255 . hence , we can employ the data structure from section [ sec : slow ] to report all colors from @xmath30 $ ] in @xmath256\cap s)^{1/2}+k)=o(k)$ ] time . if answers to all queries contain less than @xmath254 elements , then for every distinct color that occurs in @xmath30 $ ] there is an element @xmath29 such that @xmath257 $ ] , @xmath258 , or @xmath259 $ ] . by definition of @xmath260 and @xmath261 we can correctly report up to @xmath254 leftmost colors in @xmath262 or up to @xmath254 rightmost colors in @xmath263 . = 500 when a new element @xmath29 is inserted , we compute the values of @xmath233 , @xmath214 and update the values of @xmath264 , @xmath265 , where @xmath266 is the element of the same color as @xmath29 that follows @xmath29 . this can be done in the same way as in section [ sec : del ] . when a node @xmath47 on level @xmath184 is split into @xmath187 and @xmath188 , we update the values of @xmath233 and @xmath214 for @xmath267 . if @xmath268 , we examine all @xmath267 and re - compute the values of @xmath145 , @xmath233 , and @xmath214 . amortized cost of re - building nodes @xmath47 on @xmath269 lowest tree levels is @xmath36 . if @xmath270 , @xmath46 contains @xmath271 elements . we can find @xmath272 elements in @xmath273 , @xmath274 , @xmath275 , and @xmath276 using the data structure from lemma [ lemma : slowcount ] . this takes @xmath277 time . since we split a node @xmath47 one time after @xmath278 insertions , the amortized cost of splitting nodes on level @xmath279 is @xmath77 . thus the total cost incurred by splitting nodes after insertions is @xmath36 . deletions are processed in a symmetric way . thus we obtain the following result [ theor : dyn ] there exists a linear space data structure that supports one - dimensional color range reporting queries in @xmath2 time and updates in @xmath42 amortized time . 10 s. alstrup , g. s. brodal , and t. rauhe . new data structures for orthogonal range searching . in _ proc . 41st annual symposium on foundations of computer science ( focs 2000 ) _ , pages 198207 , 2000 . s. alstrup , g. s. brodal , and t. rauhe . optimal static range reporting in one dimension . in _ proc . 33rd annual acm symposium on theory of computing ( stoc ) _ , pages 476482 , 2001 . l. arge , v. samoladas , and j. s. vitter . on two - dimensional indexability and optimal range search indexing . in _ proc . 18th acm sigact - sigmod - sigart symposium on principles of database systems ( pods ) _ , pages 346357 , 1999 . l. arge and j. s. vitter . optimal external memory interval management . , 32(6):14881508 , 2003 . p. beame and f. e. fich . optimal bounds for the predecessor problem and related problems . , 65(1):3872 , 2002 . t. m. chan , s. durocher , m. skala , and b. t. wilkinson . linear - space data structures for range minority query in arrays . in _ 13th scandinavian symposium and workshops on algorithm theory ( swat ) _ , pages 295306 , 2012 . b. chazelle . filtering search : a new approach to query - answering . , 15(3):703724 , 1986 . m. l. fredman and d. e. willard . trans - dichotomous algorithms for minimum spanning trees and shortest paths . , 48(3):533551 , 1994 . h. n. gabow , j. l. bentley , and r. e. tarjan . scaling and related techniques for geometry problems . in _ proc . 16th annual acm symposium on theory of computing ( stoc 1984 ) _ , pages 135143 , 1984 . p. gupta , r. janardan , and m. h. m. smid . further results on generalized intersection searching problems : counting , reporting , and dynamization . , 19(2):282317 , 1995 . r. janardan and m. a. lopez . generalized intersection searching problems . , 3(1):3969 , 1993 . k. g. larsen and r. pagh . -efficient data structures for colored range and prefix reporting . in _ proc . 23rd annual acm - siam symposium on discrete algorithms ( soda ) _ , pages 583592 , 2012 . k. g. larsen and f. van walderveen . near - optimal range reporting structures for categorical data . in _ proc . 24th annual acm - siam symposium on discrete algorithms ( soda ) _ , page to appear , 2013 . e. m. mccreight . priority search trees . , 14(2):257276 , 1985 . b. miltersen , n. nisan , s. safra , and a. wigderson . on data structures and asymmetric communication complexity . , 57(1):3749 , 1998 . c. w. mortensen . generalized static orthogonal range searching in less space . technical report , it university technical report series 2003 - 33 , 2003 . c. w. mortensen , r. pagh , and m. patrascu . on dynamic range reporting in one dimension . in _ proc . 37th annual acm symposium on theory of computing ( stoc ) _ , pages 104111 , 2005 . s. muthukrishnan . efficient algorithms for document retrieval problems . in _ proc . 13th annual acm - siam symposium on discrete algorithms ( soda ) _ , pages 657666 , 2002 . y. nekrich . space - efficient range reporting for categorical data . in _ proc . 31st acm sigmod - sigact - sigart symposium on principles of database systems ( pods ) _ , pages 113120 , 2012 . q. shi and j. jj . optimal and near - optimal algorithms for generalized intersection reporting on pointer machines . , 95(3):382388 , 2005 . m. thorup . undirected single - source shortest paths with positive integer weights in linear time . , 46(3):362394 , 1999 . p. van emde boas , r. kaas , and e. zijlstra . design and implementation of an efficient priority queue . , 10:99127 , 1977 . d. e. willard . examining computational geometry , van emde boas trees , and hashing from the perspective of the fusion tree . , 29(3):10301049 , 2000 . in this section we describe a data structure that efficiently supports ( uncolored ) two - dimensional queries in the case when the @xmath280-coordinate @xmath281 of each point @xmath282 does not exceed @xmath44 and the query range is bounded on three sides , two sides in the @xmath283-dimension and one side in the @xmath280-dimension ; the @xmath283-coordinate @xmath284 of each @xmath282 does not exceed the value of parameter @xmath20 . this result is used in the dynamic solution of the color range reporting problem . we also believe it to be of independent interest . [ lemma : fast ] let @xmath5 be a set of two - dimensional points such that @xmath285 for all @xmath282 . there exists a linear - space data structure that reports all points @xmath286\times [ c,\log n]$ ] in @xmath2 time and supports updates in @xmath42 time , where @xmath20 is the size of the universe . we divide the points into consecutive groups @xmath287 according to their @xmath283-coordinates . each group , except the last one , contains @xmath288 points ; the last group contains @xmath76 points . for every group @xmath287 and for each value @xmath289 , we keep two values @xmath290 and @xmath291 . intuitively , we can access the value of @xmath292 and @xmath293 for any @xmath294 and @xmath139 by examining @xmath77 values @xmath290 or @xmath291 . we remark that we can not directly store the values of @xmath295 in @xmath291 because an update operation would be too costly . let @xmath296 . when a new point @xmath297 is inserted , we represent @xmath298 in base @xmath299 as a sum of powers is a power of @xmath299 . ] of @xmath299 : @xmath300 where @xmath301 and @xmath302 . let @xmath303 for @xmath304 and @xmath305 be defined as @xmath306 . let @xmath287 denote the group into which @xmath297 must be inserted . we examine the values @xmath307 and @xmath308 for all @xmath304 and @xmath305 . if @xmath309 or @xmath310 , we update their values . the deletion procedure is symmetric . in both cases , we examine and update @xmath102 values . [ fact : simp ] suppose that some @xmath311 contains at least one point @xmath297 such that @xmath312 $ ] and @xmath313 . suppose further that @xmath314 or @xmath315 . then either @xmath316 $ ] or @xmath317 $ ] . [ fact : sel ] for any @xmath294 , we can select @xmath77 values @xmath318 , such that for any @xmath139 there is at least one @xmath319 , @xmath320 , satisfying @xmath321 and at least one @xmath322 , @xmath323 , satisfying @xmath324 . let @xmath325 where @xmath301 . we set @xmath326 , @xmath327 , @xmath328 , @xmath329 , @xmath328 , @xmath330 . if @xmath331 , then there is a point @xmath297 , such that @xmath332 and @xmath333 . let @xmath334 . if @xmath335 , we consider an index @xmath336 such that @xmath337 and @xmath338 for @xmath339 . by definition of @xmath340 , @xmath341 , where @xmath342 . by our choice of @xmath343 , @xmath344 hence , @xmath345 . if @xmath346 , then @xmath347 and @xmath348 . the statement concerning @xmath349 can be proved in the same way . facts [ fact : simp ] and [ fact : sel ] suggest the following method for answering queries . for any @xmath294 , @xmath350 , we store all @xmath290 and @xmath291 in a data structure @xmath351 that supports one - dimensional point reporting queries . data structure @xmath352 contains @xmath283-coordinates of all points in @xmath5 and also supports one - dimensional reporting queries . using the result from @xcite , @xmath352 and all @xmath351 support queries in @xmath2 time and updates in @xmath102 time . for each @xmath287 , we maintain a data structure @xmath353 . for any @xmath354 and @xmath355 , @xmath353 can report all points @xmath356 , @xmath357 and @xmath358 . @xmath353 uses @xmath359 space and supports updates in @xmath77 time . we can implement @xmath287 in the same way as the data structure @xmath73 in section [ sec : statcol ] . given a query @xmath13\times [ c,\log n]$ ] , we use @xmath352 for identifying an arbitrary @xmath360 , such that @xmath361 $ ] . let @xmath362 be the group that contains @xmath363 . if all points @xmath297 such that @xmath312 $ ] are contained in @xmath362 , we use the data structure @xmath364 to answer the query . if @xmath30 $ ] spans several groups , then we generate the values @xmath365 as in the proof of fact [ fact : sel ] . we query data structures @xmath366 and identify all values @xmath367 $ ] and @xmath368 $ ] . for every such @xmath369 and @xmath370 , we visit the corresponding group @xmath287 and answer the query @xmath30\times [ c,\log n]$ ] on @xmath287 using @xmath353 . by facts [ fact : simp ] and [ fact : sel ] , queries to @xmath366 return at least one representative element from every @xmath287 such that @xmath371 and only such groups will be visited . since we ask @xmath372 queries to @xmath366 , every group @xmath287 is visted at most @xmath373 times . when a new point @xmath297 is inserted into @xmath5 , we identify the group @xmath287 where it belongs in @xmath374 time . then we re - examine the values of @xmath307 and @xmath308 for @xmath298 . if necessary , we update the data structures @xmath375 . when the number of elements in some @xmath287 becomes equal to @xmath376 , we split @xmath287 into two groups in a standard way . deletions are symmetric . thus updates are supported in @xmath377 time and queries can be answered in @xmath77 time . we obtain the result of lemma [ lemma : fast ] if we replace @xmath28 by @xmath378 into the above proof . in this section we describe a linear - space data structure that supports color reporting queries in @xmath379 time , where @xmath380|$ ] is the number of elements in the query range @xmath30 $ ] . although the query cost is high for large @xmath381 , updates are supported in @xmath27 time , where @xmath20 is the size of the universe . [ lemma : slowrep ] there exists a linear space data structure that reports all distinct colors in a query range @xmath30 $ ] . queries are supported in @xmath382 time where @xmath383\cap s|$ ] is the number of elements in the query range @xmath30 $ ] . updates are supported in @xmath42 time . let @xmath384 be a range tree of the set @xmath5 . leaves of @xmath384 contain the elements of @xmath5 in sorted order . the root of @xmath384 has @xmath385 children . each child of the root node has @xmath386 children and @xmath385 leaf descendants . a node of depth @xmath8 has @xmath387 children and @xmath388 leaf descendants where @xmath389 . thus the height of @xmath384 is @xmath390 . as before , @xmath46 denotes the set of elements stored in leaf descendants of @xmath47 . recall that @xmath145 for @xmath63 denotes the largest element @xmath146 of the same color . the set @xmath391 , @xmath45 , contains all elements @xmath51 , such that @xmath392 . we maintain a balanced tree @xmath43 and sets @xmath391 in all nodes @xmath393 . further , all elements of every @xmath391 are kept in two sorted lists , @xmath394 and @xmath395 . elements @xmath396 are sorted by @xmath145 ; elements in @xmath395 are sorted by their values . finally , we also maintain a data structure , described in @xcite , that supports reporting queries on @xmath391 in @xmath2 time and updates in @xmath42 time . to answer a query @xmath13 $ ] , we identify the leaves @xmath62 and @xmath65 that hold @xmath397 and @xmath398 respectively , where @xmath397 is the smallest element that is greater than @xmath14 and @xmath398 is the largest element that is smaller than @xmath215 . let @xmath399 be the lowest common ancestor of @xmath62 and @xmath65 . suppose that @xmath30 $ ] covers children @xmath400 , @xmath328 , @xmath401 of @xmath399 and intersects with @xmath72 and @xmath402 . let @xmath403 denote the path from @xmath399 to @xmath62 . the query answering procedure works as follows . [ list : slowstep1 ] we visit all @xmath404 and report all elements @xmath405 , @xmath64 , in each @xmath47 using @xmath395 . [ list : slowstep2 ] then we visit all right siblings @xmath187 of nodes @xmath404 except @xmath59 ; in every @xmath187 , we report colors of @xmath406 , @xmath166 , using @xmath407 . we also report all colors of @xmath408 ) , @xmath409 . [ list : slowstep3 ] we also report colors of all @xmath410 , @xmath66 , using @xmath411 . [ list : slowstep4 ] finally we visit all proper ancestors @xmath68 of @xmath399 ; in every @xmath68 we report all elements @xmath412 , @xmath82 , using the reporting data structure . correctness of our procedure can be demonstrated as follows . suppose that @xmath29 is the leftmost occurrence of some color in @xmath30 $ ] . we consider two different cases . ( 1 ) @xmath413 . then @xmath412 for some ancestor @xmath68 of @xmath399 and it will be reported when @xmath414 is queried . ( 2 ) @xmath415 where @xmath89 is the left sibling of some node @xmath47 on @xmath403 . then @xmath29 is stored in @xmath391 or in @xmath416 for a right sibling @xmath187 of @xmath47 . hence @xmath29 was reported when @xmath47 ( resp . @xmath187 ) was visited . each color is reported at most two times . if the color of an element @xmath417 is reported during step ( ii ) , then @xmath417 is the leftmost element of that color in @xmath30 $ ] because we only output elements @xmath29 such that @xmath409 . if the color of an element @xmath417 is reported during step ( i ) or step ( iv ) , than @xmath417 is likewise the leftmost element of that color . this is because @xmath418 ( resp . @xmath405 ) and the leftmost occurrence of a color in @xmath419 ( or @xmath46 ) is also the leftmost occurrence in @xmath30 $ ] . the only situation when we report the color of an element @xmath417 and @xmath417 is not the leftmost occurrence of that color is during step ( [ list : slowstep3 ] ) . we can get rid of duplicates by traversing the list of answers and removing elements @xmath29 , such that @xmath167 . the time needed to answer a query can be estimated by counting the number of visited nodes . let @xmath420 denote the highest node on @xmath403 such that at least one sibling @xmath47 of @xmath420 is visited . the number of leaves in the subtree of @xmath47 is @xmath421 , where @xmath8 is the depth of @xmath47 . we consider all nodes @xmath422 below @xmath420 on @xmath403 , the total number of siblings of such @xmath422 is bounded by @xmath423 . thus @xmath424 , where @xmath425 is the time needed to answer a query and @xmath380|$ ] is the total number of elements in @xmath30 $ ] . when an element @xmath29 is inserted into @xmath5 , we identify the greatest @xmath426 and the smallest @xmath427 such that @xmath428 and @xmath266 are of the same color as @xmath29 . we insert @xmath29 into an appropriate leaf @xmath59 and find the ancestor @xmath47 of @xmath59 , such that @xmath429 , @xmath430 . the element @xmath29 is inserted into the set @xmath391 and into lists @xmath394 , @xmath395 . suppose that @xmath266 was stored in a set @xmath431 ; we remove @xmath266 from @xmath431 , @xmath432 , @xmath433 and insert it into corresponding secondary structures in the node @xmath121 . the node @xmath121 is chosen in such way that @xmath434 and @xmath435 . when the number of elements in some node @xmath47 becomes equal to @xmath436 , where @xmath8 is the depth of @xmath47 , we split @xmath47 into @xmath187 and @xmath188 . when a node is split , we re - build the data structures in @xmath187 , @xmath188 and all their descendants . thus the total amortized cost incurred by splitting a node is @xmath437 . the total cost of an insertion is dominated by the time necessary to update the data structure that supports reporting queries on @xmath391 . deletions are symmetric . we will also need another result that uses almost the same data structure , but reports only the @xmath3 leftmost colors in the query range . [ lemma : slowcount ] there exists a linear - space data structure that reports , for any integer @xmath3 , the @xmath3 leftmost ( rightmost)colors in a query range @xmath30 $ ] . queries are supported in @xmath438 time where @xmath383\cap s|$ ] is the number of elements in the query range @xmath30 $ ] . updates are supported in @xmath27 time . the data structure from lemma [ lemma : slowrep ] can be used to determine whether the number of points in a query range exceeds a threshold value @xmath299 . to compare the number of points , with @xmath299 , we proceed as in lemma [ lemma : slowrep ] and use the fact that every color is reported at most twice . if the number of reported elements in visited nodes exceeds @xmath439 at some point , we stop processing the query and report that the number of distinct colors exceeds @xmath299 . otherwise , we answer a color reporting query and determine whether the number of colors exceeds @xmath299 . in both cases , a comparison of the number of colors with @xmath299 is performed in @xmath440 time . we identify the range @xmath441 $ ] that contains at least @xmath3 and at most @xmath442 colors by binary search . we start by setting @xmath443 and comparing the number of colors @xmath444 in @xmath445 $ ] with @xmath3 . if @xmath444 is larger than ( smaller than ) @xmath3 , we move @xmath446 to the left ( to the right ) using a standard binary search procedure . after @xmath447 iterations we obtain @xmath446 , such that @xmath448 . then we answer a color reporting query on @xmath441 $ ] as in lemma [ lemma : slowrep ] . each comparison query is answered in @xmath449 . @xmath76 iterations take @xmath438 time . our dynamic external memory data structure is based on the same approach as the data structure of theorem [ theor : dyn ] . but we need to change some of the auxiliary data structures . [ lemma : extmem3sid ] there exists a data structure that supports three - sided queries on a set @xmath5 , such that @xmath450 , in @xmath77 i / os . this data structure uses linear space and supports updates in @xmath102 i / os . if @xmath451 , the data structure from @xcite gives us the desired query and update bounds because @xmath452 in this case . if @xmath453 , we implement our data structure as external priority search tree @xmath384 . every leaf of @xmath384 contains @xmath454 points sorted by their @xmath283-coordinates . as before @xmath455 denotes the set of all points stored in the leaves of @xmath151 . let @xmath456 denote the set of @xmath454 points @xmath457 with highest @xmath280-coordinates satisfying @xmath458 for any ancestor @xmath68 of @xmath151 . each internal node has @xmath459 children . we keep a data structure @xmath460 in each internal node @xmath151 ; @xmath460 contains all points @xmath461 , where the union is over all children @xmath462 of @xmath151 , and answers three - sided queries . @xmath460 is implemented as a static data structure described in @xcite . when a set @xmath463 is updated , we re - build @xmath460 . since @xmath460 contains @xmath464 points , it can be re - built in @xmath465 i / os . the query answering procedure is the same as in the external priority tree . for a query @xmath13\times [ c,+\infty]$ ] , let @xmath403 and @xmath466 denote the search paths for @xmath14 and @xmath215 respectively . we visit all nodes @xmath47 on @xmath403 and @xmath466 and report all points in @xmath467 . then we visit relevant descendants of nodes @xmath47 on @xmath468 . in each visited node @xmath47 , we report all points in @xmath469 ; we visit a child @xmath175 of @xmath47 only if @xmath470 . details of the reporting procedure and a proof of its correctness can be found in @xcite . since the height of our priority tree is @xmath77 , we can update our data structure by updating @xmath77 structures @xmath460 . replacing @xmath28 by @xmath378 , we obtain the result of this lemma . [ lemma : extfast ] let @xmath5 be a set of two - dimensional points such that @xmath285 for all @xmath282 . there exists a linear - space data structure that reports all points @xmath286\times [ c,\log n]$ ] in @xmath171 i / os and supports updates in @xmath42 i / os . the structure of lemma [ lemma : fast ] can not be directly extended to the external memory . the reason is that we have to visit a number of groups @xmath287 when a query is processed . we would have to spend @xmath77 i / os in each group ; thus the query time would be @xmath471 . we need further modifications to obtain the desired @xmath41 query cost . our external memory solution is based on increasing the group size . essentially , we answer a three - sided query in a group @xmath287 only if we know that @xmath287 contains a sufficient number of points from the query range . otherwise we resort to the data structure that is based on range trees . as in lemma [ lemma : fast ] we divide points into groups @xmath287 . each group , except the last one , contains @xmath472 points ; the last group contains @xmath473 points . we distinguish between _ group - stored _ and _ directly stored _ points . let @xmath474=\{\,p\in g_i\,|\,p.y = h\,\}$ ] . a point @xmath297 is group - stored if @xmath356 and @xmath475=\omega(b\log n)$ ] . otherwise @xmath297 is directly stored . either all points in @xmath474 $ ] for a fixed value of @xmath294 are group - stored or all points in @xmath474 $ ] are directly stored . we maintain the values of @xmath290 and @xmath291 , defined in the same way as in lemma [ lemma : fast ] , with respect to group - stored points . let @xmath476 denote the set of directly stored points @xmath282 ; let @xmath477 denote the set of group - stored points in @xmath5 . all points in @xmath476 are kept in the standard range tree @xmath478 with node degree @xmath479 . @xmath478 allows us to reduce a three - sided query on @xmath476 to @xmath77 one - dimensional point reporting queries . we refer to @xcite for details . the space usage of @xmath478 is @xmath480 ; updates can be implemented by @xmath481 updates of one - dimensional auxiliary structures stored in the nodes of @xmath478 . we implement these one - dimensional reporting data structures using the result of mortensen et al . @xcite . thus @xmath478 supports queries and updates in @xmath77 and @xmath102 i / os respectively . all points in @xmath482 for each group @xmath287 are kept in a data structure @xmath353 that supports three - sided range reporting queries as described in lemma [ lemma : extmem3sid ] . when a new point @xmath282 is inserted into @xmath5 , we identify the group @xmath287 where it belongs . if points in @xmath475 $ ] are group - stored , then @xmath297 should be group - stored . we compute @xmath303 for @xmath298 and update the values of @xmath307 , @xmath308 as in lemma [ lemma : fast ] . if points in @xmath475 $ ] are directly stored , then we check how many points are currently in @xmath475 $ ] . if @xmath483|=(3b\log n)/2 $ ] , then points in @xmath484 $ ] will be group - stored . we remove all points of @xmath475 $ ] from the tree @xmath478 and add them to @xmath353 . then we identify the leftmost and the rightmost points in @xmath475 $ ] and update @xmath307 , @xmath308 for @xmath298 . if @xmath483| < ( 3b\log n)/2 $ ] , we continue to store @xmath475 $ ] directly . in the latter case we simply add @xmath297 to @xmath478 . deletions are symmetric to insertions . however if points in @xmath475 $ ] are group - stored and @xmath297 is deleted , we continue to group - store the points in @xmath475 $ ] if @xmath475\ge ( b\log n)/2 $ ] . for a query @xmath13\times [ c,+\infty]$ ] , we report all points in @xmath485 using @xmath478 . we identify groups @xmath287 such that @xmath486 in the same way as in lemma [ lemma : fast ] . for every such @xmath287 we report all points in @xmath487 using @xmath353 . if @xmath488 and @xmath489 for some group @xmath311 , then either @xmath490 or @xmath491 . hence , the total cost of answering queries in all @xmath353 is @xmath492 . thus a query @xmath0 is answered in @xmath41 i / os . we observe that only the case when the block size @xmath493 should be considered . if @xmath494 , then we can use the reduction of color reporting queries to three - sided queries . the data structure of arge et al @xcite supports three - sided queries and updates in @xmath495 i / os for @xmath494 . in the rest of this section we assume that @xmath493 . we will only sketch the differences of our data structure and the data structure of theorem [ theor : dyn ]

two - particle tunneling in synchronous and asynchronous regimes is studied in the framework of dissipative quantum tunneling . we show that the use of the proposed model is justified by a comparison with realistic potential energy surfaces of porphyrin and experimental dependence of the reaction rate on temperature . the critical temperature corresponding to a bifurcation of the underbarrier trajectory is determined . the effect of a heat bath local mode on the probability of two - dimensional tunneling transfer is also investigated . at certain values of the parameters , the degeneracy of antiparallel tunneling trajectories is important . thus , four , six , twelve , etc . , pairs of the trajectories should be taken into account ( a cascade of bifurcations ) . for the parallel particle tunneling the bifurcation resembles phase transition of a first kind , while for the antiparallel transfer it behaves as second order phase transition . the proposed theory allows for the explanation of experimental data on quantum fluctuations in two - proton tunneling in porphyrins near the critical temperature .

cond-mat0308516

quantum tunneling dynamics of a particle interacting with a heat bath is one of the important problems of modern condensed matter physics @xcite-@xcite . the interest to this problem is related to studies in low - temperature superconductive tunnel junctions @xcite , a dissipative quantum tunneling in crystals @xcite , and low - temperature chemical reactions @xcite . in low dimensional systems an effective mass approximation often fails . thus , quantum tunneling with dissipation becomes an important tool in the description of electron transfer @xcite . in many physical processes tunneling of _ two _ particles occurs . for example , semenov and dakhnovskii @xcite found the bistability of the tunneling trajectories in a two - proton transfer . later , benderskii and co - workers extended their investigations to different two - dimensional potentials @xcite . some features of the two - dimensional tunneling dynamics were studied in interacting josephson junctions @xcite . another example of two - particle tunneling is a two - proton transfer in porphyrin systems @xcite . porphyrins are important molecules in biology @xcite and a new area of electronics molecular wires and devices @xcite . there are experimental data @xcite that clearly indicate that a tunneling instability occurs at some critical temperature . such a behavior in the rate constant demonstrates the existence of bifurcations in two - dimensional underbarrier trajectories . apparently , this effect requires a thorough investigation when the protons interact with a thermal bath . the effect of cleavage in the temperature dependence of the current for two - dimensional systems of interacting josephson junctions was identified in ref . @xcite . however , there was no experimental confirmation of this effect . in the present paper we identify and study such an effect of unstable character observed experimentally in tunneling constant as a function of temperature for reactions with porphyrin . also , we predict a stable cleavage in the temperature dependence of current in the artificial two - dimensional system , a pair of weakly coupled charged quantum dots with single electrons tunneling to superlattice substrate , and the associated fluctuations of quantum origin . we study the emerging effect of chaotization near the critical point of unstable cleavage for porphyrin . in this paper , we continue to study bifurcation effects in two dimensional tunneling . in particular , we consider two - proton correlations of different types within the framework of dissipative quantum tunneling ( instanton ) approach . moreover , we discuss a fine structure of bifurcations for systems with various potential energy surfaces @xcite . the layout of the paper is as follows . in section ii , we introduce two - dimensional model potential energy surfaces for a pair of interacting particles . in section iii we present experimental aspects of the model . in section iv , we study the effect of temperature on the tunneling rate . in sections v and vi , we calculate the rate for the parallel and antiparallel tunneling and provide the analysis of the origin of bifurcation . the effect of a promoting mode ( an environment ) is studied in section vii . consider two charges that tunnel in two independent double - well potentials @xmath1 and @xmath2 presented as follows @xcite : @xmath3\theta(q_i ) , \quad i=1,2 , % \nonumber \\ % u(q_2 ) = \frac{1}{2}{\omega^2(q_2+a)^2}\theta(-q_2 ) % \nonumber \\ % + \left[-\delta i % + \frac{1}{2}{\omega^2(q_2-b)^2}\right]\theta(q_2),\end{aligned}\ ] ] where the sum @xmath4 determines the length of a `` link '' in the corresponding macrocluster fragment ; @xmath5 is a bias ( an asymmetry parameter of the potential ) ; @xmath6 is a step function , and @xmath7 is the frequency ( see discussion in ref . the mass is absorbed into the definition of @xmath8 . the interaction between two charges , e.g. , protons , is considered in a dipole - dipole approximation @xcite [ 2 ] v_(q_1,q_2 ) = - ( q_1-q_2)^2 , where @xmath9 is a positive constant . we use the same interaction potential as in ref . @xcite . thus , the total two - dimensional potential energy surface for parallel tunneling normalized by @xmath10 , is given by @xmath11\theta(q_1 ) + ( q_2+a)^2\theta(-q_2 ) \nonumber\\ + \left[-(b^2-a^2 ) + ( q_2-b)^2\right]\theta(q_2 ) - \frac{\alpha^*}{2}(q_1-q_2)^2 . \label{3}\end{aligned}\ ] ] here , @xmath12 is the dimensionless parameter , @xmath13 , @xmath14 , @xmath15 is the electron charge , @xmath16 is the separation distance between the reaction coordinates @xmath17 and @xmath18 of the tunneling particles ; @xmath19 is the dielectric constant . the form of potential energy surface ( [ 3 ] ) is shown in fig . [ fig1 ] . asymmetric potential energy surface ( [ 3 ] ) for parallel tunneling ; @xmath20 , @xmath21 , @xmath22 . @xmath23 and @xmath24 indicate the initial and final states of the particles , respectively . the minimum of the potential at @xmath24 is lower than that of at @xmath23 . the other two ( intermediate ) minima are lower than those of at @xmath23 and higher than at @xmath24.,scaledwidth=60.0% ] for antiparallel transfer , the two - dimensional potential energy with the interaction term can be defined as @xmath25\theta(q_1 ) + ( q_2-a)^2\theta(q_2 ) \nonumber\\ + \left[-(b^2-a^2 ) + ( q_2+b)^2\right]\theta(-q_2 ) - \frac{\tilde\alpha^*}{2}(q_1-q_2)^2 , \label{4}\end{aligned}\ ] ] where @xmath26 is a dimensionless parameter , @xmath27 . the potential ( [ 4 ] ) is depicted in fig . [ fig2 ] . the main difference between the potential energy surfaces ( [ 3 ] ) and ( [ 4 ] ) is in the initial location of the second particle , @xmath28 . asymmetric potential energy surface ( [ 4 ] ) for antiparallel tunneling ; ( a ) @xmath20 , @xmath29 , @xmath30 ( top panel ) ; ( b ) @xmath20 , @xmath29 , @xmath31 ( bottom panel ) . @xmath23 and @xmath24 indicate initial and final states of the particles . the minimum of the potential at @xmath24 is lower than that of at @xmath23 . the other two ( intermediate ) minima are higher than those of at @xmath23 and @xmath24.,title="fig:",scaledwidth=60.0% ] asymmetric potential energy surface ( [ 4 ] ) for antiparallel tunneling ; ( a ) @xmath20 , @xmath29 , @xmath30 ( top panel ) ; ( b ) @xmath20 , @xmath29 , @xmath31 ( bottom panel ) . @xmath23 and @xmath24 indicate initial and final states of the particles . the minimum of the potential at @xmath24 is lower than that of at @xmath23 . the other two ( intermediate ) minima are higher than those of at @xmath23 and @xmath24.,title="fig:",scaledwidth=60.0% ] the potential energy ( [ 3 ] ) can be referred to as `` parallel '' while the potential energy ( [ 4 ] ) as `` antiparallel '' . we also define a `` symmetric '' potential energy as a particular case of the potential ( [ 4 ] ) under the condition @xmath32 , i.e. , @xmath33 where the constant @xmath34 . the potential energy ( [ 5 ] ) is presented in fig . [ fig3 ] . symmetric potential energy surface ( [ 5 ] ) ; ( a ) @xmath20 , @xmath35 , @xmath36 ( top panel ) ; ( b ) @xmath20 , @xmath35 , @xmath37 ( bottom panel ) . @xmath23 and @xmath24 denote the initial and final states of the particles . the minimum of the potential energy at @xmath24 is equal to that of at @xmath23 . the other two ( intermediate ) minima are higher than those of at @xmath23 and @xmath24.,title="fig:",scaledwidth=60.0% ] symmetric potential energy surface ( [ 5 ] ) ; ( a ) @xmath20 , @xmath35 , @xmath36 ( top panel ) ; ( b ) @xmath20 , @xmath35 , @xmath37 ( bottom panel ) . @xmath23 and @xmath24 denote the initial and final states of the particles . the minimum of the potential energy at @xmath24 is equal to that of at @xmath23 . the other two ( intermediate ) minima are higher than those of at @xmath23 and @xmath24.,title="fig:",scaledwidth=60.0% ] as shown in refs . @xcite , such model potential energy surfaces well describe the dynamics in porphyrins ( see also sec . iii below ) . in many practical cases , the effect of a bath on particles tunneling should be also included . in the next section we consider two particles embedded into a harmonic medium and linearly interacting with the bath modes . as it was mentioned in introduction , adiabatic tunnel dissipative transfer appears to be very important in various physical and chemical processes @xcite-@xcite . it is of much interest to consider two - particle tunnel dissipative transfer of coupled charges . for example , a two - proton transfer can be characterized by a simultaneous tunneling and the changing of electronic structure which forms and breaks chemical bonds @xcite . one can consider , e.g. , transfer of two protons in the dimer of 7-azaindole @xcite ; see fig . [ fig4new ] . the dimer of 7-azaindole : ( a ) before the proton transfer , ( b ) after the proton transfer.,scaledwidth=50.0% ] a view of two equilibrium configurations , i and ii , for the porphyrin molecule with the transfer of a pair of hydrogen atoms between nitrogen atoms . the intermediate ( nonequilibrium ) configuration iii is characterized by the hydrogen atoms residing on the neighbor nitrogen atoms.,scaledwidth=50.0% ] temperature dependence of the observed and predicted @xcite rate constants in the reaction of the internal transfer of hydrogen ( or deuterium ) atoms in porphyrin molecule . open and filled circles denote experimental data for the hydrogen and deuterium atoms respectively @xcite . the lines denoted by 1h and 1d represent predicted curves @xcite for a synchronous transfer of weakly coupled hydrogen and deuterium atoms respectively . the lines denoted 2h and 2d represent predicted curves for asynchronous transfer of weakly coupled hydrogen and deuterium atoms respectively . all the coupling constants @xmath38 were scaled to reproduce the results for hydrogen atom transfer at @xmath39 k. the used parameters are : the stretching and bending frequencies for the hydrogen atom are 3318 and 2482 @xmath40 , for the deuterium atoms are 1223 and 1094 @xmath40 respectively @xcite ; h and d stretching anharmonics and reduced masses are @xmath4180 and @xmath42 @xmath40 , and 1 and 2 respectively ; the skeletal frequency is 380 @xmath40 and the reduced mass is 2.8 . the equilibrium separation n@xmath41h ( n@xmath41d ) is 400 pm . endothermic value for an asynchronous mechanism is @xmath43 ev , the coupling constant for synchronous transfer is @xmath44 ev , the coupling constant for asynchronous transfer of weakly coupled atoms is @xmath45 ev , the coupling constant for asynchronous transfer of strongly coupled atoms is @xmath46 ev.,scaledwidth=50.0% ] it is particularly interesting to consider two - dimensional tunnel dynamics of coupled protons in a symmetric case of porphyrin molecules . for this symmetric reaction an unstable interchanging of the regimes of tunneling ( synchronous regime is replaced by asynchronous one as temperature decreases ) was observed for the first time . the interchanging corresponds to the instability cleavage in the dependence of the rate constant on temperature . in some aspect this picture resembles the effect predicted for interacting josephson junctions @xcite . in fig . [ fig5new ] we schematically present the above mentioned reaction of the coupled protons transfer in porphyrin molecule . [ fig6new ] @xcite represents the experimental and the theoretically estimated temperature dependencies of the reaction rate for the transfer of internal hydrogen ( or deuteron ) atoms in porphyrin . as it was mentioned by w. siebrand ( div . of chemistry , nat research council of canada , ottawa ) , a modified and extended microscopic model could be used to describe transfer charge transfer states in molecular dimers or crystals . such states can be represented in the form @xmath47 that formally corresponds to a two - electron transfer . two types of dimers corresponding to @xmath48 and @xmath41 combinations are separated by the band which is associated to the frequency of the transfer . none of these states have a constant dipole moment but an external field can mix them and induce a quite considerable dipole moment . in experiments , the dependence of the absorption spectrum of such systems on the external field was measured . the results were interpreted in terms of charge - transfer excitons . this interpretation is valid only if the external field is sufficiently strong , namely , to be able to break the symmetry . therefore , it is of much interest to investigate splitting in the spectrum of charge - transfer dimer . in early studies one neglected an electron - electron interaction that appears to be of much importance here . the splitting seems to be much smaller than it was estimated earlier . in terms of the model presented below , electrons are attracted by coulomb force of positive molecular ions and repel each other . since the instantaneous situation when the electrons are at equal distance from the two molecules is energetically not favorable , a high barrier for the transfer is formed . an asynchronous transfer gives an intermediate state corresponding to highly excited vibronic state so that this type of transfer is not favorable due to frank - condon factors . also , this state rapidly undergoes vibronic relaxation that corresponds to a non - radiative decay of charge - transfer states @xcite . as the result , it is still important to describe the observed properties in terms of controlled processes of electron transfer . although w. siebrand expressed some doubts on the possibility of a direct use of the model presented below to the two - proton transfer in porphyrin - type systems , he pointed out that there is a possibility to use such a nonlinear two - particle model in studying experimentally observed effectively two - dimensional ( and two - particle ) electron transfer . two - dimensional adiabatic potential and the path characterized by a minimal energy for ( a ) synchronous and ( b ) asynchronous mechanisms of nh - tautometry . , title="fig:",scaledwidth=45.0% ] two - dimensional adiabatic potential and the path characterized by a minimal energy for ( a ) synchronous and ( b ) asynchronous mechanisms of nh - tautometry . , title="fig:",scaledwidth=45.0% ] the microscopic model suggested in this paper allows one to interpret the experimental data , and provides complete qualitative and good quantitative description . also , the model allows one to determine the critical point in an analytically exact manner ( in the quasiclassical approximation ) , both the temperature and the coupling coefficient , that is an interesting theoretical result which is analogous to that of ref . @xcite . as an indirect confirmation of applicability of the microscopic model we mention a comparison with the results of modelling of the potential energy surface in porphyrins @xcite . for the case of 7-azaindole dimer , the protons are initially reside closely to nitrogen atoms in the corresponding 5-chain `` rings '' of the dimer . when two coupled molecules of 7-azaindole form a dimer there may occur a cooperative ( tunnel ) two - proton transfer . the protons are moved to nitrogen atoms of 6-chain `` rings '' of the dimer . the proton transfer is accompanied by changing of the structure of orbitals of electrons . a synchronous transfer of two protons appears to be energetically preferable while the independent tunneling of the protons becomes forbidden due to the changing of the structure of orbitals . in sections below , we consider a synchronous tunnel transfer of two weakly coupled particles . for two uncoupled particles , each of the particles moves independently in the proper double - well potential . we will study the effect of coupling to oscillators ( an environment ) at independent ( _ asynchronous _ ) and cooperative ( _ synchronous _ ) tunneling of the particles . also , we will investigate fine structure of the arising two - dimensional bifurcations which occur during the changing of regimes . as it was mentioned above , the study of temperature - dependent nmr spectra of porphyrins ( compounds of this class are known to play an important role in biological processes ) revealed intermolecular regrouping ( see fig . [ fig5new ] ) . to describe this regrouping two mechanisms are suggested : ( i ) the _ synchronous _ one ( which is associated to a synchronous transfer of two nh hydrogens ) and ( ii ) the _ asynchronous _ one ( with consecutive transfer of one of the particles ) . two - dimensional potential energy surfaces ( pes ) @xmath49 have been calculated by using indo and we represent them in the following analytical way : v(r , ) = c [ a()(r - r_min())^2 + b ( ) ] . the synchronous case : @xmath50 r_min ( ) = 1.00 - 0.14(4 ) , a ( ) = 38.1(4)+50.8 /^2 . the asynchronous case : @xmath51 r_min ( ) = 1.00 - 0.14(4 ) , a ( ) = 19.05(4)+25.4 /^2 . the potential along the path characterized by a minimal energy and pes @xmath49 are illustrated in fig . [ fig7new ] . by comparing of the chosen model potential ( see fig . [ fig3 ] ) with that of fig . [ fig7new ] one can see that the chosen model potential corresponds to a realistic pes for porphyrin . more details on pes for porphyrin have been discussed in ref . we assume that two particles independently interact with a harmonic bath . such interaction is considered in a bilinear approximation . the dynamics of the environment is described by the oscillator hamiltonian ( we use @xmath52 , @xmath53 units with the oscillators masses equal to one ) , [ 6 ] h _ = _ i(p_i^2 + _ i^2q_i^2 ) . each of the tunneling particles ( electrons or effective charges ) interacts with the oscillator bath in the following way : [ 7 ] v_^(1)(q_1 , q_i ) = q_1_ic_iq_i , v_^(2)(q_2 , q_i ) = q_2_ic_iq_i , as in ref . @xcite , we are interested in the transition probability per unit time or , more precisely , only in its exponential part which can be written in the langer s form , [ 8 ] = 2 t . in such a consideration , metastable levels can be presented as [ 9 ] = -2 e , e = e_0 - i/2 . it should be emphasized that eq . ( [ 8 ] ) is valid only for temperatures below the crossover temperature while for higher temperatures a different prefactor is required @xcite . equation ( [ 8 ] ) is obtained by generalizing the expression ( [ 9 ] ) to nonzero temperatures @xcite [ 10 ] = = = . here , @xmath54 labels the energy levels in the metastable state , @xmath55 is the partition function of the system , and @xmath56 is the temperature . to calculate @xmath57 , it is convenient to present @xmath55 in the form of a path integral @xcite [ 11 ] z = _ i dq_1dq_2dq_i . here , @xmath58 denotes an underbarrier action for the total system . the imaginary part @xmath59 is due to the decay of the energy states in the initial well . the validity of this approximation requires dissipation to be strong enough so that only an incoherent decay occurs @xcite . the imaginary part in the partition function with a double well potential energy can be also explained due to the strong dissipation to the bath . indeed , the particles do not come back to their initial state . coherent oscillations can happen only if the interaction with bosons is weak enough @xcite or the bath is in nonequilibrium state @xcite . the integral ( [ 11 ] ) can be performed over phonon coordinates @xcite resulting in @xmath60[q_1(\tau')+q_2(\tau ' ) ] ) \biggr ] , \label{12}\end{aligned}\ ] ] where [ 13 ] d ( ) = _ n=-^d(_n)e^i_n , @xmath61 is an inverse temperature ( below we assume that @xmath52 and @xmath53 ) , @xmath62 is the matsubara s frequency , and [ 14 ] d(_n ) = - _ i. a trajectory that minimizes the euclidean action @xmath58 can be found from the equations of motion . in particular , we embark on the antiparallel tunneling @xmath63 \nonumber\\ + \omega^2a\theta(-q_1)-\omega^2b\theta(q_1 ) = 0,\quad \label{15}\end{aligned}\ ] ] @xmath64 \nonumber\\ - \omega^2a\theta(q_2)+\omega^2b\theta(-q_2 ) = 0 . \quad \label{16}\end{aligned}\ ] ] in eqs . ( [ 15 ] ) and ( [ 16 ] ) the kernel @xmath65 is defined by [ 17 ] k ( ) = _ n=-^_ne^i_n . here , @xmath66 is the green s function defined by eq . ( [ 14 ] ) without the zero - frequency term , [ 18 ] d(_n ) = -_i + _ n. thus , we seek for solutions of eqs . ( [ 15 ] ) and ( [ 16 ] ) by expanding the trajectories @xmath67 and @xmath68 into fourier series , [ 19 ] q_1 = _ n=-^q_n^(1 ) e^i_n , q_2 = _ n=-^q_n^(2 ) e^i_n . introducing the renormalized frequency and interaction constant , [ 20 ] _ 0 ^ 2 = ^2 - _ i - , _ 1 = - _ i , respectively , and substituting eqs . ( [ 19 ] ) into eqs . ( [ 15 ] ) and ( [ 16 ] ) , we obtain that for @xmath69 , @xmath70 and for @xmath71 , @xmath72 here , we have introduced the following notation : [ 23 ] = _ 1-_2 , _ 0 = ( _ 1+_2)/2 . the time instants @xmath73 and @xmath74 , at which the particles pass the top points of the barrier , are determined from the following equations : [ 24 ] q_1(_1)=0 , q_2(_2)=0 . equations ( [ 24 ] ) allow us to change the argument of the @xmath75-function . namely , instead of the @xmath17- and @xmath18-dependencies , we can use a time - dependent @xmath75-function . this reduces eqs . ( [ 15 ] ) and ( [ 16 ] ) to a linear form . finally , substituting the trajectory determined from eqs . ( [ 19 ] ) , ( [ 21 ] ) and ( [ 22 ] ) into eq . ( [ 12 ] ) , we arrive at the following expression for the instanton action : @xmath76.\ \label{25}\end{aligned}\ ] ] for the case of parallel particle tunneling , the euclidean action @xmath58 can be determined similarly to antiparallel tunneling [ see eqs . ( [ 15 ] ) and ( [ 16 ] ) ] . the trajectory minimizing the euclidean action ( instanton ) , can be determined from the equations of motion . as in the previous section , we seek for solutions of these equations in the form of the fourier expansion ( [ 19 ] ) . the time instants @xmath73 and @xmath74 are determined by eqs . ( [ 24 ] ) . in the case of parallel tunneling particles [ the potential energy ( [ 3 ] ) ] , the resulting euclidean action is given as follows : @xmath77 , \label{26}\end{aligned}\ ] ] where @xmath66 is defined by eq . ( [ 18 ] ) . below , we use the following notation : @xmath78 @xmath79 @xmath80 @xmath81 @xmath82 and assume that @xmath83 in the absence of interaction with an oscillator bath , i.e. , at @xmath84 , the action ( [ 26 ] ) as a function of the parameters @xmath19 and @xmath85 yields @xmath86 \nonumber\\ + \frac{(\tau\!\!-\!\!|\varepsilon|)\alpha^*}{1-\alpha^ * } \!+\ ! \mathrm{coth}\beta^ * \!-\ ! \mathrm{sinh}^{\!-\!1}\beta^ * [ \mathrm{cosh}(\beta^*\!\!-\!\!\tau)\mathrm{cosh}\varepsilon \nonumber\\ + \mathrm{cosh}(\beta^*\!\!-\!\!\tau ) - \mathrm{cosh}(\beta^*-|\varepsilon| ) ] \nonumber\\ - ( 1-\alpha^*)^{-3/2 } \bigl[- \mathrm{coth}(\beta\sqrt{1-\alpha^ * } ) \nonumber\\ + \mathrm{sinh}^{-1}(\beta\sqrt{1-\alpha^ * } ) [ \mathrm{cosh}((\beta^*\!\!-\!\!\tau)\sqrt{1-\alpha^ * } ) \nonumber\\ \times ( \mathrm{cosh}(\varepsilon\sqrt{1-\alpha^ * } ) -1 ) + \mathrm{cosh}((\beta^*-|\varepsilon|)\sqrt{1-\alpha^ * } ) ] \bigr ] \biggr\}. \quad \label{27}\end{aligned}\ ] ] as soon as the trajectory is found , eqs . ( [ 24 ] ) can be presented in the following form : @xmath87 + \nonumber \\ \frac{1}{1\!-\!\alpha^*}\ , \mathrm{sinh}(\varepsilon\sqrt{1\!-\!\alpha^ * } ) [ \mathrm{cosh}(\tau\sqrt{1\!-\!\alpha^*})\mathrm{coth } ( \beta^*\sqrt{1\!-\!\alpha^ * } ) \nonumber \\ -\mathrm{sinh}(\tau\sqrt{1-\alpha^ * } ) + \mathrm{coth}(\beta^*\sqrt{1-\alpha^ * } ) ] = 0 , \nonumber\\ 3-\frac{4}{1+b^ * } - \frac{1}{1-\alpha^ * } + \mathrm{cosh}\varepsilon [ \,\mathrm{sinh}\tau \mathrm{coth}\beta^ * -\mathrm{cosh}\tau \nonumber \\ - 1 ] + \mathrm{sinh}\tau \mathrm{coth}\beta^ * - \mathrm{cosh}\tau \nonumber \\ + \frac{1}{1\!-\!\alpha^*}\,\mathrm{cosh } ( \varepsilon\sqrt{1\!-\!\alpha^ * } ) [ \,\mathrm{sinh}(\tau\sqrt{1\!-\!\alpha^ * } ) \nonumber \\ \times \mathrm{coth}(\beta^*\sqrt{1\!-\!\alpha^ * } ) - \mathrm{cosh}(\tau\sqrt{1-\alpha^ * } ) + 1 ] \nonumber \\ -\frac{1}{1-\alpha^ * } [ \,\mathrm{sinh}(\tau\sqrt{1-\alpha^*})\ , \mathrm{coth}(\beta^*\sqrt{1-\alpha^ * } ) \nonumber \\ -\mathrm{cosh}(\tau\sqrt{1-\alpha^*})]=0.\quad \label{28}\end{aligned}\ ] ] simple analytic solutions of eqs . ( [ 28 ] ) are obtained in the particular case when @xmath88 + \frac{\beta}{4}. \label{29}\end{aligned}\ ] ] however , a complete analysis requires numerical studies . at sufficiently low temperatures , @xmath89 , for @xmath90 , and @xmath91 we finally obtain , with the exponential accuracy , @xmath92(1-\alpha^*)^{1/(1-\sqrt{1-\alpha^ * } ) } \nonumber \\ \times\left\ { 1 + ( 1-\alpha^*)^{1/(1-\sqrt{1-\alpha^*})}[-\frac{1}{1-\alpha^ * } + ( 3-\frac{4}{1+b^ * } \right . \left . -\frac{1}{1-\alpha^*})/(1-\sqrt{1-\alpha^ * } ) ] \right\}^{-1 } , \nonumber \\ e^{-\varepsilon } \simeq [ 3 - \frac{4}{1+b^ * } - \frac{1}{1-\alpha^*}]e^{\tau\sqrt{1-\alpha^ * } } + \frac{1}{1-\alpha^*}. \quad \label{30}\end{aligned}\ ] ] the solution ( [ 30 ] ) is valid at > _ c . [ 31 ] we point out that an approximate solution can be found for large values of the parameter @xmath93 ( and small @xmath94 ) . however , below we focus on the more important solution ( [ 30 ] ) . the analysis indicates that there are no perturbative solutions of eqs . ( [ 28 ] ) at low temperatures and small @xmath19 . for @xmath95 [ see eq . ( [ 29 ] ) ] , the action ( [ 27 ] ) results in @xmath96 \nonumber \\ - \frac{1}{2}\omega^2(b^2-a^2)\beta + \omega(b+a)^2 \biggl[\mathrm{cosh}\frac{\omega\beta}{2 } \nonumber \\ -\left(1+\frac{(b^*-1)^2}{(b^*+1)^2 } \mathrm{sinh^2}\frac{\omega\beta}{2}\right)^{1/2}\biggr ] \mathrm{sinh}^{-1}\frac{\omega\beta}{2}. \label{32}\end{aligned}\ ] ] the action ( [ 32 ] ) coincides ( up to a factor of two ) with that of calculated in ref . thus , we have calculated the two - particle euclidean action for the case of synchronous parallel motion of the two interacting particles . in the symmetric case ( @xmath97 ) , the action ( [ 32 ] ) reduces to s _ c=0 + a = b = 4 ^ 2 . [ 33 ] at @xmath98 , the character of the temperature dependence is almost the same . at @xmath99 , the corresponding action @xmath58 can be obtained by substituting eq . ( [ 30 ] ) into eq . ( [ 27 ] ) ( for brevity , this cumbersome expression is omitted ) . a simple analysis shows that @xmath100 . moreover , it appears that the difference @xmath101 has a maximum at @xmath89 . trajectories ( a single path is characterized by @xmath102 and a splitted path is characterized by @xmath99 ) for two parallel moving particles , at @xmath89 ; ( 1)-(4 ) label the projections of the minima of potential energy ( [ 3]).,scaledwidth=60.0% ] the tunneling paths ( [ 19 ] ) are presented by the solutions ( [ 29 ] ) and ( [ 30 ] ) . at the critical point , @xmath103 , defined by eq . ( [ 31 ] ) , a relatively abrupt change in the dynamics results in a splitting of the single trajectory ( @xmath102 ) into two ( @xmath99 ) , as shown in fig . [ fig4 ] . instanton action as a function of the interaction parameter @xmath104 of the two parallel moving particles , at @xmath105 ; @xmath106 ; ( 1 ) is the single trajectory ; ( 2 ) is the split trajectory.,scaledwidth=60.0% ] at @xmath107 , i.e. , for temperatures below the critical temperature , @xmath108 , only a split trajectory ( @xmath99 ) occurs since @xmath109 ( see fig . [ fig5 ] ) . when @xmath110 , i.e. , at @xmath111 , and @xmath112 [ see eq . ( [ 30 ] ) ] , the solution is determined by a single trajectory . we have also found that the single trajectory solution holds in the whole temperature range for a symmetric potential energy ( @xmath97 ) . the above consideration is based on the analytic solutions ( [ 29 ] ) for @xmath113 near the critical point . a complete analysis of eq . ( [ 28 ] ) requires more detail . numerical solutions of transcendental equations ( [ 28 ] ) . in addition to the studied solution @xmath114 , with larger @xmath115 there are additional solutions , @xmath116 , shown in panels ( b ) and ( c ) . panel ( b ) reveals a _ single _ additional solution corresponding to the lower value of the euclidean action . panel ( c ) shows two different additional solutions with the one corresponding to the lower value of the action.,scaledwidth=50.0% ] numerical solutions of transcendental equations ( [ 28 ] ) . in addition to the studied solution @xmath114 , with larger @xmath115 there are additional solutions , @xmath116 , shown in panels ( b ) and ( c ) . panel ( b ) reveals a _ single _ additional solution corresponding to the lower value of the euclidean action . panel ( c ) shows two different additional solutions with the one corresponding to the lower value of the action.,scaledwidth=50.0% ] numerical solutions of transcendental equations ( [ 28 ] ) . in addition to the studied solution @xmath114 , with larger @xmath115 there are additional solutions , @xmath116 , shown in panels ( b ) and ( c ) . panel ( b ) reveals a _ single _ additional solution corresponding to the lower value of the euclidean action . panel ( c ) shows two different additional solutions with the one corresponding to the lower value of the action.,scaledwidth=52.0% ] the numerical study of the set of transcendental equations ( [ 28 ] ) reveals remarkable features of two - dimensional tunneling . the value of the critical parameter @xmath117 decreases with temperature as shown in figs . [ fig6 ] , [ fig7 ] , and [ fig8 ] . the temperature dependence of @xmath117 indicates to the existence of a finite low - temperature limit . it becomes clear that the weaker the interaction between the tunneling particle , the lower temperatures are required for the synchronous tunneling . since the synchronous tunneling is valid only for @xmath118 [ see eq . ( [ 31 ] ) ] the dependence of @xmath119 on the interaction parameter reveals that the minimum occurs at @xmath120 . the nonmonotonic behavior of @xmath119 as a function of @xmath94 can be explained by the @xmath94-dependence of @xmath85 as well . parameter @xmath117 as a function of inverse temperature , at the fixed value of the frequency @xmath7 and asymmetry parameter @xmath93.,scaledwidth=60.0% ] additionally , the curve in fig . [ fig8 ] exhibits an anomalous behavior ( an increasing part of the curve ) that can be explained in the following way : a two - dimensional potential energy for parallel tunneling is evidently deformed by an increase of the interaction constant . indeed , the minima of the potential surface become lower and the distance between them is larger . small deformations in the potential energy surface change the contribution produced by the temperature increase towards a synchronous character of the tunneling . however , for sufficiently large deformations of the potential , the situation is a quite opposite . the significant deformations are in favor of synchronous transfer . thus , an increase in the interaction constant provides a similar effect when temperature increases . such a behavior takes place up to the certain value of the interaction constant , beyond which the potential energy surface becomes strongly perturbed . thus , figs . [ fig7 ] and [ fig8 ] are complimentary to each other . consequently , fig . [ fig8 ] can be viewed as a bifurcation diagram . indeed , the region below the curve corresponds to the synchronous tunneling , while the region above the curve corresponds to the asynchronous one . parameter @xmath119 is presented as a function of the interaction parameter of the two tunneling particles.,scaledwidth=60.0% ] fig . [ fig6 ] demonstrates how the critical parameter @xmath119 changes from synchronous to asynchronous values with temperature . [ fig6]b and [ fig6]c reveal the existence of the additional points , referred to as the bifurcation points , corresponding to the change of tunneling regimes . for the two bifurcation points shown in fig . [ fig6]c , only one is physical because it corresponds to the lower value of the action ( the second point is metastable ) . however , for sufficiently small differences between the values of the action at these bifurcation points , _ both _ close points contribute , thus leading to corresponding `` fluctuations '' in the system during the change of the regimes . for lower temperatures , such fluctuations become negligible because two bifurcation points contributes differently . thus , a stable character of the asynchronous tunnel transfer is achieved due to a much lower value of the action at one of the bifurcation points . for chemical reactions mentioned in introduction , the effect of the change in the tunneling regimes reveals as a _ cleavage _ in the experimental temperature dependence of the tunneling constant . fine details of this instability have not been studied yet . nevertheless , the very existence of instability at the edge of the bifurcation can be explained as the result of specific fluctuations . a numerical analysis of the bifurcations in antiparallel tunneling is given in the next section . for antiparallel tunneling of two particles [ see the potential energy ( [ 4 ] ) ] , the instanton action as a function of the parameters @xmath19 and @xmath85 is determined by eq . ( [ 25 ] ) . for @xmath84 , we obtain @xmath121^{-1 } [ \mathrm{cosh}((\beta^*-\tau)\sqrt{1-\tilde\alpha^ * } ) \nonumber \\ -\mathrm{cosh}(\beta^*\sqrt{1-\tilde\alpha^ * } ) ] \nonumber \\ \left . + \frac{\mathrm{cosh}\varepsilon -1}{\mathrm{sinh}\beta^ * } [ \mathrm{cosh}(\beta^*-\tau ) + \mathrm{cosh}\beta^ * ] \right\}. \label{34}\end{aligned}\ ] ] the parameters @xmath19 and @xmath85 are found from the following set of equations [ see eq . ( [ 24 ] ) ] : @xmath122 \nonumber \\ + \frac{1}{1-\tilde\alpha^ * } \mathrm{sinh}(\varepsilon\sqrt{1-\tilde\alpha^ * } ) [ \mathrm{coth}(\beta^*\sqrt{1-\tilde\alpha^ * } ) \nonumber \\ -\mathrm{cosh}(\tau\sqrt{1-\tilde\alpha^ * } ) \mathrm{coth}(\beta^*\sqrt{1-\tilde\alpha^ * } ) \nonumber \\ + \mathrm{sinh}(\tau\sqrt{1-\tilde\alpha^ * } ) ] = 0 , \nonumber \\ -1 - \frac{4}{(1+b^*)(1\!-\!\tilde\alpha^ * ) } + \frac{1}{1\!-\!\tilde\alpha^ * } \nonumber \\ + ( \mathrm{cosh}\varepsilon -1 ) ( \mathrm{sinh}\tau\mathrm{coth}\beta^ * -\mathrm{cosh}\tau ) \nonumber \\ + \mathrm{cosh}\varepsilon % % \nonumber \\ + \frac{1}{1-\tilde\alpha^ * } \left\ { [ \mathrm{cosh}(\varepsilon\sqrt{1-\tilde\alpha^*})+1 ] \right . \nonumber \\ \times [ \mathrm{sinh}(\tau\sqrt{1-\tilde\alpha^ * } ) \mathrm{coth}(\beta^*\sqrt{1-\tilde\alpha^ * } ) \nonumber \\ - \mathrm{cosh}(\tau\sqrt{1-\tilde\alpha^ * } ) ] \left . - \mathrm{cosh}(\varepsilon\sqrt{1-\tilde\alpha^ * } ) \right\ } = 0 . \quad \label{35}\end{aligned}\ ] ] simple analytic solutions of eqs . ( [ 35 ] ) can be obtained in the following form : @xmath123 + \frac{\beta}{4}. \label{36}\end{aligned}\ ] ] similarly to the case of parallel tunneling , we obtain that at low temperatures , @xmath89 , with the exponential accuracy , @xmath124 here , @xmath125 while @xmath126 , @xmath93 , @xmath19 , and @xmath85 are determined as for the parallel transfer . the solution ( [ 37 ] ) is valid at @xmath127 , where the lower and upper bounds @xmath128 and @xmath129 are derived from a cumbersome transcendental equation ( for brevity it is not presented here ) . particularly , in the symmetric case , @xmath97 , we obtain the condition in a simple analytic form , @xmath130 . furthermore , an approximate solution can be found for large values of the parameter @xmath131 ( and small @xmath132 ) . however , we restrict our analysis to the more important physical solution ( [ 37 ] ) . the @xmath115-dependent solution ( [ 37 ] ) is valid for @xmath133 , where @xmath134 \label{38}\end{aligned}\ ] ] at @xmath89 , the solutions of eqs . ( [ 35 ] ) can be found perturbatively ( for small @xmath19 ) with given values of the parameters @xmath135 and @xmath126 . at @xmath102 [ the solution ( [ 36 ] ) ] , the action ( [ 34 ] ) yields : @xmath136 \nonumber \\ - \frac{\omega^2\beta(b^2-a^2)}{2(1-\tilde\alpha^2 ) } + \frac{\omega(b+a)^2}{(1-\tilde\alpha^*)^{3/2 } } \left [ \mathrm{coth}\frac{\omega\beta\sqrt{1-\tilde\alpha^*}}{2 } \right . \nonumber \\ \left . -\left ( \mathrm{sinh}^{-2}\frac{\omega\beta\sqrt{1-\tilde\alpha^*}}{2 } + \frac{(b - a)^2}{(b+a)^2 } \right)^{\ ! 1/2 } \right ] . \label{39}\end{aligned}\ ] ] for the symmetric potential ( @xmath32 ) and @xmath102 , we obtain that ( see fig . [ fig7 ] ) s = . [ 40 ] we do not present here a cumbersome expression for @xmath137 which one can obtain by substituting the solutions @xmath85 and @xmath19 into eq . ( [ 34 ] ) . a simple analysis reveals that @xmath138 for @xmath139 and for relevant @xmath126 . similarly to parallel transfer , the tunnel paths can be found from eqs . ( [ 36 ] ) and ( [ 37 ] ) . these trajectories on the @xmath140-plane are shown in fig . [ fig9 ] . trajectories ( the basic path characterized by @xmath102 and the split one is characterized by @xmath99 ) at @xmath89 of two antiparallel tunneling particles ; ( 1)-(4 ) denote the projections of the minima of the potential energy surface @xmath141 defined by eq . ( [ 4]).,scaledwidth=60.0% ] as for parallel tunneling and at @xmath142 , the pair - tunneling changes from a single to double trajectory regime . in contrast to the parallel tunneling , such a splitting occurs at any values of the parameters of the potential . at @xmath142 , we have @xmath143 and , therefore , @xmath144 determines the tunneling rate . at @xmath145 , the two degenerated trajectories are transformed into a single trajectory , @xmath146 , corresponding to synchronous antiparallel transfer . for single particle tunneling , there is only a single tunneling path ( instanton ) minimizing the action . hence , there are two different types of trajectories for the pair of interacting particles . namely , the main contribution to the instanton action is determined either by the single or by the double - degenerated paths depending on the value of @xmath115 . we also point out that in the case of the parallel tunneling , the particles do not simultaneously pass the top points of the barrier , @xmath147 . this means that the tunneling transfer is asynchronous . at small values of the interaction parameter @xmath94 [ see eq . ( [ 30 ] ) ] and at temperatures such that @xmath110 [ see eq . ( [ 31 ] ) ] , there is no splitting of a single path ( @xmath148 ) . therefore , the particles pass the top of the barriers at the same instants ( @xmath114 ) . consequently , the transfer of the particles is synchronous . the temperature dependence for the antiparallel transfer action is plotted in fig . [ fig10 ] at various @xmath126 . instanton action for antiparallel tunneling ( @xmath102 , @xmath32 ) as a function of an inverse temperature at the two different values of the interaction parameter ; @xmath149 , @xmath150 ; ( 1 ) @xmath151 ; ( 2 ) @xmath152 ; ( 3 ) a dashed line corresponds to the action ( [ 33 ] ) for the parallel transition ( @xmath102 , @xmath32).,scaledwidth=60.0% ] the type of the interaction given by eqs . ( [ 2])-([5 ] ) is such that it does not affect the motion along the `` center of mass '' coordinate , @xmath148 . for this reason , the euclidean action is independent of the interaction parameter as for parallel transfer . since the state of the interacting system characterized by a maximal value of the relative coordinate , @xmath146 , is preferable ( as it provides the lower action ) , it becomes clear that the instanton action decreases with the interaction parameter in the parallel transfer along the degenerated tunnel trajectories , and increases with the interaction parameter for the antiparallel tunneling . for antiparallel tunneling , synchronous transfer ( @xmath114 ) takes place , while asynchronous transfer is forbidden due to the greater contribution to the euclidean action ( see fig . [ fig10 ] ) . the validity condition for weakly interacting instanton - antiinstanton pairs , in the adiabatic approximation , was discussed in ref . @xcite . numerical solutions of transcendental equation ( [ 35 ] ) . in addition to the studied solution @xmath114 , with respect to @xmath115 , there are the additional solutions , @xmath147 , shown in panels ( a ) , ( b ) and ( c ) , that correspond to one , four , and six ( pairs ) additional solutions , respectively.,title="fig:",scaledwidth=50.0% ] numerical solutions of transcendental equation ( [ 35 ] ) . in addition to the studied solution @xmath114 , with respect to @xmath115 , there are the additional solutions , @xmath147 , shown in panels ( a ) , ( b ) and ( c ) , that correspond to one , four , and six ( pairs ) additional solutions , respectively.,title="fig:",scaledwidth=50.0% ] numerical solutions of transcendental equation ( [ 35 ] ) . in addition to the studied solution @xmath114 , with respect to @xmath115 , there are the additional solutions , @xmath147 , shown in panels ( a ) , ( b ) and ( c ) , that correspond to one , four , and six ( pairs ) additional solutions , respectively.,title="fig:",scaledwidth=50.0% ] the numerical analysis of transcendental equation ( [ 35 ] ) reveals interesting features for a transition region between the tunneling regimes , i.e. , a fine structure near the first bifurcation point for the antiparallel transfer . the numerical results are presented in fig . [ fig11 ] . we found that , in addition to the first bifurcation point characterized by the two solutions ( fig . [ fig11]a ) , there exist additional bifurcation points at lower temperatures , i.e. , the four pairs ( fig . [ fig11]b ) , the six pairs ( fig . [ fig11]c ) , and there are even twelve pairs of additional solutions at @xmath153 @xmath154 , etc . we refer to this phenomenon as multiplication of bifurcations or a _ cascade of bifurcations_. such an effect resembles a scenario of transition to chaos . although the synchronous regime is preferred due to the minimal instanton action , in a certain temperature range its value is comparable with those of corresponding to the cascade solutions . as a result , _ quantum fluctuations _ of a nonregular character occur in contrast to the parallel transfer . the antiparallel tunneling is , thus , characterized by the instability of the transition due to the synchronous to asynchronous behavior . such instabilities are similar to continuous second - order phase transition , while the parallel tunneling is viewed as a step process similar to phase transition of a first order ( see fig . [ fig6 ] ) . the dependencies shown in figs . [ fig7 ] and [ fig8 ] , @xmath155 and @xmath156 , for the antiparallel transfer are found to be of the same character as those of the parallel transfer . in summary , our investigation reveals a quite complicated fine structure of the transition for parallel and antiparallel tunneling of two particles with different degenerate trajectories leading to the bifurcation cascade . in this section , we study the effect of a heat bath on tunneling transition of two interacting particles . in many tunneling reactions , interaction with a vibrational subsystem can often be approximated by interaction with a single vibrational mode ( a so - called promoting mode ) . as it follows from eq . ( [ 12 ] ) , a heat bath affects only the dynamics of the center of mass ( @xmath148 ) . therefore , in the case of an antiparallel motion , the medium does not affect the rate constant , while for parallel tunneling it essentially contributes to the transfer rate . for both parallel and antiparallel transfer , bilinear interaction of the particles with a single oscillator can make a qualitative change in the character of tunneling . at small values of the interaction parameter between the two tunneling particles dissipative effects become important @xcite . in two dimensions the dissipative effects are more pronounced for parallel rather than antiparallel tunneling . the latter increases with temperature with a considerable contribution to the prefactor . in the present work , we are interested in the tunneling rate assuming only exponential evolution of the transition probability . however , nonexponential evolution can occur in an nonequilibrium environment @xcite . such a case is not discussed here . accordingly , a reservoir is assumed to be in thermodynamic equilibrium , i.e. , the tunneling transition is rather slow compared to the thermodynamic relaxation . thus , we assume that dissipation affects the value of instanton action only . for the case of the antiparallel tunnel transfer , the action ( [ 25 ] ) can be calculated with the following vibronic green s function : d(_n ) = - , [ 41 ] where @xmath157 is the frequency of the vibrational mode . after some tedious calculations , one obtains the following expression for the instanton action ( [ 25 ] ) : @xmath158 \nonumber \\ -\sum\limits_{i=1}^{2}\frac{\beta(\omega_i^2-\omega_l^2 ) \mathrm{cosh}(\varepsilon\omega_i ) } { ( -1)^i4\omega_i^3(\omega_1 ^ 2-\omega_2 ^ 2 ) \mathrm{sinh}(\beta\omega_i/2 ) } \nonumber \\ \times [ \mathrm{cosh}(\omega_i(\beta/2 - 2\tau_0 ) ) + \mathrm{cosh}(\omega_i\beta/2 ) ] % % \nonumber \\ % % + \frac{\beta(\omega_1 ^ 2-\omega_l^2 ) % % \mathrm{ch}(\varepsilon\omega_1 ) } % % { 4\omega_1 ^ 3(\omega_1 ^ 2-\omega_2 ^ 2 ) % % \mathrm{sinh}(\beta\omega_1/2 ) } % % \nonumber \\ % % \times [ % % \mathrm{ch}(\omega_1(\beta/2 - 2\tau_0 ) ) % % + \mathrm{ch}(\omega_1\beta/2 ) % % ] \right\}. \label{42}\end{aligned}\ ] ] here , we have introduced the following notation : @xmath159 for the parallel tunneling transition , a corresponding action can be found in a similar way . the action ( [ 27 ] ) as a function of the parameters @xmath160 and @xmath161 , with the vibronic frequency @xmath157 and coupling constant @xmath162 , yields ( see also ref . @xcite ) : @xmath163 \right . \nonumber \\ -\mathrm{cosh}[(\frac{\beta}{2}-\varepsilon^*)\sqrt{x_i } ] + \frac{1}{2}\mathrm{cosh } [ ( \frac{\beta}{2}-2\tau^*-\varepsilon^*)\sqrt{x_i } ] \nonumber \\ \left . \left . + \frac{1}{2}\mathrm{cosh } [ ( \frac{\beta}{2}-2\tau^*+\varepsilon^*)\sqrt{x_i } ] \right ) \right ] + \frac{\omega^4(a\!+\!b)^2}{2(\omega^2\!-\!2\alpha)^{3/2 } } \nonumber \\ \times \left [ -\mathrm{coth}(\frac{\beta}{2}\sqrt{\omega^2\!-\!2\alpha } ) + \mathrm{sinh}^{-1}(\frac{\beta}{2}\sqrt{\omega^2\!-\!2\alpha } ) \right . \nonumber \\ \left(\!\ ! -\mathrm{cosh}[(\frac{\beta}{2}\!-\!2\tau^ * ) \sqrt{\omega^2\!-\!2\alpha } ] + \mathrm{cosh}[(\frac{\beta}{2}\!-\!\varepsilon^ * ) \sqrt{\omega^2\!-\!2\alpha } ] \right . + \frac{1}{2}\mathrm{cosh } [ ( \frac{\beta}{2}-2\tau^*-\varepsilon^*)\sqrt{\omega^2 - 2\alpha } ] \nonumber \\ \left . + \frac{1}{2}\mathrm{cosh } [ ( \frac{\beta}{2}-2\tau^*+\varepsilon^*)\sqrt{\omega^2 - 2\alpha } ] \right ) \right].\quad \label{43}\end{aligned}\ ] ] here , we have denoted @xmath164 for particular values of the interaction constant @xmath9 and in the absence of interaction with the oscillator bath , the critical temperature @xmath0 ( at which the synchronous and asynchronous tunnel regimes interchange ) is found from eqs . ( [ 31 ] ) and ( [ 38 ] ) . these equations can be generalized to non - zero interaction with the promoting mode . typically , the critical temperature is found to be in the range from _ 10 k _ to _ 400 k_. in glasses , @xmath0 can be very small while for chemical reactions it can be rather large . additionally , @xmath0 depends on the mean distance between the particles and , therefore , on their concentration . quantum tunneling is important @xcite when @xmath165 . therefore , the symmetry breaking effects can take place at relatively high temperatures depending on the frequency of a barrier . for example , for porphyrins the critical temperature @xmath0 is estimated to be about _ 200 k_. in the single instanton approximation , we have calculated the euclidean action ( [ 12 ] ) for the models characterized by the different adiabatic potential energy surfaces , ( [ 3 ] ) , ( [ 4 ] ) and ( [ 5 ] ) , and made a detailed comparative analysis of the tunneling rate for two interacting particles moving in parallel or antiparallel within a dissipative environment . we have also assumed exponential evolution of the transition probability @xcite . we have shown that the change in a tunneling regime from synchronous to asynchronous transfer for parallel transition occurs as a step process , similar to phase transition of a first order , while for the antiparallel transfer it resembles second order phase transition . we have explained the effect of a _ cleavage _ in the experimentally observed @xcite temperature dependence of the reaction rate for two tunneling particles . it has been shown that the effect of symmetry breaking is stable for the parallel and unstable for the antiparallel transfer , as it is observed experimentally for some porphyrin systems @xcite . we have found a complicated fine structure in the bifurcation region due to the _ quantum fluctuations _ for the parallel two - dimensional tunneling . for the antiparallel tunneling , the contribution of four , six , twelve , etc . , pairs of trajectory becomes important resembling the transition to _ chaos_. additionally , we have studied interaction of two particles with phonons . such coupling essentially modifies the antiparallel and parallel transitions in different ways . as it follows from eq . ( [ 12 ] ) , the interaction with the reservoir does not change the dynamics of the center of mass for the antiparallel motion , while it makes a significant contribution to the transfer rate for the parallel transfer . finally , eq . ( [ 31 ] ) determines the validity condition for temperatures beyond of which stable two - dimensional synchronous tunneling correlations of all the kinds occur . + the authors would like to thank a. i. larkin and b. i. ivlev for the stimulating interest in this work . v. a. benderskii , e. v. vetoshkin , and e. i. kats , zh . . fiz . * 122 * , 746 ( 2002 ) ; cond - mat/0303275 ( 2003 ) ; v. a. benderskii , e. v. vetoshkin , e. i. kats , and h. p. trommsdorff , cond - mat/0209030 ( 2002 ) . j. hennig , h. h. limbach , j. chem . faraday trans . 75 * , 752 ( 1979 ) ; h. h. limbach , j. hennig , d. gerritzen , and h. rumpel , faraday discussions chem . soc . * 74 * , 229 ( 1982 ) ; p. stibbs and m. e. mosely , j. chem . faraday trans . 76 * , 729 ( 1980 ) . sebastian et al . , chem . phys . * 61 * , 125 ( 1981 ) ; * 75 * , 105 ( 1983 ) ; bonnds et al . , * 63 * , 303 ( 1981 ) ; * 95 * , 197 ( 1985 ) ; petelenz et al . , chem . phys . * 111 * , 209 ( 1987 ) ; * 119 * , 25 ( 1988 ) ; * 131 * , 409 ( 1989 ) ; * 133 * , 199 ( 1989 ) ; chem . * 133 * , 157 ( 1987 ) ; * 147 * , 430 ( 1988 ) .

we consider a model for two lasers that are mutually coupled optoelectronically by modulating the pump of one laser with the intensity deviations of the other . signal propagation time in the optoelectronic loop causes a significant delay leading to the onset of oscillatory output . multiscale perturbation methods are used to describe the amplitude and period of oscillations as a function of the coupling strength and delay time . for weak coupling the oscillations have the laser s relaxation period , and the amplitude varies as the one - fourth power of the parameter deviations from the bifurcation point . for order - one coupling strength the period is determined as multiples of the delay time , and the amplitude varies with a square - root power law . because we allow for independent control of the individual coupling constants , for certain parameter values there is an atypical amplitude - resonance phenomena . finally , our theoretical results are consistent with recent experimental observations when the inclusion of a low - pass filter in the coupling loop is taken into account . coupled lasers , delay , hopf bifurcation , resonance 34d15 , 37g15 , 39a11 , 78a60

nlin0608015

in recent work , we presented experimental and simulation results for two mutually coupled lasers @xcite with time - delayed asymmetric coupling . the light emitted from one laser propagates through fiber - optic cables to a photodetector that generates an electronic signal proportional to the light - intensity deviations from steady state . the electronic signal may be attenuated or amplified before it modulates the pump current of the other laser . the propagation time of the signal in the optoelectronic path introduces a significant time delay , and the coupling strength in each direction can be controlled separately . our work in @xcite investigated how the time delay and asymmetric coupling led to oscillatory and pulsating laser output . in this paper , we present a more theoretical exploration of the dynamics that result from this coupling configuration . in @xcite we investigated this same configuration ( asymmetric pump coupling ) but without including the effect of delay . in addition , the coupling constant from laser-2 to laser-1 was negative , while the coupling constant from laser-1 to laser-2 was positive . that is , we assumed that the electronic coupling signal from laser-2 was inverted before applying it to the pump of laser-1 . for purely harmonic signals , having opposite - sign coupling constants is equivalent to time - delayed coupling when the delay is half of the period . we found that as the coupling constant from laser-2 to laser-1 is increased in magnitude : ( i ) there is a hopf bifurcation to oscillatory output . ( ii ) for certain parameter values there exists a small - coupling resonance such that the amplitudes of both lasers are pulsating . ( iii ) for large coupling , laser-2 acts as a small - amplitude , nearly harmonic modulation to laser-1 . laser-1 exhibits period - doubling bifurcations to chaos and complex subharmonic resonances . in this paper we include the effect of the delayed - coupling , which results in a coupled set of delay - differential equations ( ddes ) . with delay there is again a hopf bifurcation to oscillatory output . however , the delay allows for periodic oscillations not just at the laser s relaxation period as in @xcite , but at periods that are integer multiples of the delay ; we refer to the former oscillatory solutions as internal modes , " and the latter as external modes . " we show that as the coupling strength is increased , the first instability is a hopf bifurcation to the internal mode . our nonlinear analysis then shows that the amplitude of the internal mode varies as the one - fourth root of the bifurcation parameter s deviation from the hopf bifurcation point . however , in the experiments of @xcite the output was not the internal mode but an external mode . the period of the oscillatory output was a multiple of the delay , and the amplitude of the oscillations varied as the square root of the bifurcation parameter . the discrepancy is due to the fact that the optoelectronic coupling contains an intrinsic low - pass filter that attenuates the internal mode . we show that the filter selects the external mode with the greatest possible frequency passed by the low - pass filter . a nonlinear analysis then shows that the amplitude of the external modes does indeed vary with a square - root power law . for the coupled lasers without delay in @xcite , we found that the parameters could be tuned to cause a resonance - type effect with respect to the coupling ; more specifically , the coupling could be tuned to maximize the amplitude of the oscillatory output . we find that this resonance effect also occurs with the inclusion of the delay . however , there are some dramatic qualitative differences as the coupling parameters are varied . in the system with delay , the branch of periodic solutions in the bifurcation diagram can be smoothly folded to form parameter intervals of bi - stability . then there is a critical value of the same coupling parameter beyond which the bi - stability disappears suddenly and non - smoothly . our coupling scheme is an example of `` incoherent coupling '' @xcite because it depends only on the laser s intensity and not on the complex electric field . this is because the intensity of one laser affects the other only indirectly ; in our case , the intensity of one laser modulates the pump current of the other . in contrast , `` coherent coupling '' refers to when the optical field of one laser is directly injected into the cavity of the other laser . analogously , a single laser with delayed and re - injected self - feedback would be called `` coherent '' feedback . semiconductor lasers with delayed coherent feedback have been extensively studied because of the their widespread application in electronics and communication systems ; there the laser s output signal may be reflected off external surfaces back into the laser . the lang - kobayashi ddes @xcite are the canonical model for semiconductor lasers with delayed self - feedback and have been used to investigate phenomena ranging from the onset of instabilities to the development of chaotic output referred to as `` coherence collapse '' ( see @xcite for a review ) . coupled sets of lang - kobayashi or related ddes are most often used to described coherently - coupled semiconductor lasers with delay two recent studies @xcite are in the same spirit as this paper as they track the number of and properties of oscillatory solutions that appear as the coupling strength is increased . the main simplification that results from considering incoherent coupling between the lasers is that neither the laser - frequency detuning nor the electric - field phase differences affect the dynamics of the coupled system @xcite . pieroux et al . @xcite have shown that the electrooptical coupling we consider in this paper leads to an equivalent dynamical model as incoherent coupling . there have recently been a number of other investigations of lasers with incoherent coupling . the main focus of many of these works was on chaotic synchronization ( see @xcite and included references).two recent papers by vicente et al . @xcite consider a similar implementation of optoelectronic coupling to that we consider here , i.e. , the intensity of one laser modulates the pump current of the other . their investigation of the oscillatory solutions is mainly numerical and they discuss interesting phenomena such as amplitude quenching , frequency locking and routes to chaos . delay - coupled relaxation and limit - cycle oscillators have been the subject of many investigations ( see @xcite and their included references ) . these systems are self - oscillatory and the amplitude is often fixed by the intrinsic properties of the oscillator . in many cases the system may be reduced to a phase equation , or a system of phase equations if the oscillators are coupled @xcite . in contrast , for laser systems the amplitude is an important dynamical variable . this is because the lasers we consider are weakly damped , nearly conservative systems such that the steady state is the only asymptotically stable solution . oscillations must be induced by external mechanisms such as modulation , injection or coupling @xcite . thus , the amplitude of the oscillations is highly dependent upon the external mechanism , in our case the coupling , rather than the individual laser . we should mention that for coupled limit - cycle oscillators there is the amplitude instability referred to as `` amplitude death , '' where for specific values of the coupling ( and oscillator parameter values ) the amplitude becomes zero . after presenting our model , we show that the linear - stability analysis predicts a critical value of the coupling constant ( that depends upon the delay time ) such that there is hopf bifurcation to the internal mode , i.e. , oscillations with the intrinsic relaxation period . we then use multiple - scale perturbation methods , modified to account for time delays @xcite , to analyze the long - time evolution of the internal mode . the results are a pair of complex stuart - landau ddes for the oscillation amplitudes that include a slow - time delay term . we analyze the amplitude equations to determine the bifurcation properties of the internal mode . of particular note is that we allow for independent control of the coupling constants ; most other studies of coupled lasers and oscillators consider symmetric coupling , where the coupling is the same across all of the elements . the independent control is important because , as we will show , it allows for a singularity in the bifurcation equations that marks the resonance " of large amplitude solutions . the hopf bifurcation to the internal mode occurs for small coupling . however , as the coupling is increased , the external modes appear via hopf bifurcations . we have extended our multiple - scale analysis to be able to describe the bifurcation of the external modes that occur for @xmath0 coupling . the analytical challenge is that the delay terms remain a part of the leading - order problem . describing the bifurcation of the external modes is important for comparing our results to those in experiments @xcite , because the experimental system contains low - pass filters that attenuate the internal mode originating from the first hopf bifurcation . an alternative to our multiple - scale method for deriving slow - time amplitude equations is to apply center - manifold theory with averaging @xcite . however , the averaging method does not retain the delay in the slow time ; that is , the amplitude equations are ordinary - differential equations , not ddes . the distinction is not important for their or our investigation because we look for steady - state solutions of the amplitude equations . however , time - varying amplitudes may require consideration of the slow - time delay . reddy et al . @xcite considered delay - coupled stuart - landau ddes similar to the amplitude equations we derive here . however , the complex coefficients they use are appropriate for limit - cycle oscillators and not the lasers that we consider . they focus their work on the properties of synchronization and amplitude death . in the next section , we nondimensionalize the model for two - coupled lasers , with the result being the focus of the rest of the paper . we perform the linear - stability analysis in sec . [ s : linstab ] . slow - time evolution equations for small and @xmath0 coupling are derived in sec . [ s : intmodes ] and sec . [ s : extmodes ] , respectively . we close with a discussion of the results . we consider two class - b lasers @xcite modeled by rate equations as @xmath1 , \label{eq : idmodel}\end{aligned}\ ] ] where @xmath2 and @xmath3 are the scaled intensity and population inversion , respectively , of each laser . scaled time @xmath4 is measured with respect to the cavity - decay constant @xmath5 , @xmath6 , where @xmath7 measures real time . @xmath8 is a ratio of the inversion - decay rate , @xmath9 , to the cavity - decay rate , @xmath5 , and measures the relative life - time of photons to excited electrons . @xmath10 is the dimensionless pump rate and corresponds to the energy input to the laser by an external source , e.g. , another laser , an incoherent light source or an electronic current . the mass - action coupling term @xmath11 models stimulated emission " ; a photon passing through the laser cavity stimulates an excited electron to drop to the lower energy level , resulting in one less excited electron and one more photon . ( [ eq : idmodel ] ) may be derived as a reduced model from semi - classical maxwell - bloch equations @xcite ; the latter use maxwell s equations to describe the laser s electromagnetic fields coupled with the bloch equations from quantum mechanics for the amplifying media . if the pump rate is a constant , @xmath12 , then the laser relaxes to the steady state @xmath13 , @xmath14 . to facilitate further analysis , we define new variables for the deviations from the non - zero steady state @xcite as @xmath15 our goal is to investigate the effects of coupling through the pump . in addition , we account for the effect of a delay when the signal from one laser takes a non - negligible time before affecting the other . thus , the pump coupling is taken to be @xmath16 thus , we feed the intensity _ deviations _ @xmath17 from the steady - state output of laser @xmath18 to the pump of laser @xmath19 ; the strength of the coupling is controlled by @xmath20 . the pump - coupling scheme allows for easy electronic control of the coupling signal . the negative coupling results from the configuration of the electronic coupling circuits in @xcite . finally , we assume that the decay constants of the two lasers are related by @xmath21 the new rescaled equations are @xmath22 , \label{e : laserxy}\end{aligned}\ ] ] where @xmath23 for notational convenience we have let @xmath24 and dropped the subscript on @xmath25(@xmath26 . rogister et al . @xcite have considered a very similar model to ours ; however , in their case the cross - coupling term is instantaneous , while the delay appears through self - feedback of each laser s own intensity . in the new variables , the steady state is given by @xmath27 . the linear stability of the steady state is determined by studying the evolution of small perturbations , for which we obtain the characteristic equation @xmath28 [ \lambda ( \lambda + \epsilon a_2 \beta^2 ) + \beta^2 ] - \beta^2 \delta_1 \delta_2 e^{- \lambda ( \tau_1 + \tau_2 ) } = 0.\ ] ] the delay results in the exponential term @xmath29 . the transcendental form of the characteristic equation and , hence , the possibility of an infinite number of roots is typical for ddes . if either @xmath30 , then there are only decaying oscillations , which indicates that a single uncoupled laser is a weakly damped oscillator ; this is the general characteristic of `` class - b '' lasers @xcite . for @xmath31 numerical simulations indicate there is a hopf bifurcation as the @xmath32 are increased . to identify the hopf bifurcation point , we let @xmath33 and separate the characteristic equation into real and imaginary parts . after some algebra we obtain a single equation for the frequency and another equation for the value of @xmath34 at the bifurcation point . the equations are @xmath35 \\ & & + \epsilon \omega ( 1 - \omega^2 ) ( a_1 + a_2 \beta^2 ) , \nonumber \\ \beta^2 \delta_h^2 & = & [ ( 1 - \omega^2 ) ( \beta^2 - \omega^2 ) \label{e : dh}\\ & & - \epsilon^2 a_1 a_2 \beta^2 \omega^2 ] ^2 + [ \epsilon \omega ( 1 - \omega^2 ) ( a_1 + a_2 \beta^2 ) ] ^2 \nonumber , \end{aligned}\ ] ] where @xmath36 is the `` round - trip '' delay time . the transcendental equation for @xmath37 can be solved numerically and its solution is then substituted into eq . ( [ e : dh ] ) to determine the value of the coupling at the bifurcation point @xmath38 . to simplify further discussion we consider @xmath39 this implies @xmath40 , which is a specific relationship between the lasers decay constants , @xmath41 and @xmath42 , and the pump rate , @xmath43 ; if @xmath44 and @xmath45 then the lasers are identical . the results are qualitatively the same for nearly identical lasers with @xmath46 . to simplify notation , we define @xmath47 and @xmath48 and have @xmath49 + \epsilon \omega ( 1 - \omega^2 ) c_1 , \label{eq : omega}\\ \delta_h^2 & = & [ ( 1 - \omega^2 ) ^2 - \epsilon^2 c_2 \omega^2 ] ^2 + [ \epsilon \omega ( 1 - \omega^2 ) c_1 ] ^2 , \label{eq : dh}\end{aligned}\ ] ] for @xmath50 the leading approximation to the frequency is given by @xmath51 or @xmath52 , @xmath53 an integer . we will refer to the former as the _ internal mode _ because this is the scaled laser - relaxation frequency ; more specifically , an oscillatory solution of eq . ( [ e : laserxy ] ) with period @xmath54 is called the internal mode . similarly , oscillatory solutions with period @xmath52 , @xmath53 an integer , are called _ external modes _ because their periods are determined by the delay . for any fixed value of the delay @xmath55 there is the internal mode and an infinite number of external modes . from eq . ( [ eq : dh ] ) we see that for almost all values of the delay , as the coupling is increased from zero , the bifurcation to the internal mode occurs at a lower value of the @xmath56 than the external mode ( this is true except when @xmath57 , @xmath58 even , which we will discuss in later paragraphs ) . more specifically , for the internal mode with @xmath59 the hopf bifurcation occurs when @xmath60 . for the external mode with @xmath61 the value of the coupling at the hopf bifurcation is @xmath62 . this is illustrated in fig . [ f : deltavsomega ] , where we have plotted eq . ( [ eq : dh ] ) . we will determine more precise approximations for these bifurcation points below ; however , we will first make additional qualitative observations . ( solid curve ) for hopf bifurcations described by eq . ( [ eq : dh ] ) . the solid circle at @xmath59 indicates the internal mode . the crosses ( + ) at @xmath63 indicate the external modes . because @xmath57 , then @xmath64 for the external modes . ( @xmath65),width=384 ] let us fix the delay to be a multiple of @xmath66 , i.e. , @xmath57 , @xmath58 an integer . then @xmath51 , corresponding to the internal mode , is an exact solution to the frequency equation eq . ( [ eq : omega ] ) and @xmath67 . the @xmath68 is taken if @xmath58 is odd and the @xmath69 is taken if @xmath58 is even ( the sign association comes from the real and imaginary parts of the characteristic equation before @xmath37 and @xmath70 were separated ) . in the present paper , we will consider @xmath71 and @xmath72 ; results for @xmath73 are qualitatively the same . thus , we have the exact solution @xmath51 , @xmath74 when @xmath75 , @xmath58 odd . an analysis of eq . ( [ eq : dh ] ) shows this occurs at a minimum in the curve @xmath76 as illustrated in fig . [ f : deltavsomega ] . if @xmath55 is _ not _ a multiple of @xmath66 , say @xmath77 , the hopf bifurcation to the internal mode still occurs before the bifurcation to any of the external modes . the frequency of the internal mode is @xmath59 ( instead of @xmath78 ) , and @xmath79 but still @xmath80 . thus , if we plot @xmath56 as a function of the delay @xmath55 , the _ minimum _ of the curves @xmath81 occur when @xmath57 , @xmath58 odd . this is illustrated in fig . [ f : linstabfig2 ] and will be discussed further below . now consider the external modes with @xmath63 . for negative couplings when @xmath71 ( see eq . ( [ eq : negcoupling ] ) ) and @xmath72 an analysis of the characteristic equation shows that @xmath53 _ must be even_. thus , the external modes have frequencies @xmath82 etc . in fig . [ f : deltavsomega ] we illustrate three cases when @xmath75 , @xmath83 and @xmath84 . the external modes then have @xmath85 , @xmath53 even . finally , we consider the case when @xmath57 , for @xmath58 _ even_. as discussed in the previous paragraph , for negative couplings the external modes with frequency @xmath86 require that @xmath87 . thus , for each external mode there is a value of @xmath57 , @xmath58 even , such that @xmath59 similar to the internal mode . this is illustrated in the numerical data of fig . [ f : linstabfig2]a for @xmath88 , while in fib . [ f : linstabfig2]b we see that the bifurcations to these two modes occur at nearly the same value of the coupling . referencing fig . [ f : deltavsomega ] , the two modes would have frequencies on opposite sides of @xmath78 such that @xmath56 is the same for each . this is referred to as a double hopf " @xcite and may lead to more complicated bifurcation scenarios that we will not pursue in the present paper . we also note that at the double hopf bifurcation , the designation of internal mode " versus external mode " is ambiguous because both have @xmath59 . indeed , following the numerical data in fig . [ f : linstabfig2]a , as @xmath55 is varied through @xmath89 , we see that @xmath90 internal mode becomes the @xmath91 external mode , while the @xmath92 external mode becomes the @xmath93 internal mode . as discussed above , for any delay @xmath55 when the coupling is increased , the first periodic solutions to appear correspond to the internal mode and will have frequency @xmath59 . for arbitrary delay we can find an approximation for the frequency by letting @xmath94 in eq . ( [ eq : omega ] ) to find that @xmath95 satisfies @xmath96 - 2 c_1 \omega_1 = 0 , \label{eq : omegaint}\ ] ] which is a simple quadratic for @xmath95 . the bifurcation point is then approximated as @xmath97 , \nonumber\\ & = & \epsilon^4 ( 4 \omega_1 ^ 2 + a_1 ^ 2 ) ( 4 \omega_1 ^ 2 + a_2 ^ 2 ) . \label{eq : dhint}\end{aligned}\ ] ] we can obtain simpler results than eqs . ( [ eq : omegaint ] ) and ( [ eq : dhint ] ) if we require that @xmath98 , @xmath58 odd ; that is , in fig . [ f : linstabfig1 ] we examine locally to the critical values @xmath99 and the minimum of the curves @xmath81 . if @xmath100 , then @xmath101 and coupling @xmath34 at the hopf bifurcation point as a function of the delay @xmath36 . the solid curve is the asymptotic results of eq . ( [ eq : omegaint ] ) and ( [ eq : dhint ] ) , while the + are the result of numerically evaluating eq . ( [ eq : omega ] ) and ( [ eq : dh ] ) . ( @xmath102,width=384 ] but with a wider range in @xmath37 and @xmath70 . the @xmath68 indicate the numerical results . the thick curves are the analytical approximations of the internal modes ( same as fig . [ f : linstabfig1 ] ) . the thin curves are the approximations of the external modes.,width=384 ] for the external modes we let @xmath103 with @xmath104 , @xmath53 even , and find that @xmath105 and the hopf bifurcation point is approximated as @xmath106 the external mode results are valid for @xmath55 such that @xmath107 ; thus , @xmath55 must be away from the double hopf point when @xmath108 . in figs . [ f : linstabfig1 ] and [ f : linstabfig2 ] we plot the linear stability given by : ( i ) numerically evaluating eqs . ( [ eq : omega ] ) and ( [ eq : dh ] ) ( marked with + ) and ( ii ) the asymptotic approximations from eqs . ( [ eq : omegaint])-([eq : dhint ] ) and eqs . ( [ eq : omegaext])-([eq : dhext ] ) ( solid curves ) . fig . [ f : linstabfig1 ] focuses on the bifurcation of the internal mode . for any fixed value of the delay time @xmath55 , we increase @xmath70 until the curve is crossed at @xmath38 ; the frequency is @xmath59 . in fig . [ f : linstabfig2 ] we pan out so that both the internal and external modes are visible . from here we see that as @xmath70 is increased , the bifurcation to the internal mode always occurs first . as @xmath70 is increased further , we cross secondary curves indicating the bifurcations to the external modes . more specifically , suppose that @xmath109 ( @xmath90 ) . after the internal mode , the external modes with frequencies @xmath110 , @xmath111 appear in sequence . for longer delay times , the order in which the external modes appear will be different ; this can be seen in fig . [ f : linstabfig2 ] for @xmath112 , where the bifurcation to the external mode with @xmath113 occurs before the mode with @xmath114 . alternatively , in fig . [ f : deltavsomega]b when @xmath115 , as @xmath70 is increased the bifurcation to the periodic solutions occur in the order @xmath116 , which corresponds to the internal mode , followed by the external modes @xmath117 , @xmath118 , @xmath119 . we use the method of multiple scales to analyze the oscillatory solutions that appear at the hopf bifurcation points . for the uncoupled lasers , oscillations decay on an @xmath120 time scale , which suggests that we introduce the slow time @xmath121 ; time derivatives become @xmath122 . we analyze the nonlinear problem using perturbation expansions in powers of @xmath123 , e.g. , @xmath124 ; the relevant nonlinear terms and the slow - time derivative then balance at @xmath125 . finally , for simplicity we set @xmath44 indicating identical lasers ; however , relaxing this assumption to nearly identical lasers ( @xmath126 ) does not qualitatively change the results . we now consider the effect of the two - time scale assumption on the delay term . with the additional slow time , the delay term becomes @xmath127 if @xmath128 , then consideration of the delay term is postponed until higher order and appears as part of the solvability condition for the slowly varying amplitude . however , if @xmath129 , then the leading - order problem will contain the delay term and analytical progress is much more difficult . if @xmath130 ( large delay ) , then @xmath131 and the delay term must be retained in the slow argument . however , if @xmath132 , then the slow argument can be expanded as @xcite @xmath133 now the delay of the slow argument is postponed to higher order . we shall see that this leads to simpler slow - evolution equations for @xmath134 ( see eq . ( [ eq : aonetau ] ) ) , and is necessary to make any progress at all for @xmath135 ( see sec . [ s : extmodes ] ) . it should be noted that care must be taken when using a series expansion of a delay term in a differential equation . the taylor series may itself be justified but doing so can change the stability of the differential equation . a simple example is given in @xcite , while @xcite provide more theoretical discussions concerning restrictions on the size of the delay . in our presentation we will check the validity of our approximations by comparing our analytical and numerical results . from the linear - stability analysis , we know that the first bifurcation will be to the internal mode with @xmath59 when @xmath136 ( we assume that @xmath137 , @xmath58 even , and thus do not consider the case of the double - hopf bifurcation ) . we present the case that both coupling constants are of the same order @xmath138 . however , as long as @xmath139 the bifurcation results are qualitatively the same . proceeding with the multiple - scale analysis , we find that at the leading order , @xmath140 , we obtain the solutions @xmath141 which exhibit oscillations with radial frequency 1 on the @xmath4 time scale . to find the slow evolution of @xmath142 , we must continue the analysis to @xmath125 . then , to prevent the appearance of unbounded secular terms , we determine `` solvability conditions '' for the @xmath142 . due to the scalings of @xmath32 and @xmath143 , the delay terms @xmath144 are @xmath125 and contribute to the solvability condition . the final result is @xmath145 @xmath146 , @xmath147 , and where @xmath148 . the effect of the delay in the slow time appears explicitly in the delay terms @xmath149 . the delay in the fast time has resulted in the exponential terms @xmath150 . ( [ eq : alongtau ] ) is valid for arbitrary values of the delay ( we have not simplified the delay term @xmath151 ) . however , according to eq . ( [ eq : dt - expansion ] ) , if @xmath132 , then @xmath152 and delay in the slow time is postponed until higher order . this results in a simpler solvability condition where the delay in the slow time does not appear . @xmath153 periodic solutions to the original laser equations , eqs . ( [ eq : xymodel ] ) , correspond to @xmath154-independent solutions to the solvability conditions . the conditions are the same for both eqs . ( [ eq : alongtau ] ) and ( [ eq : aonetau ] ) because the delay terms in eq . ( [ eq : alongtau ] ) become constants . the full @xmath154-dependent solvability conditions , including delay , are required only to analyze the stability of the periodic solutions . to determine the amplitude and phase of the periodic solutions , we let @xmath155 , define the phase difference @xmath156 and set the time derivatives to zero . we obtain @xmath157 we find that @xmath158 where @xmath159 and @xmath160 and the amplitudes are found from the implicit equations @xmath161 of particular note is that just as in the linear - stability analysis , the amplitudes of the periodic solutions depend on the sum of the delays @xmath162 . this corresponds to the effective round - trip time of information from laser - j returning to laser - j . in app . [ s : s1s2 ] we present explicit solutions of eq . [ e : s1s2 ] that specify @xmath163 as a function of the parameters and the delay @xmath55 . the expressions are rather complicated but are easily evaluated numerically . to obtain simpler expressions , we consider @xmath164 , where @xmath165 and @xmath58 an odd integer . that is , we tune the delay time @xmath55 to be near one of the minimums of the neutral stability curves in fig . [ f : linstabfig1 ] . the bifurcation equation is @xmath166 where @xmath167 and @xmath168 is specified in app . [ s : s1s2 ] . we compare the results of numerically computed bifurcation diagrams @xcite and our analytical results in figs . [ f : bd_int_n5]-[f : bd_int_n53 ] . in figs . [ f : bd_int_n5 ] and [ f : bd_int_n53 ] , @xmath115 and @xmath169 , respectively , and hence @xmath170 in eq . ( [ e : bifequ_int ] ) . the analytical bifurcation equation describes the numerical results quite well even for @xmath171 far away from the bifurcation point . an expansion of eq . ( [ e : bifequ_int ] ) near the bifurcation point shows that @xmath172 with @xmath173 , and @xmath174 . numerical data points indicated by @xmath175 . the solid curve is the bifurcation equation determined by eqs . ( [ e : bifequ_nearmin])-([e : omega_nearmin ] ) for @xmath170 , or eqs . ( [ e : r2bifequ])-([e : r1bifequ ] ) . because @xmath176 , @xmath177 so the bifurcation equations are equivalent ( @xmath178).,width=384 ] ( @xmath118 ) except that @xmath179 . thus , @xmath180 , @xmath181 and the delay is tuned to the right of the minimum of the neutral - stability curve . the solid curves are the approximate bifurcation equations based on @xmath165 given by eqs . ( [ e : bifequ_nearmin])-([e : omega_nearmin ] ) . the dashed curves are the more general bifurcation equations given by eqs . ( [ e : r2bifequ])-([e : r1bifequ ] ) . all other parameters are the same as in fig . [ f : bd_int_n5].,width=384 ] except that @xmath182.,width=384 ] we also note that eq . ( [ e : bifequ_int ] ) is identical to the bifurcation equation we derived in @xcite for two coupled lasers with opposite sign coupling . the coupled laser equations in @xcite are identical to eq . ( [ e : laserxy ] ) except that in @xcite : ( i ) there is no delay such that the coupling is instantaneous and ( ii ) we allowed the sign of one of the coupling constants to be positive . near the bifurcation point , the small - amplitude periodic oscillations are nearly harmonic as indicated by eq . ( [ e : leadingordery ] ) . thus , a positive coupling constant corresponds to an effective phase shift of half the period or , more generally , a phase shift of @xmath183 ( see app . [ s : delayandneg ] ) . thus , we expect that when @xmath55 is tuned to the minimum of the neutral stability curves in fig . [ f : linstabfig1 ] , the bifurcation equations of the two cases should be equivalent . in fig . [ f : bd_int_n5 - 2 ] we show the case when @xmath184 . here we have tuned the delay @xmath55 to be to the right of the minimum with @xmath185 . the dashed curve is the result of numerically evaluating the full bifurcation equations , eqs . ( [ e : s1s2 ] ) . the solid curve is the approximation eq . ( [ e : bifequ_int ] ) and it shows very good agreement with the numerical results . in general , we maintain good agreement as we tune @xmath55 away from the minimum up to @xmath186 , @xmath58 odd . the full bifurcation equation , eq . ( [ e : s1s2 ] ) , where we have not made any assumption on the delay , maintains good agreement out to at least @xmath187 ; beyond that however , it loses fidelity as we approach the double - hopf bifurcations at @xmath188 ( @xmath57 , @xmath58 even ) . finally , we note that the first correction to the bifurcation equation eq . ( [ e : bifequ_int ] ) is linear in @xmath189 . thus , the bifurcation equation is not symmetric about the minimums of the neutral - stability curves ; that is , the amplitude will be larger on one side of the minimum than the other for a given perturbation of @xmath70 above the minimun . the bifurcation equation eq . ( [ e : bifequ_int ] ) is singular when its denominator is zero ( @xmath190 ) , or @xmath191 ( @xmath34 ) . if the parameters are such that the singular point is before the hopf point , @xmath192 , then the singularity can be ignored because the vanishing denominator will not occur when periodic solutions exist for @xmath193 . however , if @xmath194 , the bifurcation equation will exhibit the pole - type singularity . near the singular point , the bifurcation equation predicts that the amplitude of the oscillations will become large corresponding to a resonance . the condition that @xmath194 corresponds to @xmath195 . thus , in laser-2 the increase in the population inversion provided by pump coupling is greater than the effective losses . in @xcite , when the lasers were coupled without delay , we investigated this resonance in detail and the bifurcation equation matched the result of simulations very well . for the present case with delay coupling , however , we find that there are some dramatic differences . in fig . [ f : resonance0p010]a , we compare the analytical and numerical bifurcation branches . for @xmath196 , the bifurcation equation predicts a strong resonance peak . the numerical data for @xmath197 ( @xmath175 ) exhibits a small resonance peak and there is good agreement between the analytical and numerical curves both before and after the singularity . in fig . [ f : resonance0p010]b we increase @xmath198 from @xmath199 to @xmath200 to follow the deformation of the bifurcation branch from an isolated resonance peak to a curve that forms a loop before continuing to higher values of @xmath198 ; this folding of the bifurcation branch is new to the system with delay and was not observed in @xcite . we want to make clear that the bifurcation branch is _ not _ intersecting itself . the apparent intersection results from projecting the full bifurcation curve of @xmath198 versus @xmath201 , which does not intersect , onto the @xmath202 plane . because the multiple - scale method is a local analysis and we assumed small - amplitude solutions , it is not unexpected that the analytical and numerical results match well near the bifurcation point . similarly , in the vicinity of the resonance peak , the amplitude of the solutions becomes @xmath0 and so it is not surprising that the analytical results fail to match the numerical results . that said , in our previous work with coupled lasers without delay @xcite , the analytically determined bifurcation equation matched the numerically computed resonance peak throughout the parameter range , i.e. , even for @xmath0 amplitudes ; that was a bit surprising . we can surmise that in the present case it is the folding of the bifurcation branch that causes the mismatch . it may be that if we continued our analysis to higher order and derived additional nonlinear corrections to the bifurcation equation , we might better describe the folded branch . this would be a non - trivial and tedious calculation ; but without doing so our supposition remains speculation . in @xcite we used a different asymptotic method better suited to analyzing periodic solutions with @xmath0 amplitudes with only partial success . its application to the present problem with delay would be quite complicated and we have not made the attempt . in fig . [ f : resonance0p010]b we increase @xmath198 to 5.0 and see that there is qualitative change in the bifurcation branch ; there is no longer a loop , and the bifurcation branch continues to negative values of @xmath198 . in fig . [ f : resonance0p010]b we see that the change in the nature of the bifurcation curve apppears quite abruptly at a critical value of @xmath198 . typically , one expects smooth folding and unfolding of bifurcation branches as a parameter is changed @xcite . in the present case , we have not been able to isolate an interval of @xmath198 over which such an unfolding might occur . to our knowledge , such a discontinuous unfolding of a branch of solutions in a bifurcation diagram has not been previously described . with @xmath203 . ( a ) the solid curve is the predicted bifurcation branch from eqs . ( [ e : bifequ_nearmin ] ) when @xmath198 is tuned such that the singularity in the bifurcation equation occurs after the hopf bifurcation . the @xmath175 are the numerical bifurcation data for @xmath196 . ( b ) numerical bifurcation branches as they deform between @xmath204 and @xmath205 . ( c ) there is a critical value of @xmath198 , where on either side of the critical value the bifurcation branch has a loop and proceeds to higher values of @xmath198 , or bends back without a loop to negative values of @xmath198 . all other parameters are the same as in fig . [ f : bd_int_n5],width=480 ] . [ f : resonance0p010 ] finally , if @xmath198 is increased further the bifurcation branch folds back and continues for positive @xmath171 ; there also appear to be secondary bifurcations but we have not explored these in any detail . we mention that as the delay @xmath55 increases , or as @xmath206 increases , the effect of the singularity is diminished . conversely , with reduced values of @xmath206 , the bifurcation branch may exhibit more complicated turns and folds . it is intuitively understandable that increasing the damping ( @xmath206 ) will diminish any type of resonance phenomena . the role of the longer delay is unclear . in this section , we describe the external modes that emerge via hopf bifurcations as the coupling is increased . we fix @xmath207 and use @xmath208 as the bifurcation parameter with @xmath209 . we again have the slow time @xmath210 and the perturbation expansion of @xmath211 and @xmath212 in powers of @xmath213 . the multiple - scale analysis is more complicated because the @xmath214 couplings result in a leading - order problem that contains the delay terms . due to the two time scales , the delay terms are of the form @xmath215 . to make analytical progress we need to remove the slow delay from the leading - order problem this requires that we assume @xmath216 such that by using eq . ( [ eq : dt - expansion ] ) the slow delay is postponed to higher order . we then have a restriction on the size of the delay such that @xmath217 . thus , we find that our results fit well with the external modes corresponding to the case @xmath115 , fig . [ f : deltavsomega]b , but are inaccurate in the case @xmath218 , fig . [ f : deltavsomega]c . finally , to simplify the presentation , we consider the case of equal delays @xmath219 , such that @xmath220 ; analysis of the general case with unequal delays can be carried out in the same way . the leading order @xmath221 problem is @xmath222 where @xmath223 @xmath224 we look for oscillatory solutions of the form @xmath225 , where @xmath226 is a slowly varying scaler amplitude ( to be determined from a solvability condition at @xmath227 ) . @xmath228 is a vector that is determined by substituting our ansatz into eq . ( [ eq : o1dde ] ) to obtain @xmath229 for a nonzero solution @xmath228 , we require @xmath230 . this results in the same condition obtained from the leading - order linear - stability problem ; specifically , @xmath231 thus , we have that @xmath232 for later reference it will be useful to note that @xmath233 where @xmath234 if @xmath235 is odd ( e.g. @xmath236 and @xmath237 if @xmath238 ( e.g. @xmath239 ) is even . finally , we find that @xmath240 at @xmath120 the problem is @xmath241 @xmath242 because the homogeneous problem is the same as the @xmath243 problem we can , without loss of generality , set the homogeneous solution to 0 . the inhomogeneous term @xmath244 is proportional to @xmath245 so that the solution is @xmath246 where @xmath247 is specified in app . [ s : extmode_coeff ] . at @xmath227 we find the solvability condition that determines the slow - evolution equation for @xmath226 . the @xmath227 problem is @xmath248 where @xmath249 the vector @xmath250 has terms proportional to @xmath251 and @xmath252 and the former will lead to solutions of the form @xmath253 . the secular terms @xmath254 must be eliminated to prevent unbounded solutions for large @xmath4 , which implies that a solvability condition must be imposed on @xmath250 . the solvability condition is formulated as follows : we look for a solution to eq . ( [ eq : o3dde ] ) of the form @xmath255 and at the same time identify the terms @xmath244 in @xmath250 proportional to @xmath256 . we then obtain a algebraic system of equations for the vector @xmath257 as @xmath258 where @xmath259 for @xmath257 to have a non - zero solution , the fredholm alternative requires that @xmath260 , where @xmath261 is the solution to @xmath262 ( the superscript @xmath263 refers to hermitian ) . we find that @xmath264 , and the resulting condition for the amplitude @xmath226 is @xmath265 where @xmath266 and @xmath267 are given in app . [ s : extmode_coeff ] . to analyze the solvability condition given by + eq . ( [ e : solvability_ext ] ) , we let @xmath268 to obtain @xmath269 the equation for @xmath270 determines the frequency correction as a function of the amplitude . the bifurcation equation is determined by considering steady - state solutions to the equation for @xmath271 , and we find that @xmath272 . \label{e : bifequ_ext}\ ] ] the onset of oscillations occurs when @xmath273 and determines @xmath274 at the hopf bifurcation point ; this result matches exactly that obtained in the linear - stability analysis . , @xmath113 with @xmath173 , and @xmath174 . numerical data points are indicated by @xmath175 . the solid curve is the asymptotic approximation based on eq . ( [ e : bifequ_ext ] ) . ( @xmath178),width=384 ] except @xmath275 . all other parameter values are the same.,width=384 ] in figs . [ f : bd_ext_n5m4 ] and [ f : bd_ext_n5m6 ] , we compare the bifurcation equation eq . ( [ e : bifequ_ext ] ) to the numerically computed result . in each figure we have @xmath276 ( see figs . [ f : bd_int_n5 ] and [ f : bd_int_n5 - 2 ] for analysis of the internal mode ) . in fig . [ f : bd_ext_n5m4 ] the external mode corresponds to @xmath113 and its direction of bifurcation is subcritical . in fig . [ f : bd_ext_n5m6 ] we have @xmath275 and the bifurcation is supercritical . in each case , there is good agreement between the numerical and analytical result local to the bifurcation point , where the multiple - scale analysis validity is strongest . the direction of bifurcation ( super- or subcritical ) in eq . ( [ e : bifequ_ext ] ) is controlled by the sign of the constant @xmath277 , which is given in eq . ( [ e : r2parameter ] ) . analysis of @xmath277 shows that @xmath278 in the interval @xmath279 , where @xmath280 is shown in fig . [ f : r2parameter ] ( the lower bound is a zero of the denominator of @xmath277 , while the upper bound is the sole real zero of the numerator ) . thus , for external modes with frequencies within this interval , the bifurcation will be subcritical ; this is the case for the external mode in fig . [ f : bd_ext_n5m4 ] . for all other modes @xmath281 and the direction of the bifurcation will be supercritical . as a function of @xmath198 giving a sign change for parameter @xmath277 in eq . ( [ e : r2parameter]).,height=192 ] for reference , in fig . [ f : bd_intext_n5 ] we show the bifurcation diagrams of the internal mode ( @xmath282 ) and the external modes ( @xmath85 , @xmath113 and @xmath275 ) for the case @xmath118 ; that is , we have combined the bifurcation diagrams of figs . [ f : bd_int_n5 ] , [ f : bd_ext_n5m4 ] and [ f : bd_ext_n5m6 ] . the subcritical bifurcation for the external mode @xmath113 folds to provide an interval of hysteresis . the bifurcation branches are projections onto planes so that the intersections of the curves are not relevant . the periodic solutions corresponding to the external modes are unstable as they bifurcate from the unstable branch of steady - state solutions . the internal mode is stable until @xmath283 , when a period - doubling sequence to chaos begins . and the left - most branch corresponds to the internal mode shown in fig . [ f : bd_int_n5 ] ; periodic solutions are stable until @xmath283 , when there is a period - doubling bifurcation . the middle and right - most branches are the continuation of the external modes shown in fig . [ f : bd_ext_n5m4 ] ( @xmath113 ) and fig . [ f : bd_ext_n5m6 ] ( @xmath275 ) , respectively ; periodic solutions are unstable along these branches ( @xmath284).,width=384 ] finally , we note that for the external modes @xmath285^{1 / 2 } .\ ] ] the amplitude of the external modes varies as the square root of the distance from the bifurcation point . in contrast , the internal mode varies as 1/4 the power of the deviation . we have analyzed the output of two mutually coupled lasers , where the light intensity deviations from steady state modulated the pump of the other laser . the coupling strength in one direction is held fixed , while we examine the effect of increasing the coupling strength in the other direction . the signal - propagation time through the optical fiber and the electronic circuit causes a delay leading to a model that is a system of delay - differential equations . linear - stability analysis finds that the steady - state solution becomes unstable at a hopf bifurcation ; we call the resulting periodic solution the internal mode because it oscillates at the laser s relaxation - oscillation frequency . as the coupling is increased , subsequent instabilities occur with frequencies determined by the round - trip delay time @xmath286 , @xmath287 ; we call these periodic solutions external modes . using a multiple - scale analysis , we derive bifurcation equations for both the internal and external modes . we find that the amplitude of the internal mode increases with the 1/4-root of the deviation from the bifurcation point and is supercritical ; i.e. , @xmath288 . the amplitude of the external modes increase with the 1/2-root of the deviation and may be super- or subcritical ; i.e. , @xmath289 . both the initial instability and the bifurcation results depend on the product of the coupling constants @xmath34 . for the analysis of the internal mode , we assumed that both coupling constants were of the same relative size , @xmath290 . however , other scalings satisfy the hopf condition , e.g. , @xmath291 and @xmath292 . we have found that this does not change the qualitative properties of the bifurcation . we have focused our analysis on the onset of oscillatory instabilities and just beyond . we have not considered the stability of the internal and external modes as the coupling is increased further . however , numerical simulations indicated period - doubling bifurcations of individual modes as well as multimode behavior . with independent control of the individual coupling strengths , we have observed an atypical resonance phenomena . as in our previous work on two mutually coupled lasers without delay @xcite , there is an interval of the coupling parameter @xmath208 over which the amplitude of the oscillations becomes large ; we have referred to this effect as a resonance . " the bifurcation equation , eq . ( [ e : bifequ_int ] ) , can describe the amplitude of the oscillations for @xmath208 above and below resonance . a singularity in the bifurcation equation indicates the existence of the resonance , but eq . ( [ e : bifequ_int ] ) does not describe the amplitude within the resonance well . the parametric form of the singularity indicates that physically , the coupling term provides an effective negative - damping that cancels with the lasers self - damping and hence provides a resonance effect . what is different from the results in @xcite is that the bifurcation branch folds with a change in the coupling parameter @xmath293 ; in figs . [ f : resonance0p010 ] we see that the resonance peak forms a loop in the @xmath294 plane . we also find that instead of a smooth unfolding of the loop , there appears to be a critical value of the parameter @xmath198 where the folded portion of the bifurcation branch abruptly disappears . to our knowledge , such a discontinuous unfolding of a branch of solutions in a bifurcation diagram has not been previously described . we do not understand the mechanism or manner by which the abrupt change occurs , but this will be the focus of future work . the application of multiple - scale perturbation techniques to ddes is , to our knowledge , a relatively recent development @xcite . for our analysis of the internal modes , the delay terms appeared only as part of the solvability condition ; thus , the multiple - scale analysis was relatively straightforward . in contrast , for the external modes that bifurcate when the coupling constants are @xmath0 , the delay terms are included in the leading - order problem . however , by looking for periodic solutions we are able to continue the analysis and formulate a solvability condition . finally , as mentioned in the text , multiple - scale expansion of the delay term in eq . ( [ eq : dt - expansion ] ) amounts to a taylor series expansion that , while it may be justified in and of itself , can lead to erroneous results for the dde @xcite . thus , it is important to compare our analytical results to those from numerical simulations as an important check of the work . as discussed in the introduction , the method of averaging has also been used to analyze the weakly - nonlinear characteristics of delay problems . averaging and the multiple - scale technique will lead to a similar slow - time evolution equation for the amplitudes . however , the multiple - scale technique accounts for the delay in the slow time where averaging does not . because we focused on the existence of periodic solutions , the slow - time delay is removed and both averaging and multiple - scales give equivalent results . however , stability of the periodic solutions , or the investigation of more complicated phenomena such as quasiperiodicity , would require the slow - delay from the multiple - scale analysis . finally , we finish with a discussion relating the results of our analysis to experimental results observed in @xcite and @xcite . the coupling circuit in the experimental system has two important characteristics . first , the signal is inverted , which results in the negative signs that appear in front of the coupling constants in eq . ( [ e : laserxy ] ) . second , the circuit acts as a low - pass filter that suppresses coupling of the relaxation - oscillations . we discuss the effect of each of these below . the signal inversion of the coupling circuit results in negative coupling constants in eq . ( [ e : laserxy ] ) . however , both the linear - stability and the leading - order bifurcation results depend on the product of the coupling constants @xmath295 . thus , local to the hopf bifurcation it does not matter if both @xmath32 are negative or both are positive . more generally , our results depend only on whether @xmath70 is positive or negative , not on the signs of the individual coupling constants . this is effectively a symmetry result , because for small amplitudes the oscillations are nearly harmonic . this means a sign change is merely a phase shift . however , we have observed in numerical simulations that when the amplitudes become larger such that the intensity is pulsating , the symmetry is lost such that positive coupling results in different system output from negative coupling . as mentioned above , the optoelectronic coupling circuit in @xcite and @xcite acts as a low - pass filter on the coupling signal . we do not account for this in our model of the system ( recently , illing and gauthier @xcite have analyzed a dde where they explicitly account for the bandlimited response of their feedback system ) . however , because our results address the linear and nonlinear dynamics of both the internal and external modes , we can make a comparison between theory and experiment . the low - pass filter characteristic of the experimental coupling circuit attenuates the high - frequency relaxation - oscillations , corresponding to the internal mode . the result is that the experimental system oscillates at one of the low - frequency external modes . this is illustrated in fig . [ f : pervsdel ] , where we indicate the experimentally observed external mode with a ( @xmath296 ) for different values of the delay . the ( @xmath68 ) indicate the theoretical value of @xmath76 for a bifurcation to an external mode . linear stability predicts that as the frequency increases the coupling strength , @xmath56 , required for a hopf bifurcation decreases ( see also fig . [ f : deltavsomega ] for small @xmath37 ) . however , there is a filter - cutoff frequency @xmath297 such that the relaxation oscillations and higher - frequency external modes are attenuated and only the external modes with frequencies @xmath298 are observed . for example , in @xcite we observed oscillations with period equal to the round - trip delay time ; specifically , @xmath299 corresponding to the external mode with @xmath91 . this indicates that the filter - cutoff frequency , @xmath297 , is such that @xmath300 . if @xmath297 was greater than @xmath301 then we would have expected to observe the external mode with @xmath113 because the value of the coupling at the hopf - bifurcation point decreases as @xmath302 increases , i.e. , it will bifurcate at a lower value of the coupling . more recent experiments @xcite confirm that with longer delays , external modes with @xmath303 are exhibited ; this is shown in the lower two plots in fig . [ f : pervsdel ] . that is , as the delay increases , the external modes with larger @xmath53 will be below the cutoff frequency , @xmath304 . and it is always the mode with largest frequency , but still below cutoff , that is exhibited because @xmath76 is least for that frequency . taking all five plots together , the experimental data suggest that the low - pass filter cutoff frequency is @xmath305 because all of the observed modes have frequency less than @xmath297 . finally , we add that it was observed in @xcite that the amplitude of the oscillations followed a square - root power law as a function of the coupling . this is exactly as predicted by the bifurcation equation eq . ( [ e : bifequ_ext ] ) for the external modes . , @xmath53 even , indicated by @xmath68 . the frequencies of the external modes have been plotted on the linear - stability curve @xmath76 as in fig . [ f : deltavsomega ] . external modes with higher frequency bifurcate first because @xmath76 is decreasing . for each value of the delay , the ( @xmath296 ) indicates the specific external mode observed in experiment . all of the observed modes are less than a cutoff frequency due to the low - pass filter characteristic of the coupling circuit.,width=384 ] our comparison between the theory and the experiment does have limitations . we do not know the detailed filter characteristics of the optoelectronic coupling circuit . it is known that the filter profile is most certainly not a step function but is instead frequency dependent ; it depends on the properties of the optical cable , the electronic coupling circuit and the frequency response of the laser to the modulated pump . in addition , for large values of the delay the frequency difference between the external modes becomes very small and many external modes are excited at nearly the same value of the coupling . for these cases the linear stability theory may be insufficient to identify the mode that is observed in the experiments . i.b.s acknowledges the support of the office of naval research . @xmath306 where @xmath307\\ & & -3\,a_1d_1 ^ 3 d_2 ^2 \sin^3\tau_s + 3 a_1 ^ 2 d_1 ^ 2d_2 ( 3 a_2 + a_1 ) \sin \tau_s \cos\tau_s \nonumber\\ & & + 6a_1 ^ 3 a_2 d_1(a_1 + a_2)\sin \tau_s,\nonumber\\ d & = & q \sin\tau_s\left [ \,a_1 ^ 2d_1 -\,d_1 d_2 ^ 2\sin^2 \tau_s + 2\,a_1 a_2 d_2 \cos \tau_s \right]\\ & & + d_1 ^ 2 d_2 ^ 3 \sin^4 \tau_s -4\,a_1 a_2 d_1 d_2 ^ 2 \cos \tau_s \sin^2 \tau_s -4\,a_1 ^ 2 a_2 ^ 2 d_2 \sin^2\tau_s , \nonumber\\ & & + 2\,a_1 ^ 3 a_2 d_1 \cos\tau_s - a_1 ^ 2 d_1 ^ 2 d_2 \sin^2\tau_s + 2\,a_1 ^ 2 a_2 ^ 2 d_2 \nonumber\\ q & = & \sqrt { d_1 ^ 2 d_2 ^ 2\sin^2\tau_s -4\,a_1 a_2 d_1 d_2 \cos \tau_s -4\,a_1 ^ 2 a_2 ^ 2 } , \end{aligned}\ ] ] and @xmath308 when the delay is tuned to be near one of the minimums of the neutral stability curves , i.e. , @xmath309 , @xmath165 , we have @xmath310,\nonumber\\ d_1 & = & \frac{d_1d_2}{a_1a_2}-1,\nonumber\\ d_2 & = & \frac{a_2d_2}{a_1d_1}-1 . \label{e : bifequ_nearmin}\end{aligned}\ ] ] @xmath311+o(\xi^2),\ ] ] @xmath312+o(\xi^2 ) . \label{e : omega_nearmin}\ ] ] consider a negative coupling constant such that @xmath313 if @xmath314 is harmonic with frequency @xmath78 and given by @xmath315 , then @xmath316 where @xmath57 is the total round - trip " or system - delay because the other coupling constant , @xmath293 , is negative . sukow , a. gavrielides , t. erneux , m.j . baracco , z.a . parmenter , and k.l . blackburn . two - field description of chaos synchronization in diode lasers with incoherent optical feedback and injection . , 5722:256258 , 2005 . f. rogister , d. pieroux , m. sciamanna , p. megret , and m. blondel . anticipating synchronization of two chaotic laser diodes by incoherent optical coupling and its application to secure communications . , 207:295306 , 2002 .

we investigate the mixed dirichlet - neumann boundary value problems for the laplace - beltrami equation on a smooth bounded surface with a smooth boundary in non - classical setting in the bessel potential space for , . to the initial bvp we apply a quasi - localization and obtain a model bvp for the laplacian . the model mixed bvp on the half plane is reduced to an equivalent system of mellin convolution equation ( mce ) in sobolev - slobodekii space ( potential method ) . mce is ivestigated in both bessel potential and sobolev - slobodekii spaces . the symbol of the obtained system is written explicitly and is responsible for the fredholm properties and the index of the system . an explicit criterion for the unique solvability of the initial bvp in the non - classical setting is derived as well . * key words : * boundary value problem , mixed boundary conditions , potential method , fredholm criteria , symbol , banach algebra of operators , mellin convolution equation , meromorphic kernel . bessel potential space , besov space . * ams subject classifications : * primary 35j57 , secondary 45e10 , 47b35 * funding : * the research was supported by shota rustaveli national science foundation grants no . 13/14 and 31/39 . -5 mm

1503.04578

let @xmath4 be some closed orientable surface , bordering a compact inner @xmath5 and outer @xmath6 domains . by @xmath7 we denote a subsurface of @xmath8 , which has two faces @xmath9 and @xmath10 and inherits the orientation from @xmath8 : @xmath10 borders the inner domain @xmath5 and @xmath9 borders the outer domain @xmath11 . @xmath7 has the smooth boundary @xmath12 , which is decomposed into two closed parts @xmath13 , consisting each of finite number of smooth arcs , having in common only endpoints . let @xmath14 , @xmath15 be the unit normal vector field on the surface @xmath7 and @xmath16 be the normal derivative . let us consider the laplace - beltrami operator in @xmath0 written in terms of the gnter s tangent derivatives ( see @xcite for more details ) @xmath17 let @xmath18 , @xmath19 , be the unit normal vector field on the boundary @xmath20 , which is tangential to the surface @xmath7 and directed outside of the surface . and , finally , let @xmath21 be the normal derivative on the boundary of the surface , which is the outer tangential derivative on the surface . we study the following mixed boundary value problem for the laplace - beltrami equation @xmath22 u^+(\tau)=g(\tau ) , \qquad & \tau\in\gamma_d , \\[0.2 cm ] ( \partial_{{{\boldsymbol{\nu}}}_\gamma}u)^+(\tau)=h(\tau),\qquad & \tau\in\gamma_n . \end{array}\right.\end{aligned}\ ] ] where @xmath23 and @xmath24 denote respectively the dirichlet and the neumann traces on the boundary . we need the bessel potential @xmath25 , @xmath26 , @xmath27 and sobolev - slobodekii @xmath28 spaces , where @xmath8 is a closed smooth surface ( without boundary ) , which contains @xmath7 as a subsurface , @xmath29 . the bessel potential space @xmath30 is defined as a subset of the space of schwartz distributions @xmath31 endowedp with the norm ( see @xcite ) @xmath32 where @xmath33 is the bessel potential and @xmath34 , @xmath35 are the fourier transformations . for the definition of the sobolev - slobodekii space @xmath36 see @xcite . the space @xmath37 coincides with the trace space of @xmath38 on @xmath8 and is known that @xmath39 for @xmath40 , @xmath3 ( see @xcite ) . we use , as common , the notation @xmath41 and @xmath42 for the spaces @xmath43 and @xmath44 ( the case @xmath45 ) . the spaces @xmath46 and @xmath37 are defined by a partition of the unity @xmath47 subordinated to some covering @xmath48 of @xmath8 and local coordinate diffeomorphisms ( see @xcite for details ) @xmath49 the space @xmath50 is defined as the subspace of @xmath46 of those functions @xmath51 , which are supported in the closed sub - surface @xmath52 , whereas @xmath53 denotes the quotient space @xmath54 , and @xmath55 is the complemented sub - surface . for @xmath56 the space @xmath53 can be identified with the space of those distributions @xmath57 on @xmath58 which admit extensions @xmath59 , while @xmath53 is identified with the space @xmath60 , where @xmath61 denotes the restriction from @xmath8 to the sub - surface @xmath7 . it is worth noting that for an integer @xmath62 the sobolev spaces @xmath63 and @xmath64 coincide and the equivalent norm is defined with the help of the gnter s derivatives ( see @xcite ) : @xmath65^{\frac1p } , \quad\mbox { where } \quad { { \mathcal d}}^\alpha:={{\mathcal d}}^{\alpha_1}_1{{\mathcal d}}^{\alpha_2}_2{{\mathcal d}}^{\alpha_3}_3\ ] ] and the gnter s derivatives @xmath66 are defined in . let us also consider @xmath67 , a subspace of @xmath68 , orthogonal to @xmath69 @xmath70 consists of those distributions on @xmath8 , belonging to @xmath68 which have their supports just on @xmath20 and @xmath68 can be decomposed into the direct sum of subspaces : @xmath71 the space @xmath72 is non - empty ( see @xcite ) and excluding it from @xmath68 is needed to make bvps uniquelly solvable ( cf . @xcite and the next theorem [ t0.1 ] ) . the lax - milgram lemma applied to the bvp gives the following result . [ t0.1 ] the bvp has a unique solution in the classical weak setting : @xmath73 from theorem [ t0.1 ] we can not even conclude that a solution is continuous . if we can prove that there is a solution @xmath74 for some @xmath75 , we can enjoy even a hlder continuity of @xmath76 . it is very important to know maximal smoothness of a solution as , for example , while designing approximation methods . to this end we will investigate the solvability properties of the bvp in the following non - classical setting @xmath77 and find necessary and sufficient conditions of solvability . note , that the constraint @xmath2 is necessary to ensure the existence of the trace @xmath23 on the boundary . to formulate the main theorem of the present work we need the following definition . [ t0.2 ] the bvp , is fredholm if the homogeneous problem @xmath78 has a finite number of linearly independent solutions and only a finite number of orthogonality conditions on the data @xmath79 ensure the solvability of the bvp . we prove below the following theorem ( see the concluding part of 5 ) . [ t0.3 ] let @xmath3 , @xmath2 . the bvp is fredholm in the non - classical setting if and only if : @xmath80 in particular , the bvp has a unique solution @xmath76 in the non - classical setting if @xmath81 note , that conditions and are independent of the parameter @xmath82 . the proof of the foregoing theorem [ t0.3 ] in [ sect5 ] is based on the theorem [ t0.3a ] and theorem [ t0.4 ] . [ t0.3a ] let @xmath3 , @xmath2 . let @xmath83 and @xmath84 be some fixed extensions of the boundary data @xmath85 and @xmath86 ( non - classical formulation ) , initially defined on the parts of the boundary @xmath13 . a solution to the bvp is represented by the formula @xmath87 here @xmath88 , @xmath89 and @xmath90 are the newton s , double and single layer potentials , defined below @xmath91see @xmath92 and @xmath93 , @xmath94 in are solutions to the following system of pseudodifferential equations @xmath95 { \displaystyle}\frac12\psi_0+r_d{{\boldsymbol w}}^*_{\gamma,0 } \psi_0-r_d{{\boldsymbol v}}_{\gamma,+1}\varphi_0=h_0\qquad & \text{on}\quad\gamma_d , \end{array}\right . \end{array}\\[2 mm ] \label{e0.7 } \begin{array}{c } \varphi_0\in{\widetilde}{{\mathbb{w}}}^{s-1/p}_p(\gamma_n),\quad \psi_0\in{\widetilde}{{\mathbb{w}}}^{s-1 - 1/p}_p(\gamma_d),\\[3 mm ] g_0\in{\mathbb{w}}^{s-1/p}_p(\gamma_n),\qquad h_0\in{\mathbb{w}}^{s-1 - 1/p}_p(\gamma_d ) , \end{array}\end{aligned}\ ] ] where @xmath96 and @xmath97 are given functions and the participating pseudodifferential operators are defined in 1 below . vice versa : if @xmath76 is a solution to the bvp , @xmath98 , @xmath99 and @xmath100 , @xmath101 are some fixed extensions of @xmath102 anf @xmath103 to @xmath20 , then @xmath104 , @xmath105 are solutions to the system . the system of boundary pseudodifferential equations has a unique pair of solutions @xmath106 and @xmath107 in the classical setting @xmath45 , @xmath108 . the proof of theorem [ t0.3a ] is exposed in [ sect1 ] . for the system we can remove the constraint @xmath2 and prove the following result for arbitrary @xmath109 . [ t0.4 ] let @xmath3 , @xmath110 . the system of boundary pseudodifferential equations is fredholm in the sobolev - slobodekii space setting @xmath111 g_0\in{\mathbb{w}}^r_p(\gamma_n),\qquad h_0\in{\mathbb{w}}^{r-1}_p(\gamma_d ) \end{array}\end{aligned}\ ] ] and also in the bessel potential space setting @xmath112 g_0\in{\mathbb{h}}^r_p(\gamma_n),\qquad h_0\in{\mathbb{h}}^{r-1}_p(\gamma_d ) \end{array}\end{aligned}\ ] ] if the following condition holds : @xmath113 in particular , the system has a unique solution in both settings and if : @xmath114 the proof of the foregoing theorem [ t0.4 ] in [ sect5 ] is based on the auxiliary theorem [ t0.5 ] . to formulate the theorem consider the following model system of boundary integral equations ( bies ) @xmath115 \psi(t ) + { { \boldsymbol k}}^1_{-1}{\varphi}(t)=h(t),\qquad & t\in{\mathbb{r}}^+,\end{array}\right.\\[2 mm ] & & \varphi,\psi\in{\widetilde}{{\mathbb{w}}}^{s-1 - 1/p}_p({\mathbb{r}}^+),\qquad g,\ ; h\in{\mathbb{w}}^{s-1 - 1/p}_p({\mathbb{r}}^+ ) , \nonumber\end{aligned}\ ] ] where @xmath116 is a mellin convolution operator with the kernel homogeneous of order @xmath117 ( see @xcite ) . [ t0.5 ] let @xmath3 , @xmath110 . + the system of boundary pseudodifferential equations is fredholm in the sobolev - slobodekii and bessel potential space settings if the system of boundary integral equations is locally invertible at @xmath118 in the sobolev - slobodekii @xmath119 and the bessel potential space @xmath120 settings , respectively . [ r0.7 ] theorem [ t0.5 ] is proved at the end of 1 . for the proof we apply a quasi - localization of the bvp with some model bvps on the half space ( see lemma [ l1.4 ] and lemma [ l1.5 ] ) . the constraint @xmath110 is due to this approach , since the boundary value problems are involved . in a forthcoming paper will be proved directly the local quasi - equivalence of the equation and the system at the points where the dirichlet and neumann boundary conditions collide and some simpler equations , which are uniquely solvable , at all other points . then the constraint @xmath110 can be dropped and replaced by @xmath109 . correspondingly , theorem [ t0.4 ] is also valid for all @xmath109 and the condition acquires the form @xmath121 a quasi - localization means `` freezing coefficients '' and `` rectifying '' underling contours and surfaces . for details of a quasi - localization we refer the reader to the papers @xcite and @xcite , where the quasi - localization is well described for singular integral operators and for bvps , respectively . we also refer to @xcite , where is exposed a short introduction to quasi - localization . in the present case under consideration we get 3 different model problems by localizing the mixed bvp to : * inner points of @xmath7 . * inner points on the boundary @xmath122 and @xmath123 . * points of the boundary @xmath20 where different boundary conditions collide ( endpoints of @xmath123 and @xmath122 ) . the model bvps obtained by a quasi - localization , are well investigated in the first two cases and such model problems have unique solutions without additional constraints . in the third case we get a mixed bvp on the half plane for the laplace equation ( cf . the system is related to this model mixed problem just as bvp is related to the system ( cf . lemma [ l1.5 ] below ) . the investigation of the boundary integral equation system is based on recent results on mellin convolution equations with meromorphic kernels in bessel potential spaces ( see r. duduchava @xcite , r. duduchava and v. didenko @xcite ) . the symbol @xmath124 of the system is a continuous function on some infinite rectangle @xmath125 and is responsible for the fredholm property and the index of the system . this provides necessary and sufficient conditions for the fredholm property of which is then used to prove the solvability of the original bvp in the non - classical setting . a rigorous analysis of solvability of the above and similar problems with dirichlet , neumann , mixed and impedance boundary condition for the helmholtz and other other elliptic equations are very helpful for a general understanding of elliptic boundary value problems in conical domains ( see @xcite ) . in @xcite the authors suggest another approach to the investigation of the model mixed problem for the helmholtz equation : they write explicit formulae for a solution with two different methods . but the setting is classical only ( the case @xmath45 ) and the approach can not be applied to the non - classical setting . other known results are either very limited to special situations such as the rectangular case @xcite or apply rather sophisticated analytical methods @xcite , or are missing a precise setting of appropriate functional spaces ( see , e.g. , @xcite ) . for the historical survey and for further references we recommend @xcite . there is another approach , which can also be applied is the limiting absorption principle , which is based on variational formulation and lax - milgram lemma and its generalizations . such approach is presented , e.g. , in @xcite . but again , these results are for the classical setting . in 1960 s there was suggested to solve canonical diffraction problems in sobolev spaces , based on the recent development in pseudodifferential equations in domains with corners and , more generally , with a lipschitz boundary . it was popularized by e. meister @xcite , e. meister and f .- o . speck @xcite , w.l . wendland @xcite , a. ferreira dos santos @xcite and their collaborators in the 1980 s . also see the book of vasilev @xcite with a considerable list of references . the results are also restricted to the classical setting . let @xmath8 be a closed , sufficiently smooth orientable surface in @xmath126 . we use the notation @xmath127 for either the bessel potential @xmath128 or the sobolev - slobodekii @xmath129 spaces for @xmath130 closed or open and a similar notation @xmath131 for @xmath130 open . consider the space @xmath132 where @xmath133 denotes the duality pairing between the adjoint spaces . it is obvious , that @xmath134 does not contain constants : if @xmath135 than @xmath136 and @xmath137 . moreover , @xmath138 decomposes into the direct sum @xmath139 and the dual ( adjoint ) space is @xmath140 the following is a part of theorem 10 proved in @xcite . [ t1.1a ] let @xmath130 be @xmath141-smooth @xmath142 , @xmath3 and @xmath143 . let @xmath134 be the same as in - . the laplace - beltrami operator @xmath144 is invertible between the spaces with detached constants @xmath145 i.e. , has the fundamental solution @xmath146 in the setting . let @xmath147 be a subsurface with a smooth boundary @xmath148 . with the fundamental solution @xmath146 of the laplace - beltrami operator at hand we can consider the standard newton , single and double layer potentials on the surface @xmath7 : @xmath149 { { \boldsymbol v}}_\gamma v(x):={\displaystyle}\int_{\gamma}{{\mathcal k}}_{{\mathcal s}}(x,\tau)v(\tau)d\tau,\\[3 mm ] { { \boldsymbol w}}_\gamma v(x):={\displaystyle}\int_{\gamma}{\partial}_{{{\boldsymbol{\nu}}}_\gamma(\tau)}{{\mathcal k}}_{{\mathcal s}}(x,\tau)v(\tau)d\tau , \qquad x\in{{\mathcal c}}. \end{array}\end{aligned}\ ] ] the potential operators , defined above , have standard boundedness properties @xmath150 and any solution to the mixed bvp in the space @xmath151 is represented as follows : @xmath152^+(x ) \qquad u\in{\mathbb{h}}^1_\#({{\mathcal c } } ) , \quad x\in{{\mathcal c}}\end{aligned}\ ] ] ( see @xcite ) . densities in represent the dirichlet @xmath23 and the neumann @xmath153^+$ ] traces of the solution @xmath76 on the boundary . since @xmath154 , we can extend layer potentials to the entire space as follows : @xmath155 \text{we set}\quad { { \boldsymbol v}}_\gamma{\varphi}={{\boldsymbol v}}_\gamma{\varphi}_0+c , \quad { { \boldsymbol w}}_\gamma{\varphi}={{\boldsymbol w}}_\gamma{\varphi}_0+c , \quad { { \boldsymbol n}}_{{\mathcal c}}{\varphi}={{\boldsymbol n}}_{{\mathcal c}}{\varphi}_0+c , \end{array } \end{aligned}\ ] ] i.e. , by setting @xmath156 . [ l1.1 ] the representation formula remains valid for a solution in the space @xmath157 , provided the potentials are extended as in . * proof : * indeed , since @xmath158 , @xmath159 , @xmath160 , we apply the representation formula for a solution in the space @xmath151 , formula , and get the representation formula for a solution in the space @xmath157 : @xmath161^+(x)+c\\[3 mm ] & \hskip-3mm=&\hskip-3 mm { { \boldsymbol n}}_{{\mathcal c}}f(x)+{{\boldsymbol w}}_\gamma ( u - c)^+(x)-{{\boldsymbol v}}_\gamma[{\partial}_{{{\boldsymbol{\nu}}}_\gamma } ( u - c)]^+(x)+c\\[3 mm ] & \hskip-3mm=&\hskip-3 mm { { \boldsymbol n}}_{{\mathcal c}}f(x)+{{\boldsymbol w}}_\gamma u^+(x)-{{\boldsymbol v}}_\gamma[{\partial}_{{{\boldsymbol{\nu}}}_\gamma } u]^+(x ) , \qquad u\in{\mathbb{h}}^1({{\mathcal c } } ) , \quad x\in{{\mathcal c}}. \end{array}\end{aligned}\ ] ] -9 mm @xmath162 * proof of theorem [ t0.3a ] : * let us recall the plemelji formulae @xmath163 ( { \partial}_{{{\boldsymbol{\nu}}}_\gamma}{{\boldsymbol v}}_\gamma v)^\pm(t)=\mp{\displaystyle}\frac12v(t)+{{\boldsymbol w}}^*_{\gamma,0}v(t),\quad ( { { \boldsymbol v}}_\gamma v)^\pm(t)={{\boldsymbol v}}_{\gamma,-1}v(t ) , \end{array } \end{aligned}\ ] ] where @xmath164 and @xmath165 { { \boldsymbol w}}_{\gamma,0 } v(t)&\hskip-3mm:=&\hskip-3mm{\displaystyle}\int_\gamma({\partial}_{{{\boldsymbol{\nu}}}_\gamma(\tau)}{{\mathcal k}}_{{\mathcal s}})(t,\tau ) v(\tau)d\tau,\\ { { \boldsymbol w}}^*_{\gamma,0}w(t)&\hskip-3mm:=&\hskip-3mm{\displaystyle}\int_\gamma({\partial}_{{{\boldsymbol{\nu}}}_\gamma(t)}{{\mathcal k}}_{{\mathcal s}})(t,\tau ) w(\tau)d\tau,\\ { { \boldsymbol v}}_{\gamma,+1}w(t)&\hskip-3mm:=&\hskip-3mm{\displaystyle}\int_\gamma({\partial}_{{{\boldsymbol{\nu}}}_\gamma(t)}{\partial}_{{{\boldsymbol{\nu}}}_\gamma(\tau ) } { { \mathcal k}}_{{\mathcal s}})(t,\tau)w(\tau)d\tau,\qquad\qquad t\in\gamma , \end{array } \end{aligned}\ ] ] are pseudodifferential operators on @xmath20 , have orders @xmath117 , @xmath118 , @xmath118 and @xmath166 , respectively , and represent the direct values of the corresponding potentials @xmath90 , @xmath89 , @xmath167 and @xmath168 . let @xmath83 and @xmath84 be some fixed extensions of the boundary conditions @xmath85 and @xmath86 ( non - classical formulation ) , initially defined on the parts of the boundary @xmath13 . since the difference between such two extensions belong to the spaces @xmath169 and @xmath170 respectively , let us look for two unknown functions @xmath171 and @xmath172 , such that for @xmath173 and @xmath174 the boundary conditions in hold on the entire boundary @xmath175 g_0(t)+\varphi_0(t)\quad & { \rm if}\quad t\in\gamma_n , \end{array}\right.\\ \\ ( { \partial}_{{{\boldsymbol{\nu}}}_\gamma } u)^+(t)=h_0(t)+\psi_0(t)=\left\{\begin{array}{ll } h_0(t)+\psi_0(t ) \quad & { \rm if}\quad t\in\gamma_d,\\[3 mm ] h(t)\quad & { \rm if}\quad t\in\gamma_n , \end{array}\right . \end{array}\end{aligned}\ ] ] provided f @xmath176 is a solution to the bvp . by introducing the boundary values of a solution to the bvp into the representation formula ( see lemma [ l1.1 ] ) we get the following representation of a solution : @xmath177(x)-{{\boldsymbol v}}_\gamma[h_0+\psi_0](x),\qquad x\in{{\mathcal c}},\end{aligned}\ ] ] where @xmath178 by applying the plemelji formulae to and taking into account we get the following : @xmath179(t)-{{\boldsymbol v}}_{\gamma,-1}[h_0+\psi_0](t),\\[2 mm ] h_0(t)+\psi_0(t)=({\partial}_{{{\boldsymbol{\nu}}}_\gamma } u)^+(t)={({\partial}_{{{\boldsymbol{\nu}}}_\gamma}{{\boldsymbol n}}_{{\mathcal c}}f)^+}+{{\boldsymbol v}}_{\gamma,+1}[g_0+\varphi_0](t)\\ \hskip20mm+{\displaystyle}\frac12(h_0(t ) + \psi_0(t))-{{\boldsymbol w}}^*_{\gamma,0}[h_0+\psi_0](t),\qquad t\in\gamma . \end{array}\right . \end{array}\ ] ] if we apply the restriction operator @xmath180 to @xmath122 to the first equation in the obtained system and the restriction operator @xmath181 to @xmath123 to the second one , we obtain the system , where @xmath182\in{\mathbb{w}}^{s-1/p}_p(\gamma_n),\\[2 mm ] h_0:=r_d\left[({\partial}_{{{\boldsymbol{\nu}}}_\gamma}{{\boldsymbol n}}_{{\mathcal c}}f)^+-{\displaystyle}\frac12h_0+{{\boldsymbol v}}_{\gamma,+1 } g_0-{{\boldsymbol w}}^*_{\gamma,0}h_0\right]\in{\mathbb{w}}^{s-1 - 1/p}_p(\gamma_d ) . \end{array}\end{aligned}\ ] ] thus , we have proved the inverse assertion of theorem [ t0.3a ] : if @xmath76 is a solution to the bvp , the functions @xmath183 and @xmath184 are solutions to the system . the direct assertion is easy to prove : * the function in represented by the potentials , satisfies the equation . * if @xmath183 and @xmath184 are solutions to the system , using plemelji formulae it can easily be verified that @xmath76 in satisfies the boundary conditions in . the existence and the uniqueness of a solution to the bvp in the classical setting is stated in theorem [ t0.3 ] , while for the system it follows from the equivalence with the bvp . @xmath162 the remainder of the paper is devoted to the proof of solvability properties of the system in the non - classical setting . consider the following equation on the 2-dimensional euclidean space @xmath185 also the model dirichlet @xmath186 u^+(t)=g_0(t ) , \qquad & t\in\partial{\mathbb{r}}^2_+={\mathbb{r } } , \\[0.2 cm ] \end{array}\right.\end{aligned}\ ] ] the model neumann @xmath187 -(\partial_2u)^+(t)=h_0(t ) , \qquad & t\in\partial{\mathbb{r}}^2_+={\mathbb{r } } , \end{array}\right.\end{aligned}\ ] ] and the model mixed @xmath188 u^+(t)=g_1(t ) , \qquad & t\in{\mathbb{r}}^-:=(-\infty,0 ) , \\[0.2 cm ] -(\partial_{2}u)^+(t)=h_1(t),\qquad & t\in { \mathbb{r}}^+:=(0,\infty ) , \end{array}\right.\end{aligned}\ ] ] boundary value problems for the laplace equation on the upper half plane @xmath189 , where @xmath190 is the normal derivative on the boundary of @xmath191 . the bvps and will be treated in the non - classical setting : @xmath192 1<p<\infty , \qquad s>{\displaystyle}\frac1p \end{array}\ ] ] and the bvp will be treated in the non - classical setting : @xmath193 1<p<\infty , \qquad s>{\displaystyle}\frac1p . \end{array}\ ] ] [ p1.7 ] the bvps , have unique solutions in the setting and the laplace equation in the setting has a unique solution as well . * proof : * the assertion is a well - known classical result , available in many textbooks on partial differential equations ( see e.g. @xcite ) . @xmath162 as a paticular case of theorem [ t0.1 ] ( can easily be proved with the lax - milgram lemma ) we have the following . [ p1.4 ] the mixed bvp has a unique solution @xmath76 in the classical weak setting @xmath194 [ l1.4 ] the bvp is fredholm in the non - classical setting if the model mixed bvp is locally fredholm ( ie . , is locally invertible ) at @xmath118 in the non - classical setting . * proof : * we apply quasi - localization of the boundary value problem in the more general non - classical setting , which includes the classical setting as a particular case ( see @xcite ) for details of quasi - localization of boundary value problems and also @xcite for general results on localization and quasi - localization . by quasi - localization at the point @xmath15 we first localize to the tangential plane @xmath195 ( tangential half plane @xmath196 ) to @xmath7 at @xmath197 ( at @xmath198 , respectively ) . the differential operators remain the same @xmath199 but the normal vector @xmath200 to the tangent plane @xmath201 and the normal vector @xmath202 to the boundary of the tangent plane @xmath203 are now constant . next we rotate the tangent planes @xmath195 and @xmath196 to match them with the plane @xmath201 and @xmath191 . the normal vector fields @xmath200 will transform into @xmath204 and @xmath205 . the rotation is an isomorphism of the spaces @xmath206 , @xmath207 , @xmath208 etc . and transforms the operators in into the operators @xmath209 and we get , , , as a local representatives of bvp . for the bvp in the non - classical setting we get the following local quasi - equivalent equations and bvps at different points of the surface @xmath15 : * the equation at @xmath118 if @xmath197 is an inner points of the surface ; * the dirichlet bvp in the non - classical setting at @xmath118 if @xmath210 ; * the neumann bvp in the non - classical setting at @xmath118 if @xmath211 ; * the mixed bvp in the non - classical setting at @xmath118 if @xmath212 is one of two points of collision of different boundary conditions . the main conclusion of the present theorem on fredholm properties of bvps and follows from proposition [ p1.7 ] and the general theorem on quasi - localizaion ( see @xcite ) : _ the bvp , is fredholm if all local representatives , , and in non - classical settings are locally fredholm ( i.e. , are locally invertible)_. @xmath162 now we concentrate on the model mixed bvp .to this end let us recall that the function @xmath213 is the fundamental solution to the laplace s equation in two variables @xmath214 \delta=\partial^2_1+\partial^2_2=\partial^2_{{\boldsymbol{\nu}}}+\partial^2_\ell . \end{array } \end{aligned}\ ] ] from follows the equality @xmath215 which we use to prove the following : @xmath216 applying the latter equality , we represent the hypersingular operator @xmath217 as follows @xmath218 since @xmath219 on @xmath220 and for the tangential differential operator @xmath221 on arbitrary smooth contour @xmath20 the following `` partial integration '' formula is valid ( see @xcite ) : @xmath222 we can define standard layer potential operators , the newton , the single and the double layer potentials respectively ( cf . ) @xmath223 { { \boldsymbol v}}_{{\mathbb{r } } } v(x)&\hskip-3mm:=&\hskip-3mm{\displaystyle}\frac1{2\pi}{\displaystyle}\int_{{\mathbb{r}}}\ln|x-\tau| v(\tau)d\tau,\nonumber\\ { { \boldsymbol w}}_{{\mathbb{r } } } v(x)&\hskip-3mm:=&\hskip-3mm-{\displaystyle}\frac1{2\pi}{\displaystyle}\int_{{\mathbb{r}}}{\partial}_{2 } \ln|(x_1,x_2)-(\tau , y_2)|\big|_{y_2=0 } v(\tau)d\tau\\ & \hskip-3mm=&\hskip-3mm-{\displaystyle}\frac1{2\pi}{\displaystyle}\int_{\mathbb{r}}{\partial}_{2 } \ln\sqrt{(x_1-\tau)^2+(x_2-y_2)^2}\big|_{y_2=0 } v(\tau)d\tau\nonumber\\[3 mm ] & \hskip-3mm=&\hskip-3mm{\displaystyle}\frac1{2\pi}{\displaystyle}\int_{\mathbb{r}}\frac{x_2 v(\tau)d\tau}{(x_1-\tau)^2 + x_2 ^ 2 } , \qquad x=(x_1,x_2)^\top\in{\mathbb{r}}^2_+.\nonumber\end{aligned}\ ] ] the pseudodifferential operators on @xmath224 , @xmath225 , @xmath226 and @xmath217 , associated with the layer potentials ( see ) , acquire the form @xmath227 { { \boldsymbol w}}_{{\mathbb{r}},0 } v(t)&\hskip-3mm:=&\hskip-3mm\lim\limits_{x_2\to0}{\displaystyle}\frac1{2\pi}{\displaystyle}\int_{\mathbb{r}}\frac{x_2 v(\tau)d\tau}{(x_1-\tau)^2+x_2 ^ 2}=0,\qquad { { \boldsymbol w}}^*_{{\mathbb{r}},0}v(t)=0,\\[3 mm ] \end{array } \end{aligned}\ ] ] by using the representation we find the following : @xmath228 & \hskip-3mm=&\hskip-3mm - v(t)+{\displaystyle}\frac1{2\pi}{\displaystyle}\int_{{\mathbb{r}}}{\displaystyle}\frac{t-\tau}{(t-\tau)^2}{\partial}_\tau v(\tau)d\tau\\[3 mm ] & \hskip-3mm=&\hskip-3mm - v(t)+{\displaystyle}\frac1{2\pi}{\displaystyle}\int_{{\mathbb{r}}}{\displaystyle}\frac{{\partial}_\tau v(\tau)d\tau}{t-\tau } , \quad t\in{\mathbb{r}}\end{array } \end{aligned}\ ] ] and the plemelji formulae acquire the form @xmath229 -({\partial}_{y_2}{{\boldsymbol w}}_{{\mathbb{r}}}v)^\pm(t)={{\boldsymbol v}}_{{\mathbb{r}},+1}v(t)\qquad ( { { \boldsymbol v}}_{{\mathbb{r } } } v)^\pm(t)={{\boldsymbol v}}_{{\mathbb{r}},-1 } v(t ) \qquad t\in{\mathbb{r}}. \end{array}\ ] ] now we prove the following . [ l1.5 ] let @xmath3 , @xmath2 . let @xmath230 and @xmath231 be some fixed extensions of the boundary conditions @xmath232 and @xmath233 @xmath91non - classical formulation @xmath92 , initially defined on the parts of the boundary @xmath234 . a solution to the bvp is represented by the formula @xmath235 ( cf . for the potential operators ) and @xmath236 and @xmath237 are solutions to the system of pseudodifferential equations @xmath238 { \displaystyle}\frac12\psi^0 + r_-{{\boldsymbol w}}^*_{{\mathbb{r}},0}\psi^0 - r_-{{\boldsymbol v}}_{{\mathbb{r}},+1}\varphi^0=h_1 \qquad \text{on}\quad { \mathbb{r}}^- , \end{array}\right . \end{array } \\[2 mm ] \label{e1.23 } & & \varphi^0 , { { \boldsymbol r}}_*\psi^0\in{\widetilde}{\mathbb{w}}^{s-1 - 1/p}_p({\mathbb{r}}^+),\quad g_1 , { { \boldsymbol r } } _ * h_1\in{\mathbb{w}}^{s-1 - 1/p}_p({\mathbb{r}}^+),\end{aligned}\ ] ] where @xmath239 and @xmath240 are the restriction operators from the axes @xmath220 to the semi- axis @xmath241 and @xmath242 . the system of boundary pseudodifferential equations has a unique pair of solutions @xmath236 and @xmath237 in the classical setting @xmath45 , @xmath108 . * proof : * by repeating word by word the proof of theorem [ t0.3a ] , we prove the equivalence via the representation formulae of the bvp in the non - classical setting and of the system . the existence and uniqueness of a solution to the bvp in the classical setting is stated in proposition [ p1.4 ] , while for the system it follows from the proved equivalence with the bvp . @xmath162 [ l1.5a ] let @xmath3 , @xmath2 . the system of boundary pseudodifferential equations is locally invertible at @xmath118 if and only if the system is locally invertible at @xmath118 in the non - classical setting and the space parameters are related as follows : @xmath243 . * proof : * due to the equalities @xmath244 , @xmath245 and the equation in acquires the form @xmath246 { \displaystyle}\frac12\psi^0(t ) -{\displaystyle}\frac1{2\pi}{\displaystyle}\int_{{\mathbb{r}}^+}\frac{({\partial}_\tau\varphi^0)(\tau ) d\tau}{t-\tau}=h_1(t ) , \qquad t\in{\mathbb{r}}^- . \end{array}\right.\end{aligned}\ ] ] multiply both equations by 2 , apply to the first equation the differentiation @xmath247 , replace @xmath248 , apply to the second equation the reflection @xmath249 and replace @xmath250 , also under the integral . we get the following @xmath251 \psi(t ) + { \displaystyle}\frac1\pi{\displaystyle}\int_{{\mathbb{r}}^+}\frac{\varphi(\tau)d\tau}{t+\tau}=2h_1(-t)=:h(t ) , \qquad t\in{\mathbb{r}}^+ \end{array}\right.\end{aligned}\ ] ] and the obtained equation coincides with the system . to prove the local equivalence at @xmath118 of the systems and note , that the multiplication by @xmath252 and the reflection @xmath253 are invertible operators since @xmath254 and @xmath255 and , therefore , are locally invertible at @xmath118 . the differentiation @xmath256 is locally invertible at any finite point @xmath257 because the operators @xmath258 are isomorphisms ( represent bessel potentials , see theorem [ t4.8 ] below , ( * ? ? ? * lemma 5.1 ) and @xcite ) . on the other hand , the embeddings @xmath259 are locally compact due to the sobolev s embedding theorem and the compact perturbation does not influences the local invertibility . @xmath162 * proof of theorem [ t0.5 ] : * by theorem [ t0.3a ] the system is fredholm in the sobolev - slobodekii space setting if the bvp is fredholm in the non - classical setting . on the other hand , by lemma [ l1.4 ] the bvp is fredholm in the non - classical setting if the bvp is locally invertible at @xmath118 in the non - classical setting . and , finally , by lemma [ l1.5 ] and lemma [ l1.5a ] the bvp is locally invertible in the non - classical setting if the system of boundary integral equations is locally invertible at @xmath118 in the sobolev - slobodekii space setting . this accomplishes the proof of the first part of the assertion , concerning the solvability in the sobolev - slobodekii space settings and . the second part of the assertion , concerning the solvability in the bessel potential space settings and , follows from the first part and proposition [ p4.3 ] , exposed below and proved in @xcite , which states that these solvability properties are equivalent . @xmath162 to formulate the next theorem we need to introduce fourier convolution and bessel potential operators . for the spaces of scalar , vector and matrix functions we will use the same notation if this will not lead to a confusion . for example , @xmath261 might be the space of locally bounded functions either scalar , but also vector or matrix valued functions ; this will be clear from the context . let @xmath262 be a locally bounded @xmath263 matrix function . the fourier convolution operator ( fco ) with the symbol @xmath264 is defined by @xmath265 here @xmath266 is the fourier transform and @xmath267 is its inverse transform . if the operator @xmath268 is bounded , we say that @xmath264 is an @xmath269-multiplier of order @xmath270 and use `` @xmath269-multiplier '' if the order is @xmath118 . the set of all @xmath269-multipliers of order @xmath270 ( of order @xmath118 ) is denoted by @xmath271 ( by @xmath272 , respectively ) . let @xmath273 note , that @xmath274 and @xmath275 are independent of @xmath276 because , due to interpolation theorem @xmath277 for all @xmath278 . for an @xmath269-multiplier of order @xmath270 , @xmath279 , the fourier convolution operator ( fco ) on the semi - axis @xmath280 is defined by the equality @xmath281 where @xmath282 is the restriction operator to the semi - axes @xmath280 . we did not use the parameter @xmath283 in the definition of the class of multipliers @xmath271 . this is due to the fact that @xmath271 is independent of @xmath284 : if the operator @xmath285 in is bounded for some @xmath283 , it is bounded for all other values of @xmath284 . another definition of the multiplier class @xmath271 is written as follows : @xmath279 if and only if @xmath286 , where @xmath287 . this assertion is one of the consequences of theorem [ t4.8 ] below . consider the bessel potential operators defined as follows @xmath288 \mathbf{\lambda}_{-\gamma}^r = r_+w^0_{\lambda^r_{-\gamma}}\ell\;:\;\mathbb{h}^s_p ( \mathbb{r}^+)\rightarrow\mathbb{h}^{s - r}_p(\mathbb{r}^+)\ , , \\[3 mm ] \lambda^r_{\pm\gamma}(\xi):=(\xi\pm\gamma)^r,\qquad\xi\in\mathbb{r } , \qquad { \rm im}\,\gamma>0 \end{array}\ ] ] for a non - negative @xmath289 . here @xmath290 is some extension operator . in there is no need of any extension operator since the space @xmath291 is automatically embedded in @xmath292 provided functions are extended by @xmath118 . for a negative @xmath293 the bessel potential operators @xmath294 are defined by the duality between the spaces . [ t4.8 ] let @xmath3 . then : 1 . for any @xmath295 , @xmath296 , @xmath297 the bessel potential operators arrange isomorphisms of the corresponding spaces @xmath91see @xcite@xmath92 and are independent of the choice of an extension operator @xmath298 . + + 0.15 cm 2 . for any operator @xmath299 of order @xmath270 , the following diagram is commutative @xmath300 + the diagram provides an equivalent lifting of the operator @xmath301 of order @xmath270 to the operator @xmath302 of order @xmath118 . + 0.15 cm 3 . for any bounded convolution operator @xmath303 of order @xmath270 and for any pair of complex numbers @xmath304 such that @xmath305 , @xmath306 , the lifted operator @xmath307 a_{\mu,\nu}(\xi):=(\xi-\gamma_1)^\mu a(\xi)(\xi+\gamma_2)^\nu \end{array}\ ] ] is again a fourier convolution . + in particular , the lifted operator @xmath308 in @xmath269-spaces , @xmath309 has the symbol @xmath310 [ r4.12 ] for any pair of multipliers @xmath279 , @xmath311 the corresponding convolution operators on the full axes @xmath312 and @xmath313 have the property @xmath314 . for the corresponding wiener - hopf operators on the half axis a similar equality @xmath315 is valid if at least one of the following conditions hold : the function @xmath316 has an analytic extension in the lower half plane or the function @xmath317 has an analytic extension in the upper half plane ( see @xcite ) . note , that actually is a consequence of . let @xmath318 denote the one point compactification of the real axis @xmath220 and @xmath319-the two point compactification of @xmath220 . by @xmath320 ( by @xmath321 , respectively ) we denote the space of continuous functions @xmath322 on @xmath220 which have the same limits at the infinity @xmath323 ( limits at the infinity might differ @xmath324 ) . by @xmath325 is denoted the space of piecewise - continuous functions on @xmath326 , having limits @xmath327 at all points @xmath328 , including infinity . [ p2.5 ] let @xmath3 , @xmath329 , @xmath330 and @xmath331 . then the operators @xmath332 are compact . moreover , these operators are compact in all bessel potential and besov spaces , where they are bounded , due to the krasnoselskij interpolation theorem for compact operators . [ p2.7 ] let @xmath3 and let @xmath264 and @xmath333 satisfy at least one of the following conditions : * @xmath334 , @xmath335 , + 0.15 cm * @xmath336 , @xmath337 . then the commutants @xmath338 $ ] are compact operators in the space @xmath339 and also , due to krasnoselskij interpolation theorem for compact operators , in all bessel potential and besov spaces , where they are bounded . in this section we expose auxiliary results from @xcite ( also see @xcite ) , which are essential for the investigation of boundary integral equations from the foregoing section . let @xmath316 be a @xmath340 matrix function @xmath341 , continuous on the real axis @xmath342 with the only possible jump at infinity . consider a mellin convolution operator @xmath343 with the symbol @xmath264 in the bessel potential spaces @xmath344 where @xmath345 \mathcal{m}^{-1}_\beta u(t):=\displaystyle\frac1{2\pi } \displaystyle\int_{-\infty}^{\infty}t^{i\xi-\beta } u(\xi)d\xi,\qquad t\in\mathbb{r}^+ , \end{array } \end{aligned}\ ] ] are the mellin transformation and the inverse to it . the most important example of a mellin convolution operator is an integral operator of the form @xmath346 with @xmath347 matrix coefficients and @xmath347 matrix kernel @xmath348 then @xmath343 is a bounded operator in the weighted lebesgue space of vector functions @xmath349 { \beta}:=\frac{1+\gamma}p , \quad 1<p<\infty,\quad -1<\gamma < p-1,\nonumber \end{aligned}\ ] ] endowed with the norm @xmath350^{1/p}\ ] ] ( cf . the symbol of the operator is the mellin transform of the kernel @xmath351 obviously,@xmath352 for @xmath353 . [ t2.1 ] let @xmath3 and @xmath354 ( or @xmath355 provided @xmath356 in ) . the following three properties are equivalent : * operator @xmath357 in is fredholm ; * the symbol of the operator is invertible ( is elliptic ) @xmath358 * the operator is invertible and the inverse operator is @xmath359 . [ p2.7a ] let @xmath3 and let @xmath264 and @xmath333 satisfy at least one of the following conditions : * @xmath334 , @xmath335 , + 0.15 cm * @xmath336 , @xmath337 . then the commutants @xmath360 $ ] are compact operators in the space @xmath339 and also , due to krasnoselskij interpolation theorem for compact operators , in all bessel potential and besov spaces , where they are bounded . things are different in the bessel potential spaces if compared with the lebesgue spaces . let us recall some results from @xcite . consider meromorphic functions in the complex plane @xmath361 , vanishing at infinity @xmath362 with poles at @xmath363 , complex coefficients @xmath364 and @xmath365 . [ d2.2 ] we call a kernel @xmath366 in admissible if for those poles @xmath367 which belong to the positive semi - axes @xmath368 , the corresponding multiplicities are one , i.e. , @xmath369 . for example : the mellin convolution operator @xmath370 has an admissible kernel for arbitrary @xmath62 if @xmath371 as soon as @xmath372 is real @xmath373 . [ p2.4 ] let @xmath3 and @xmath354 ( or @xmath374 provided @xmath356 in ) and @xmath366 in be an admissible kernel . then the mellin convolution @xmath375 is a bounded operator in the lebesgue space @xmath376 and , also , in the bessel potential spaces @xmath377 for all @xmath378 . the next theorem provides the lifting of the mellin convolution operator from a pair of bessel potential spaces to the lebesgue spaces . [ t4.7 ] let @xmath379 , @xmath380 and @xmath295 , @xmath3 . then the operator @xmath381 is lifted equivalently to the operator @xmath382 where @xmath383 if only @xmath384 . if @xmath385 , choose any @xmath386 such that @xmath387 and @xmath388 ( such a choice of @xmath389 is possible since @xmath372 is not a real constant @xmath390 ) . then @xmath391 where @xmath392 is a compact operator . the results of the foregoing two sections together with results on a banach algebra generated by mellin and fourier convolution operators ( see @xcite ) allow the investigation of lifted mellin convolution operators . for this we need to write the symbol of a model operator @xmath393 where @xmath394 are admissible mellin convolution operators . to expose the symbol of the operator , consider the infinite clockwise oriented `` rectangle '' @xmath395 , where ( cf . figure 1 ) @xmath396 ( 300,140 ) ( -00,40 ) ( 80,50)(0,0)[lc]@xmath397 ( 80,130)(0,0)[lc]@xmath398 ( 80,33)(0,0)[lc]@xmath399 ( 80,145)(0,0)[lc]@xmath400 ( -13,90)(0,0)[lc]@xmath401 ( 8,90)(0,0)[lc]@xmath402 ( 203,90)(0,0)[lc]@xmath403 ( 160,90)(0,0)[lc]@xmath404 ( -10,145)(0,0)[lc]@xmath405 ( 170,35)(0,0)[lc]@xmath406 ( -10,35)(0,0)[lc]@xmath407 ( 170,145)(0,0)[lc]@xmath408 ( 0,15)(0,0)[lc]the domain @xmath125 of definition of the symbol @xmath409 . according to @xcite the symbol @xmath409 of the operator @xmath301 is @xmath410 where @xmath411 \left(\displaystyle\frac{\eta-\gamma}{\eta+\gamma}\right)^{\mp s } , & \omega=(\eta,\pm\infty)\in\gamma^\pm_2 , \\[1ex ] e^{\pi si } , & \omega=(0,\xi)\in\overline{\gamma}_3 , \qquad \xi,\eta\in\mathbb{r},\end{cases}\\[2ex ] & & \hskip-20 mm g^s_{-\gamma,\gamma , p}(\infty,\xi):=\frac{e^{2\pi si}+1}2 + \frac{e^{2\pi si}-1}{2i}\cot\pi\big(\frac1p - i\xi\big)=e^{\pi si}\frac{\sin\pi\big(\frac1p+s - i\xi\big ) } { \sin\pi\big(\frac1p - i\xi\big ) } , \quad \xi\in\mathbb{r},\nonumber \end{aligned}\ ] ] @xmath412 0 , & \omega=(\eta,\pm\infty)\in\gamma^\pm_2,\\[1ex ] \displaystyle\frac{e^{-i\pi(\frac1p - i\xi-1)}c^{\frac1p - i\xi - s-1 } } { \sin\pi(\frac1p - i\xi)},&\omega=(0,\xi ) \in\overline{\gamma}_3,\end{cases}\\[1.5ex ] 0<\arg\,c<2\pi,\quad 0<|\arg(c\,\gamma)|<\pi,\quad 0<\arg\gamma<\pi \nonumber \end{aligned}\ ] ] and @xmath413 , @xmath414 . note , that the mellin convolution operator @xmath415 , @xmath416 which we encounter in applications ( see and lemma [ l1.5 ] ) , has a rather simple symbol in the bessel potential space @xmath417 : from follows that : @xmath418 the image of the function @xmath419 , @xmath420 is a closed curve in the complex plane ( easy to check analyzing the symbol in - ) . hence , if the symbol is elliptic , i.e. if @xmath421 the increment of the argument @xmath422 when @xmath423 ranges through @xmath125 in the direction of orientation , is an integer . it is called the winding number or the index of the curve @xmath424 and is denoted by @xmath425 . propositions [ p4.1]-[p4.3 ] , exposed below , are well known and will be applied in the next section in the proof of main theorems . [ p4.1 ] let @xmath3 , @xmath283 . the operator @xmath426 defined in is fredholm if and only if its symbol @xmath409 defined in , , is elliptic . if @xmath301 is fredholm , then @xmath427 the operator @xmath301 in is locally invertible at @xmath118 if and only if its symbol @xmath409 is elliptic on the set @xmath400 only : @xmath428 . [ p4.2 ] let @xmath3 , @xmath283 and let @xmath301 be defined by . if the operator @xmath429 is fredholm ( is invertible ) for all @xmath430 and @xmath431 , where @xmath432 , @xmath433 , then @xmath434 is fredholm ( is invertible , respectively ) in the sobolev - slobodekii space setting @xmath435 and has the same index @xmath436 [ p4.3 ] let two pairs of parameter - dependent banach spaces @xmath437 and @xmath438 , @xmath439 , have intersections @xmath440 dense in @xmath441 and in @xmath442 for all @xmath306 , @xmath443 . if a linear bounded operator @xmath444 is fredholm for all @xmath445 , it has the same kernel and co - kernel for all values of this parameter @xmath445 . in particular , if @xmath444 is fredholm for all @xmath445 and is invertible for only one value @xmath446 , it is invertible for all values of this parameter @xmath445 . the proof of theorem [ t0.4 ] ( see below ) is based , besides theorem [ t0.5 ] , on the following theorem . the system of the boundary pseudodifferential equations is fredholm in the sobolev - slobodekii space setting and in the bessel potential space setting if and only if the condition holds . the system has a unique solution in both settings and if the condition holds . @xmath447,\\ \label{e5.2b } & \phi:=\left(\begin{array}{c}{\varphi}\\ \psi\end{array}\right)\in \widetilde{{\mathbb{w}}}{}^r_p({\mathbb{r}}^+),\qquad { \bf f}:=\left(\begin{array}{c}g\\ h\end{array}\right)\in{\mathbb{w}}^r_p({\mathbb{r}}^+),\\ \label{e5.2c } & \phi:=\left(\begin{array}{c}{\varphi}\\ \psi\end{array}\right)\in \widetilde{{\mathbb{h}}}{}^r_p({\mathbb{r}}^+),\qquad { \bf f}:=\left(\begin{array}{c}g\\ h\end{array}\right)\in{\mathbb{h}}^r_p({\mathbb{r}}^+)\end{aligned}\ ] ] due to formulae and the symbol of @xmath448 on @xmath400 reads @xmath449 { \displaystyle}\frac { e^{-\pi r i}}{\sin\pi\xi } & e^{\pi r i}{\displaystyle}\frac{\sin\pi(\xi+r)}{\sin\pi\xi } \end{array}\right],\quad \omega=(\infty,\xi)\in\overline{\gamma_1 } , \end{aligned}\ ] ] where @xmath450 , @xmath451 , @xmath452 . we have dropped the information about the symbol @xmath453 on the contours @xmath454 and @xmath399 because , due to theorem [ t0.5 ] , we are interested only in the local invertibility of the operator @xmath448 at @xmath118 . this information , due to the concluding part of the proposition [ p4.1 ] , is contained in the symbol @xmath453 on the contour @xmath400 only . the symbol is non - elliptic if @xmath458 the latter equation has the following solutions @xmath459 because for @xmath460 the equation @xmath461 has no solution . equation decomposes into the following two equations for even and odd @xmath462 : @xmath463 from follows that if conditions hold , the operator @xmath448 is fredholm in both settings and . on the other hand , for the values @xmath45 , @xmath464 , which also satisfy the conditions , the operator @xmath448 is invertible ( see the concluding assertion in lemma [ l1.5 ] ) . then , due to proposition [ p4.3 ] , @xmath448 is invertible in both settings and for all those @xmath270 and @xmath82 which satisfy . from follows that , if conditions hold , the operator @xmath465 , corresponding to the system , is fredholm in both settings and . on the other hand , for the values @xmath45 , @xmath464 , which also satisfy the conditions , the operator @xmath466 is invertible ( see the concluding assertion in theorem [ t0.3a ] ) . then , due to proposition [ p4.3 ] , @xmath466 is invertible in both settings and for all those @xmath270 and @xmath82 which satisfy . * proof of theorem [ t0.3 ] : * due to theorem [ t0.3a ] and theorem [ t0.4 ] the bvp is fredholm if the system in the non - classical setting is , provided @xmath467 , i.e. , if the condition holds with @xmath467 ( cf . the condition ) , which is the same condition as . from follows that if conditions hold , the bvp is fredholm in the non - classical setting . on the other hand , for the values @xmath45 , @xmath108 , which also satisfy the conditions , the bvp has a unique solution ( see theorem [ t0.1 ] ) . then , due to proposition [ p4.3 ] , the bvp has a unique solution in the non - classical setting for all those @xmath284 and @xmath82 which satisfy . bonnet - ben dhia , l. chesnel and p. ciarlet , jr . , t - coercivity for scalar interface problems between dielectrics and metamaterials , _ esaim : mathematical modelling and numerical analysis _ * 46 * , 2012 , 1363 - 1387 . castro , r. duduchava and f .- o . speck , localization and minimal normalization of some basic mixed boundary value problems . _ in factorization , singular operators and related problems , proceedings of the conference in honour of professor georgii litvinchuk at funchal _ , portugal 2002 ( eds . s. samko et alt . ) , kluwer , dordrecht 2003 , 73 - 100 . r. duduchava , general singular integral equations and basic theorems of the plane theory of elasticity , _ trudi tbiliskogo mathematicheskogo instituta academii nauk gruzinskoi ssr _ * 82 * , 1986 , 45 - 89 ( russian ) . r. duduchava , mellin convolution operators in bessel potential spaces with admissible meromorphic kernels , http://arxiv.org/pdf/1502.02756.pdf , 55 pages ( a corrected version of the paper in _ memoirs on differential equations and mathematical physics _ * 60 * , 2013 , 135 - 177 . r. duduchava , m. tsaava and t. tsutsunava , mixed boundary value problem on hypersurfaces . _ international journal of differential equations _ , hindawi publishing corporation * 2014 * , article i d 245350 , 8 pages . g. i. eskin , boundary value problems for elliptic pseudodifferential equations . translated from the russian by s. smith . translations of mathematical monographs _ , * 52*. _ american mathematical society , providence , r.i . _ , 1981 . e. meister , some multiple - part wiener - hopf problems in mathematical physics , in : _ mathematical models and methods in mechanics _ , banach center publications * 15 * , pwn - polish scientific publishers , warsaw , 1985 , 359 - 407 . * r. duduchava * , _ a.razmadze mathematical institute , tbilisi state university , tamarashvili str . 6 , tbilisi 0177 , georgia . _ email : [email protected] + + * m. tsaava * , _ a.razmadze mathematical institute , tbilisi state university , tamarashvili str . 6 , tbilisi 0177 , georgia . _ email : [email protected]

we report evidence for an fe k fluorescence line feature and disk reflection in the very high , high , and low state x - ray spectra of the galactic microquasar during its june 1998 outburst . spectral analyses are made on data gathered throughout the outburst by the _ rossi x - ray timing explorer _ proportional counter array . gaussian line , relativistic disk emission line , and ionized disk reflection models are fit to the data . in the very high state the line profile appears strongly redshifted , consistent with disk emission from the innermost stable orbits around a maximally rotating kerr black hole . in the high state the line profile is less redshifted and increasingly prominent . the low state line profile is very strong (.5 kev equivalent width ) and centered at 6.7.10 kev ; disk line emission model fits indicate that the inner edge of the disk fluctuates between and in this state . the disk reflection fraction is traced through the outburst ; reflection from an ionized disk is preferred in the vhs and hs , and reflection from a relatively neutral disk is preferred in the ls . we discuss the implications of our findings for the binary system dynamics and accretion flow geometry in . [ firstpage ]

astro-ph0008118

quasars and other active galactic nuclei ( agn ) are generally thought to be powered by accretion onto supermassive black holes at the centers of galaxies . they are well - studied across the full astronomical bandpass , from radio to gamma - ray wavelengths . at x - ray energies , attention in recent years has focused on the broad fe k@xmath0 fluorescence line seen in many agn of the seyfert 1 type ( for a recent review , see fabian 2000 ) . at the high spectral resolutions first achieved with the _ asca _ satellite , the line in these sources is revealed to have a strongly asymmetric profile that is consistent with emission from the innermost radii of a relativistic accretion disk ( tanaka 1995 ; mushotzky 1995 ; nandra 1997 ) . the theory for the emission mechanism holds that the x - rays originate by inverse - compton processes acting in a hot corona above a cold ( weakly ionized ) accretion disk that produces thermal emission in the optical and ultraviolet ; thus the theory is bolstered by observations of a reflected continuum above @xmath110 kev that flattens the intrinsic inverse - compton power - law . within our galaxy , radio and x - ray observations over the course of the past six years have identified a population of x - ray binaries , the galactic microquasars , that are similar to agn in a number of important respects . first , these sources exhibit relativistic radio - emitting outflows ( e.g. , mirabel & rodriguez 1994 ) analogous to the megaparsec - scale radio jets of agn . second , the sources are thought to harbor black hole primaries : they exhibit the canonical x - ray spectral states of black hole candidates ( bhcs ; see discussion below ) , and when dynamical mass estimates have been obtained they have supported this hypothesis ( e.g. , orosz & bailyn 1997 ) . finally , the sources exhibit strong x - ray variability on a broad range of timescales ( hundreds of hertz to hundreds of days ) , also analogous to agn . the name `` microquasar '' refers approximately to the relative mass difference between the two classes of object : the @xmath110 masses of the microquasars , on the one hand , and the 10@xmath410@xmath5 masses of agn , on the other . this relative mass scale in turn is expected to set the relative variability timescale . in addition to the properties mentioned above , an important common characteristic of the microquasars ( without yet an analogue in agn ) are x - ray quasi - periodic oscillations ( qpos ) in the 30300 hz frequency range . these qpos have been identified in the sources grs 1915@xmath6105 ( morgan , remillard , & greiner 1997 ) , gro j1655@xmath740 ( remillard 1997 ) , xte j1550@xmath7564 ( remillard 1999 ) , ( fox & lewin 1998 ; revnivtsev , trudolyubov , & borozdin 1999 ) , and 4u 1630@xmath747 ( cui 1999a ) . the phenomenology of these qpos is remarkably rich and continues to challenge both the observers seeking to characterize them ( e.g. , reig 2000 ; sobczak 2000 ) , and the theorists seeking to explain them ( e.g. , dimatteo & psaltis 1999 ; lehr , wagoner , & wilms 2000 ) . not surprisingly , the microquasars have also been found to be a rich arena in which to study accretion physics and geometry via x - ray spectral features . _ rossi x - ray timing explorer _ ( _ rxte _ ; bradt , rothschild , & swank 1993 ) proportional counter array ( pca ; jahoda 1996 ) observations of xte j1550@xmath7564 are well - fit by including a gaussian line feature at 6.5 kev with fixed width of 1.2 kev ( fwhm ) ( sobczak 1999 ) . two line features , one at @xmath15.7 kev and one at @xmath17.7 kev , were observed in the 1996 outburst of 4u 1630@xmath747 ( cui , chen , & zhang , 1999 ) . the two lines are attributed to doppler shifting of neutral fe k@xmath0 in either a keplerian accretion disk or bipolar outflow . balucinska - church & church ( 2000 ) report likely fe k@xmath0 emission from four spectra taken from gro j1655@xmath740 . was discovered with the _ rxte _ all sky monitor ( asm ; levine 1996 ) on 4 june 1998 ( smith 1998 ) . radio observations undertaken with the very large array ( vla ) revealed an unresolved radio source with position coincident to that of the x - ray source ( r.a . @xmath8 , dec.@xmath9 , j2000 ; hjellming 1998a ) . subsequent radio observations revealed an extended source , exhibiting components with proper motion of 2040 mas day@xmath10 ( rupen , hjellming , & mioduszewski 1998 ) . based on a 21-cm hi absorption measurement the distance to the source was estimated to be @xmath118 kpc , implying that the intrinsic velocity of the jet components must be higher than 0.93@xmath12 ( hjellming 1999b ) . qpos with centroid frequencies of @xmath10.5 hz and @xmath132 hz were reported in the power density spectrum ( pds ) from 6 june 1998 ( fox and lewin , 1998 ) . the combination of relativistic radio jets and x - ray qpos secured the identification of as a galactic microquasar . multiple observations of the source were made with _ ( revnivtsev , trudolyubov , & borozdin 1999 ) and _ asca_(kotani 2000 ) ; the source flux decayed below the asm detection limit in september 1998 . the evolution of black hole x - ray binary ( bhxb ) outbursts is generally described in terms of several canonical spectral states , distinguished by the relative contribution from low energy and high energy x - rays within the 110 kev band . typically , low energy x - rays ( 12 kev ; the soft component ) dominate the emission of a bhxb outburst near maximum , and as the outburst evolves the soft component grows weaker . higher energy x - rays ( 510 kev ; the hard component ) become stronger and dominate as the overall flux decreases in the late stages of a bhxb outburst . the canonical bhxb spectral states are the very high state ( vhs ) , the high or soft state ( hs ) , the intermediate state ( is ) , the low or hard state ( ls ) , and the quiescent state ( qs ) . in the ls , the x - ray spectrum is nearly a pure power - law , with photon index 1.52.5 , often extending to hundreds of kev ( tanaka & lewin 1995 ) . in the hs , the 110 kev flux is an order of magnitude higher than in the ls due to the addition of a strong soft component . in the vhs the soft component is stronger still and may be accompanied by a substantial hard component ; in addition , qpos are often observed in this state . the is is a transitional state between the ls and the hs ( mendez & van der klis , 1997 ) , not observed in all sources . finally , the qs state is typified again by a non - thermal spectrum , with a flux level several orders of magnitude below the lowest ( ls ) outburst flux . although the temporal and spectral behavior of bhxbs has been widely studied , the mechanism driving the state transitions is still unknown . the quasi - thermal radiation present in the vhs , hs , and is is typically modeled as thermal emission from an optically - thick , geometrically - thin , multi - temperature accretion disk ( mitsuda 1984 ) . the power - law component , meanwhile , is generally attributed to inverse - compton processing of lower - energy disk photons in an optically - thin corona above the accretion disk ( sunyaev & titarchuk , 1980 ) . if signs of hardening in the spectrum above @xmath110 kev are present and accompanied by a fluorescent fe k@xmath0 line , then much as in agn these can be interpreted as the signature of compton reflection from the cold matter in the disk ( george & fabian 1991 ) . reflection features in the ls of bhxbs , however , are generally much less prominent than in agn ( zycki 1999 ) , so it has been argued that at least in the ls , the geometrically - thin disk does not extend inward to the innermost stable orbit , but is truncated at hundreds or thousands of schwarzschild radii ( esin 1997 ) . in the hs , the optically thick disk is postulated to move inwards so that the majority of the dissipated energy emerges in the form of a blackbody - like spectrum . measuring reflection spectra in bhxbs may therefore be helpful in understanding the physical properties of bhxb accretion geometries . previous analysis of _ rxte _ observations of revealed fe k@xmath0 emission in the ls spectra of this source ( revnivtsev , trudolyubov , & borozdin 2000 ) . our fits to the same data revealed a strong ls iron feature , and in addition we found that it was possible to trace the feature backwards into the hs , and even the vhs . here we present the results of 22 x - ray spectral observations spanning @xmath1120 days of the 1998 outburst of . we investigate relativistically - skewed emission features and disk reflection in the spectra from this source . we include all of the publicly - available _ rxte _ too data between peak bolometric luminosity and quiescence in our analysis . this embodies 22 pointed observations , effectively sampling the different spectral states across the outburst ; see table 1 for times and obsids of these observations . the standard ftools v.4.2 package was used to reduce the standard-2 pca data from these observations . all observations were dead - time corrected , and the background was calculated using the very large events ( or bright source ) option within pcabackest . the _ rxte _ pca instrument ( jahoda 1996 ) consists of 5 individual proportional - counter detectors , the proportional counter units ( pcus ) . due to gas exchange between propane and xenon layers in the detectors , the gain of the individual pcus drifts . pcus 0 , 1 , and 4 have very similar gain drift patterns , and a better overall performance than pcus 2 and 3 ( sobczak 1999 ; balucinska - church & church 2000 ) . we have chosen to fit data from all layers ( all propane layers and all xenon layers ) of pcus 0 , 1 , and 4 simultaneously . analysis of the data from an may 1998 observation of the crab nebula with the response matrices calculated via the ftool pcarmf ( v3.5 ) shows substantial residual trends . these deviations are thought to represent defects in the response matrices themselves as the crab is known to have a featureless power - law spectrum . we use response matrices generated by pca calibration specialist keith jahoda that eliminate these residuals . they are available through the goddard space flight center ftp archive at ` ftp://lheaftp.gsfc.nasa.gov/pub/keith/response_matrices_v2.2.1_v80.tar ` . a comparison of the data / model ratio using pcarsp - generated and jahoda - generated ( hereafter `` static '' ) response matrices , presented in fig . 1 , reveals that the static response matrices yield a much cleaner fit to the crab data in the 3.510.0 kev band . for this reason , we use the static response matrices for all spectral fits . we have analyzed every non - slewing observation of the crab nebula from may to october 1998 to check for variations in the performance of the static response matrices over time . these crab fits were used to select an appropriate energy range for precision analysis of the data from . the bins between 2.0 and 3.5 kev varied drastically and randomly in fits to these crab observations : the @xmath13 statistic for these bins consistently fell in the range of @xmath140100 . therefore , we set a lower energy limit of 3.5 kev for our analysis . above @xmath120 kev , neither type of response matrix satisfactorily fits the crab . as this energy is high enough to constrain reflection features fit to the data , we fix 20 kev as the upper limit of our fitting range . the addition of 1% systematic errors has become a standard practice when reducing pca spectral data ( see , e.g. , cui , chen , & zhang , 1999 ; sobczak 1999 ) . fits to the set of crab observations mentioned above with this 1% systematic error produced artificially low @xmath13 values . we find that a systematic error of 0.75% yields fits to the crab observations with @xmath14 ; we therefore adopt this systematic error value for our fits . this step adds validity to the measures of statistical merit based on @xmath15 that we employ in our analysis . onboard the pca are several pellets of radioactive americium , @xmath16am , that are used for in - flight calibration . data from these sources is accumulated every 128 seconds and telemetered as part of the standard1 data . as the line emission energies of the source are precisely known , any drift in the measured energy of the lines over the lifetime of the satellite can be directly attributed to changes in the detector response . we collected all of the calibration source data from the first ( 4 june ) and last ( 26 september ) days on which was observed ; analyzing this data in channel space we find that the centroids and widths of the @xmath16am lines agree to @xmath171% and @xmath172% , respectively , for these two days . we therefore expect that the pca energy scale drift is less than 1% over the time _ rxte _ conducted these observations . assuming a linear scale , this represents a systematic drift of @xmath180.06 kev for a line feature centered at 6.0 kev . as we determined , _ rxte _ s effective sensitivity range does not extend below 3.5 kev during the time of our observations . we therefore rely on _ asca _ observations to measure the column density reliably . kotani ( 2000 ) report @xmath20 for a power - law index of @xmath21 , and @xmath19 at @xmath22 for @xmath23 , each with an error of approximately @xmath230% . both of these asca - measured power - law indices agree with the power - law indices we measure using _ rxte _ data . we therefore fix the value of @xmath24 for all of our fits . our spectral fits progressed through three increasingly complex spectral models , a process which we illustrate for two representative observations ( obs . 3 and 19 , from the vhs and ls , respectively ) in figure 2 . first , we fit the data with the canonical multicolor blackbody disk plus power - law model ( top and second panels in figs . 2a and 2b ) . the most striking feature in the data / model ratio plots for this model ( second panel ) is a broad excess between 4 and 7 kev for obs . 3 , and a distinct 6.7 kev line for obs . our second model therefore adds one ( fig . 2b ) or two ( fig . 2a ) gaussian components to the underlying continuum ( third panel ) . in the case of obs . 19 ( and other ls observations ) , the fit is further improved by addition of a gaussian at 8 kev ( fourth panel in fig . 2b ) ; we discuss this feature in more detail in sect . [ res : line ] below . two spectral features are expected to result from the reflection of hot x - rays from an accretion disk : fluorescent line emission , and a broad compton - upscattering excess often referred to as a `` reflection bump . '' to be self - consistent , our final model ( bottom panels in figs . 2a and 2b ) therefore consists of the multicolor blackbody disk ( diskbb ) component , two gaussians , and a pexriv component ( magdziarz & zdziarski 1995 ) . pexriv incorporates an incident power - law source of photons , and the reflection of these photons off an ionized disk . for all fits with pexriv , we fix the power - law cutoff energy at 200 kev , and the disk inclination at 45 degrees ( as we do for disk line profiles ) . we report measurements of the disk reflection fraction ( @xmath25 ) and ionization ( @xmath26 ) . pexriv reduces to a simple power - law in the absence of reflection . we report the results of our continuum and gaussian - line fits in tables 2 and 3 , respectively . models for emission line propagation from a relativistic spacetime generally predict a double - peaked line profile ( see , e.g. , dabrowski 1997 , martocchia 2000 ) . noting that the double - peaked nature of the soft emission excess in the vhs might be due to relativistic skewing , we attempted to fit these spectra with a relativistic line feature , the laor model from xspec ( laor , 1991 ) . this is a general model and may be applied to kerr geometries but reduces to a schwarzschild geometry in the minimal angular momentum limit . in the absence of optical or radio data to constrain the disk inclination angle , we fix this parameter at 45 degrees . in addition , the disk emissivity profile is fixed at @xmath27 , the inner emission radius of the disk to @xmath28 ( as for a nearly - maximally rotating kerr black hole ) , and the outer emission radius to @xmath29 . the gravitational radius @xmath3 is @xmath30 , where @xmath31 is newton s gravitational constant , @xmath32 is the mass of the compact object , and @xmath12 is the speed of light ( @xmath3 is half of the schwarzschild radius ) . measurements of the inner disk radius in the vhs are reported in table 4 . in the ls , models for accretion flow around black holes ( esin 1997 ) suggest that the inner edge of the accretion disk does not extend as far inwards as in the vhs . measurements of the inner disk radius derived from the normalization of the multicolor blackbody disk component suggest the same phenomenology . it follows that emission from the inner disk regions in the ls would not suffer the same degree of relativistic skewing . thus , an additional fit is made to ls spectra with a model consisting of the `` diskline '' line emission model within xspec ( fabian 1989 ) and pexriv . for these fits , we fix the disk emission profile to @xmath27 and the disk inclination to 45 degrees ; and the outer disk radius to @xmath33 . we fit for inner disk radius , line centroid energy , and the component normalization . the results of fits using this model are reported in table 5 . the total 3.520.0 kev flux measured using our models peaks at 11.4@xmath34 ergs @xmath35 s@xmath10 between outburst days 2 and 8 , and is dominated by the hard power - law component . the soft multicolor disk blackbody flux is slightly less than half of the hard power - law flux in observations 3 and 4 , but a smaller fraction in observations 58 ( see table 2 ) . the normalization of the blackbody disk component is tightly constrained in this state , and is relatively constant . these relative flux measurements and very high overall flux level signify that is in the vhs during observations 38 . the disk blackbody temperature is well - measured , and decreases very slightly through this state , from 1.57 to 1.26 kev . the normalization of the hard power - law component is well - constrained ; the spectral index hardens through the vhs ( see table 2 , figure 4a ) . between observations 8 and 9 , the hard power - law flux falls and the 3.520.0 kev spectrum becomes dominated by the soft component . the multicolor blackbody disk flux increases by a factor of @xmath14 between these observations . we interpret this change in continuum component flux as the vhs to hs transition . the blackbody disk temperature decreases from 1.29 to 0.91 kev through this state ( see table 2 ) . the normalization of the soft component peaks mid - way through the hs , dipping most dramatically in observation 15 , just prior to the hs ls transition . the power - law spectral index is softer than in the vhs at the beginning of the hs , but the index steadily hardens through the hs . whereas the soft blackbody disk normalization peaks mid - way through this state , the hard power - law component normalization falls steadily throughout ( see table 2 ) . the multicolor blackbody disk temperature falls by half between observations 15 and 16 , and the soft flux is nearly extinguished . this marks the hs to ls transition . ginga observations of gs 1124@xmath768 ( ebisawa 1994 ) and gs 2023@xmath6338 ( zycki 1999 ) yield ls disk temperatures of @xmath10.3 kev . we measure a disk component with similarly low temperature in observations 16 , 17 , and 18 , but find no evidence of a disk flux in subsequent observations . it is therefore possible that the disk cools to a temperature well below the range of sensitivity of the pca . the power - law spectral index is uniformly harder ( @xmath36 ) in the ls than in any observation in the vhs or hs . the normalization of the power - law component decreases throughout the ls , as does the overall flux . the power - law index measured with pexriv hardens steadily , slightly increasing in the last two observations ( see table 2 , figure 4b ) . the multicolor blackbody disk model yields an inner accretion disk radius , with component normalization , source distance , and disk inclination as parameters . fits with this model indicate that the inner disk radius is relatively constant at @xmath110 km in the vhs , and @xmath119 km for every observation made in the hs ( see table 4 ) . figures 3a and 3b show data / model ratios for the vhs and hs , and ls , respectively . the ratios are off - set from each other , but on the same relative scale . the models used here are the multicolor blackbody disk plus pexriv for the vhs and hs , and pexriv only for the ls . these figures serve to illustrate how the equivalent width , width , and centroid energy of the line features detailed in table 3 vary across the outburst . the bottom data / model ratio plot in figure 3b corresponds to observation 23 , just before the source slipped beneath rxte detection limits . the line feature is very strong , broadened , and single - peaked . tracing backwards to the beginning of the ls ( top of figure 3b ) , it is readily apparent that the line centroid energy is approximately constant , and the equivalent width is diminished in earlier observations . the width of the line feature is measured to be @xmath10.30 kev throughout , indicating that the line is resolved by the pca in the ls ( see table 3 ) . the resolved line in the ls allows for tightly - constrained measurements via fits with the diskline model . centroid energy values fall within @xmath20.1 kev of 6.7 kev , and the line equivalent width increases steadily from @xmath37 ev to @xmath38 ev . these measurements are in good agreement with the values obtained by fitting the line with a gaussian profile . most notably , the diskline model measures the inner edge of the accretion disk , which remains relatively close to the innermost radius in the ls . our fits indicate the inner edge of the disk is moving closer to the bh in observations 16 , 17 , and 18 , from @xmath39 to @xmath40 . the inner edge then recedes back to @xmath41 in observation 19 . following this sharp recession , we measure the inner disk radius to move steadily inward in subsequent observations , to @xmath42 in observation 23 . fits to the ls are improved by adding a higher - energy gaussian line component to the spectral model at @xmath18.3 kev . this unresolved line feature is suggested at @xmath4368% confidence throughout the ls , and is consistent with fluorescent ni k@xmath0 emission from a highly - ionized species . the equivalent width of this line varies between 10260 ev from observation 16 to observation 23 . the presence of this line feature is visible in figure 2b once the more intense line at @xmath16.7 kev has been fit , but is not visible in the plots in figure 3b as it is far less intense than the lower energy line . following the data / model ratio from the beginning of the ls ( top of figure 3b ) to the end of the hs ( bottom of figure 3a ) , the @xmath16.7 kev line feature is still very prominent , and can in fact be traced all the way back to the vhs ( top of figure 3a ) . although the line feature is much weaker in the vhs than in the hs or ls , the constancy of the centroid energy and smooth evolution of the equivalent width clearly tie the line emission feature near @xmath16.7 kev in the hs and vhs to that in the ls . a lower - energy line feature is present in the hs that is not seen in the ls . we distinguish these hs and vhs features as the `` red '' and `` blue '' lines . fitting a model including two gaussian lines to the hs , we find that the blue line feature is significant at greater than 90% confidence in all but one observation in the hs ( obs the centroid energy of the blue line slowly increases from 6.4 to 6.7 kev from the beginning to the end of the hs . the equivalent width also increases steadily , from 39 to 347 ev ( see table 2 , fig . the red line is not significant at 90% confidence in obs . 10 , but is measured in the three observations at the beginning of the hs . again tracing backwards in time ( bottom to top ) in figure 3a , the rising equivalent width of the red component and the smeared profiles of both red and blue line features is clearly seen . in the vhs the line emission feature may be characterized as smeared and possibly double - peaked . fits to the vhs spectra with a model including two gaussian line emission features find the red line is significant at @xmath4399.9% confidence for every observation . the blue line is significant at @xmath4399.9% in all but obs . 6 . generally , the equivalent width and flux of the red line are greater than that of the blue line in the vhs . the red line centroid energy decreases slightly through the vhs , but is consistent with 4.6 kev ( see fig . the centroid energy of the blue line varies in the vhs , steadily increasing from 5.8 to 6.6 kev ( see table 3 ) . the red line width is inconsistent with zero in the vhs except in observations 3 and 6 ; the blue line except in observation 7 . the red line is likely at the limit of the pca resolution in the vhs ; it is unresolved in the hs . the blue line is especially broad ( 0.8 kev ) at the beginning of the vhs , becoming narrower throughout ( to 0.4 kev in obs . . fits with the laor relativistically - skewed line emission model show that the addition of a line feature emitted from a region between @xmath44 and @xmath45 , with centroid energy of 5.7 kev , is significant at 90% confidence in observations 3 , 4 , and 8 . later observations can not be fit as well with the laor model . indeed , we do not expect that the resolution of the pca is sufficient to constrain laor model parameters ; fits with this model serve primarily as a consistency check . the continuity of the changes in line parameters and profiles as we search backwards from the ls to the vhs indicates that the line - like excesses seen in figures 3a and 3b are very likely manifestations of the same line emission feature . tracing the line feature(s ) backwards in time from the ls , the equivalent width is observed to steadily decrease . it is important to note that the flux in the red line feature in the vhs is greater than the flux in the single - peaked / blue line feature in the ls ( see table 3 ) . considering the states separately , the flux of the lines is highest at the beginning of each spectral state , and gradually decreases throughout ( see table 3 , fig . 4 ) . more generally , the lines have the highest flux in the vhs at the beginning of the outburst . the line profile is possibly double - peaked in the vhs . as the outburst moves into the hs , the line profile becomes single - peaked and increases in equivalent width , though the line flux decreases steadily . finally , as the outburst moves into the ls , the line equivalent width becomes very large ( the soft blackbody disk continuum component disappears ) and the line flux continues to decrease steadily . revnivtsev , trudolyubov , & borozdin ( 1999 ) only find evidence for a line component in the ls spectrum of , and suggest that the line is from diffuse galactic emission as the line flux is stable relative to the decline in the power - law flux . we disagree with this interpretation . our finding that the line flux is highest at the beginning of each spectral state and decays until the end of the state demands that the spectral lines are produced by . the power - law continuum component pexriv models the reflection of hot x - rays impinging on an ionized , optically - thick gas , often assumed to be an accretion disk . in the vhs , the reflection fraction @xmath46 is small but non - zero at @xmath4390% confidence ( except obs . 6 ; see table 2 , fig . measurements indicate that the disk is very ionized at the beginning of the vhs ( @xmath47 in obs . 3 ) . the ionization is constrained only in observations 3 , 4 , and 5 . thereafter in the vhs , and throughout the ls , the ionization is not tightly constrained but a highly - ionized disk is preferred at 90% confidence . we therefore fixed the ionization ( @xmath48 ) in fitting observations 615 . across the vhs hs transition , @xmath46 varies discontinuously from @xmath49 in observation 8 to @xmath50 in observation 9 . throughout the hs , @xmath46 increases to a value of @xmath51 in obs . 15 . perhaps due to the dominance of the soft component in the hs , @xmath46 is consistent with no reflection throughout . the reflection fraction @xmath46 again varies sharply across the hs ls transition , from @xmath51 in observation 15 to @xmath52 in observation 17 . the value of @xmath46 falls sharply in subsequent observations , to @xmath49 in observation 21 , and becomes consistent with zero in obs . 23 ( see table 2 , fig . measurements of @xmath53 in the ls are consistent with a neutral disk , marking a sharp contrast to the highly - ionized disk required in the vhs and hs . based on the broadened , redshifted line emission profile observed by iwasawa in mcg@xmath76 - 30 - 15 ( 1997 ) , and on the profiles and equivalent widths calculated by many researchers ( e.g. magdziarz & zdziarski 1995 , fabian 1989 , george & fabian 1991 , dabrowski 1997 , martocchia 2000 ) , we interpret the red and blue line features in the vhs of as wings of a doppler- and gravitationally - shifted fe k@xmath0 line emission profile . the line profiles ( figs . 2a and 3a ; table 3 ) are consistent with those calculated by martocchia ( 2000 ) for emission from the innermost region of an accretion disk orbiting a bh of high angular momentum , viewed at a modest inclination angle , with a source of hot x - rays located @xmath54 above the bh . the equivalent widths measured in the vhs are broadly consistent with those predicted by the model for a disk with outer emission radius near @xmath55 , and viewed at an angle slightly below the 45 degree inclination we assumed . our fits to the vhs using the laor model indicate that in this spectral state the relativistically - skewed line is consistent with emission from an inner radius extending down to the marginally stable orbit for a maximally rotating kerr black hole , @xmath44 ( see table 4 ) , in obs . 3 and obs . 4 . measurements of the inner disk radius via the multicolor blackbody disk model in the vhs confirm that the disk may extend to the innermost stable orbit in a kerr geometry ( see table 4 ) . throughout the hs , the red wing of the profile becomes less prominent , and the blue wing strengthens . in particular , the centroid energy of the blue line shifts from 5.8 kev ( beginning of the vhs ) to 6.7 kev ( end of the hs ) . throughout the ls , the line profile is single - peaked , and measured at [email protected] kev . this value is consistent with highly ionized species of iron . we suggest that this line is likely a complex of lithium - like , helium - like , and hydrogenic iron species ( fe xxiv - xxvi ) . the shift of line profile to higher centroid energies may be produced in two ways , either by an evolving ratio of iron ionizations contributing to the line profile , or by the disk mildly recessing from the bh . given the inner disk radii we measure ( see table 2 ) across the outburst and the high ionization preferred in the vhs and hs , it is likely that the latter process dominates , but both may contribute . although unlikely given the relatively continuous evolution of the line profile ( in contrast to the discontinuous nature of the jet , which is only reported in the vhs ) , it is nevertheless possible that the lines originate within the jet itself . it is also possible that the lines are produced within a hot coronal region , and the measured reflection is from a cold geometry apart from the disk . the reflection fractions we measure in the vhs and hs are smaller than the same parameter measured in observations of other bhxbs and agn . done , madejski , & zycki ( 2000 ) measure @xmath46 between @xmath56 in seyfert 1 galaxy ic4329a . zycki , done , & smith ( 1998 ) measure @xmath57 in the vhs , and @xmath58 in the hs of the 1991 outburst of gs 1124@xmath768 . zycki ( 1999 ) measure @xmath59 during the decline of gs 2023@xmath6338 ( this source was dominated by the hard power - law component through its decline , unlike and gs 1124@xmath768 ) . the reflection fraction measured in the early ls observations of is comparable to that measured in the ls of cygnus x-1 by gierlinski ( 1999 ) @xmath60 . the low reflection fractions we measure are likely due to the high disk ionizations strongly preferred in the vhs and hs . the contrasting high reflection and negligible ionization at the beginning of the ls is not surprising ; a similar phenomenology is noted by zycki , done , & smith ( 1998 ) for the hs - ls transition in gs 1124@xmath768 . in a geometry where the disk is illuminated by a central source of hot x - rays , the disk should be most ionized when the hard x - ray flux is highest and when the disk is closest to the central source . the hard x - ray fluxes and inner disk radii measured in support this picture . moreover , the lines that we measure in the vhs and hs , though often significant at 99.9% confidence , are still weaker than those seen in agn spectra . therefore , the strength of the two components predicted for reflection spectra fluorescent line production and disk reflection , are at least qualitatively consistent . ross , fabian , and young ( 1999 ) note that even the small departures from uniformity that may occur in a highly ionized disk can account for the small reflection fraction often observed is such scenarios , without invoking reflection from the outer disk . this research , and that of nayakshin , kazanas , & kallman ( 1999 ) finds that the hottest outer layers of the accretion disk are important in accurately determining disk reflection fractions . these studies find that pexriv may be too basic in its treatment of compton processes in these hot outer layers . at present , pexriv is the best code publicly available . we look forward to the release of more sophisticated codes and to high resolution data from chandra and xmm - newton to constrain new models . we also note that we are only able to obtain measurements for the reflection fraction inconsistent with zero when the hard flux dominates the 3.5 - 20.0 kev bandpass we fit . this condition is met throughout the vhs , in obs . 15 in the hs , and throughout the ls . with the exception of obs . 15 , we are not able to place tight constraints on reflection in the hs . many models for disk reflection are based upon agn spectra ( the codes we mention above are notable exceptions ) , which generally do not contain a soft component in the x - ray band . the extent to which the physical process of soft x - ray emission from the disk is responsible for confusing disk reflection measurements , relative to merely poor mathematical modeling , is unknown . dimatteo & psaltis ( 1999 ) establish a possible connection between qpo frequency and inner disk radius in bhxbs based on similar work on neutron star ( ns ) systems : @xmath61 where @xmath62 is the qpo frequency , @xmath63 is the inner disk radius , @xmath3 the gravitational radius , @xmath32 the mass of the black hole , and @xmath64 a solar mass . for a measured qpo frequency , and a measured or assumed black hole mass , the inner radius can be determined via this equation . inserting the 32 hz qpo ( fox & lewin , 1998 ) measured in the vhs of , and assuming a black hole mass of 10 , this equation predicts an upper limit for the accretion disk inner radius of @xmath65 . this value supports the measurements we obtain via the laor line model , and our comparison to the models of martocchia ( 2000 ) . the multicolor blackbody disk continuum model also provides a measure of the inner disk radius . a central black hole with a mass of 10@xmath64 has a gravitational radius of @xmath115 km ; a central black hole with mass 5@xmath64 a gravitational radius of @xmath17.5 km . fits using the multicolor blackbody disk model measure the inner disk radius in the vhs to be @xmath110 km . fits with the laor disk line emission model in the vhs are consistent with an inner disk radius extending down to 1.2@xmath3 ; this corresponds to @xmath118 km for @xmath66 , and @xmath19 km for @xmath67 . finally , the connection noted by dimatteo & psaltis ( 1999 ) suggests an inner disk radius of @xmath68 , or 41.3 km and 20.6 km for 10@xmath64 and 5@xmath64 black holes , respectively . although these methods do not yield exactly the same inner disk radii , they all indicate that the inner edge of the accretion disk extends below the marginally stable orbit for a non - rotating schwarzschild black hole ( @xmath69 km ; for @xmath70 , @xmath71 km , for @xmath72 ; see table 4 for @xmath63 results ) . narayan and yi ( 1994 ) describe a solution for bh accretion geometry wherein a hot , optically thin , inefficiently - cooled plasma advects energy directly onto the bh . originally developed to describe the presence of hard emission within the spectra of agn , these advection - dominated accretion flow ( adaf ) models successfully describe a number of agn ( see narayan 1997 for a review ) . a natural application of this model is soft x - ray transients in quiescence , where @xmath73 is sufficiently small . bhxbs v404 cygnus and a0620 - 00 are well fit by adaf models in quiescence ( narayan , mcclintock , & yi 1996 ; narayan , barret , & mcclintock 1997a ) . esin , mcclintock , and narayan ( 1997 ) describe an adaf / thin - disk composite model tuned to explain the full spectral state evolution of bhxbs over a range of luminosities and @xmath73 . the parameter which drives the spectral transitions is the location of the adaf / disk transition radius , @xmath74 . when @xmath74 is close to the bh , the emission is dominated by the disk . as the outburst evolves , cooling becomes less efficient , @xmath73 falls , the inner region becomes an adaf , and @xmath74 increases . esin predict that in the vhs and hs , the disk may extend to the innermost stable orbit , but in the ls and qs @xmath74 is likely to be near @xmath75 . this model is applied to the 1991 outburst of gs 1124@xmath768 , and describes all but the vhs . subsequent application of the model to cygnus x-1 in its ls indicates @xmath76 ( esin 1998 ) . zycki , done , & smith ( 1998 ) constrain the hot advection zone to lie within @xmath77 in the ls of gs 1124@xmath768 , based on smearing of the compton upscattering excess observed above @xmath110 kev . the inner radii we measure in the ls of via the multicolor blackbody disk model and the diskline relativistic line model also constrain the inner disk edge to lie in this range . we therefore conclude that a large ( @xmath33 ) inner advection region of the kind predicted by esin , mcclintock , and narayan is not consistent with the ls of but must be within 100 @xmath3 ( similar to cygnus x-1 ) if such a geometry exists . we make note of work by merloni , fabian , & ross ( 2000 ) , which finds that the multicolor blackbody disk model systematically underestimates the real inner disk radius , and can suggest inner disk motion at high accretion rates when in fact the disk is stable at the marginally stable orbit . in the vhs , we find that the multicolor blackbody disk radius is stable and corresponds to the marginally stable orbit for a kerr black hole ( see table 4 ) . the radii we measure via this model in the vhs are supported by radii derived via dimatteo & psaltis ( 1999 ) , but may be an important caveat for hs measurements . combining the relation discussed by dimatteo & psaltis with the multicolor blackbody disk normalization in the vhs and the radio - band distance estimate , we obtain the following constraint for the mass of the central object in : @xmath78 where @xmath32 is the mass of the compact object , @xmath64 the solar mass , @xmath79 the distance in kiloparsecs , @xmath80 is the normalization of the diskbb component ( see table 2 ) , @xmath81 the disk inclination angle , and @xmath82 the qpo frequency . although this calculation is subject to numerous uncertainties , including the uncertain color correction ( ebisawa 1994 ) , taken at face value it implies a black hole primary ( @xmath83 ) in for a large parameter space including all likely values of @xmath79 ( @xmath438 kpc from radio measurements ) and @xmath84 ( @xmath171 ) . cui , chen , & zhang ( 1999 ) find two co - moving line features placed quasi - symmetrically about 6.5 kev in the 1996 outburst of 4u 1630 - 47 . it is suggested that each may be due to doppler - shifted fe k@xmath0 emission ( not principally gravitational redshifting , in the case of the reddened line ) . balucinska - church & church ( 2000 ) report on four spectra from gro j1655@xmath740 , and find evidence for gravitationally redshifted and doppler - broadened emission . our finding that the red emission wing is more prominent than the blue emission wing early in the vhs of is qualitatively similar to these results . additionally , fits with pexriv to these observations of gro j1655@xmath740 find limits for disk reflection ( @xmath85 1% ) and ionization ( @xmath86 ) . like the hs of these observations occur when the flux is dominated by the soft disk component , likely complicating efforts to measure any reflection . unfortunately , 4u 1630@xmath747 and gro j1655@xmath740 are both known to be `` dipping '' sources ( kuulkers 1988 ) to experience periods of markedly different absorption during outburst and this complicates efforts to trace spectral lines across a full outburst and thereby constrain accretion flow geometries . we have analyzed the 22 publicly - available observations between maximum x - ray luminosity and quiescence from the outburst of galactic microquasar bhxb . the spectra can be categorized into very high , high , and low states based on the relative flux contributions from hard and soft continuum components . we trace emission lines through these spectral states , and find evidence for an fe k@xmath0 line that is gravitationally redshifted and doppler shifted in the vhs , with a profile that is consistent with emission from the innermost stable orbit for a maximally rotating kerr black hole . the line is less redshifted but increasingly prominent in the hs . finally , in the ls , the line is centered at [email protected] kev , and we measure a profile that is consistent with emission from a disk with inner radius between @xmath87 and @xmath88 . we fit each observation with a power - law continuum model ( pexriv ) that includes compton reflection of hot incident x - rays from an ionized disk component , manifested spectrally as a broad excess above @xmath110 kev . this work represents the first trace of the disk reflection fraction across the full outburst of a bhxb . the fractions we measure in the vhs and hs are much smaller than those often measured in agn , but well - constrained . we note that the pexriv model may be too basic for accurate reflection measurements , and urge the public release of new codes . the fe k@xmath0 line serves as a direct diagnostic of the accretion geometry . we consider our findings in the context of the adaf model developed by esin , mcclintock , & narayan ( 1997 ) to describe the spectral states of bhxbs . as the disk in is measured to be a few tens of @xmath3 from the bh in the ls , rather than @xmath89-@xmath33 , we find that the predictions of this adaf model are inconsistent with the accretion geometry of . we look forward to data from chandra and xmm to obtain higher - resolution spectra of a bhxb in outburst . special thanks is due taro kotani for his communication of asca @xmath90 measurements in advance of publication . whgl gratefully acknowledges support from nasa . tdm and rw were supported by nasa through chandra fellowship grant numbers pf9 - 10005 and pf-9 - 10010 , respectively , which are operated by the smithsonian astrophysical observatory for nasa under contract nas8 - 39073 . this research has made use of data and resources obtained through the heasarc online service , provided by the nasa - gsfc . iwasawa , k. , fabian , a. c. , young , a. j. , h. inoue , & matsumoto , c. , 1999 , mnras 306l , 19 + kuulkers , e. , wijnands , r. , belloni , t. , mendez , m. , van der klis , m. , van paradijs , j. , 1998 , apj 494 753 + jahoda , k. , swank , j. h. , giles , a. b. , stark , m. j. , strohmayer , t. , zhang , w. , & morgan , e. h. , 1996 , spie , 2808 , 59 lrllllll no . & obs.id & date , ut & start time & pca exp . ( s ) + 1 & 30188 - 05 - 01 - 00 & 04/06/98 & 20:05:04 & 1760 + 2 & 30188 - 05 - 02 - 00 & 05/06/98 & 03:03:44 & 768 + 3 & 30171 - 02 - 01 - 00 & 06/06/98 & 09:41:20 & 2655 + 4 & 30185 - 01 - 01 - 00 & 07/06/98 & 07:56:32 & 2944 + 5 & ... -02 - 00 & 08/06/98 & 06:23:28 & 3027 + 6 & ... -03 - 00 & 09/06/98 & 12:48:00 & 3729 + 7 & ... -04 - 00 & 10/06/98 & 03:38:24 & 7721 + 8 & ... -05 - 00 & 11/06/98 & 12:52:16 & 3439 + 9 & ... -06 - 00 & 13/06/98 & 12:51:28 & 3114 + 10 & ... -07 - 00 & 15/06/98 & 04:53:36 & 1795 + 11 & ... -08 - 00 & 18/06/98 & 20:55:12 & 2327 + 12 & ... -09 - 00 & 22/06/98 & 22:30:08 & 3210 + 13 & ... -10 - 00 & 27/06/98 & 11:39:28 & 1647 + 13.1 & ... -10 - 01 & 02/07/98 & 18:18:24 & 1280 + 14 & ... -11 - 00 & 08/07/98 & 16:21:52 & 1295 + 15 & ... -12 - 00 & 13/07/98 & 06:44:00 & 2056 + 16 & ... -13 - 00 & 18/07/98 & 04:00:32 & 10585 + 17 & ... -14 - 00 & 30/07/98 & 09:44:48 & 6841 + 18 & ... -15 - 00 & 05/08/98 & 18:25:20 & 4333 + 19 & ... -16 - 00 & 13/08/98 & 10:17:36 & 1565 + 20 & ... -17 - 00 & 20/08/98 & 16:41:20 & 1704 + 21 & ... -18 - 00 & 25/08/98 & 03:32:00 & 1785 + 22 & ... -19 - 00 & 14/09/98 & 08:17:36 & 886 + 23 & ... -20 - 00 & 26/09/98 & 03:29:04 & 10287 + lllllllllllll obs . & @xmath91 & norm . & @xmath92 & flux@xmath93 & @xmath94 & @xmath46 & f - stat . & @xmath53 & norm . & flux@xmath95 & flux@xmath96 & @xmath97 + & ( kev ) & & ( km ) & @xmath98 & & @xmath99 & & @xmath89 & & @xmath98 & @xmath98 & + + 3 & @xmath100 & @xmath101 & 9.0(1 ) & @xmath102 & 2.728(1 ) & @xmath103 & 0.079 & @xmath104 & @xmath105 & @xmath106 & @xmath107 & 1.02 + 4 & @xmath108 & @xmath109 & 9.7(1 ) & @xmath110 & 2.736(1 ) & @xmath111 & 0.036 & @xmath112 & @xmath113 & @xmath114 & @xmath115 & 0.64 + 5 & @xmath116 & @xmath117 & 9.9(3 ) & @xmath118 & 2.631(1 ) & @xmath119 & 0.001 & @xmath120 & @xmath121 & @xmath122 & @xmath123 & 0.49 + 6 & @xmath124 & @xmath125 & 10.6(5 ) & @xmath126 & 2.515(1 ) & @xmath127 & 0.403 & @xmath128 & @xmath129 & @xmath130 & @xmath131 & 0.45 + 7 & @xmath132 & @xmath133 & 10.2(4 ) & @xmath126 & 2.571(1 ) & @xmath134 & @xmath170.001 & @xmath128 & @xmath135 & @xmath136 & @xmath137 & 0.34 + 8 & @xmath138 & @xmath139 & 8.9(2 ) & @xmath140 & 2.466(1 ) & @xmath141 & @xmath170.001 & @xmath128 & @xmath142 & @xmath143 & @xmath144 & 0.51 + + 9 & @xmath145 & @xmath146 & 18.7(2 ) & @xmath147 & 2.73(2 ) & @xmath148 & 1.0 & @xmath128 & @xmath149 & @xmath150 & @xmath151 & 0.42 + 10 & @xmath152 & @xmath153 & 18.3(2 ) & @xmath154 & 2.94(3 ) & @xmath155 & 1.0 & @xmath128 & @xmath156 & @xmath157 & @xmath158 & 0.69 + 11 & @xmath159 & @xmath160 & 19.1(1 ) & @xmath161 & 2.90(6 ) & @xmath162 & 1.0 & @xmath128 & @xmath163 & @xmath164 & @xmath165 & 0.55 + 12 & @xmath166 & @xmath167 & 19.9(1 ) & @xmath168 & 2.69(7 ) & @xmath169 & 1.0 & @xmath128 & @xmath140 & @xmath170 & @xmath171&0.70 + 13 & @xmath172 & @xmath173 & 20.0(1 ) & @xmath174 & 2.58(1 ) & @xmath175 & 1.0 & @xmath128 & @xmath176 & @xmath177 & @xmath178 & 0.69 + 13.1 & @xmath179 & @xmath180 & 20.2(3 ) & @xmath181 & 2.47(5 ) & @xmath182 & 1.0 & @xmath128 & @xmath183 & @xmath184 & @xmath185 & 0.45 + 14 & @xmath186 & @xmath187 & 20.0(3 ) & @xmath126 & 2.41(7 ) & @xmath188 & 1.0 & @xmath128 & @xmath189 & @xmath190 & @xmath191 & 1.13 + 15 & @xmath192 & @xmath193 & 18.8(3 ) & @xmath194 & 2.44(5 ) & @xmath195 & @xmath17 0.001 & @xmath128 & @xmath196 & @xmath197 & @xmath198 & 0.94 + + 16 & @xmath199 & @xmath200 & @xmath201 & @xmath202 & 1.86(2 ) & @xmath203 & @xmath170.001 & @xmath204 & 0.32(1 ) & @xmath205 & @xmath126 & 0.75 + 17 & @xmath206 & @xmath207 & @xmath208 & @xmath209 & 1.98(1 ) & 45(2 ) & @xmath170.001 & @xmath204 & 0.24(1 ) & @xmath210 & @xmath211 & 0.62 + 18 & @xmath212 & @xmath213 & @xmath208 & @xmath202 & 1.95(2 ) & @xmath214 & @xmath170.001 & @xmath204 & 0.20(1 ) & @xmath215 & @xmath216 & 1.02 + 19 & & & & & 2.02(1 ) & @xmath217 & @xmath170.001 & @xmath204 & 0.21(1 ) & @xmath218 & @xmath219 & 0.85 + 20 & & & & & 1.90(1 ) & @xmath220 & 0.254 & @xmath221 & 0.11(1 ) & @xmath222 & @xmath223&0.78 + 21 & & & & & 1.91(1 ) & @xmath224 & 1.0 & @xmath225 & 0.10(1 ) & @xmath226 & @xmath227 & 0.72 + 22 & & & & & 2.17(1 ) & @xmath228 & @xmath170.001 & @xmath204 & 0.17(1 ) & @xmath226 & @xmath227 & 0.77 + 23 & & & & & 2.08(1 ) & @xmath140 & 1.0 & @xmath204 & 0.10(1 ) & @xmath229 & @xmath230 & 0.40 + lllllllll obs . & centroid energy & width & eq.width & flux & @xmath13/dof & @xmath13/dof & @xmath231-stat . + & ( kev ) & ( kev ) & ( ev ) & @xmath232 & w / o line & w/ line & + + + 3 & @xmath233 & @xmath234 & @xmath235 & @xmath236 & 4.19 & 1.02 & @xmath170.001 + 4 & @xmath233 & @xmath237 & @xmath238 & @xmath239 & 5.62 & 0.64 & @xmath170.001 + 5 & @xmath240 & @xmath241 & @xmath242 & @xmath243 & 2.97 & 0.49 & @xmath170.001 + 6 & @xmath244 & @xmath245 & @xmath246 & @xmath247 & 1.87 & 0.45 & @xmath170.001 + 7 & @xmath244 & @xmath248 ) & @xmath249 & @xmath247 & 1.96 & 0.34 & @xmath170.001 + 8 & @xmath250 & @xmath251 & @xmath252 & @xmath253 & 4.62 & 0.50 & @xmath170.001 + + 9 & @xmath244 & @xmath254 & @xmath255 & @xmath256 & 1.35 & 0.42 & @xmath170.001 + 10@xmath257 & @xmath258 & @xmath259 & @xmath260 & @xmath220 & 1.00 & 0.69 & 0.14 + 11 & @xmath244 & @xmath259 & @xmath261 & @xmath262 & 0.83 & 0.45 & 0.04 + 12 & & & & & & & + 13 & & & & & & & + 13.1 & & & & & & & + 14 & & & & & & & + 15 & & & & & & & + + + + 3 & @xmath263 & @xmath264 & @xmath265 & @xmath266 & 5.42 & 1.02 & @xmath170.001 + 4 & @xmath267 & @xmath268 & @xmath269 & @xmath270 & 2.49 & 0.64 & @xmath170.001 + 5 & @xmath271 & @xmath272 & @xmath273 & @xmath274 & 1.57 & 0.49 & @xmath170.001 + 6 & @xmath275 & @xmath127 & @xmath276 & @xmath277 & 1.25 & 0.45 & 0.002 + 7 & @xmath275 & @xmath278 & @xmath279 & @xmath280 & 1.09 & 0.34 & @xmath170.001 + 8 & @xmath281 & @xmath282 & @xmath283 & @xmath280 & 2.00 & 0.50 & @xmath170.001 + + 9 & @xmath275 & @xmath284 & @xmath285 & @xmath220 & 1.20 & 0.42 & 0.002 + 10@xmath257 & @xmath286 & @xmath287 & @xmath288 & @xmath262 & 1.07 & 0.69 & 0.103 + 11 & @xmath289 & @xmath290 & @xmath291 & @xmath292 & 1.03 & 0.55 & 0.036 + 12 & @xmath293 & @xmath294 & @xmath295 & @xmath262 & 1.31 & 0.70 & 0.036 + 13 & @xmath158 & @xmath296 & @xmath297 & @xmath292 & 2.19 & 0.69 & @xmath170.001 + 13.1 & @xmath298 & @xmath299 & @xmath300 & @xmath301 & 3.13 & 0.45 & @xmath170.001 + 14 & @xmath302 & @xmath299 & @xmath303 & @xmath304 & 5.96 & 1.13 & @xmath170.000 + 15 & @xmath305 & @xmath306 & @xmath307 & @xmath308 & 23.7 & 0.94 & @xmath170.001 + + 16 & @xmath309 & @xmath310 & @xmath311 & @xmath312 & 16.9 & 0.75 & @xmath170.001 + 17 & @xmath313 & @xmath314 & @xmath315 & @xmath185 & 44.6 & 0.62 & @xmath170.001 + 18 & @xmath316 & @xmath317 & @xmath318 & @xmath319 & 50.1 & 1.02 & @xmath170.001 + 19 & @xmath320 & @xmath321 & @xmath322 & @xmath185 & 32.6 & 0.85 & @xmath170.001 + 20 & @xmath323 & @xmath324 & @xmath325 & @xmath181 & 27.0 & 0.78 & @xmath170.001 + 21 & @xmath326 & @xmath327 & @xmath328 & @xmath191 & 25.8 & 0.72 & @xmath170.001 + 22 & @xmath329 & @xmath330 & @xmath331 & @xmath332 & 30.4 & 0.77 & @xmath170.001 + 23 & @xmath333 & @xmath334 & @xmath335 & @xmath181 & 30.5 & 0.40 & @xmath170.001 + lllll obs . & @xmath336 & @xmath337 & @xmath338 & @xmath339 + & ( hz ) & ( @xmath3 ) & ( @xmath3 ) & ( @xmath3 ) + 3 & 31.6(2 ) & 4.38(1 ) & 2.8(1 ) & 1.2(1 ) + 4 & 31.3(2 ) & 4.39(1 ) & 2.8(1 ) & 1.3(1 ) + 5 & 23.7(1 ) & 4.84(1 ) & 3.0(1 ) & 1.3(1 ) + 6 & 20.2(1 ) & 5.12(1 ) & 3.2(1 ) & 1.4(1 ) + 7 & 20.0(1 ) & 5.14(1 ) & 3.2(1 ) & 1.4(1 ) + 8 & 22.6(2 ) & 4.92(1 ) & 3.1(1 ) & 1.2(1 ) + lllllll obs . & @xmath340 & @xmath63 & ew & flux(325 kev ) & @xmath13/dof & @xmath13/dof + & ( kev ) & ( @xmath3 ) & ( ev ) & @xmath232 & ( gaussian ) & ( diskline ) + 16 & @xmath309 & @xmath341 & @xmath342 & @xmath343 & 0.75 & 0.55 + 17 & @xmath313 & @xmath344 & @xmath345 & @xmath332 & 0.62 & 0.77 + 18 & @xmath346 & @xmath347 & @xmath348 & @xmath349 & 1.02 & 0.81 + 19 & @xmath350 & @xmath351 & @xmath352 & @xmath349 & 0.85 & 0.91 + 20 & @xmath353 & @xmath354 & @xmath355 & @xmath312 & 0.78 & 0.89 + 21 & @xmath333 & @xmath356 & @xmath357 & @xmath312 & 0.72 & 0.80 + 22 & @xmath329 & @xmath358 & @xmath359 & @xmath360 & 0.77 & 0.96 + 23 & @xmath361 & @xmath362 & @xmath363 & @xmath364 & 0.40 & 0.80 +

we develop a ginzburg - landau theory for the jaynes - cummings - hubbard model which effectively describes both static and dynamic properties of photons evolving in a cubic lattice of cavities , each filled with a two - level atom . to this end we calculate the effective action to first - order in the hopping parameter . within a landau description of a spatially and temporally constant order parameter we calculate the finite - temperature mean - field quantum phase boundary between a mott insulating and a superfluid phase of polaritons . furthermore , within the ginzburg - landau description of a spatio - temporal varying order parameter we determine the excitation spectra in both phases and , in particular , the sound velocity of light in the superfluid phase .

1201.6060

in many fields of physics , especially in the fields of information processing , material science , superfluidity , and the relatively new field of quantum information theory , a profound understanding of strongly correlated quantum many - body systems is of striking importance in order to further improve existing applications and invent new ones @xcite . this is due to the fact that these research fields mainly use solid - state systems in which strongly correlated systems appear quite naturally . however , it is experimentally challenging to access the microscopic properties of such systems due to the short time and length scales involved . therefore , motivated by feynman s conjecture of the quantum simulator @xcite , artificial structures have been considered to create effective many - body systems , which can be investigated under much more controllable and tunable experimental conditions . the first attempts to build up artificial many - body structures used josephson junction arrays @xcite , which proved to be capable of simulating the dynamic properties of bose - hubbard systems including the generic quantum phase transition of this model @xcite . additionally , over the last two decades , the advances in preparing and controlling ultra - cold atoms led to new experimental realizations , which raised a huge amount of interest and research in this field . for example one investigated the interference of bec clouds @xcite , studied rotating bec s @xcite , observed spinor condensates @xcite , where bec occurs in different hyperfine states , analyzed bose - fermi mixtures @xcite , where a pure bec is contaminated with fermions , described bec s in disordered potentials @xcite , dipolar bec s @xcite , and , more recently , tried to probe the properties of bec s in zero gravity @xcite . + of particular interest for the simulation of strongly correlated quantum many - body systems has been the realization of bec s trapped in optical lattices @xcite . since their experimental realization , optical - lattice systems have initiated intensive studies and led to a multitude of new applications such as entanglement of atoms @xcite , quantum teleportation @xcite , bell state experiments @xcite , disorder @xcite , and ultra - cold molecules @xcite , to name but a few . unfortunately , the experimental approaches discussed so far face some crucial limitations . on the one hand , it is necessary to cool down the considered system to some nano kelvin above absolute zero and , on the other hand , it is experimentally challenging to control and access single sites individually , however , recently developed experimental techniques also allow for single site addressability @xcite . nevertheless , these experiments still need ultra - cold temperatures , a restriction which could be circumvented by using cavity lattices . encouraged by the latest progress in the fabrication and manipulation of micro cavities @xcite , philippe grangier and others @xcite proposed a new experimental setup using cavity quantum electrodynamics ( qed ) schemes . the underlying idea behind this new approach is to build up a lattice from micro cavities and place some real or artificial atoms in each cavity , for example josephson junctions or quantum dots . subsequently , light is coupled into the system in such a way , that it interacts with the atoms . as a result , the coupling between the light field and the atoms leads to the formation of quasi - particles , so called polaritons . these quasi - particles behave like real bosonic particles on the lattice . in fact , bose - einstein condensation of polaritons was recently experimentally achieved in semi - conductor cavities filled with quantum wells @xcite and even superfluidity could be observed @xcite . this new idea for a quantum simulator based on cavity qed does not share the above mentioned limitations of the optical lattice approach . due to relatively huge distances between the cavities , local control and accessibility emerges quite naturally for these systems . hence , it is possible to analyze these systems without destroying them , in contrast to the time - of - flight imaging technique used for ultracold quantum gases . since the atoms are trapped inside the cavities right from the start and their thermal motion does not quantitatively disturb the polariton dynamics @xcite , bec experiments with cavity qed setups can thus be performed even at room temperatures @xcite . however , in order to facilitate stable experiments with this setup , one needs a strong coupling between light and matter in order to reduce the losses induced by the spontaneous emission . fortunately , over the past few years , this so called _ strong - coupling regime _ @xcite has become experimentally accessible for a large number of different setups @xcite . + however , a drawback of the strong coupling is the short polariton lifetime which prevents to reach thermal equilibrium . thus , in a real experiment the polaritons decay faster then they can equilibrate via phonon emission and polariton - polariton scattering and therefore the system needs to be continuously pumped by an external laser . recently , f. nissen et al . theoretically investigated the coherently driven and dissipative jchm @xcite . they could show that the important photon - blockade effect prevails in this system for weak hopping . this effect guarantees a strong polariton - polariton interaction which allows the system to reach a quasi thermal equilibrium @xcite . in principal one can modify the method presented in this paper to include external baths in order to account for this non - equilibrium situation . subsequently tracing out the bath degrees of freedom yields an effective theory for the reduced system . nevertheless , we will in this paper focus on the quasi equilibrium situation with a steady state polariton density in order to demonstrate the theoretical feasibility . this paper is structured as follows . in section @xmath0 we will introduce the underlying hamiltonian for the jaynes - cummings hubbard ( jch ) model and briefly analyze its dynamical properties in some special limits . in section @xmath1 we will derive an effective ginzburg - landau action for the model . to this end we will use the approach of refs . @xcite , that has already been successfully applied to analyze collective excitations of bose gases in optical lattices @xcite , and transfer it to a cavity qed system . in contrast to other methods , this procedure yields a consistent thermodynamic theory for finite temperatures in the whole quantum phase diagram . therefore our ginzburg - landau approach is of relevance in order to develop a thermometer for the jchm . for instance , to establish a thermometer for the usual bose - hubbard model turned out to be a difficult task @xcite . the respective results for both static and dynamic properties as well as their temperature dependence obtained from the effective action will be presented in section @xmath2 . in particular we focus on the excitation spectra and on the sound velocity of light and discuss their dependence on the experimentally accessible detuning parameter . for our lattice model we consider a hamiltonian of the form @xmath3 using the convention @xmath4 throughout this paper , the local part of this hamiltonian is given by @xmath5\ ] ] and consists of a sum over local jaynes - cummings hamiltonians in the _ rotating wave approximation _ ( rwa ) @xcite . here @xmath6 denotes the frequency of the local monochromatic photon fields associated with the annihilation and creation operators @xmath7 and @xmath8 . the parameter @xmath9 corresponds to the detuning between the local mode frequency @xmath6 and the energy splitting @xmath10 of each two - level system . the operators @xmath11 and @xmath12 are the ladder operators of the two - level systems and @xmath13 is the real coupling strength between each two - level system and the bosonic mode . the operator @xmath14 represents the on - site polariton number operator and @xmath15 is the total polariton number operator . as it commutes with @xmath16 , the number of these polaritons , which are coupled excitations of the two - level system and the local light fields , is a conserved quantity in this model . finally , working in the grand - canonical ensemble , yields an additional term in the hamiltonian which is proportional to the polariton number operator and the chemical potential @xmath17 . + the jaynes - cummings model is well known in the literature and the underlying hamiltonian can be exactly diagonalized within the rwa @xcite leading to the energy eigenvalues @xmath18,\;n>1 , \\e_{0}&=&0,\ ; n=0.\end{aligned}\ ] ] + here , we introduced the generalized rabi frequency @xmath19 and the effective chemical potential @xmath20 . the eigenvalues @xmath21 correspond to the energy eigenstates [ e : dressedstatesdefinition ] @xmath22 with the mixing angle @xmath23 . the vacuum state is given by @xmath24 . we note that the energy spectrum for a fixed polariton number @xmath25 naturally splits into an upper and a lower branch , where the lower branch is always lower in energy than the upper branch . + the possibility for photons to tunnel between next neighboring cavities is modeled by a hubbard - like hopping hamiltonian of the form @xmath26 where the sum runs over all next neighbor lattice sites . as the tunnel matrix elements @xmath27 exponentially decay with increasing distance between the lattice sites @xmath28 and @xmath29 , it is justified to assume that @xmath30 if @xmath28 and @xmath29 are next neighbors and @xmath31 otherwise . + as a first analysis of the ground - state of the model described by eq . we consider the physically relevant extremes of both the atomic and the hopping limit . in the atomic limit @xmath32 , eq . simplifies to @xmath33 which decomposes into purely local contributions with eigenvalues . obviously , in this regime the photons can not move in the lattice and , thus , all excitations are pinned to their respective lattice sites . therefore , the ground state wave function of the whole lattice is simply a direct product of the local on - site ground state wave functions . for this reason the ground state of the whole system is reached when each on - site jaynes - cummings system is in the lowest energy state @xmath34 or one of the lower - branch states @xmath35 . however , decreasing the difference @xmath36 one eventually reaches a point when @xmath37 and , hence , adding a polariton excitation becomes energetically favorable . the successive repetition of this argument leads to a complete set of such degeneracy points @xmath38 , which are characterized by the explicit relations : @xmath39,\;&n>1\\ \frac{\mu_{\rm{eff}}}{g}&=\frac{1}{2\,g}\left[\delta - r_{1}(\delta)\right],\;&n=0.\end{aligned}\ ] ] from the above discussion follows that , in the atomic limit , the local polariton number is fixed at each lattice site for a given set of parameters . when the total polariton number equals an integer multiple of the number of cavities , this regime is called the mott insulating phase which has been predicted @xcite and experimentally observed in the bose - hubbard model @xcite . + as a second interesting limit we investigate the ground - state in the regime when @xmath40 , i.e. where the photon hopping dominates the system dynamics . we additionally assume that all two - level systems are in their respective ground - state as the system minimizes the energy . hence , we can drop all atomic contributions , which leads to the hopping - limit hamiltonian @xmath41 this hamiltonian can be diagonalized in fourier space leading to @xmath42 with the energy dispersion @xmath43 , where @xmath44 is the lattice constant of a simple @xmath45-dimensional cubic lattice . thus , the hamiltonian of the hopping limit is local in fourier space . this situation corresponds to the superfluid phase of the system . signatures of this phase have already been observed in the interference patterns of time - of - flight experiments with bose - hubbard systems @xcite . in the following section we derive an effective action for the jch model from the free energy . explicitly calculating the lowest hopping order contributions of the effective action amounts effectively to a resummation of infinitely many hopping contributions . therefore , this effective action allows us to determine the quantum phase transition as well as calculate the excitation spectra , energy gap , effective mass , and sound velocity for finite temperatures in the mott phase and in the superfluid phase . following the approach used for example in refs . @xcite we additionally introduce source currents @xmath46 , in the system hamiltonian leading to the new hamiltonian @xmath47=\hat{h}+\sum_{i}\left[j_{i}^{*}(\tau)\,\hat{a}_{i}+j_{i}(\tau)\,\hat{a}_{i}^{\dagger}\right],\ ] ] which is now a functional of the currents . these artificial currents , which explicitly depend on the imaginary - time variable @xmath48 , will be used to artificially break the @xmath49 symmetry that is responsible for the quantum phase transition in the model @xcite . since all physical results are obtained in the limit of vanishing currents , we can treat all terms in the hamiltonian , which are proportional to the source currents , as small quantities . furthermore , for small hopping amplitudes @xmath50 all off - diagonal contributions in eq . become small . hence , we decompose the new system hamiltonian into the form @xmath51=\hat{h}_{0}+\hat{h}_{1}(\tau)\left[j , j^{*}\right],\ ] ] where the local part @xmath16 from eq . is exactly solvable and the remaining part @xmath52=\hat{h}_{\rm{h}}+\sum_{i}\left[j_{i}^{*}(\tau)\,\hat{a}_{i}+j_{i}(\tau)\,\hat{a}_{i}^{\dagger}\right]\ ] ] with the hopping hamiltonian @xmath53 from eq . can be treated as a perturbation . + we aim at establishing a thermodynamic perturbation theory in the present section . therefore , it is convenient to switch from the schrdinger picture to the imaginary - time dirac interaction picture . this leads to a reformulation of the partition function as a perturbation series , involving just the perturbative part of the full hamiltonian of the system . introducing the abbreviation @xmath54 with the inverse temperature @xmath55 for the thermal average with respect to the unperturbed system , the partition function takes on the following form @xmath56 here , the partition function of the unperturbed system is given by @xmath57 and the imaginary - time evolution operator in the dirac picture is defined as @xmath58\right\rbrace.\ ] ] note that all operators , which depend on imaginary - time variables , have to be taken in the imaginary - time dirac interaction picture , i.e. @xmath59 . + using the above definition together with eqs . and we see that the partition function also becomes a functional of @xmath60 and @xmath61 . splitting the grand - canonical partition functional into the respective perturbative contributions @xmath62=\mathcal{z}_{0}\left\lbrace 1+\sum_{n=1}^{\infty}\mathcal{z}_{n}\left[j , j^{*}\right]\right\rbrace , \ ] ] the free energy defined by @xmath63 can be written as @xmath64=\mathcal{f}_{0}-\frac{1}{\beta}\ln\left\{1+\sum_{n=1}^{\infty}\mathcal{z}_{n}\left[j , j^{*}\right]\right\}\,.\ ] ] here we introduced the free energy of the unperturbed system as the usual expression @xmath65 . in the next step we expand the free energy functional in a power series of the perturbation parameters @xmath66 . in this paper we will focus just on the lowest order contributions from the hopping and thus neglect all terms of higher than first order in @xmath50 . furthermore , according to the landau theory @xcite , one needs to consider all terms at least up to fourth order in the order parameter to describe the thermodynamic properties of a second order phase transition . since we will see later on that the source currents @xmath67 are of the order of the ginzburg - landau order parameter for the considered system , we thus have to calculate the power series up to fourth order in @xmath67 . hence , we expand the logarithm in expression and keep all terms up to fourth order in @xmath29 and @xmath68 and first order in @xmath69 . + this procedure leads to an expansion of the free energy functional in terms of imaginary time integrals over sums of products of thermal green functions with respect to the unperturbed system of increasing order . the @xmath70th order thermal green function with respect to the unperturbed system is defined as @xmath71\right\rangle _ { 0}\,.\end{aligned}\ ] ] in principle , one could now make use of the definition and calculate the expansion coefficients of the free energy straightforwardly . however , with increasing order of the thermal green function the calculation becomes more and more complex due to the increasing number of space- and time - index permutations . therefore , we use another approach to calculate the thermal green functions , which automatically takes care of the emerging problems . usually one would apply in field theory the wick theorem to decompose @xmath70-point correlation functions into sums of products of @xmath0-point correlation functions @xcite . unfortunately , this is not possible for the considered system , since the wick theorem just holds for systems , where the unperturbed hamiltonian is linear in the occupation number operator . instead , in our case one has to use the so called _ cumulant expansion _ , which was originally developed for the hubbard model as is reviewed by metzner @xcite . it states that the logarithm of the partition function is given by the sum of all connected green functions . the power of this approach lies in the fact that these connected green functions can subsequently be derived from a single generating functional by performing functional derivatives with respect to the currents . due to the fact that the unperturbed hamiltonian decomposes into a sum over local contributions , the generating functional decomposes into products of purely local cumulants : @xmath72=&\underset{i}{\prod}\ln\left\langle\hat{t}e^{-\underset{0}{\overset{\beta}{\int}}d\tau\left[j_{i}(\tau)\hat{a}_{i}^{\dagger}(\tau)+j_{i}^{*}(\tau)\hat{a}_{i}(\tau)\right]}\right\rangle _ { 0}.\end{aligned}\ ] ] the local cumulants then follow from @xmath73}{\delta j_{i_{1}^{\prime}}\left(\tau_{1}^{\prime}\right)\ldots\delta j_{i_{n}^{\prime}}\left(\tau_{n}^{\prime}\right)\delta j_{i_{1}}^{*}\left(\tau_{1}\right)\ldots\delta j_{i_{n}}^{*}\left(\tau_{n}\right)}\right|_{j = j^{*}=0}.\notag\end{aligned}\ ] ] due to the local structure of all cumulants vanish unless the site indexes are all equal which yields the relation @xmath74 hence , we just have to calculate the _ local _ cumulants @xmath75 . performing the calculations according to formula and rearranging the resulting terms yields the cumulant decomposition for each thermal green function . the lowest order green functions read @xmath76 with this cumulant decomposition we find the following expansion of the free energy functional @xmath77=\mathcal{f}_{0}-\frac{1}{\beta}\sum_{i , j}\int_{0}^ { \beta}d\tau_{1}\,\int_{0}^ { \beta}d\tau_{2}\,\bigg\ { \bigg[a_{2}^{(0)}\left(i;\tau_{1}|\tau_{2}\right)\delta_{ij}+a_{2}^{(1)}\left(i;\tau_{1}|\tau_{2};j\right)\bigg]\,j_{i}(\tau_{1})\,j_{j}^{*}(\tau_{2})\notag\\ % & + \frac{1}{4}\int_{0}^ { \beta}d\tau_{3}\,\int_{0}^ { \beta}d\tau_{4}\bigg[a_{4}^{(0)}\left(i;\tau_{1},\tau_{3}|\tau_{2},\tau_{4}\right)\delta_{ij}+2\,a_{4}^{(1)}\left(i;\tau_{1},\tau_{3}|\tau_{2},\tau_{4};j\right)\bigg]\,j_{j}(\tau_{1})\,j_{i}(\tau_{3})\,j_{i}^{*}(\tau_{2})\,j_{j}^{*}(\tau_{4})\notag\\ % & + \frac{1}{2}\int_{0}^ { \beta}d\tau_{3}\,\int_{0}^ { \beta}d\tau_{4}\,\tilde{a}_{4}^{(1)}\left(i;\tau_{1},\tau_{3}|\tau_{2},\tau_{4};j\right)\,j_{j}(\tau_{1})\,j_{i}(\tau_{3})\,j_{i}^{*}(\tau_{2})\,j_{j}^{*}(\tau_{4})\bigg\},\end{aligned}\ ] ] where the introduced expansion coefficients are defined as @xmath78 due to the locality of @xmath16 in eq . the cumulants @xmath79 do not depend on the site indexes @xmath80 . for this reason , we will drop the site index in the following calculations for convenience . finally , we notice that the form of the above coefficients @xmath81 , and @xmath82 can be further simplified by going into frequency space . therefore we perform the matsubara transformation @xmath83 with the matsubara frequencies @xmath84 at first , we calculate the coefficient @xmath85 in matsubara space . due to frequency conservation the following relation has to hold @xmath86 this coefficient can be derived from the expression with the help of relations and by performing a matsubara transformation . using the polariton mapping introduced in ref . @xcite to calculate the thermal expectation values , we obtain the following result : @xmath87 @xmath88\bigg\}.\label{e : a20inmatsubaraspace}\end{aligned}\ ] ] the coefficients @xmath89 in the above expression stem from the fact that there exist two kind of polariton species , where the lower branch is labeled by @xmath90 and the upper branch by @xmath91 . these coefficients are defined as [ e : polaritonmappingtransitionamplitudes ] @xmath92 with mixing angle dependent amplitudes given by @xmath93 with the help of this result we can also determine the higher hopping corrections . by using frequency conservation again we find the relation @xmath94 and @xmath95+\int_{0}^{\beta}d\tau_{1}\ldots d\tau_{4}\notag\\ % & \times\,\left\langle \hat{t}\left[\hat{a}^{\dagger}(\tau_{1})\,\hat{a}^{\dagger}(\tau_{3})\,\hat{a}(\tau_{2})\,\hat{a}(\tau_{4})\right]\right\rangle _ { 0}\notag\\ % & \left.\times\,e^{i\left(-\omega_{\rm{m1}}\tau_{1}+\omega_{\rm{m2}}\tau_{2}-\omega_{\rm{m3}}\tau_{3}+\omega_{\rm{m4}}\tau_{4}\right)}\right\rbrace.\end{aligned}\ ] ] the calculation of the latter expression is complicated and rather lengthy . therefore , we put a detailed calculation of this quantity in the appendix . nevertheless , from general considerations like frequency conservation and integral properties in matsubara space , we can deduce right away that the first order hopping correction is of the form @xmath96 within this section we finally derive the ginzburg - landau action for the jaynes - cummings - hubbard model , which is the proper thermodynamic potential to describe the quantum phase transition of this system . since the symmetry - breaking currents @xmath67 are no physical quantities , one has to transform them into physical fields in terms of the order parameter . this is accomplished by means of a legendre transformation of the free - energy functional to an effective action as has already been explored in detail in the context of the thermal phase transitions @xcite . in order to do this in a concise way , we first rewrite the grand - canonical free energy in the following form in matsubara space @xmath97=\mathcal{f}_{0}-\frac{1}{\beta}\sum_{i , j}\sum_{\omega_{\rm{m1}},\omega_{\rm{m2}}}\left\ { m_{ij}\left(\omega_{\rm{m1}},\omega_{\rm{m2}}\right)j_{i}\left(\omega_{\rm{m1}}\right)\right.\notag\\ % & \times\ ; j_{j}^{*}\left(\omega_{\rm{m2}}\right)+\sum_{k , l}\sum_{\omega_{\rm{m3}},\omega_{\rm{m4}}}n_{ijkl}\left(\omega_{\rm{m1}},\omega_{\rm{m2}},\omega_{\rm{m3}},\omega_{\rm{m4}}\right)\notag\\ % & \left.\times\ ; j_{i}(\omega_{\rm{m1}})j_{j}(\omega_{\rm{m3}})j_{k}^{*}(\omega_{\rm{m2}})j_{l}^{*}(\omega_{\rm{m4}})\right\}+\ldots\,,\end{aligned}\ ] ] where we have introduced the abbreviations @xmath98\delta_{\omega_{\rm{m1}},\omega_{\rm{m2}}}\,,\end{aligned}\ ] ] and @xmath99\delta_{\omega_{\rm{m1}}+\omega_{\rm{m3}},\omega_{\rm{m2}}+\omega_{\rm{m4}}}\,.\end{aligned}\ ] ] now we define the legendre transformation by self - consistently introducing the order parameter field @xmath100 according to @xmath101 note , that this ginzburg - landau order parameter field differs from the landau order parameter by being space _ and _ time dependent . inserting expression into equation yields the following relation for the ginzburg - landau order parameter field @xmath102 furthermore , relation motivates to introduce the effective action @xmath103=&\mathcal{f}\left[j , j^{*}\right]-\frac{1}{\beta}\sum_{i,\omega_{\rm{m}}}\left[\psi_{i}\left(\omega_{\rm{m}}\right)j_{i}^{*}\left(\omega_{\rm{m}}\right)\right.\notag\\ % & \left . + \;\psi_{i}^{*}\left(\omega_{\rm{m}}\right)j_{i}(\omega_{\rm{m}})\right]\,,\end{aligned}\ ] ] where @xmath104 and @xmath29 are conjugate variables satisfying the legendre relations @xmath105 using the fact , that physical situations correspond to vanishing currents @xmath106 , yields the following _ equations of motion _ @xmath107 thus , the effective action is stationary with respect to the order parameter field . now , in order to determine the explicit form of the effective action , we need to express all symmetry - breaking currents @xmath29 by the ginzburg - landau order parameter field @xmath104 . therefore , we recursively invert relation up to first order in the hopping strength @xmath69 , which yields @xmath108+\ldots\,,\end{aligned}\ ] ] where we define the abbreviations @xmath109 and @xmath110.\ ] ] inserting the above relations into the legendre transformation gives the explicit expression for the effective action up to the desired order in the hopping strength @xmath111= \mathcal{f}_{0}+\frac{1}{\beta}\sum_{i}\sum_{\omega_{\rm{m1}}}\left\ { \frac{\left|\psi_{i}\left(\omega_{\rm{m1}}\right)\right|^{2}}{a_{2}^{(0)}\left(\omega_{\rm{m1}}\right)}\right.\notag\\ % & -\sum_{j}\kappa_{ij}\,\psi_{j}\left(\omega_{\rm{m1}}\right)\psi_{i}^{*}\left(\omega_{\rm{m1}}\right)-\sum_{\omega_{\rm{m2}},\omega_{\rm{m3}},\atop\omega_{\rm{m4}}}\psi_{i}\left(\omega_{\rm{m1}}\right)\psi_{i}^{*}\left(\omega_{\rm{m2}}\right)\notag\\ % & \times\left.\frac{a_{4}^{(0)}\left(\omega_{\rm{m1}},\omega_{\rm{m3}}|\omega_{\rm{m2}},\omega_{\rm{m4}}\right)\psi_{i}\left(\omega_{\rm{m3}}\right)\psi_{i}^{*}\left(\omega_{\rm{m4}}\right)}{4\,a_{2}^{(0)}\left(\omega_{\rm{m1}}\right)a_{2}^{(0)}\left(\omega_{\rm{m2}}\right)a_{2}^{(0)}\left(\omega_{\rm{m3}}\right)a_{2}^{(0)}\left(\omega_{\rm{m4}}\right)}\right\}\notag\\ % & + \,\ldots.\;. % \end{aligned}\ ] ] equation is the thermodynamic potential for the jaynes - cummings hubbard model up to the desired accuracy in both the order parameter and the hopping parameter . however , as should be clear from the approximations performed so far , this expression can , in principle , be extended to include higher order corrections . in the next section this result is used to analyze several properties of the considered system . having derived the effective action in the previous chapter , we now use this result to extract both thermodynamic and dynamic properties of the jaynes - cummings - hubbard model . the starting point for this analysis is the equation of motion . inserting yields @xmath112\psi_{i}\left(\omega_{\rm{m}}\right)-\sum_{\omega_{\rm{m1}},\omega_{\rm{m2}},\omega_{\rm{m3}}}\\ % \times&\frac{a_{4}^{(0)}\left(\omega_{\rm{m1}},\omega_{\rm{m3}}|\omega_{\rm{m2}},\omega_{\rm{m}}\right)\psi_{i}\left(\omega_{\rm{m1}}\right)\,\psi_{i}\left(\omega_{\rm{m3}}\right)\,\psi_{i}^{*}\left(\omega_{\rm{m2}}\right)}{2\ , a_{2}^{(0)}\left(\omega_{\rm{m1}}\right)a_{2}^{(0)}\left(\omega_{\rm{m2}}\right)a_{2}^{(0)}\left(\omega_{\rm{m3}}\right)a_{2}^{(0)}\left(\omega_{\rm{m}}\right)}.\notag\end{aligned}\ ] ] using a particular ansatz for the order parameter field allows to examine both static and dynamic order parameter fields @xmath113 . first , we consider an equilibrium situation , where the order parameter field is constant in both space and time @xmath114 inserting this ansatz in the stationarity condition yields the following relation for the equilibrium order parameter @xmath115^{3}}{a_{4}^{(0)}\left(0,0|0,0\right)}\left[1-a_{2}^{(0)}\left(0\right)\,\kappa\ , z\right]\,,\ ] ] where @xmath116 represents the coordination number of the @xmath45-dimensional cubic lattice . since the order parameter field is zero in the mott insulator phase and takes on finite values in the superfluid regime , we can extract the quantum phase boundary from the condition that the equilibrium order parameter has to vanish . thus the quantum phase boundary is defined via the relation @xmath117 together with the result this equation yields a phase diagram which is pictured in fig . [ f : quantumphaseboundary ] for vanishing detuning @xmath118 . here we plot the effective hopping strength versus the effective chemical potential leading to a lobe structure where each lobe is associated to a specific mean on - site polariton number . the regions within these lobes correspond to the mott insulator phase , whereas the exterior region corresponds to the superfluid regime . first we note that the phase boundary for zero temperature is consistent with the results from refs . moreover , our ginzburg - landau theory also yields the phase boundary for finite temperatures . from fig . [ f : quantumphaseboundary ] we can see that increasing the temperature leads to a smeared out phase boundary . thermal fluctuations mostly affect the region between two neighboring mott lobes , whereas the middle of each mott lobe almost does not change . this effect is stronger for lobes with higher polariton number as these system configurations are rather unstable . . interior of the lobes for @xmath119 corresponds to the mott insulator phase whereas the exterior corresponds to the superfluid phase . for finite temperatures the mott insulator vanishes and becomes a mixture of normal and mott phase . note that the drop of the phase boundary at @xmath120 is a numerical remnant resulting from a summation cutoff in eq . . a full summation would lead to an infinite sequence of mott lobes approaching @xmath121.,scaledwidth=48.0% ] since a genuine mott insulator is defined by a vanishing compressibility @xmath122 we find that the mott insulator phase is only present for zero temperature . for finite temperature it becomes a mixture of normal and mott insulating phase instead @xcite . note that , in order to validate our results obtained so far , we follow ref . @xcite and explicitly compare our expressions for @xmath123 with a corresponding mean - field calculation . we get back the mean - field free energy @xmath124 by formally identifying @xmath125 , @xmath126 and @xmath127 . with this we reproduce the mean - field results of refs . @xcite including the dependence of the quantum phase boundary on the detuning parameter @xmath128 . if the system is tuned out of resonance , i.e. @xmath129 , all mott lobes with higher mean particle number than one shrink in size and are shifted to smaller values of the effective chemical potential @xmath130 irrespective of the sign of the detuning . the mott - lobe with mean particle number @xmath131 grows in size for negative detuning and shrinks for positive detuning , whereas the mott - lobe with mean particle number @xmath132 covers the rest of the phase diagram and , thus , is the only mott insulator region that does not form a closed lobe . this special behavior of the first two mott regions stems from the composite nature of the polaritons and has no analogue in the bose - hubbard model . within this section we analyze the dynamic behavior of the jch model . we especially focus on signatures within excitation spectra such as energy gaps , effective masses and the sound velocity of polaritons . these system properties are experimentally accessible via emission and transmission spectroscopy @xcite.in order to derive these properties we investigate the dynamic behavior of the effective action around the equilibrium fields . thus , introducing the vector @xmath133 we insert the ansatz @xmath134 in the equation of motion and its conjugate complex . this yields a system of two coupled equations corresponding to the equations of motion for the elongations @xmath135 around the order field . in order to find a non - trivial solution of the equation of motion , we obtain the relation @xmath136_{\mathbf{\psi}=\mathbf{\psi}_{\rm{eq}}}\notag.\end{aligned}\ ] ] inserting eq . we see that , due to the effective hopping amplitude @xmath50 , the second partial derivatives of the action still depend on the site distance @xmath137 . this suggests to further apply a spatial fourier transformation in order to simplify the calculations . additionally , switching from matsubara frequencies to continuous frequencies within a wick rotation , yields the following explicit expression for eq . @xmath138 with the abbreviations @xmath139 ^ 2 \left [ a_2^{(0)}(\mathbf{k},0)\right]^2},\end{aligned}\ ] ] and @xmath140 ^ 2 a_2^{(0)}(\mathbf{k},\omega)a_2^{(0)}(\mathbf{k},-\omega)}.\end{aligned}\ ] ] assuming a simple three - dimensional cubic lattice with lattice constant @xmath44 yields @xmath141 . taking a closer look at eq . we see that it implicitly defines the dispersion relation @xmath142 . however , due to the complex expression found for @xmath143 , a full evaluation of the above equations can only be done numerically . just in the mott insulator regime , where the @xmath2-point correlation @xmath144 drops out due to @xmath145 , it is possible to derive analytic expressions for the dispersions . in this special case we have to solve @xmath146 leading to the dispersion relations @xmath147\right\ } ^{2}\right.\notag\\ % & -4\bigl\ { j(\mathbf{k})\ , e_{(n+1)-}t_{n--}^{2}-e_{n-}^{2}+e_{(n-1)-}\notag\\ % & \times\left(e_{n-}-e_{(n+1)-}+j(\mathbf{k})\ , t_{(n+1)--}^{2}\right)+e_{n-}\bigl[e_{(n+1)-}\notag\\ % & \left.\left .- j(\mathbf{k})\left(t_{n--}^{2}+t_{(n+1)--}^{2}\right)\bigr]\bigr\ } ^{\frac{1}{2}}\right)\right].\end{aligned}\ ] ] at @xmath148 with zero detuning @xmath118 and @xmath149 ( solid ) , @xmath150 ( dashed ) and @xmath151 ( dotted ) . the right picture represents the effective mass of both the hole ( lower branch / light ) and particle ( upper branch / dark ) excitations . ] in order to clarify the physical meaning of equation we set @xmath152 , which yields the simple relations @xmath153 from these relations it is clear that @xmath154 ( @xmath155 ) is the energy needed to remove ( add ) a lower - branch polariton from a lattice site , which is occupied by @xmath70 lower - branch polaritons . therefore , we refer to these energies as the lower - branch - hole ( @xmath156 ) and lower - branch - particle excitation ( @xmath157 ) , respectively . the same analysis can be applied to the full expression leading to additional dispersion relation for upper - branch excitations @xmath158 and @xmath159 as well as mixed excitations @xmath160 , @xmath161 , @xmath162 and @xmath163 . however , these additional excitation channels are hardly of any interest for our considerations because they occur at much higher energies compared to the lower - branch polariton excitations . the lowest dispersion relation involving upper polariton states is @xmath160 , which lies at an energy around @xmath164 for @xmath165 . therefore , we will just focus in all further calculations on the lower - branch excitations as they completely determine the low - temperature physics of the system . in order to extract more detailed quantities of interest , we subsequently expand the dispersion relations as follows @xmath166 from this expansion we are able to derive , within the mott insulator phase , the energy gap @xmath167 and the effective mass of the particle and hole excitations @xmath168 , respectively . as an example we evaluated their temperature dependence for the first mott lobe @xmath169 with @xmath170 and zero detuning in fig . [ f : subgapt ] . we find that , both the pair excitation energy gap as well as the effective mass of particle and hole excitations increase with higher temperatures . this effect becomes stronger as one approaches the critical hopping strength at the tip of the lobe . + moreover , we expect to find in the superfluid regime , apart from the gapped mode , also a linear excitation mode which is associated with the broken symmetry in the superfluid regime according to the nambu - goldstone theorem @xcite . the dispersion relation for this mode reads @xmath171 the results obtained from a numerical evaluation of the above formulas is presented in fig . [ f : dynamicresults ] for a mean particle density @xmath165 at zero temperature and vanishing detuning @xmath118 . the pictures are arranged in a table in such a way that each line represents a specific physical quantity , for instance the second line fig . [ f : subgap ] shows plots for the energy gap . furthermore , the plots in each column correspond to a fixed effective chemical potential . the left column shows plots for an effective chemical potential below the critical one , the middle column represents the critical effective chemical potential , and the right column corresponds to an effective chemical potential above the critical one . in the first line in fig . [ f : subspec ] we show the particle and hole excitation spectra in @xmath172 direction . in the mott phase and at the phase boundary we always observe two excitation modes corresponding to the particle ( dotted ) and hole ( dashed ) excitations , respectively . these modes are always gapped in the mott - insulator phase . by approaching the phase boundary at least one gap vanishes and thus for @xmath173 the particle mode becomes gapless at the phase border , whereas for @xmath174 the hole mode becomes gapless at the phase border . when approaching the phase boundary exactly at the lobe tip both particle and hole modes become gapless . going further into the superfluid regime we find a gapped excitation mode as well as the anticipated gapless linear mode . in fig . [ f : submass ] we plot the corresponding effective masses . in the mott phase we observe the masses of both the particle ( dotted ) and hole ( dashed ) excitations , whereas in the superfluid phase also only one massive mode survives . additionally , we depict the sound velocity of the polariton excitations in the superfluid phase in figs . [ f : subsoundk ] and [ f : subsoundd ] . figure [ f : subsoundk ] shows the dependence of the sound velocity on the hopping strength . we find that it approaches a finite value at the tip of the mott lobe , but vanishes at all other points of the mott lobe . this behaviour shows that the jch model has a dynamical critical exponent of @xmath175 which has been recently confirmed in a large - scale quantum monte - carlo simulation by m. hohenadler et al . @xcite . entering the superfluid phase the sound velocity increases steadily . however , as pictured in fig . [ f : subsoundd ] , if the system is tuned out of resonance , i.e. @xmath129 , the sound velocity drops significantly . finally , we note that our results figs . [ f : dynamicresults ] and [ f : subsoundd ] are in good qualitative agreement with the results from ref . however , due to the restriction to the lowest hopping order in the effective action , our results lose validity deep in the superfluid phase . for this reason we do not obtain a shift of the maximum of the sound velocity in fig . [ f : subsoundd ] , as is observed in ref . @xcite . in order to obtain better results in the superfluid regime higher hopping corrections must be considered . corresponding perturbative approaches for higher order corrections for the jch model have been numerically calculated in ref . in summary , we successfully applied the cumulant expansion approach from ref . @xcite to derive a ginzburg - landau theory for the jaynes - cummings - hubbard model up to fourth order in the symmetry - breaking currents and up to first order in the hopping strength . from the resulting effective action we extracted the phase diagram of the inherent quantum phase transition of the jch model for finite temperature . in the case of vanishing temperature our result is in accordance with the results found from mean - field calculations @xcite . subsequently , we derived the excitation spectra , energy gaps , and effective masses of the lower - branch polariton - excitations in the mott insulator phase as well as in the superfluid phase . we investigated the temperature dependence of both the pair excitation energy gap and the effective mass of the particle and hole excitations in the mott phase . furthermore , we analyzed how the sound velocity in the superfluid phase depends at zero temperature on the hopping parameter and the detuning parameter . finally , we point out that the the ginzburg - landau approach of this paper can be generalized to describe the real - time dynamics of the jchm . this has already been shown for the real - time dynamics of the bose - hubbard model in refs . + we thank n. g. berloff , c. ciuti , and s. schmidt for useful discussions and especially m. hayn for valuable suggestions . here we evaluate the fourth order coefficient , which involves the expectation values of time ordered products of four operators . first we notice that , for the time - ordered product of two annihilation and two creation operators , there are @xmath176 distinct permutations leading to different expectation values . each of these orderings itself has @xmath2 time variable permutations corresponding to @xmath177 and @xmath178 . thus , overall one finds @xmath179 terms for the expectation value . luckily , the integrals over different time - variable permutations yield the same result , and thus , they just lead to a fixed pre - factor @xmath2 . for this reason , one just needs to determine the @xmath176 different thermal averages for one specific time - ordering . furthermore , these expectation values are local quantities and , therefore , we drop the site indexes in the following calculations . thus , one has to determine the following expressions : @xmath180 with the help of the polariton mapping introduced in ref . @xcite , one can calculate these averages straightforwardly . hence , we find for example the following expression for the expectation value @xmath181 subsequently , performing a matsubara transformation according to yields a formal integral of the form @xmath182 with the solution @xmath183.\end{aligned}\ ] ] the variables @xmath184 correspond to differences of energy eigenvalues . due to energy conservation these variables have to fulfill the condition @xmath185 and thus the above solution has a pole in the first term . therefore , we have to take the limit @xmath186 since this pole arises for all expectation values we always have to consider this particular limit . taking this result into account , the explicit expression for the expectation value in matsubara space is given by however , this expression still possesses some poles for special choices of matsubara frequencies . fortunately , all these poles can be eliminated by investigating the corresponding limits analogous to equation . further care has to be taken considering the occurrence of the ground - state energy due to its uniqueness . similar expressions can be calculated for the other expectation values . 3 a. kay and j. k. pachos , _ new j. phys . _ * 6 * , 126 , ( 2004 ) p. treutlein , t. steinmetz , y. colombe , b. l. lev , p. hommelhoff et al . , _ fortschr . phys . _ * 54 * , 702 ( 2006 ) r. p. feynman , _ int . * 21 * , 467 ( 1982 ) h. s. j. van der zant , f. c. fritschy , w. j. elion , l. j. geerligs , and j. e. mooij , _ phys . _ * 69 * , 2971 ( 1992 ) a. van oudenaarden and j. e. mooij , _ phys lett . _ * 76 * , 4947 ( 1996 ) m. p. a. fisher , p. b. weichman , g. grinstein , and d. s. fisher , _ phys . b _ * 40 * , 546 ( 1989 ) m. naraschewski , h. wallis , a. schenzle , j. i. cirac , and p. zoller , _ phys a _ * 54 * , 2185 ( 1996 ) m. r. andrews , _ science _ * 275 * , 637 ( 1997 ) k. w. madison , f. chevy , w. wohlleben , and j. dalibard , _ phys . * 84 * , 806 ( 2000 ) j. r. abo - shaeer , c. raman , j. m. vogels , and w. ketterle , _ science _ * 292 * , 476 ( 2001 ) s. kling and a. pelster , _ phys . a _ * 76 * , 023609 ( 2007 ) s. kling and a. pelster , _ laser physics _ * 19 * , 1072 ( 2009 ) d. m. stamper - kurn , m. r. andrews , a. p. chikkatur , s. inouye , h .- j . miesner , j. stenger , and w. ketterle , _ phys . lett . _ * 80 * , 2027 ( 1998 ) f. schreck , l. khaykovich , k. corwin , g. ferrari , t. bourdel et al . lett . _ * 87 * , 080403 ( 2001 ) s. rthel and a. pelster , _ eur . j. b _ * 59 * , 343 ( 2007 ) j. billy , v. josse , z. zuo , a. bernard , b. hambrecht et al . _ nature _ * 453 * , 891 ( 2008 ) g. roati , c. derrico , l. fallani , m. fattori , c.fort et al . _ nature _ * 453 * , 895 ( 2008 ) r. graham and a. pelster , _ int . j. bif . chaos _ * 19 * , 2745 ( 2009 ) a. griesmaier , j. werner , s. hensler , j. stuhler , and t. pfau , _ phys . rev . lett . _ * 94 * , 160401 ( 2005 ) a. r. p. lima and a. pelster , _ phys . a _ * 81 * , 021606 ( 2010 ) t. van zoest , n. gaaloul , y. singh , h. ahlers , w. herr et al . , _ science _ * 328 * , 5985 ( 2010 ) d. jaksch , c. bruder , j.i . cirac , c. w. gardiner , and p. zoller , _ phys . rev . lett . _ * 81 * , 3108 ( 1998 ) m. greiner , o. mandel , t. esslinger , t. w. hnsch , and i. bloch , _ nature _ * 415 * , 6867 ( 2002 ) d. jaksch , h .- j . briegel , j. i. cirac , c. w. gardiner , and p. zoller , _ phys . lett . _ * 82 * , 1975 ( 1999 ) o. mandel , m. greiner , a. widera , t. rom , t. w. hnsch et al . , _ nature _ * 425 * , 6961 ( 2008 ) m. riebe , h. hffner , c. f. roos , w. hnsel , j. benhelm et al . , _ nature _ * 429 * , 6993 ( 2004 ) c. roos , g. lancaster , m. riebe , h. hffner , w. hnsel et al . lett . _ * 92 * , 220402 ( 2004 ) l. sanchez - palencia and l. santos , _ phys a _ * 72 * , 053607 ( 2005 ) v. ahufinger , l. sanchez - palencia , a. kantian , a. sanpera , and m. lewenstein , _ phys . a _ * 72 * , 063616 ( 2005 ) d. clment , a. varn , m. hugbart , j. retter , p. bouyer et al . , _ phys . rev * 95 * , 170409 ( 2005 ) k. v. krutitsky , a. pelster , and r. graham , _ new j. phys . _ * 8 * , 187 ( 2006 ) v. a. yurovsky , a. ben - reuven , p.s . julienne , and c.j . williams , _ phys a _ * 62 * , 043605 ( 2000 ) s. j. j. m. f. kokkelmans , h. m. j. vissers , and b. j. verhaar , _ phys . a _ * 63 * , 031601 ( 2001 ) p. wrtz , t. langen , t. gericke , a. koglbauer , and h. ott , _ phys lett . _ * 103 * , 080404 ( 2009 ) n. gemelke , x. zhang , c .- l . hung , and c. chin , _ nature _ * 460 * , 995 ( 2009 ) j. f. sherson , c. weitenberg , m. endres , m.cheneau , i. bloch et al . , _ nature _ * 467 * , 68 ( 2010 ) w. s. bakr , a. peng , m. e. tai , r. ma , j. simon et al . , _ science _ * 329 * , 547 ( 2010 ) k. j. vahala , _ nature _ * 424 * , 6950 ( 2003 ) g. khitrova , h. m. gibbs , m. kira , s. w. koch , and a. scherer , _ nature physics _ * 2 * , 81 ( 2006 ) t. m. benson , s. v. boriskina , p. sewell , a. vukovic , s. c. greedy et al . , _ frontiers of planar lightwave circuit technology _ , pp . 39 ( springer , berlin , 2005 ) p. grangier , g. reymond , and n. schlosser , _ fortschr . phys . _ * 48 * , 859 ( 2000 ) m. bayindir , b. temelkuran , and e. ozbay , _ phys . b _ * 61 * , 11855 ( 2000 ) m. j. hartmann , f. g. s. l. brando , and m. b. plenio , _ nature physics _ * 2 * , 849 ( 2006 ) a. d. greentree , c. tahan , j. h. cole , and l. c. l. hollenberg , _ nature physics _ * 2 * , 856 ( 2006 ) d. g. angelakis , m. f. santos , and s. bose , _ phys a _ * 76 * , 031805 ( 2007 ) m. bajcsy , s. hofferberth , v. balic , t. peyronel , m. hafezi et al . , _ phys . _ * 102 * , 203902 ( 2009 ) h. deng , g. weihs , c. santori , j. bloch , and y. yamamoto , _ science _ * 298 * , 199 ( 2002 ) j. kasprzak , m. richard , s. kundermann , a. baas , p. jeambrun et al . , _ nature _ * 443 * , 7110 ( 2006 ) r. balili , v. hartwell , d. snoke , l. pfeiffer , and k. west , _ science _ * 316 * , 1007 ( 2007 ) a. amo , d. sanvitto , f. p. laussy , d. ballarini , e. del valle et al . , _ nature _ * 457 * , 291 ( 2009 ) a. amo , j. lefrre , s. pigeon , c. adrados , c. ciuti et al . , _ nature physics _ * 5 * , 805 ( 2009 ) a. z. muradyan and g. a. muradyan , _ j. phys . b _ * 35 * , 3995 ( 2002 ) j .- q . liao , z. r. gong , l. zhou , y .- x . liu , c. p. sun et al . a _ * 81 * , 042304 ( 2010 ) t. carmon , t. j. kippenberg , l. yang , h. rokhsari , s. spillane et al . , _ optics express _ * 13 * , 3558 ( 2005 ) f. p. laussy , e. del valle , a. gonzalez - tudela , e. cancellieri , d. sanvitto et al . , _ `` nanostructures : physics and technology '' _ * 298 * ( 2010 ) s. m. spillane , t. j. kippenberg , o. j. painter , and k. j. vahala , _ phys . rev . lett . _ * 91 * , 043902 ( 2003 ) a. wallraff , d. i. schuster , a. blais , l. frunzio , r. s. huang et al . , _ nature _ * 431 * , 162 ( 2004 ) t. j. kippenberg , s. m. spillane , d. k. armani , and k. j. vahala , _ optics letters _ * 29 * , 1224 ( 2004 ) m. trupke , j. goldwin , b. darquie , g. dutier , s. eriksson , j. ashmore , and e. a. hinds , _ phys . lett . _ * 99 * , 63601 ( 2007 ) m. bajcsy , s. hofferberth , v. balic , t. peyronel , m. hafezi et al . lett . _ * 102 * , 203902 ( 2009 ) f. s. f. brossard , x. l. xu , d. a. williams , m. hadjipanayi , m. hopkinson et al . lett . _ * 97 * , 111101 ( 2010 ) a. laucht , j. m. villas - bas , s. stobbe , n. hauke , f. hofbauer et al . b _ * 82 * , 075305 ( 2010 ) g. ctistis , a. hartsuiker , e. van der pol , j. claudon , w. l. vos et al . b _ * 82 * , 195330 ( 2010 ) f. nissen , s. schmidt , m. biondi , g. blatter , h. e. treci , and j. keeling , arxiv:1202.1961 b. bradlyn and f. e. a. dos santos , _ phys . rev . a _ * 79 * , 013615 ( 2009 ) f. e. a. dos santos and a. pelster , _ phys . a _ * 79 * , 013614 ( 2009 ) t. d. grass , f. e. a. dos santos , and a. pelster , _ laser physics _ * 21 * , 1459 ( 2011 ) t. d. grass , f. e. a. dos santos , and a. pelster , _ phys . rev . a _ * 84 * , 013613 ( 2011 ) a. hoffmann and a. pelster , _ phys . rev . a _ * 79 * , 053623 ( 2009 ) d. mckay and b. demarco , _ rep . prog . phys . _ * 74 * , 054401 ( 2011 ) d. f. walls and g. j. milburn , _ quantum optics _ ( springer , berlin , 1995 ) e. t. jaynes and f. w. cummings , _ p. ieee _ * 51 * , 89 ( 1963 ) s. trotzky , l. pollet , f. gerbier , u. schnorrberger , i. bloch et al . , _ nature physics _ * 6 * , 998 ( 2010 ) f. gerbier , s. trotzky , s. foelling , u. schnorrberger , j. d. thompson et al . , _ phys . * 101 * , 155303 ( 2008 ) s. sachdev , _ quantum phase transitions _ ( cambridge university press , cambridge , 1999 ) l. d. landau and e. m. lifschitz , _ course of theoretical physics , volume 5 _ ( akademie verlag , berlin , 1970 ) j. w. negele and h. orland , _ quantum many - particle systems _ ( westview press , boulder , 1998 ) w. metzner , _ phys . rev . b _ * 43 * , 8549 ( 1991 ) j. koch and k. le hur , _ phys . rev . a _ * 80 * , 023811 ( 2009 ) h. kleinert and v. schulte - frohlinde , _ critical properties of @xmath188 theories _ ( world scientific publishing company , singapore , 2001 ) j. zinn - justin , _ quantum field theory and critical phenomena _ ( oxford university press , oxford , 2002 ) s. schmidt and g. blatter , _ phys . lett . _ * 103 * , 086403 ( 2009 ) p. buonsante and a. vezzani , _ phys . a _ * 70 * , 033608 ( 2004 ) i. carusotto , d. gerace , h. e. tureci , s. de liberato , c. ciuti et al . lett . _ * 103 * , 033601 ( 2009 ) j. goldstone , _ phys . rev . _ * 127 * , 965 ( 1962 ) m. hohenadler , m. aichhorn , s. schmidt , and l. pollet , _ phys . a _ * 84 * , 041608 ( 2011 ) s. schmidt and g. blatter , _ phys . lett . _ * 104 * , 216402 ( 2010 ) c. hei and w. von der linden , arxiv:1105.2418

residuals in regression models are often spatially correlated . prominent examples include studies in environmental epidemiology to understand the chronic health effects of pollutants . i consider the effects of residual spatial structure on the bias and precision of regression coefficients , developing a simple framework in which to understand the key issues and derive informative analytic results . when unmeasured confounding introduces spatial structure into the residuals , regression models with spatial random effects and closely - related models such as kriging and penalized splines are biased , even when the residual variance components are known . analytic and simulation results show how the bias depends on the spatial scales of the covariate and the residual : one can reduce bias by fitting a spatial model only when there is variation in the covariate at a scale smaller than the scale of the unmeasured confounding . i also discuss how the scales of the residual and the covariate affect efficiency and uncertainty estimation when the residuals are independent of the covariate . in an application on the association between black carbon particulate matter air pollution and birth weight , controlling for large - scale spatial variation appears to reduce bias from unmeasured confounders , while increasing uncertainty in the estimated pollution effect . .

1011.1139

spatial confounding is likely present in many of the applied contexts in which residuals are spatially correlated , particularly in public health and social science . consider the motivating example of thehealth effects of exposure to ( spatially varying ) air pollution , an important public health issue . many variables that explain variability in the response , including potential confounding variables that may be correlated with exposure , also vary spatially . for example , large - scale regional patterns in air pollution may be correlated with regional patterns in diet , income and other risk factors for a health outcome of interest . small - scale patterns in air pollution from local sources may be correlated with risk factors as well , for example , if lower - income people live nearer to busy roads or industrial sources . if confounding variables are not measured , it will be difficult to distinguish the effect of air pollution from residual spatial variation in the health outcome . i use the term spatial confounding to characterize this situation . researchers have modeled the spatial structure in the outcome with the apparent goal of reducing confounding bias ( e.g. , clayton , bernardinelli and montomoli , @xcite ; @xcite , @xcite ; @xcite , @xcite ; @xcite , @xcite ) . however , the statistical mechanism for reducing bias does not appear to be well understood nor investigated rigorously in the statistical or applied literature . to consider the problem formally , start with simple linear regression with spatial structure : @xmath0\\[-8pt ] { \mathbf{e } } & \sim & { \mathcal{n}}({\mathbf{0}},\mathit{\bolds{\sigma}}),\nonumber\end{aligned}\ ] ] where each outcome , @xmath1 , is associated with a spatial location , @xmath2 . @xmath3 is the corresponding value of a univariate regressor of interest , which may also vary spatially , in which case we would represent @xmath3 as @xmath4 . @xmath5 is the vector of errors , whose covariance matrix , @xmath6 , captures any residual spatial correlation , as well as independent variation . the regression coefficients , @xmath7 , are unknown , and estimation of @xmath8 is of primary interest . spatial statistics and regression texts note that the ordinary least squares ( ols ) estimator for @xmath8 in this setting is unbiased but inefficient , and the usual ols variance estimator is incorrect . assuming known @xmath6 , the generalized least squares ( gls ) estimator is the most efficient estimator . however , little appears to be known about how the spatial scales of the residual variability and of @xmath9 affect inference . spatial structure in @xmath9 is very common in applications and complicates the problem because @xmath9 and the residual spatial structure compete to explain the variability in the response @xcite . furthermore , it would not be surprising if the spatial correlation in the residuals were caused by an unmeasured spatially varying confounder ; i next introduce another representation of ( [ eq : basic ] ) to enable exploration of confounding . motivated by the air pollution example , i will refer to @xmath9 as the `` exposure . '' one can obtain the basic spatial regression model ( [ eq : basic ] ) using a simple mixed model , @xmath10 with random effects , @xmath11 , andwhite noise errors , @xmath12 . suppose the random effects are spatially correlated , with @xmath13 , where @xmath14 is a spatial correlation matrix parameterized by @xmath15 , a spatial range parameter , and @xmath16 is the variance of the random effects . marginalizing over @xmath17 gives the marginal likelihood,@xmath18\\[-8pt ] & \sim&\mathcal{n}\bigl(\beta_{0}{\mathbf{1}}+\beta_{x}{\mathbf{x}},\sigma_{g}^{2}\mathit{{\mathbf{r}}(\theta_{g})}+\tau^{2}{\mathbf{i}}\bigr),\nonumber\end{aligned}\ ] ] where @xmath19 is an @xmath20-vector of ones , @xmath21 is the identity matrix , and @xmath22 . here @xmath6 in ( [ eq : basic ] ) is explicitly decomposed into spatial and nonspatial components . an alternative formulation would specify the unknown spatial function , @xmath23 , as a penalized spline , where a penalty parameter plays the role of @xmath24 in the marginal likelihood in penalizing complexity of the spatial structure . the exposure may itself be spatially correlated . for example , if @xmath25 is a gaussian process , then @xmath26 , with parameters analogous to those for @xmath17 . to demonstrate processes operating at different spatial scales , figure [ fig : scales ] shows simulated spatial surfaces as one varies the spatial range parameter , @xmath27 , in a gaussian process model . the spatial statistics literature assumes that the error , @xmath28 in ( [ eq : basic ] ) , is independent of the covariate(s ) ( @xcite , @xcite ; @xcite , @xcite ) , with little or no discussion of the possibility that the error involves variation from unmeasured confounders . henceforth , i will refer to the errors as residuals because of the common use of the term `` spatial residual '' to refer to unexplained spatial variability . to explore the possibility of confounding , let s consider @xmath29 to be induced as the effect , @xmath30 , of an unmeasured variable , @xmath31 , on the outcome . @xmath32 may also be spatially correlated , for example , @xmath33 , such that @xmath34 , where @xmath35 is again a spatial range parameter . if @xmath31 and @xmath9 are dependent , then @xmath31 is an unmeasured spatial confounder . derivation of the marginal likelihood should be done by integrating over the ( unknown ) conditional distribution of @xmath36 given @xmath37 , whereas the integration leading to ( [ eq : basic ] ) ignores the dependence . note that if @xmath37 and @xmath36 are considered fixed , then association between @xmath37 and @xmath38 is known as concurvity ( @xcite , @xcite ; @xcite , @xcite ) . in the applied literature , practitioners often recognize the need to consider residual spatial structure in the outcome , with language of `` control '' or `` accounting '' for autocorrelation , and they fit models ( such as kriging or spatial random effects ) that implicitly assume independence of the residual andthe exposure ( @xcite , @xcite ; @xcite , @xcite ; @xcite , @xcite ; @xcite , @xcite ; @xcite , @xcite ; @xcite , @xcite ; @xcite , @xcite;@xcite , @xcite ) . with the recent exception of @xcite , formal statements of the goals and properties of fitting such spatial models are generally absent . however , much of the interest appears to lie in using the spatial residual structure to try to account for spatial confounding , with the implicit assumption that such models reduce or eliminate confounding bias(e.g . , @xcite , @xcite;@xcite , @xcite ; @xcite , @xcite ; @xcite , @xcite ; @xcite , @xcite ) . one approach is to explicitly consider the spatial scales involved , hoping that accounting for variation at a relatively large spatial scale allows for identification of the parameter of interest based on exposure heterogeneity at a smaller spatial scale ( e.g. , @xcite , @xcite ; @xcite , @xcite ; @xcite , @xcite ) . this smaller scale variation may be less prone to confounding in a given application . however , this consideration of spatial scale is often not explicit , and effects of scale on bias reduction , while sometimes hinted at , have not been developed formally . in the analogous context of time series modeling of air pollution , @xcite attempt to attribute all the temporally correlated variability in the outcome to the residual term in order to identify the effect of exposure based on the temporally uncorrelated ( and presumably unconfounded ) heterogeneity in the exposure . @xcite provide no guidance in the scenario that the exposure can not be decomposed into autocorrelated and uncorrelated components . this issue also applies to the approach of @xcite , who filter out the dependence between fixed and random effects . in the spatial setting , in which measurements can not be made at all locations , accurate estimation of the uncorrelated component , if such a component even exists , is rare : consider atmospheric phenomena such as temperature and air pollution . a common situation in which fine - scale heterogeneity is not resolved involves prediction of spatially varying exposure values using averages of nearby measurements or spatial smoothing techniques . hence , i seek to address the problem when all of the measured components of variation in exposure are spatial . in this paper i address estimation in simple regression models with spatial residual structure . i focus on the properties of penalized models , using a simple mixed model fit by gls , equivalent to universal kriging , to analyze the effects of the spatially correlated residual structure on fixed effect estimators . section [ sec : bias ] focuses on bias from spatial confounding . i report analytic results when the full covariance structure is known and supporting simulations when the covariance ( or the amount of smoothing in penalized spline models ) is estimated from the data . i assess the use of sensitivity analysis approaches based on spline models that explicitly consider the bias - variance tradeoff involved in choosing the spatial scale at which to model the residual variation . section [ sec : precision ] focuses on precision of estimators when there is no association between exposure and residual ( no spatial confounding ) . i close with a case study of the effects of air pollution on birthweight ( section [ sec : case - study :- massachusetts ] ) . a key consideration in the basic model ( [ eq : random - model ] ) is identifiability of @xmath8 and @xmath23 . a closely - related question is how the estimation procedure attributes variability between the exposure and the spatial residual term ( the random effects ) . in the simple linear model , attribution of variability to the covariates rather than the error term is favored because this allows the estimate of the error variance to decrease , with the normalizing constant of the likelihood favoring smaller error variance . in the spatial model , if the spatial term , @xmath17 , is unconstrained , then @xmath39 and @xmath17 are not identifiable in the likelihood : one could remove the covariate from the model and redefine @xmath40 with no change in the likelihood . identifiability comes through constraints on @xmath17 , either by ( 1 ) penalizing lack of smoothness in @xmath23 , ( 2 ) considering @xmath17 to be a random effects term , or ( 3 ) having a prior on @xmath17 . these approaches give higher penalized likelihood , marginal likelihood or posterior density , respectively , when variability is attributed to the unpenalized fixed effects term rather than to the spatial term . in the spatial confounding context this dynamic causes bias in estimation of @xmath8 , for example , as seen in the simulations of @xcite . an alternative constraint is to represent @xmath17 in a reduced dimension basis , say , as a regression spline . in this case the model is identifiable if there is a component of variability in @xmath37 that can not be explained by the spline structure , that is , if @xmath37 is not perfectly collinear with the columns of the chosen basis matrix . to consider bias from unmeasured spatially varying confounders , take the following model as the data - generating mechanism , @xmath41 with the notation as in section [ sec : introduction ] . for each location , @xmath42 , suppose the correlation of @xmath25 and @xmath43 over repeated sampling at the location is @xmath44 , so that @xmath31 is a confounder . suppose further that @xmath31 is not observed and that one models the residual spatial structure in the outcome through spatially correlated random effects , @xmath45 as in ( [ eq : random - model ] ) . finally , suppose that one ignores the correlation between @xmath38 and @xmath37 and integrates over themarginal distribution for @xmath17 , giving ( [ eq : basic - marg - lik ] ) . equivalently , @xmath46 . the induced correlation between @xmath9 and @xmath47 violates the usual regression assumption that the error is independent of the covariate , leading to bias . from the random effects perspective , we have ( incorrectly ) assumed that the random effects are independent of the covariate , a key ( but often unstated ) assumption of mixed effects models ( @xcite , @xcite ; @xcite , @xcite , page 170 ) . the treatment of @xmath25 and @xmath43 as random naturally induces spatial structure . however , in a given data set the most plausible repeated sampling framework may suggest that @xmath37 and @xmath36 reflect spatial structure that does not arise from a stochastic data generating process . rather , one might consider @xmath25 and @xmath43 to be fixed unknown functions , particularly when @xmath37 and @xmath36 vary at large scales , which mimics the partial spline / partial linear setting . this also is consistent with the treatment of large - scale variation in the mean term in traditional kriging . consider the case when there is concurvity between the two fixed functions , reflected in a nonzero empirical correlation , @xmath48 between @xmath37 and @xmath38 as calculated over the collection of locations ( e.g. , the concurvity inthe simulations of @xcite , @xcite ; @xcite , @xcite;@xcite , @xcite ) . in the partial linear / partial spline setting it is well known that such association between the exposure and the nonparametric smooth term causes bias ( @xcite , @xcite , equation 28 ; @xcite , @xcite ) . in any real data set , the orthogonality needed for @xmath49 seems particularly unlikely if both @xmath37 and @xmath36 vary at large scale relative to the size of the domain ( though @xmath50 may be as much a possibility as @xmath51 ) . the stochastic generative model is still useful under this framework of fixed functions because realizations of @xmath25 and @xmath43 give plausible values for @xmath37 and @xmath36 that could arise in real applications for which there is no reasonable stochastic mechanism . i choose to treat @xmath37 and @xmath36 stochastically , and i use @xmath52 to quantify explicitly the strength of association between the residual spatial variation and the exposure . this approach allows for some simple , useful analytic results and is further justified in that the variation that an unmeasured @xmath36 induces in @xmath53 is necessarily treated stochastically as part of the residual in actual applications . in some cases i report results conditional on @xmath37 , and in others i also average over the stochastic variability in @xmath37 and over variability in the spatial locations of the observations . since @xmath31 represents an unmeasured confounder , i assess the inferential properties of fitting a regression model by maximizing the marginal likelihood ( [ eq : basic - marg - lik ] ) using gls , thereby ignoring correlation between the residual and the exposure . i assess bias as a function of the spatial scales of @xmath25 and @xmath43 , which i suppose to be generated as gaussian processes with matrn spatial correlation function @xmath54 where @xmath55 is the euclidean distance between two locations , @xmath27 is the spatial range parameter , and @xmath56 is the modified bessel function of the second kind , whose order is the smoothness parameter , @xmath57 . i fix @xmath58 , which gives continuous and differentiablegaussian process realizations . this reflects an assumption of some smoothness in the spatial processes under consideration , but i also consider results based on an exponential correlation function ( i.e. , @xmath59 ) . the model ( [ eq : basic - marg - lik ] ) is equivalent to both a mixed model and a universal kriging model if one knows the variance and spatial dependence parameters . furthermore , given the extensive use of penalized splines in applications , and the connection between penalized splines and mixed models@xcite , i also consider the use of a penalized spline to represent @xmath23 . in the nonspatial context , one would generally try to adjust for confounding by including relevant covariates as fixed effects ; in the spatial context one could include spatial regression spline terms . the basic question that i explore in the remainder of section [ sec : bias ] is the extent to which inclusion of a spatial random effect term or a penalized spline can adjust for unmeasured spatial confounding , given that these approaches do not involve a projection in the way that a regression spline does . the random effects and penalized spline approaches do estimate the residual spatial variation based on a bias - variance tradeoff ( e.g. , claeskens , krivobokova and opsomer , @xcite ) , and the penalized spline is a regression spline in the limit as the penalty goes to zero . so it seems plausible that these approaches may reduce bias by at least partially adjusting for the unmeasured spatial confounder . i will show that the spatial scales involved are critical . this section considers bias when i suppose that the variance parameters are known and only the regression coefficients , @xmath60 and @xmath8 , are unknown . the initial results concern the situation when the exposure , @xmath9 , and the unmeasured confounder , @xmath31 , vary at the same spatial scale . i then assess what happens when @xmath9 varies at two scales and one is the same scale as the single - scale confounder . finally , i consider the possibility that there is additional variability in the outcome at another scale , but uncorrelated with @xmath9 . to start , suppose that @xmath25 and @xmath43 share the same spatial correlation range , @xmath27 , but may have different marginal variances , namely , @xmath61 and @xmath62 and @xmath63 . straightforward conditional normal calculations give@xmath64_{2}\nonumber\\ & = & \beta_{x}+ \left[\rule{0pt}{23pt}(\bolds{\mathcal{x}}^{t}\mathit{\bolds{\sigma}}^{-1}\bolds{\mathcal{x}})^{-1}\bolds{\mathcal{x}}^{t}\bolds{\sigma}^{-1 } \bolds{\mathcal{x}}\nonumber \right . \\[-8pt]\\[-8pt ] & & { } \left . \hspace*{54pt}\cdot \pmatrix { \displaystyle \mu_{z}-\rho\frac{\sigma_{z}}{\sigma_{x}}\beta_{z}\mu_{x}\vspace*{4pt}\cr \displaystyle \rho\frac{\sigma_{z}}{\sigma_{x}}\beta_{z } } \right]_{2}\nonumber\\ & = & \beta_{x}+\rho\frac{\sigma_{z}}{\sigma_{x}}\beta_{z},\nonumber\end{aligned}\ ] ] where @xmath65 $ ] , @xmath66_{2}$ ] indicates the second element of the 2-vector , and @xmath67 . the resulting bias , @xmath68 , is the same as if @xmath37 and @xmath36 were not spatially structured and is also equal to the bias under ols . this demonstrates that we have not adjusted for confounding at all by fitting the model that includes spatial structure . as with ols , the model attributes as much of the variability as possible to the exposure , rather than to the spatially correlated residual term , including all of the variability in @xmath36 that is related to @xmath37 . if @xmath69 , the bias is zero in ( [ eq : same - scale ] ) this occurs because we average over stochastic variability in @xmath36 , so any nonorthogonality between @xmath37 and @xmath36 in individual realizations contributes to variance rather than bias . this contrasts with the bias terms in @xcite and @xcite , which are caused by nonorthogonality of the fine - scale variation in @xmath37 and the nonparametric component of the model , since neither is treated stochastically . next , i keep the same data - generating and model - fitting framework , but explore the situation in which the exposure varies at two scales . i suppose that @xmath25 is a multi - scale process and introduce correlation between @xmath43 and one of the components of @xmath25 . let @xmath70 be decomposed into a component , @xmath71 , that is at the same scale as the confounder , @xmath36 , and a component at a different scale , @xmath72 , which is independent of @xmath71 and @xmath36 . specifically , take @xmath73 , @xmath74 and @xmath75 . after some straightforward algebra and matrix manipulations , we have @xmath76_{2}\\ & = & \beta_{x}+c({\mathbf{x}})\rho\frac{\sigma_{z}}{\sigma_{c}}\beta_{z},\nonumber\end{aligned}\ ] ] where @xmath77_{2}p_{c } , \\ \bolds{\sigma}^{*}&\equiv&\frac{\beta_{z}^{2}\sigma_{z}^{2}\mathit{{\mathbf{r}}(\theta_{c})}+\tau^{2}{\mathbf{i}}}{\beta_{z}^{2}\sigma_{z}^{2}+\tau ^{2}}= \bigl((1-p_{z}){\mathbf{i}}+p_{z}\mathit{{\mathbf{r}}(\theta_{c } ) } \bigr ) , \\ { \mathbf{m}}&\equiv&\bigl(p_{c}{\mathbf{i}}+(1-p_{c})\mathit{{\mathbf{r}}(\theta_{u})}\mathit{{\mathbf{r}}(\theta_{c})}^{-1}\bigr)^{-1}\end{aligned}\ ] ] and @xmath78 . we see that the bias term is proportional to that in the single - scale setting , multiplied by an additional term , @xmath79 , that modulates the bias . @xmath79 necessarily includes an extra multiplicative factor , @xmath80 , that quantifies the magnitude of the confounded component of @xmath37 relative to the total variation in @xmath37 . while the term @xmath79 is complicated , we can explore its dependence on the spatial scales ( @xmath81 and @xmath82 ) and the magnitudes of the variance component ratios ( @xmath83 and @xmath84 ) to see how the bias compares to the same - scale setting . in the following results i average over the variability in @xmath37 . for a grid of @xmath85 locations on the unit square , figure [ fig : bias - exact ] shows the average of @xmath79 over 1000 simulations as a function of @xmath81 and @xmath82 , for combinations of @xmath84 and @xmath83 , where the empirical average approximates the expectation with respect to the distribution of @xmath37 . there is a simple pattern to the bias modification relative to the same - scale setting . for @xmath86 ( the diagonal elements on the @xmath87 line ) , we do not need simulation : @xmath88 , which is equivalent to the same - scale result ( [ eq : same - scale ] ) , after accounting for the proportion of variability in @xmath37 that is confounded , @xmath84 . note that if one estimates the analogous bias to ( [ eq : diff - scale ] ) for ols applied to spatial data , it is nearly constant regardless of the spatial scales [ @xmath89 ; not shown ] . only when @xmath90 , and particularly when @xmath91 , do we see less bias than in the same - scale setting , with clear potential for bias reduction from modeling the residual spatial variation in the outcome ( recall figure [ fig : scales ] to interpret the values of @xmath27 ) . above the diagonal , for @xmath92 , @xmath93 , indicating more bias when the scale of confounding is smaller than the scale of unconfounded variability . this situation may be of limited practical interest , because it s not clear that there are real applications in which the unconfounded variation in the exposure occurs at larger scales than the confounded variation . however , it does show that there are circumstances in which bias is larger than under ols , a point also made in @xcite . note that the patterns in figure [ fig : bias - exact ] are qualitatively similar regardless of the values of @xmath84 and @xmath83 . quantitatively , for larger values of @xmath84 , corresponding to a larger proportion of the variation in the exposure being confounded variation , bias is larger . for larger values of @xmath83 , corresponding to a larger proportion of the residual variation being the contribution of the confounder , the effects of the spatial scales are more distinct . the results highlight that inclusion of the spatial residual does not give unbiased estimates and bias is substantial in many scenarios even when the covariance parameters are known . results are very similar when i sample locations uniformly on the unit square or in a clustered fashion ( using a poisson cluster process ) . the results in figure [ fig : bias - exact ] correct for the complication that the sample variance of spatial process values ( calculated over the domain ) decreases as @xmath27 increases . this occurs because the sample variance over the domain in a single spatial replicate underestimates population variability ; see figure [ fig : scales](b)(c ) for examples . i want to have fixed ratios of average sample variances , @xmath94 and @xmath95 , for all values of @xmath81 and @xmath82 , thereby avoiding the introduction of artifacts caused solely by having ratios of sample variances change with the spatial ranges . here @xmath96 , @xmath97 and @xmath98 are the sample variances of @xmath99 @xmath71 and @xmath72 , respectively . to achieve this , i generate @xmath100 and @xmath101 and modify the calculation of @xmath79 in ( [ eq : diff - scale ] ) accordingly . @xmath102 and @xmath103 are functions of @xmath81 and @xmath82 , respectively , that are chosen such that @xmath104 and @xmath105 , where @xmath106 is the sample variance of @xmath71 for a given realization under @xmath81 and analogously for @xmath107 . the expectations are taken with respect to the distribution of the subscripted random vector . these manipulations allow me to present bias for scenarios that correspond to specific ratios of average sample variability of @xmath71 , @xmath72 , @xmath36 and @xmath108 over the spatial domain . to have only a single scale of residual spatial variability is not very realistic . therefore , i carried out an additional simulation study with residual spatial variability in the outcome that is independent of the exposure and at a smaller scale than thescale of @xmath43 . i suppose that the data - generating model is@xmath109 and that @xmath110 , independent of @xmath37 , @xmath36 and @xmath108 , with all of the other details as before . under this data - generating model and again supposing that all variance parameters are known , simulation estimates of @xmath111 indicate that bias is somewhat smaller than that seen in figure [ fig : bias - exact ] for @xmath112 and somewhat larger for @xmath113 ( not shown ) . note that if the additional small - scale variability is correlated with the exposure , then one is back in the situation of having common scales for the exposure and the confounder , which is considered at the beginning of this section . to generalize the results of section [ sub : bias - with - known ] , which supposed known variance and spatial dependence parameters , i set up a simulation study to assess the impact of estimating those parameters . in addition to maximum likelihood estimation of a mixed effects / kriging model based on the marginal likelihood ( [ eq : basic - marg - lik ] ) , i consider the use of penalized likelihood to fit the model ( [ eq : random - model ] ) with a penalized thin plate spline spatial term for @xmath23 . i implemented the penalized spline using gam ( ) in r , which uses generalized cross - validation ( gcv ) for data - driven smoothing parameter estimation @xcite . for the core simulations , i set the following parameter values , @xmath114 , @xmath115 , @xmath116 , @xmath117 , and sample 100 spatial locations uniformly from the unit square . for a range of values of @xmath81 and @xmath82 , i simulate 2000 data sets for each pair \{@xmath118}. for each simulated data set , i generate new spatial locations and new values of @xmath37 and @xmath36 ; i then generate @xmath53 using ( [ eq : generative - model ] ) . again we have to account for the reduced empirical spatial variability as @xmath27 increases ; these simulations have effective values of @xmath119 and @xmath120 . with regard to bias , the simulation results for the mixed / kriging model [ figure [ fig : sims - bias](b ) ] reasonably match the theoretical values with known variance parameters [ figure [ fig : sims - bias](a ) ] . however , when @xmath91 , the bias is generally larger than with known variance parameters , because the fitted model sometimes estimates little or no spatial structure in the residuals , pushing bias results toward the larger bias seen under ols [ figure [ fig : sims - bias](d ) ] . results for the penalized spline model [ figure [ fig : sims - bias](c ) ] show smaller bias for @xmath90 than the mixed model , presumably caused by the difference between estimating the amount of smoothing by gcv compared to maximum likelihood . in either case , spatial scales are critical , and bias is smaller than with ols only when the scale of confounding is larger than the scale of the unconfounded variability . additional simulations indicate that as the correlation of confounder and exposure increases , or the magnitude of variation in the confounder increases , or the effect size decreases , relative bias increases ( not shown ) . in such scenarios , substantial bias reduction occurs only for very small spatial scales in the exposure and large scales of confounding . figure [ fig : sims ] compares the mixed model with the penalized spline in the context of a bias - variance tradeoff . there is a substantial bias - variance tradeoff , with the smaller bias of the penalized spline model ( for @xmath113 ) trading off for increased variance . the result is increased mean squared error ( mse ) in @xmath121 , except when @xmath82 is very small . both model variance estimates ( third column ) understate the variability in the coefficient estimates ( second column ) , with particular underestimation of uncertainty and low coverage for the mixed / kriging model , and with lower coverage as one moves away from the region of @xmath122 . of course the bias causes much of the poor coverage . fitting the mixed / kriging model by restricted maximum likelihood ( reml ) rather than maximum likelihood produces moderate improvement in coverage , with the average variance estimate more similar to the variance of the estimated coefficients . using @xmath59 ( i.e. , an exponential spatial correlation function ) in the fitting rather than the true @xmath58 has little effect on results . however , when i generate the unconfounded variability , @xmath72 , based on @xmath59 , bias is substantially smaller than the core results ( particularly note that there is reduced bias relative to ols when @xmath86 ) , apparently because the nondifferentiable sample paths of processes with exponential covariance play the role of very fine - scale , unconfounded variability . there is little change in results when using spatial locations simulated using a poisson cluster process with an average of seven children per cluster and cluster kernel standard deviation of @xmath123 . finally , simulations with @xmath69 , that is , no confounding , indicate no bias for either model , as expected . our bias results when @xmath44 are analogous to the bias seen with penalized spline models in @xcite and @xcite . there , concurvity ( i.e. , @xmath124 ) between the smooth temporal term ( analogous to our spatial residual ) and the exposure emerged from the fixed basis coefficients chosen based on empirical dataexamples ( r. peng , personal communication;@xcite , @xcite ) . similar results are seen in the spatial settings of @xcite . the presence of small - scale independent variation in the residual ( [ eq : multiscale - resid ] ) reduces bias for @xmath90 ( not shown ) , relative to the results presented above . this occurs through an increase in the number of degrees of freedom estimated from the data to capture residual variability , that is , undersmoothing with respect to the variation at the @xmath81 scale , analogous to undersmoothing in the partial spline setting ( @xcite , @xcite ; @xcite , @xcite ) . this scenario seems quite likely in applications : if there is large - scale residual spatial structure , there is likely to be finer - scale structure as well . thus , analyses that attempt to best fit the data may in the process reduce bias from confounding at the larger scales . we have seen that even when all covariance parameters are known and the scale of confounding is much larger than the scale of unconfounded variability in @xmath37 , bias remains , albeit at a much reduced level . in principle , if the structure at the confounded scale could be exactly fit using a set of basisfunctions , such as a regression spline ( e.g.,@xcite , @xcite ) , then the exposure effect estimate would be unbiased , as in any multiple regression . the partial residual kernel smoothing approach of @xcite reduces bias in a similar fashion , albeit without using a projection , through the technique of twicing . however , in a real application , one has to choose the basis functions , and if the basis functions do not fully explain the confounded , large - scale variability , even with a basis of seemingly sufficient dimension , this will induce a bias . one could instead consider a penalized spline approach with penalty parameter chosen in advance to give the desired effective degrees of freedom ( e.d.f . ) . for fixed e.d.f . , since the penalized spline smoother is not a true projection ( @xcite , @xcite ; @xcite , @xcite ) , one would expect the penalized spline approach to have more bias than the regression spline approach . heuristically , bias in this approach occurs because the estimated spatial term does not fully explain the confounded component of the variability in the outcome , causing a bias analogous to that seen in the partial spline setting ( @xcite , @xcite ; @xcite , @xcite ) . however , we would expect the penalized spline to be less sensitive to the exact form of the basis functions and number and placement of knots , as is seen in the example ( section [ sec : case - study :- massachusetts ] ) . furthermore , one can always undersmooth to reduce the bias , following the recommendation in the partial spline literature ( @xcite , @xcite ; @xcite , @xcite ) . thus , using a penalized spline seems reasonable , albeit without the clean interpretation of a projection . i show below that simulations comparing regression spline and penalized spline models support these theoretical results from the literature , in the spatial context considered here , with the regression spline having reduced bias and increased variance relative to penalized modeling . the primary issue in an application is choosing the amount of smoothing to reduce bias , since inference about @xmath8 is the goal rather than best fitting the data . data - driven smoothing might reduce bias ( if there is small - scale residual correlation ) or might have little effect on bias ( if the data suggest only large - scale residual correlation ) . thus , the reduction in bias will depend on the scales involved and the actual amount of smoothing done , and the analysis will reveal little about the sensitivity of estimation to scale . instead , one could explicitly assess the bias - variance tradeoff by varying the amount of smoothing and assessing the sensitivity of the exposure effect inference . one approach is a spatial analogue to the sensitivity analysis approaches of @xcite : fit a model with spatial basis functions and vary the e.d.f . ( e.g. , @xcite , @xcite ) . plotting @xmath121 and uncertainty intervals as a function of e.d.f . ( or some other metric ) provides an assessment of the robustness of results to potential spatial confounding at various scales . if one is concerned about confounding at a particular scale , then one can report the results for an e.d.f . that would undersmooth with respect to that scale to reduce bias , accepting the tradeoff of increased uncertainty . motivated by this analysis strategy , i set up a simulation under the settings of section [ sub : sims ] , using a regression spline ( i.e. , unpenalized fixed effects ) and varying the e.d.f . by changing the dimension of the basis in gam ( ) in r. figure [ fig : sims - fixdf](a ) shows relative bias as a function of the spatial scales involved . as before , i focus on the results below the @xmath87 diagonal ( @xmath86 ) , as this is the scenario of practical interest . by choosing a large number of e.d.f . , one can decrease bias more effectively than when estimating the amount of smoothing from the data [ i.e. , figure [ fig : sims - bias](c ) ] . however , with moderate and large scale variability , the variance of the estimates in this fixed effects model increases dramatically [ figure [ fig : sims - fixdf](b ) ] . this causes a concordant increase in the mse ( not shown ) , highlighting the bias - variance tradeoff . in contrast , using a penalized spline with fixed e.d.f . [ fixing the smoothing parameter in gam ( ) in r ] shows much more stable results . as expected , for a given e.d.f . bias is not reduced as much as with a regression spline [ figure [ fig : sims - fixdf](a ) ] , but there is much less variability [ figure [ fig : sims - fixdf](b ) ] . a diagnostic approach to understanding whether the residual may include variation from an unmeasured confounder is to assess the correlation between the residual and the exposure . not knowing @xmath8 , one might use a variety of plausible values of @xmath8 to estimate @xmath17 and then calculate the correlation with @xmath37 ( and potentially with filtered versions of @xmath37 that exclude small - scale variation ) . if one accounts for large - scale variation as a means of reducing bias , there may still be small - scale residual variation , such as fine - scale correlation in health outcomes related to residential sorting . as i have shown , one can reduce potential confounding bias from this fine - scale variation through explicit spatial modeling only if there is variability in the exposure at an even smaller scale . if there is not , then one is effectively assuming that the fine - scale variation is uncorrelated with the exposure . given this assumption , one may need to account for the fine - scale residual spatial variation so that uncertainty estimation for @xmath8 is not compromised ( but note the results of section [ sub : prec - ques3 ] ) . one possibility would be to use an analysis robust to misspecification of the residual variance , for example , using an estimating equation with uncertainty based on the sandwich estimator , with regression spline terms in the mean to account for large - scale spatial confounding bias . alternatively , one could fit a penalized model with the amount of smoothing determined from the data . this has two effects . first , it naturally accounts for the effect of the spatial structure on uncertainty estimation . second , in the presence of small - scale residual variability , the model will naturally undersmooth with respect to large - scale variability that may cause confounding , thereby reducing bias from confounding at the larger scale , as discussed previously . in this section i suppose that the residual and the exposure are independent ( @xmath69 in the framework of section [ sec : bias ] ) , which results in unbiased estimation of @xmath8 . i consider effects of spatial scale on the following questions about efficiency of estimators for @xmath8 ( henceforth simply @xmath125 ) and quantification of uncertainty : 1 . given a fixed amount of residual variation , how is efficiency affected by the proportion of that variation that is spatial ? what is the magnitude of the improvement in efficiency when accounting for residual spatial variation , relative to ols ? 3 . if one uses the naive estimator for the variance of the ols estimator , @xmath126 , what is the magnitude of the error in uncertainty estimation compared to the correct variance estimator for @xmath126 ? the first question does not appear to have been raised in the literature . with regard to the second , while we know that gls is the most efficient estimator when the residuals are correlated , here i investigate the magnitude of this efficiency advantage as a function of the spatial scales involved . regarding the third , the conventional wisdom in the statistical and applied literature appears to be that not accounting for spatial structure leads to underestimation of uncertainty ( e.g. , @xcite , @xcite ; @xcite , @xcite ; @xcite , @xcite , page 324 ) . however , i have not seen a formal quantification of this underestimation for a regression coefficient , in contrast to our understanding of the potentially severe underestimation of uncertainty for the meanof a spatial process ( @xcite , @xcite , section 1.3;@xcite , @xcite , section 1.5 ) . note that there are three variance estimators ( i.e. , estimators for the sampling variability of the estimated regression coefficient ) under consideration here : the true gls variance estimator , and the true and naive ols variance estimators . when @xmath69 , ols is unbiased , so it makes sense to consider ols for estimation , provided we adjust the usual ols variance estimator to account for the residual spatial correlation . while actual applications will likely involve more complicated modeling , consideration of these questions in this simple setting , and with known variance components , helps to understand the basic issues . given a fixed amount of residual variation , how is efficiency affected by the proportion of variation that is spatial ? i quantify efficiency in terms of precision rather than variance , as this allows for closed form derivations . [ lem ] consider the model ( [ eq : basic - marg - lik ] ) and suppose that all parameters are known except @xmath60 and @xmath127 . the expectation of the precision of @xmath128 , with respect to the sampling distribution of @xmath37 , is @xmath129 where @xmath130 , @xmath131 , and the remaining notation follows that in previous sections . see the for the proof . note that the term in parentheses is an effective sample size , analogous to @xmath132 in the nonspatial problem . here the adjustment is for spatial structure in residual and exposure , with the second component in the parentheses analogous to the degree of freedom lost for estimating a mean . figure [ fig : prec - gls](a ) shows monte carlo estimates of the expected precision as a function of @xmath133 and @xmath15 , averaging ( [ eq : gls - prec ] ) over 500 sets of @xmath85 locations simulated uniformly on the unit square . i report the expected precision divided by a baseline of @xmath134 , which is the expected precision in the nonspatial setting , supposing that the total residual variation , @xmath135 , remains constant . compared to the nonspatial setting , lower precision occurs unless the exposure varies at small spatial scale . when the exposure varies at small spatial scale and the residual at larger spatial scale , precision can be substantially greater than in the nonspatial setting . the model is able to account for part of the residual variance through the spatial structure , as if the spatial structure were an additional covariate to which variation in the response is attributed . gls implicitly estimates the process , @xmath17 in ( [ eq : random - model ] ) , that gives rise to the marginalized model ( [ eq : basic - marg - lik ] ) . this reduces the remaining `` unexplained '' residual variability and thereby improves efficiency relative to having independent errors but equivalent overall residual variability . in contrast , when @xmath37 varies only at large spatial scales , then efficiency decreases because of difficulty in distinguishing @xmath136 from @xmath23 . results are similar using points on a regular grid or clustered based on a poisson cluster process . here i consider how spatial scale affects the relative efficiency of spatial and nonspatial estimators , comparing the precisions of the ols andgls estimators . since the true ols variance,@xmath137_{2,2}$ ] , is acomplicated function , it is difficult to derive closed form expressions for efficiency relative to the gls estimator . instead i conduct a small simulation study . for a regular grid of values of @xmath133 and @xmath15 , i carry out 500 simulations for each pair of values , with @xmath85 observations whose spatial locations are drawn uniformly over the unit square domain . note that i consider the ratio of the gls precision to the ols precision , so the values of @xmath138 and @xmath139 cancel out of the ratio and do not affect the results . figure [ fig : prec - gls](b ) shows the monte carlo estimates of the expected relative precision , as a function of the spatial scales , @xmath15 and @xmath133 , and the proportion of the residual variability that is spatial . when little of the residual variability is spatial ( @xmath140 ) , there is little gain in precision , as expected . when more is spatial , the gains in precision are small when @xmath17 varies at a small scale , but substantial when @xmath17 varies at a large scale . unfortunately , this is also precisely the case in which one would be concerned about spatial confounding . if we suppose that the large - scale structure in the residual has been controlled for in an effort to reduce the potential for bias , then with the remaining residual variability being fine scale , there is limited gain in precision regardless of the spatial scale of the exposure . with locations on a regular grid , the gains in precision are slightly less for small values of @xmath15 , while with poisson cluster process sampling , the gains are somewhat larger for small values of @xmath15 . see also @xcite for similar simulation results when a markov random field structure induces the correlation . applied analyses often ignore residual spatial correlation , raising the question of how strongly uncertainty estimates are affected . one can express the ratio of the true ols variance to the incorrect naive ols variance as follows . first define @xmath141 where @xmath142 . after expressing @xmath143_{2}=(\tilde{\mathbf{x}}{}^{t}{\mathbf{\tilde{x}}})^{-1}\tilde{\mathbf{x}}{}^{t}{\mathbf{y}}$ ] where @xmath144 , we have @xmath145 averaging over the sampling distribution of @xmath37 , we have @xmath146 so for @xmath147 or @xmath148 , that is , when either @xmath15 or @xmath133 is close to zero , we expect the ratio to be near one . note also that with spatial correlation functions that are nonnegative , the only negative contribution to the ratio can be from negative covariances induced by standardizing @xmath37 . such negative covariances should diminish as the sample size increases , so we expect the ratio to generally be no smaller than one , indicating that the naive variance does underestimate uncertainty . finally , the largest values of the ratio would occur with large positive correlations in corresponding elements of @xmath149 and @xmath150 , which is to be expected when both @xmath17 and @xmath37 show large - scale variation . figure [ fig : prec - gls](c ) supports these heuristic results , showing the average ratio of variances in simulations , where the simulations are conducted as in section [ sub : prec - ques2 ] . the ratio is close to one when either of the spatial terms has fine - scale variability and far from one when both have large - scale behavior . this result is similar to that of @xcite for inference about a correlation coefficient and to ( @xcite , @xcite , page 178 ) under serial autocorrelation in a regression setting . as expected , when the proportion of residual variability is smaller ( moving from the bottom left to bottom right panels ) , the expected ratio gets closer to one . this indicates that when nonspatial variation dominates the residual and the spatial structure in the residual or exposure is not too large in scale , the naive variance estimator may be reasonable . a lack of large - scale residual structure might result from having accounted for large - scale variation in attempting to reduce spatial confounding bias . results with gridded locations show ratios slightly closer to one , and with clustered locations , ratios further from one . note that the uncertainty estimate in any given naive analysis may be larger than when fitting a spatial model because the more sophisticated model both corrects the variance estimate , which increases the estimated uncertainty , and uses a more efficient estimator , which decreases the fundamental uncertainty . simple simulations with spatial ranges and sampling designs specific to an analysis could be easily carried out for further guidance in a given setting , allowing one to assess whether ignoring the spatial structure has substantial impact on uncertainty estimation . accounting for small - scale spatial correlation requires estimation of the spatial structure and is often computationally burdensome , so an assumption of independence can have an important practical benefit . of course in some analyses , any underestimation of variability may be cause for concern , in which case use of the naive variance estimator would not be tenable . chronic health effects of ambient air pollution in developed countries involve small relative risks , but are of considerable public health importance because of widespread exposure . epidemiologic studies attempt to estimate a small effect from data with high levels of variability and stronger effects from other covariates , including potential confounders such as socioeconomic status , so spatial confounding bias is of critical concern . i reanalyze data on the association between ambient air pollution ( estimates of black carbon , a component of particulate matter ) and birthweight in eastern massachusetts ( @xcite , @xcite ; @xcite , @xcite ) . these analyses found significant negative effects of traffic proxy variables and black carbon , respectively , on birthweight . @xcite used several methods to try to account for effects of measurement error in the predicted black carbon concentrations , which are based on a regression model that accounts for spatial and temporal structure and key covariates . i follow these analyses in using an extensive set of covariates to try to account for potential confounding . i use smooth terms for mother s age , gestational age and mother s cigarette use , to account for nonlinearities , a linear term for census tract income , and categorical variables for the following : presence of a health condition of the mother , previous preterm birth , previous large birth , sex of baby , year of birth , index of prenatal care and maternal education . the exposure of interest is the estimated nine - month average black carbon concentration at the geocoded address of the mother , based on a black carbon prediction model @xcite . following @xcite , for simplicity , i exclude the 13,347 observations with any missing covariate values , giving 205,713 births . in @xcite we found no evidence of residual spatial correlation based on a spatial semivariogram . further analysis here indicates that there is significant residual spatial variation but that nonspatial variation overwhelms the magnitude of this variation . figure [ fig : resids](a ) is a semivariogram showing no evidence of spatial structure , while a spatial smooth of model residuals [ figure [ fig : resids](b ) ] indicates clear spatial structure . while individual nonspatial variability among babies swamps the spatial variation ( hence the flat semivariogram ) , it is large relative to the estimated pollution effect ( note the surface values in the range of @xmath15140 to 40 , for comparison with effect estimates in figure [ fig : fullresults ] ) . thus , if the residual spatial variation is caused by spatially varying confounders , it could bias estimation of the pollution effect . to include a spatial term in models of birthweight , i consider a regression spline , an unpenalized approach , and a penalized spline , both with e.d.f . chosen in advance ( see section [ sub : bias - variance ] ) , as well as a penalized spline with data - driven smoothing parameter estimation based on gcv , all implemented in gam ( ) in r , using the thin plate spline basis . note that the thin plate regression spline approach implemented in gam ( ) should minimize sensitivity to knot placement @xcite . i first add a spatial term to the model with the full set of covariates to assess whether some of the estimated effect may be biased by spatial confounding . figure [ fig : fullresults](a ) shows how the estimated effect of black carbon varies with the e.d.f . and the spatial smoothing approach . the estimate attenuates somewhat as more e.d.f . are used to account for the spatial structure . for the penalized spline , as more than about 10 e.d.f . are used , the upper confidence limit exceeds zero , and for larger e.d.f . , the upper limit increases further . gcv chooses 157 e.d.f . , indicating fairly small - scale spatial structure in the data . for context note that with 129 e.d.f . in figure [ fig : resids](b ) we see spatial features at the scale of individual towns . while the regression spline approach implemented here avoids having to choose the knots , the empirical results are still very sensitive to e.d.f . , in contrast to the stability of the penalized spline solution as the e.d.f . varies . for both penalized and regression splines , there is a clear bias - variance tradeoff , with increasing variance as the number of e.d.f . however , for this problem with a very large sample size , the confidence intervals do not increase drastically , nor is there much difference in the uncertainty between the regression and penalized spline approaches . the spatial confounding assessment suggests that while we have somewhat reduced confidence in the black carbon effect , the effect estimate is reasonably stable even when using a spatial term with a large number of degrees of freedom . next i consider what might have happened if most of the covariates ( particularly the ones related to socioeconomic status ) were not measured , potentially inducing serious confounding . figure [ fig : fullresults](b ) indicates that without any spatial term in the model , the effect estimate is @xmath15123.0 with a 95% confidence interval of ( @xmath15126.8 , @xmath15119.2 ) , indicating a much more substantial effect of black carbon than the fully adjusted model . as soon as one accounts for spatial structure , even with a small number of e.d.f . , the estimate attenuates , approaching the fully adjusted estimate , with the upper confidence limit rising above zero . the reduced model appears to suffer from serious confounding , with the estimated pollution effect apparently driven by large - scale association of pollution and birthweight . the spatial analysis is able to account for much of this apparent confounding , substituting for a rich set of covariates . ideally one would fit a model that accounts for fine - scale spatial structure to improve one s confidence in the uncertainty estimation . however , with 205,713 observations , this is a computational challenge that i do not take up here . given the results in section [ sub : prec - ques3 ] that indicate that large - scale structure causes most of the variance underestimation , one can hope that the uncertainty at the larger values for the spatial e.d.f . in figure [ fig : fullresults ] may reasonably approximate the true uncertainty . considerations of scale are critical in spatial regression problems . standard spatial regression , which use spatial random effects , kriging specifications or a penalized spline to represent the spatial structure , are penalized models with inherent bias - variance tradeoffs in estimating the smooth function . under unmeasured spatial confounding , the bias carries over into estimating the coefficient for the exposure of interest , but the degree of bias depends on the spatial scales involved . inclusion of a spatial residual term accounts for spatial correlation in the sense of reducing bias from unmeasured spatial confounders only when there is unconfounded variability in the exposure at a scale smaller than the scale of the confounding . if the variation in exposure is solely at large scales , there is little opportunity to reduce spatial confounding bias , but with a component of small - scale exposure variability , large - scale spatial confounding bias can bereduced substantially . accounting for large - scaleresidual correlation is also important for improving precision of regression estimators and for correctly estimating uncertainty . in contrast , when residual correlation occurs at small scales , there is little opportunity for reducing spatial confounding bias at those scales or improving regression estimator precision . however , under the assumption of no small - scale confounding , fitting such residual structure can reduce bias from larger scale confounding by causing undersmoothing with respect to the large - scale structure . while the results here are limited to the simple setting of linear regression / additive models with a single covariate and single unmeasured confounder , i expect that the qualitative results and principles hold in more complicated settings , with no reason to believe that the bias results would improve in more complicated models . sensitivity analyses that show the bias - variance tradeoff as a function of the scale at whichthe spatial residual structure is modeled(@xcite , @xcite ; @xcite , @xcite ) offer one approach that helps to frame the issue of bias in the context of the spatial scales involved . in choosing a spline formulation to carry out such an analysis , while a regression spline has an appealing interpretation and in theory result in less bias in estimating the effect of interest , a penalized spline with a fixed effective degrees of freedom may give more stable results . of course the sensitivity analysis approach does not answer the question of how to get a single estimate of the effect of interest . one might also consider an approach similar to that of @xcite and explicitly decompose the exposure into multiple scales , including exposure at each scale as a separate covariate and focus causal interpretation on the effect estimates for the smaller scales ( e.g. , @xcite , @xcite ) . @xcite use matching estimators for each pair of observations and assess how effect estimates vary with spatial lag between the pairs to assess sensitivity . note that estimating equation approaches are not capable of reducing bias from unmeasured spatial confounding because the marginal variance is assumed to be unrelated to the exposure and variation is not attributed to a spatial term . from the econometric perspective , spatialconfounding bias might be seen as a type of endogeneity bias , with exposure the endogenous variable and the unconfounded component of exposure , or some proxy for it , an exogenous variable . since the unconfounded component is not measured directly , some sort of scale decomposition appears necessary . standard endogenous variable techniques such as two - stage least squares and instrumental variable methods @xcite do not appear directly useful but do share commonalities with approaches mentioned above . others have noted the identifiability problems in spatial models , with sensitivity of effect estimates to inclusion of a spatial residual term when thecovariates vary spatially ( @xcite , @xcite ; @xcite , @xcite;@xcite , @xcite ; @xcite , @xcite , page 187;@xcite , @xcite ; @xcite , @xcite ) . a different methodologic perspective than that presented here has been taken by @xcite and @xcite , who estimate the effect of exposure , @xmath37 , by forcing the spatial residual to be orthogonal to @xmath37 , attributing as much variability as possible to @xmath37 . this approach makes a very strong assumption of no confounding to avoid overadjustment bias from accidentally accounting for some of the effect of the covariate in the residual . note that the residuals and covariates are not orthogonal under gls estimation ( schabenberger and gotway , @xcite , page 349 ) . @xcite confront a similar problem of modeling systematic residual confounding in a context with identifiability problems , finding that imposing structure through a prior distribution in a nonidentified model can help account for a portion of the confounding , improving bias and precision of estimators . note that measurement error in the exposure is of critical concern , because reducing bias relies on estimating variability in exposure at scales smaller than the confounding . in many contexts , measurement error becomes an increasing concern at small scalesbecause of limitations in measurement resources . in contrast , large - scale exposure variation may be well estimated using spatial smoothing and regression models , thereby inducing berkson - type errorthrough what is effectively regression calibration@xcite . to the extent to which accounting for bias forces one to rely on exposure estimates more likely contaminated by classical measurement error , one may find oneself reducing bias from confounding only to increase it from measurement error . to the extent small - scale variation is affected by berkson error , one would increase variance but not incur bias by relying on the small - scale variation . finally note that in many settings one has aggregated exposure and outcome data , so one has limited ability to identify effects of exposure based on fine - scale variation because the aggregation eliminates the fine - scale variation ( e.g. , janes , dominici and zeger , @xcite ) . this suggests that accounting for spatial confounding with areal data , for whichresearchers often use standard conditionalauto - regressive models , is likely to be ineffectivewhen aggregating over large areal units , which is consistent with the bias seen in @xcite . in work concurrent with that presented here , @xcite have investigated bias in the areal setting under a variety of perspectives on the spatial random effects , also making the case for the approach taken in @xcite . from the definition of the gls estimator , we have @xmath152_{2,2}^{-1 } \\ & = & \frac{{\mathbf{1}}^{t}\mathit{\bolds{\sigma}}^{-1}{\mathbf{1}}}{{\mathbf{1}}^{t}\mathit{\bolds{\sigma}}^{-1 } { \mathbf{1}}{\mathbf{x}}^{t}\mathit{\bolds{\sigma}}^{-1}{\mathbf{x } } -{\mathbf{x}}^{t}\mathit{\bolds{\sigma}}^{-1}{\mathbf{1}}{\mathbf{1}}^{t}\mathit{\bolds{\sigma}}^{-1}{\mathbf{x}}}.\end{aligned}\ ] ] using the definitions of @xmath149 and @xmath153 , and taking the reciprocal , we have @xmath154 conclude by taking the expectation with respect to the sampling distribution of @xmath37 , using the expectation of a quadratic form , and rearranging the matrices inside the second trace to give a scalar : @xmath155 the author thanks louise ryan and francesca dominici for feedback and encouragement , andyhouseman , eric tchetgen and brent coull for comments , ben armstrong and john rice for thought - provoking discussions , joel schwartz for access to the birthweight data , alexandros gryparis and steve melly for assistance with the birthweight data , and brent coull for funding through niehs r01 grant es01244 . this work was also funded by niehs center grant es000002 and nci p01 grant ca134294 - 01 . augustin , n. , lang , s. , musio , m. von wilpert , k. ( 2007 ) . a spatial model for the needle losses of pine - trees in the forests of baden wurttemberg : an application of bayesian structured additive regression . _ j. roy . statist . c _ * 56 * 2950 . biggeri , a. , bonannini , m. , catelan , d. , divino , f. , dreassi , e. lagazio , c. ( 2005 ) . bayesian ecological regression with latent factors : atmospheric pollutants , emissions , and mortality for lung cancer . _ environmental and ecological statistics _ * 12 * 397409 . burden , s. , guha , s. , morgan , g. , ryan , l. , sparks , r. young , l. ( 2005 ) . spatio - temporal analysis of acute admissions for ischemic heart disease in nsw , australia . _ environmental and ecological statistics _ * 12 * 427448 . burnett , r. , ma , r. , jerrett , m. , goldberg , m. , cakmak , s. , pope iii , c. krewski , d. ( 2001 ) . the spatial association between community air pollution and mortality : a new method of analyzing correlated geographic cohort data . _ environmental health perspectives _ * 109 * 375380 . cakmak , s. , burnett , r. , jerrett , m. , goldberg , m. , pope iii , c. , ma , r. , gultekin , t. , thun , m. krewski , d. ( 2003 ) . spatial regression models for large - cohort studies linking community air pollution and health . _ journal of toxicology and environmental health , part a _ * 66 * 18111823 . cerd , m. , tracy , m. , messner , s. , vlahov , d. , tardiff , k. galea , s. ( 2009 ) . misdemeanor policing , physical disorder , and gun - related homicide : a spatial analytic test of `` broken - windows '' theory . _ epidemiology _ * 20 * 533541 . gryparis , a. , coull , b. , schwartz , j. suh , h. ( 2007 ) . latent variable semiparametric regression models for spatio - temporal modeling of mobile source pollution in the greater boston area . c _ * 56 * 183209 . hodges , j. reich , b. ( 2010 ) . adding spatially - correlated errors can mess up the fixed effect you love . technical report 2010 - 002 , division of biostatistics , univ . available at http://www.biostat.umn.edu/ftp/pub/2010/rr2010-002.pdf . houseman , e. , coull , b. shine , j. ( 2006 ) . a nonstationary negative binomial time series with time - dependent covariates : enterococcus counts in boston harbor . _ j. amer . statist . assoc . _ * 101 * 13651376 . lombarda , m. j. sperlich , s. ( 2007 ) . multi - level regression between fixed effects and mixed effects models . technical report , georg - august - univ . available at http://www.zfs.uni-goettingen.de/index.php?id=54 . lu , y. zeger , s. ( 2007 ) . decomposition of regression estimators to explore the influence of `` unmeasured '' time - varying confounders . technical report 159 , dept . biostatistics , johns hopkins univ . , baltimore , md . available at http://www.bepress.com/jhubiostat/paper159 . pope iii , c. , burnett , r. , thun , m. , calle , e. , krewski , d. , ito , k. thurston , g. ( 2002 ) . lung cancer , cardiopulmonary mortality and long - term exposure to fine particulate air pollution . _ journal of the american medical association _ * 287 * 11321141 . zeger , s. , dominici , f. , mcdermott , a. samet , j. ( 2007 ) . mortality in the medicare population and chronic exposure to fine particulate air pollution . technical report 133 , dept . biostatistics , johns hopkins univ . , baltimore , md . available at http://www.bepress.com/jhubiostat/paper133 . zeka , a. , melly , s. schwartz , j. ( 2008 ) . the effects of socioeconomic status and indices of physical environment on reduced birth weight and preterm births in eastern massachusetts . _ environmental health _ * 7 * 60 .

the source responsible for reionizing the universe at remains uncertain . while an energetically adequate population of star - forming galaxies may be in place , it is unknown whether a large enough fraction of their ionizing radiation can escape into the intergalactic medium . attempts to measure this escape - fraction in intensely star - forming galaxies at lower redshifts have largely yielded upper limits . in this paper we present new hst cos and archival fuse far - uv spectroscopy of a sample of eleven lyman break analogs ( lbas ) , a rare population of local galaxies that strongly resemble the high - z lyman break galaxies . we combine these data with sdss optical spectra and spitzer photometry . we also analyse archival fuse observations of fifteen typical uv - bright local starbursts . we find evidence of small covering factors for optically - thick neutral gas in three cases . this is based on two independent pieces of evidence : a significant residual intensity in the cores of the strongest interstellar absorption - lines tracing neutral gas and a small ratio of extinction - corrected h to uv plus far - ir luminosities . these objects represent three of the four lbas that contain a young , very compact ( pc ) , and highly massive ( m ) dominant central object ( dco ) . these three objects also differ from the other galaxies in showing a significant amount of blueshifted ly emission , which may be related to the low covering factor of neutral gas . all four lbas with dcos in our sample show extremely high velocity outflows of interstellar gas , with line centroids blueshifted by about 700 km s and maximum outflow velocities reaching at least 1500 km s . we show that these properties are consistent with an outflow driven by a powerful starburst that is exceptionally compact . we speculate that such extreme feedback may be required to enable the escape of ionizing radiation from star forming galaxies .

1101.4219

following the epoch of recombination at z @xmath61000 , the next major event in the history of the universe was its reionization by an early generation of massive stars and/or black holes . one of the major goals of modern cosmology is to understand this process : what sources were responsible and when did it occur ? observations of the cosmic microwave background with wmap show that reionization occurred at a mean redshift of [email protected] ( dunkley et al . 2009 ) , while observations of absorption due to the lyman @xmath1 forest in quasars show that reionization was not quite complete until a redshift of about six ( e.g. fan et al . 2006 ) . based on the observed luminosity functions of active galactic nuclei and star - forming galaxies at the relevant redshifts , the latter appear to be the more energetically plausible source for reionization ( e.g. bouwens et al . 2010 ) . however , a major uncertainty in the contribution of star forming galaxies is the poorly constrained fraction of ionizing photons that are able to leak out of such galaxies into the intergalactic medium . bouwens et al . ( 2010 ) estimate that an escape fraction of 20% ( 60% ) is required for consistency with the wmap results at the 2@xmath8 ( 1 @xmath8 ) level . many attempts have been made to measure this escape - fraction in star - forming galaxies at low - redshift ( leitherer et al . 1995 ; hurwitz et al . 1997 ; deharveng et al . 2001 ; heckman et al . 2001 - hereafter h01 ; bergvall et al . 2006 ; grimes et al . 2009 - hereafter g09 ) , intermediate redshift ( malkan et al . 2003 , siana et al . 2007,2010 ; cowie et al . 2009 ) and high- redshift ( steidel et al . 2001 ; giallongo et al . 2002 ; fernandez - soto et al . 2003 ; inoue et al . 2005 ; vanzella et al . 2010 ) . for the most part , these measurements have only yielded upper limits , although significant escaping lyman continuum emission has been detected in 17 out of a sample of 198 lyman break and lyman @xmath1 galaxies at @xmath9 ( iwata et al . 2009 ) , consistent with earlier results for a smaller sample of lyman break galaxies ( shapley et al . in contrast , vanzella et al . ( 2010 ) found evidence for escaping lyman continuum in only 1 of 100 lyman break galaxies at z @xmath6 4 . why do only a minority of the high - z galaxies and none of the low - z galaxies show a significant leakage of ionizing radiation ? in this regard , it is instructive to note that an hi column density of only 1.6 @xmath10 @xmath11 is required to produce unit optical depth at the lyman edge . the area - averaged column densities of cold gas in star - forming galaxies are much larger , ranging from @xmath12 @xmath11 in ordinary low - redshift galactic disks to as high as @xmath13 @xmath11 in local starbursts and high - z star - forming galaxies ( kennicutt 1998 ; genzel et al . the escape of ionizing radiation must then be determined by the topology of the interstellar medium , which will in turn be strongly influenced by feedback from massive stars . it may be that rather exceptional circumstances are required to produce the required channels of very low column - density neutral gas . such processes have been extensively studied from a theoretical perspective ovre the past decade ( e.g. dove et al . 2000 ; clarke & oey 2002 ; fujita et al . 2003 ; gnedin & kravtsov 2008 ; wise & cen 2009 ; razoumov & somer - larson 2010 ; yajima et al . 2010 ) . these considerations provide the motivation of the present paper : to find local examples of galaxies in which a potentially significant fraction of the ionizing radiation may be escaping , and to then exploit the relative ease of observing such bright local objects to gain insight into the physical processes that facilitate this escape . to be able to assess the cosmological implications of the results of such an approach , it is important to observe local galaxies that most closely resemble the type of uv - bright high - redshift galaxies that are candidates for the reionization of the universe . we began such an investigation in overzier et al . ( 2009 - hereafter o09 ) , where we reported on spitzer photometry , hst imaging , and sloan digital sky survey ( sdss ) optical spectroscopy of a sample of 31 lyman break analogs ( lbas ) . these lbas are a very rare population of local galaxies that strongly resemble the high - z lyman break galaxies ( heckman et al . 2005 ; hoopes et al . 2007 ; overzier at al . 2008;2010 ; basu- zych et al . 2007;2009 ) . they are defined to have a range in far - uv luminosity ( @xmath14 l@xmath4 ) and far - uv effective surface brightness ( @xmath15 l@xmath4 kpc@xmath16 ) that are similar to typical lyman break galaxies . we showed in o09 that the morphologies of lbas as revealed in multi - band hst images fall into two distinct categories . the first class ( 25 objects ) consists of galaxies in which multiple bright complexes of clumps and knots reside in a galaxy whose disturbed morphology is suggestive of a major merger or strong tidal interaction . the second class ( six objects ) consists of galaxies whose uv image is dominated by a single extremely massive ( @xmath17 one - to - several billion @xmath18 ) and compact ( radius of - order 100 pc ) central object ( a dominant central object , or dco ) . we pointed out that the star - formation rates derived for these six galaxies based on the extinction- corrected h@xmath1 luminosities were systematically smaller than those derived from the spitzer mid - ir luminosity ( by factors of @xmath6 2 to 5 ) , and speculated that a possible explanation was that these galaxies were leaking ionizing radiation into the intergalactic medium . in this paper we test this idea using new hubble space telescope ( hst ) far - uv observations with the cosmic origins spectrograph ( cos ) of a sample of eight lbas from o09 . we will combine these new data with archival far ultraviolet spectroscopic explorer ( fuse ) data for three other lbas and for fifteen more typical uv - bright local starbursts . we will describe the sample selection in section 2 and the observations and data analysis in section 3 . we will present our results and consider their interpretation in section 4 , and summarize these results and their implications in section 5 . we have previously used the hopkins ultraviolet telescope and fuse to search for the direct evidence of the escape of ionizing radiation from local starbursts ( leitherer et al . 1995 ; h01 ; g09 ) , but the modest sensitivity of these facilities was sufficient only to observe three lbas ( none of which contained a dco ) . with the recent installation of the cosmic origins spectrograph ( cos ) on the hubble space telescope ( hst ) , we are now in a position to obtain high - quality far - uv spectra of a significantly larger sample of these galaxies , including some with dcos . for the redshifts of the lbas in our hst program ( @xmath19 0.1 to 0.25 ) , the sensitivity of cos is poor in the rest - frame lyman continuum . the sensitivity is much better at longer wavelengths , so in this paper we follow h01 and g09 and use the residual intensity in the core of the strongest far - uv interstellar absorption - lines from the neutral phase of the interstellar medium as a probe of the intensity in the region below the lyman edge . we will describe this in more detail in section 3 below . this approach requires the observations to be obtained through an aperture that encloses the bulk of the starburst s emission ( so that a global constraint is obtained ) . it also requires that the width of instrumental spectral line - spread function is significantly smaller than the characteristic velocity dispersion in the starburst ( so that the interstellar absorption lines are well resolved ) . finally , the data need to have a high signal - to - noise ratio in the far- uv continuum so that useful constraints are derived . based on these considerations , we will consider two data sets in this paper . the first consists of the sample of eighteen galaxies studied by h01 and g09 . these are local starburst or star - forming galaxies observed with the far - ultraviolet spectroscopic explorer ( fuse - moos et al . 2000 ) , each having signal - to - noise better than 4.6 per 0.078 ( @xmath6 25 km s@xmath5 ) spectral element in the fuse lif1a channel ( see g09 ) . the observations and properties of this sample are described in table 1 . three of these galaxies are in fact lbas without a dco : mrk 54 ( deharveng et al . 2001 ) , haro 11 ( grimes et al . 2007 ) , and vv 114 ( grimes et al . 2006 ) . however , most of these galaxies have considerably lower uv luminosities and star - formation rates than the lbas . they also span a much broader range in galaxy mass and metallicity ( see g09 and o09 , and compare table 1 and 2 ) . the second is a sample of eight lbas observed with the cosmic origins spectrograph ( froning & green 2009 ) on the hubble space telescope ( program 11727 : pi t. heckman ) . these are members of a sample of 31 lbas with hst uv images discussed by o09 and were selected for spectroscopy based on a high uv flux through the cos aperture and a compact uv size ( so that the cos line - spread function is not significantly degraded ) . these observations and the properties of this sample are listed in table 2 and hst images are shown in figure 1 . in the discussion to follow we will refer to the eight lbas with hst cos data and the three lbas with fuse data as the lba sample . we will refer to the other fifteen galaxies with fuse data as the local starburst sample . the observations and data reduction of the fuse sample have been described in detail in h01 and g09 . we refer the reader to these papers . all the data were obtained through the 30 x 30 arcsec lwrs aperture , except for the cases of ngc 5253 and ngc 7714 ( which used the 4 x 20 arcsec mdrs aperture ) . as shown by g09 these apertures encompass most or all of the starburst in the far - uv . these spectra cover the observed wavelength range from 905 to 1187 ( see table 1 for the corresponding range in the rest - frame ) . depending on the angular size of the starburst in the far - uv , the instrumental spectral resolution is @xmath20 5000 to 14,000 , corresponding to a velocity dispersion of @xmath21 9 to 25 km s@xmath5 ( g09 ) . in all cases the interstellar absorption lines in the starbursts are well resolved . for the hst - cos sample we have used the cos g130 m and g160 m gratings to obtain spectra of our eight targets . as can be seen in figure 1 , the cos aperture encompasses most or all of the galaxy . we have retrieved these data from the hst mast archive after they have been processed through the standard cos pipeline . the merged spectra cover a range from about 1160 to 1780 with the corresponding range in the rest- frame wavelength given for each sample member in table 2 . as with the fuse data , the instrumental spectral resolution depends on the angular size of the far - uv continuum within the cos aperture . our hst uv images ( o09 ) yield sizes of 0.1 to 0.5 arcsec ( fwhm ) , and we estimate that the resulting spectral resolution is @xmath22 3000 to 13,000 , corresponding to a velocity dispersion of @xmath21 10 to 43 km s@xmath5 ( similar to the fuse data ) . for the new cos spectra we have fit each merged g130 m and g160 m spectrum with a starburst99 ( leitherer et al . 1999 ) model spectrum for a young stellar population with a metallcity similar to that of the corresponding lba ( table 2 ) . we did not attempt to obtain a rigorous best - fit , but simply adopted a standard chabrier ( 2003 ) imf and assumed a constant rate of star - formation . we varied the duration of the star - formation until we obtained ( by eye ) a satisfactory fit to the strong stellar wind features . we then took the ratio of the model and the data and fit the result with a low - order polynomial ( avoiding spectral regions with strong emission or absorption features ) . we then divided the data by the polynomial to obtain a normalized spectrum . an example of this is shown in figure 2 . the opacity of the interstellar medium to lyman continuum photons is due to two primary sources : the photoelectric opacity of the neutral hydrogen and the opacity of dust . the latter significantly affects the entire uv continuum in typical local starbursts and high - z galaxies . we will focus first on measuring constraints on the photoelectric opacity . later we will comment on the importance of dust . the only direct probe of the photoelectric opacity is the direct measurement of the relative intensity in the continuum shortward of the lyman edge . the fuse data provide clean access to this spectral region and adequate signal - to - noise in only 5 targets . g09 have used these data to derive upper limits to the escape of ionizing radiation for these galaxies ( see also deharveng et al . 2001 ; bergvall et al . the cos data do not sample this spectral region in any galaxy . thus , to gain insight into the escape of ionizing radiation for our two samples we will follow h01 and g09 and use as probes the residual intensity in the core of the strongest interstellar absorption lines that primarily arise in the neutral interstellar medium . for the fuse data the strongest such line is [email protected] and for the hst - cos data the strongest such line is [email protected] . thus in both samples we will use the same ionic species as a tracer of the neutral ism.1277.2 line should be comparable in strength to the two cii lines in this case , however it is extremely weak in our cos spectra and in the spectra of other local starburst galaxies ( e.g. heckman et al . 1998 ; leitherer et al . 2002 ; schwartz et al . 2006 ; leitherer et al . 2010 ) and high - z lyman break galaxies ( shapley et al . 2003 ; pettini et al . 2002 ; steidel et al . 2010 ; quider et al . 2009;2010 ; dessauges - zavadsky et al . ] we have measured the relative residual intensity in the core of the cii absorption line , following the same methodology as in h01 and g09 . the results are listed in tables 1 ( fuse ) and 2 ( cos ) . lccccccccc haro 11 & y & 6175 & 887 to 1163 & @xmath24 0.08 & @xmath24 0.006 & 26 & 10.2 & -180 & -450 + vv114 & y & 6016 & 887 to 1164 & @xmath24 0.02 & @xmath240.001 & 66 & 10.8 & -150 & -650 + ngc 1140 & n & 1501 & 900 to 1181 & @xmath24 0.04 & @xmath24 0.014 & 0.83 & 9.4 & -30 & -400 + sbs 0335 - 052 & n & 4043 & 893 to 1171 & @xmath24 0.18 & @xmath24 0.18 & 0.32 & 7.8 & 30 & -130 + tol 0440 - 381 & n & 12291 & 869 to 1140 & @xmath24 0.04 & @xmath24 0.016 & 5.0 & 10.0 & -120 & -450 + ngc 1705 & n & 632 & 903 to 1184 & @xmath24 0.01 & @xmath24 0.008 & 0.16 & 8.6 & -12 & + ngc 1741 & n & 4037 & 893 to 1171 & @xmath24 0.01 & @xmath24 0.002 & 6.0 & 9.7 & -35 & -500 + izw 18 & n & 751 & 903 to 1184 & @xmath24 0.1 & @xmath24 0.1 & 0.016 & 7.1 & -3 & -150 + ngc 3310 & n & 983 & 902 to 1183 & @xmath24 0.05 & @xmath24 0.005 & 2.8 & 9.8 & -170 & -600 + haro 3 & n & 963 & 902 to 1183 & @xmath24 0.005 & @xmath24 0.001 & 0.41 & 8.9 & 10 & -150 + ngc 3690 & n & 3121 & 896 to 1175 & @xmath24 0.01 & @xmath24 0.001 & 82 & 10.9 & -120 & -900 + ngc 4214 & n & 291 & 904 to 1186 & @xmath24 0.01 & @xmath24 0.002 & 0.13 & 9.0 & -40 & -250 + mrk 54 & y & 13450 & 866 to 1136 & @xmath24 0.09 & @xmath24 0.05 & 21 & 10.4 & -90 & -550 + m 83 & n & 517 & 903 to 1185 & @xmath24 0.06 & @xmath24 0.01 & 3.5 & 10.7 & 9 & + ngc 5253 & n & 405 & 904 to 1185 & @xmath24 0.06 & @xmath24 0.02 & 0.33 & 9.3 & 17 & + iras 19245 - 4140 & n & 2832 & 897 to 1176 & @xmath24 0.21 & @xmath240.11 & 2.1 & 9.1 & -60 & -350 + ngc 7673 & n & 3392 & 895 to 1174 & @xmath24 0.1 & @xmath24 0.017 & 4.8 & 10.0 & -80 & -350 + ngc 7714 & n & 2803 & 897 to 1176 & @xmath24 0.03 & @xmath24 0.01 & 6.9 & 10.2 & -60 & -500 + lccccccccccccc 0055 - 00 & n & 0.16744 & 970 to 1513 & @xmath240.05 & @xmath240.01 & 29 & 23 & 9.7 & -230 & -640 & -0.21 & -1.5 & -0.41 + 150 + 13 & n & 0.14668 & 1022 to 1572 & @xmath240.1 & @xmath240.01 & 56 & 19 & 10.3 & -150 & -480 & -0.78 & -1.6 & -0.30 + 0213 + 12 & y & 0.21902 & 945 to 1440 & 0.35 & 0.05 & 55 & 6.7 & 10.5 & -770 & -1500 & 0.36 & -0.9 & 0.05 + 0808 + 39 & y & 0.09123 & 1038 to 1607 & 0.3 & 0.12 & 11 & 3.7 & 9.8 & -680 & -1500 & 0.27 & -1.0 & 0.05 + 0921 + 45 & y & 0.23499 & 948 to 1430 & 0.3 & 0.04 & 84 & 26 & 10.8 & -650 & -1500 & 0.75 & -1.4 & -0.02 + 0926 + 45 & n & 0.18072 & 968 to 1515 & 0.6 : & 0.4 : & 10 & 13 & 9.1 & -300 & -700 & 0.14 & -2.3 & -0.60 + 0938 + 54 & n & 0.10208 & 1029 to 1614 & @xmath240.1 & @xmath240.05 & 13 & 13 & 9.4 & -90 & -610 & -0.17 & -2.1 & -0.50 + 2103 - 07 & y & 0.13689 & 1030 to 1585 & @xmath240.05 & @xmath240.01 & 63 & 41 & 10.9 & -720 & -1500 & 0.04 & -0.5 & 0.01 + before considering constraints on the escape of ionizing radiation , it is important to establish that an ionizing population of massive stars ( o stars ) is in fact present in our targets . the strongest spectroscopic signature of these stars is provided by the broad p - cygni profiles in the resonance transitions of highly ionized species that arise in the stellar winds ( e.g. robert et al . 1993 ; 2003 ) . the strongest such feature that is present in all our cos spectra is nv@xmath231240 . in figure 3 we show the nv profile for our eight targets and overplot the generic examples of starburst99 model spectra described above . we defer to a future paper a detailed discussion of the properties of the stellar populations in the lbas . here we merely point out that the far - uv spectra of the lbas are all consistent with a normal population of ionizing stars expected for a strong starburst . in figure 4 we show the [email protected] absorption - line profiles for our cos data on eight lbas . the corresponding profiles for the [email protected] line from the fuse data can be found in h01 and g09 . a striking result is that there is a significant residual intensity in the absorption - line profiles in four lbas ( lba0213 + 12 , lba0808 + 39 , lba0921 + 45 , and lba0926 + 45 ) . the first three objects all contain a dco . these results contrast strongly with the results in h01 and g09 in which the [email protected] absorption - line in the fuse data is black or nearly black at line center in all cases . to understand our approach in using the cii line as a probe of the lyman continuum , let us consider two limiting idealized cases . in the first , we assume a picket - fence model in which the far - uv source is surrounded by a population of clouds that are optically thick in both the absorption line and the lyman continuum and that cover a fraction @xmath25 of the sky as seen by the source . in this case the residual relative intensity at the core of the line ( @xmath26 ) is just @xmath27 . thus , a significant residual intensity in the line core implies a significant escape of lyman continuum radiation . in the second case we assume that the far - uv source is surrounded by a uniform shell with unit covering factor . as discussed in h01 , in this case the optical depth in the ism just below the lyman edge is related to the optical depth at the core of the absorption - line by : + 1 ) @xmath28 km s@xmath29 + where @xmath30 is the gas - phase carbon abundance scaled to the solar value of @xmath31 ( asplund et al . 2009 ) , we assume that cii is the dominant ionization state in the neutral gas , @xmath32 is the velocity dispersion in the neutral gas , and @xmath33 17 and 12 for the 1036.3 and 1334.5 lines respectively ( morton 2003 ) . since the optical depth at the lyman edge is over an order- of - magnitude larger than in the line core , in this idealized case a significant residual intensity in the core of the absorption lines is a necessary but not sufficient condition for the escape of ionizing radiation . to test these two models , we have used the set of siii absorption - lines available in the cos data . like cii , siii is the dominant ion of silicon in the neutral ism . by using lines that all arise in the same ionic species , but which span a range in oscillator strengths , a comparison of the absorption profiles among these lines allows us to estimate the optical depth and covering factor of the gas ( e.g. heckman et al . 2000 ; rupke et al . 2002 ; quider et al . 2009 ; erb et al . 2010 ) . in the case of optically - thick clouds that do not fully cover the far - uv source , the line profiles of the different transitions should look the same . in the case of a uniform shell , the residual intensity will increase with decreasing oscillator strength @xmath34 ( since @xmath35@xmath36@xmath34 ) . we will use the following transitions ( with the respective values of @xmath37@xmath34 from morton 2003 ) : [email protected] ( -1.32 ) , 1193.3 ( -1.65 ) , and 1190.4 ( -1.95).1301.2,1303.3 stellar photospheric features ( walborn et al . 1995 ; massa 1989 ) and the adjacent [email protected] interstellar line ) . ] for each galaxy we simultaneously fit a single gaussian profile to the three different siii lines , forcing the three lines to share a common centroid and width but allowing the normalization ( depth ) of each line to vary freely . in all four galaxies the results strongly favor the picket fence model with an implied covering factor of optically - thick clouds of about 40% for lba0213 + 12 and lba0926 + 45 , 50% for lba0808 + 39 , and 60% for lba0921 + 45 . we show these fits for the case of lba0921 + 45 in figure 5 . in principle , the sightlines that avoid the clouds could be optically - thin to the metal lines but still be optically - thick to the lyman continuum ( equation 1 ) . thus , confirmation that there is significant escaping ionizing radiation will require direct observations below the lyman edge in these cases . in the mean time we can consider whether there is other evidence for the escape of ionizing radiation in these four lbas . in o09 we highlighted the relative weakness of the extinction - corrected h@xmath1 emission - line in most of the lbas with dcos , and speculated that a significant fraction of the ionizing radiation might be escaping . this information is listed in table 2 where we compare the star - formation rates derived from the extinction - corrected h@xmath1 emission - line ( @xmath38 ) to that obtained from the sum of the total - ir and far- uv luminosity ( @xmath39 ) . it is noteworthy that three of the four objects above in which we have inferred an incomplete covering of the far - uv source by neutral clouds have the highest ratios of @xmath40 : lba0213 + 12 ( a ratio of 8.2 ) , lba0808 + 39 ( 3.0 ) , and lba0921 + 45 ( 3.2 ) . all three of these contain a dco . in contrast , the mean ratio of @xmath40 for the other five galaxies is 1.4 . _ thus , based on two independent lines of evidence , we suggest that there is a significant leakage of ionizing radiation from three of the eight lbas in our sample . _ how do these results compare to what is seen at high redshift ? there are four gravitationally - lensed high - z galaxies that have been observed with spectral resolution and signal - to - noise similar to our sample . ) , so that the line profiles are not well - resolved in individual objects ( and could miss gas with a high covering factor but low velocity dispersion ) . the stacking process itself could also smear out black , narrow components found in the spectra of individual galaxies . ] of these , the residual relative intensities of [email protected] , oi @xmath231302.2 , and siii @xmath23 1260.4 ( the strongest metal lines tracing the neutral ism ) are very small in ms 1512-cb58 ( pettini et al . 2002 ) , the cosmic eye ( quider et al . 2010 ) , and the 8 oclock arc ( dessauges - zavadsky et al . 2010 ) , but large ( @xmath640% ) in the cosmic horseshoe ( quider et al . in fact , the last object appears to a good example of the type of picket fence situation seen in the dcos in our sample . as discussed above , dust can also be a significant source of opacity to ionizing radiation in galaxies . following shapley et al . ( 2006 ) and g09 , we can distinguish between the absolute and relative escape fractions . the former is the actual fraction of ionizing photons that escape the galaxy , including the effect of dust . the latter neglects the effect of dust and is defined as the ratio of the fraction of escaping ionizing photons to escaping non - ionizing far - uv photons ( e.g. it measures only the effect of the photoelectric opacity of the gas ) . our measurements discussed above provide information about the relative escape fraction . given that the ratio of far - ir to far - uv fluxes for the galaxies with dcos are of - order ten , the implied absolute escape fractions will be proportionately smaller . we list our estimates for both the relative and absolute escape fractions in tables 1 ( fuse ) and 2 ( cos ) . as can be seen in figure 4 , the [email protected] absorption - line profiles in all the lbas are blueshifted with respect to the galaxy systemic velocity , implying a large - scale outflow of gas . further evidence for outflows is provided by the broad blue - asymmetric wings seen on the optical emission - line profiles ( o09 ) . the absorption - line profiles in figure 4 show evidence for exceptionally high outflow speeds in the dcos . to further investigate this , we turn to the si iii @xmath231206.5 feature . this is the strongest metal line that is accessed in all four dco spectra , and arises in the ionized gas . as shown in figure 6 , outflowing gas is detected at extraordinarily high velocities in all four galaxies with dcos . the flux - weighted line centroid is blueshifted by about 700 km s@xmath5 in these objects ( table 2 ) . it is difficult to determine the maximum outflow velocity because of blending with the ni@xmath231200 triplet , which lies @xmath61600 km s@xmath5 blueward of the siiii line . conservatively , the maximum outflow speed seen in the dcos reaches at least 1500 km s@xmath5 . these velocities are much higher than in the local starburst sample and in the lbas without a dco . this is shown in figure 7 where we plot the outflow speeds in the ionized gas in local starburst and lba samples _ vs. _ the star formation rate and galaxy mass ( see tables 1 and 2 ) . for galaxies with fuse data we use the [email protected] and/or [email protected] and/or [email protected] lines . the outflow velocities in the dcos are also significantly larger than those typically seen in high - z galaxies . steidel et al . ( 2010 ) find line centroids that are blueshifed on - average by 164 km s@xmath5 and maximum outflow speeds that are typically @xmath6 800 km s@xmath5 for galaxies with star formation rates of @xmath41 to @xmath42 year@xmath5 . outflows from intensely star - forming galaxies are believed to be produced as gas clouds are accelerated by the ram pressure of a hot and fast wind driven by the collective thermal / kinetic energy supplied by supernovae and massive stellar winds ( e.g. heckman et al . 2000 ; veilleux et al . 2005 ; marcolini et al . 2005 ; strickland & heckman 2010 ) , and/or by radiation pressure acting on dust ( murray et al . 2005 ; 2010 ) . can the unusually high velocities seen in the dcos be explained in this way ? we consider a simple idealized model of a cloud with a column density @xmath43 accelerated outward from an initial radius @xmath44 by the ram pressure of a wind that carries momentum at a rate @xmath45 into a solid angle @xmath46 . then the terminal velocity of this cloud will be : + 2 ) @xmath47 km s@xmath5 + here the momentum flux is in units of @xmath48 dynes , the initial radius is in units of 100 pc and the cloud column density is in units of @xmath12 @xmath11 . the dco star formation rates imply momentum fluxes of this order ( leitherer & heckman 1995 ) , while the hst images yield typical dco radii of 100 pc ( o9 ) . as discussed below in section 4.6 , a column density of @xmath12 @xmath11 is a reasonable estimate for the ionized gas in these objects . the above neglects gravitational forces ( cf . wang 1995 ; martin 2005 ; murray et al . 2010 ; strickland & heckman 2010 ) . this is justified here because the typical sizes and radii of the dcos imply virial velocities of only about 200 km s@xmath5 ( much smaller than the maximal outflow speeds we observe ) . note that murray et al . ( 2010 ) conclude that outflows driven out of massive star clusters by radiation pressure will only have outflow speeds of - order the local escape velocity . this is not the case in the dcos . these considerations imply that a galactic wind driven by an extreme starburst can in principle accelerate clouds to the high velocities that we see in the dcos . the significantly higher velocities in these galaxies compared to typical starbursts can be naturally explained by the combination of a large number of massive stars confined within an unusually small radius . this leads to a very large wind ram pressure and/or radiation pressure at the launch point of the clouds . an alternative form of extreme feedback could be provided by an agn . as discussed in overzier et al . ( 2009 ) the optical spectra of the lbas with dcos may imply a composite of both an intense starburst and an obscured ( type 2 ) agn . the cos data show that the far - uv radiation in these objects is dominated by light produced by hot massive stars ( figures 2 and 3 ) , which does not rule out the presence of a type 2 agn . recent observations with xmm - newton ( jia et al . 2010 ) show that lbas with such apparently composite optical spectra are also over - luminous in the 2 - 10 kev x - ray band by factors of roughly three to thirty compared to pure starbursts of the same far - ir luminosity . they conclude that it is likely that at least some of these objects do harbor an agn , but that the overall bolometric luminosity of the galaxy is primarily due to an intense starburst . could a type 2 agn be responsible for the unusually high outflow speeds and/or the apparently low covering factor of neutral gas due to a highly ionized state of the interstellar medium ? very high outflow speeds ( many thousands of km s@xmath5 ) are commonly observed in _ unobscured _ ( type 1 ) agn ( e.g. rupke et al . 2005 ; krug et al . however these same authors show that outflows similar to those we see in the dcos ( maximium outflow speeds of at least 1500 km s@xmath5 ) are extremely rare in composite obscured - agn / starburst systems and not seen at all in `` pure '' obscured agn . they conclude that the agn can drive very high - speed outflows , but these flows are confined to a small region very near the black hole ( and are hence undetectable in type 2 agn ) . they also conclude that the agn plays little or no role in driving the large - scale outflows seen in the composite systems . it therefore seems unlikely that the high velocity outflows in the dcos are driven by an obscured agn . tremonti et al . ( 2007 ) have detected outflows at velocities ranging from 500 to 2000 km s@xmath5 in massive post - starburst galaxies at z @xmath6 0.6 . they speculate that such high velocities may require an agn - driven outflow . it is also possible that the these objects are more massive versions of the dcos in this paper . this could be confirmed by hst imaging of their galaxies . the properties of the ly@xmath1 profile are highly sensitive to the kinematics and distribution of the outflowing gas and dust in star - forming galaxies ( e.g. steidel et al . 2010 ; hansen & oh 2006 ; verhamme et al . 2006 , 2008 ; kornei et al . in figure 8 we plot the ly@xmath1 profiles for the eight lbas with cos data , and highlight one very suggestive trend . in the five cases in which the inferred relative escape fraction is small , there is very little ( if any ) ly@xmath1 emission blueward of the systemic velocity . instead , the profiles show ly@xmath1 emission redward of the systemic velocity and either pure absorption or a mix of absorption and weak emission to the blue . such profiles are typical of high - z star forming galaxies ( e.g. shapley et al . 2003 ; steidel et al . in contrast , for the three objects in which we infer a high relative escape fraction ( lba0213 + 12 , 0808 + 39 , and 0921 + 45 ) a significant fraction of the ly@xmath1 emission is blueshifted with respect to the systemic velocity . to quantify this , we have measured the net equivalent widths ( emission - absorption ) blueward and redward of the systemic velocity in these eight lbas . we define net emission ( absorption ) to have a positive ( negative ) equivalent width . we then take the ratio of the red and blue equivalent widths ( @xmath49 ) . we list the results in table 2 . for the five galaxies with low inferred escape fractions @xmath49 is either negative ( redshifted emission and blueshifted absorption ) or small and positive ( strong redshifted emission and weak blueshifted net emission ) : lba0055- 00 ( @xmath50 -0.21 ) , 0150 + 13 ( -0.78 ) , 0926 + 45 ( 0.14 ) , 0938 + 54 ( -0.17 ) , and 2103 - 07 ( 0.04 ) . for the three galaxies with high inferred escape fractions @xmath49 is of - order one and positive ( similar amounts of red- and blue - shifted emission ) : 0213 + 12 ( @xmath50 0.36 ) , 0808 + 39 ( 0.27 ) , and 0921 + 45 ( 0.75 ) . profiles with significant blueshifted emission are uncommon but not unknown in star- forming galaxies at @xmath19 2 - 3 . verhamme et al . ( 2008 ) discuss two such cases and interpret them as arising due to radiative transfer effects in a static medium . this model can not be correct for the three dcos in figure 8 because the metal lines demonstrate that the gas is rapidly outflowing ( figures 4 and 6 ) . erb et al . ( 2010 ) discuss the case of a high - z galaxy with both an outflow detected in the metal lines and a significant amount of blueshifted ly@xmath1 emission . they argue that in this case the optical depth to ly@xmath1 photons is unusually low in the foreground outflowing material and that this is naturally related to the low covering factor of neutral gas they infer in this object ( precisely the situation we see in our three galaxies ) . it is important to note that the observed uv spectral energy distributions ( seds ) of the dcos are significantly redder than the intrinsic sed of a starburst ( overzier et al . 2011 , and see table 2 ) . for a normal starburst or smc dust attenuation law ( calzetti 2001 ; leitherer et al , 2002 ) these seds imply roughly two magnitudes of dust extinction along the line - of- sight to the dco in the far - uv . this then raises a very interesting question . how can there be this much dust when the observations of interstellar absorption - lines from the neutral gas imply that such gas only covers a fraction of the far - uv source ? the answer must be that this dust is associated with _ ionized _ gas that covers most or all of the dco . this picture requires that the cores of the strongest absorption - lines due to the ionized gas should be nearly black ( unlike the case of lines from the neutral phase ) . examination of the profiles in figure 6 shows that this is the case in lba0808 + 39 ( a relative residual intensity in the [email protected] line of 10% ) , lba0921 + 45 ( 7% ) , and lba2103- 07 ( @xmath243% ) . in the case of lba0213 + 12 , the relative residual intensity in this line is high ( @xmath640% ) . however , the redshift of this galaxy is large enough to shift the [email protected] line into the cos g130l spectrum . this should be the strongest interstellar line due to ionized gas , and as such provides the most sensitive probe . although the data are noisy , the relative residual intensity in this line is small ( 10% or less ) . thus , in this galaxy the siiii line is optically thin , while the ciii line is optically thick . we therefore conclude that the ionized gas has a near - unit covering factor in the dcos . one plausible physical model would be that the neutral gas represents the denser cores of clouds being accelerated by the wind , while the ionized gas represents a lower density halo of gas and dust surrounding the cloud core and which is being ablated by the wind and photoionized by the intense radiation field ( e.g. marcolini et al . 2005 ) . can the dco photoionize a column of gas with the required amount of dust ? elementary considerations of photoionization equilibrium imply that a source of radiation can photoionize a stromgren slab with a column density given by : + 3 ) @xmath51 @xmath11 + where @xmath52 is the dimensionless ionization parameter defined as the ratio of the density of ionizing photons to electrons in the slab . in order for a significant fraction of ionizing photons to escape , the actual gas column must be less than this value ( to provide matter - bounded conditions ) . the optical depth due to dust in the far - uv ( 1500 ) is related to the gas column density by : + 4 ) @xmath53 @xmath11 + where the coefficients refer to an smc ( starburst ) dust attenuation law ( leitherer et al . 2002 ) and we assume that the dust - to - gas ratio is proportional to the metallicity . the lbas with dcos all have gas - phase oxygen abubdances close to solar ( o09 ) and the line - of - sight reddening implies @xmath54 2 . thus , to explain the line - of - sight reddening towards the dcos in a matter - bounded situation we need @xmath55 . alternatively , the optical depth at the lyman edge due to neutral hydrogen in ( mostly ) ionized gas with a column density @xmath56 is given by : + 5 ) @xmath57 + for @xmath58 @xmath11 , we require @xmath59 for @xmath35@xmath60 . is this plausible ? heckman , armus , & miley ( 1990 ) and lehnert & heckman ( 1996 ) showed that the observed radial gas density profiles seen in the optical emission - line nebulae in starburst winds are consistent with theoretical expectations for photoionized clouds exposed to the ram - pressure of the hot wind fluid ( chevalier & clegg 1985 ) . it is straightforward to show that in this case the ionization parameter in the clouds is given by : + 6 ) @xmath61 + where @xmath62 is the wind momentum flux in units of 10@xmath63 dynes and @xmath64 is the starburst ionizing photon production rate ( in units of 10@xmath65 sec@xmath66 ) . the ionizing photons are produced by the most massive stars with lifetimes @xmath24 4 myr , while the momentum in the starburst wind derives primarily from supernovae ( whose progenitors have lifetimes of @xmath24 40 myr ) . in equilibrium ( e.g. a constant rate of star formation for longer than 40 myr ) , @xmath67 0.5 and so @xmath68 . more generally , this ratio will be time dependent . for example , values of @xmath69 are associated with constant star formation for a duration less than 20 myr and for an age @xmath24 4 myr in an instantaneous burst . these ages are consistent with current constraints on the dcos ( o09 and see figure 3 ) . we therefore conclude that it is feasible for the dcos to photoionize a matter - bounded column of gas that is sufficient to contain enough dust to produce the observed reddening and the inferred attenuation of the far - uv continuum and yet have a photoelectric opacity at the lyman edge that is less than one . one of the major goals in cosmology is to understand the process by which the universe was reionized . the most plausible source for the ionizing photons is an early generation of hot massive stars in forming galaxies . the primary uncertainty at present is the unknown fraction of such ionizing photons that are able to escape from the galaxies and reach the neutral hydrogen in the inter - galactic medium . the column densities of gas associated with such intense star formation correspond to extremely high optical depths in the lyman continuum . it seems nearly certain that some form of feedback from the massive stars is required to produce the pathways through which the ionizing photons can escape . our understanding of how this might occur would be greatly improved if we could find local examples of galaxies from which significant ionizing radiation is escaping , and could then investigate in some detail the physical processes that enable this to occur . this will be most germane to the early universe if these local galaxies are good analogs to intensely star - forming galaxies at high - redshift . accordingly , in this paper we used the hubble space telescope cosmic origins spectrograph ( cos ) and the far - ultraviolet spectroscopic explorer ( fuse ) to investigate the far - uv properties of a sample of eleven lyman break analogs ( lbas ) . these rare local objects are very similar to high - z lyman break galaxies in all their properties studied to date ( heckman et al 2005 ; hoopes et al 2007 ; basu - zych et al.2007 , 2009 ; goncalves et al . 2010 ; overzier et al . 2009 - o09 ; overzier et al . 2008 , 2010 , 2011 ) . we have compared these rare objects to a sample of fifteen typical uv - bright local starbursts . following heckman et al . 2001 ( h01 ) and grimes et al . 2009 ( g09 ) we have used the relative residual intensity in the cores of the strongest interstellar absorption - lines arising in the neutral interstellar medium ( the [email protected] and [email protected] lines ) as a probe of the effects of the photoelectric opacity of the ism on the lyman continuum . this approach is made possible by the high spectral resolution ( @xmath70 ) and good signal - to - noise in our data . we find that there is a significant residual intensity ( @xmath6 30 to 50% ) in the cii line core in four out of eleven of the lbas but in none of the fifteen typical local starbursts . we show that in all four of these cases the residual intensity reflects an incomplete covering of the far - uv source by optically thick clouds ( a picket fence model ) . independent evidence for a significant escape of ionizing radiation is provided by the anomalously weak h@xmath1 emission ( relative to the sum of the far - uv and far - ir emission ) in three of these four galaxies . _ it is therefore plausible that a significant fraction of the lyman continuum is escaping in three of the lbas in our sample_. assuming that the residual intensity in these lines does indicate the escape of ionizing radiation , then following shapley et al . ( 2006 ) and g09 , the relative escape fraction in these objects ( defined as the ratio of fractions of escaping ionizing to non - ionizing far- uv photons ) is @xmath630 to 40% . after accounting for the effect of dust attenuation , the absolute escape fractions are @xmath6 3 to 10% . we have also examined the profiles of the ly@xmath1 line in the cos data of the eight lbas for additional clues . we find that there is strong emission both blueward and redward of the systemic velocity in the three objects in which a high escape fraction is likely . this strongly suggests an incomplete covering of the fuv source by neutral clouds ( in agreement with the properties of the metal lines ) . the other five cases show the profiles typically seen in high - z galaxies : ly@xmath1 emission redward of the systemic velocity and either pure absorption or a mix of absorption and weak emission blueward . in all three cases with high inferred relative escape fractions the hst images reveal that most of the far - uv emission originates in a single dominant central object ( dco ) . these are highly compact ( radii @xmath2 pc ) and massive ( one to a few @xmath71 m@xmath4 ) young starbursts ( o09 ) . the interstellar absorption lines in all four of the lbas with dcos show extremely high outflow speeds : the line centroids are blueshifted by about 700 km s@xmath5 and the gas reaches maximum outflow velocities of at least 1500 km s@xmath5 . these are significantly higher than the velocities seen in other low redshift starbursts or in typical high redshift galaxies . we have shown that these high speeds can be explained in principle by the acceleration of gas clouds in a starburst - driven wind . the exceptionally high velocities would be due to a large amount of momentum deposition in such an unusually small region , leading to an extremely high wind ram pressure and/or radiation pressure at the launch point for the clouds . the far - uv emission from the dcos is significantly reddened and attenuated by dust . we have argued that this dust must reside in the ionized gas and have shown that the ionized gas fully covers the far - uv source ( the dco ) . this situation is made possible by the intense ionizing radiation field in these galaxies ( leading to significant column densities of dusty ionized gas that is optically thin in the lyman continuum ) . based on the above considerations , it is tempting to link the escape of ionizing radiation to the extreme feedback associated with the dcos . the combination of a high rate of energy deposition ( kinetic energy and ionizing radiation ) within such a small volume may be the key to excavating channels through the interstellar medium and also intensely irradiating these channels so that they are fully ionized and translucent to ionizing radiation . it is interesting that only 19% ( @xmath72 ) of the sample of lbas imaged by o09 contain a dco . together with the present results this would suggest that large relative escape fractions are present in only 13% of lbas . this is similar to the fractions seen in high- redshift galaxies ( shapley et al . 2006 ; iwata et al . 2009 ) . could the presence or absence of dcos in these galaxies explain these results as well ? it is important to note that in the sample of lbas studied by overzier et al . ( 2009 ) , the dcos were preferentially found in the most massive galaxies ( m@xmath73 to @xmath74 m@xmath4 ) . galaxies this massive are exceedingly rare at @xmath75 7 , when the universe was reionized , and so the dco host galaxies will not resemble the population of low - mass systems believed to be responsible for producing most of the ionizing radiation at these early epochs ( e.g. bouwens et al . 2009 ) . we are instead suggesting that objects similar to the dcos in this paper may exist in these early galaxies and facilitate the escape of ionizing radiation . a number of theoretical papers over the past decade have investigated the processes that regulate the escape of lyman continuum radiation from galaxies ( e.g. dove et al . 2000 ; clarke & oey 2002 ; fujita et al . 2003 ; gnedin & kravtsov 2008 ; wise & cen 2009 ; razoumov & somer - larson 2010 ; yajima et al . 2010 ) . these studies are at least qualitatively consistent with what we have observed in the dcos . finally , we note that o09 have shown that the masses and sizes of the dcos are consistent with the regions of excess mass ( cusps ) seen in the centers of typical present - day m@xmath76 elliptical galaxies ( hopkins et al . if so , then such galaxies may have all passed through an early stage during which they were sources of a significant amount of ionizing radiation that was able to reach the intergalactic medium . asplund , m. grevasse , n. , sauval , a. , & scott , p. 2009 , ara&a , 47 , 481 basu - zych , a. et al . 2007 , apjs , 173 , 457 basu - zych , a. et al . 2009 , apj , 699 , l118 bell , e. & de jong , r. 2001 , apj , 550 , 212 bergvall , n . , zackrisson , e . , andersson , b .- g . , arnberg , d . , masegosa , j . , & ostlin , g . 2006 , a&a , 448 , 513 bouwens et . 2010 , arxiv1006.4360 calzetti , d. 2001 , pasp , 113 , 1449 chabrier , g. 2003 , pasp , 115 , 763 clarke , c. , & oey , m.s . 2002 , mnras , 337 , 1299 cowie , l.l . , barger , a.j . , & trouille , l . 2009 , 697 , l122 deharveng , j .- m . , buat , v . , lebrun , v . , milliard , b . , kunth , d . , shull , j.m . , & gry , c . 2001 , a&a , 375 , 805 dessauges - zavadsky , m . , dodorico , s . , schaerer , d . , modigliani , a . , tapken , c . , & vernet , j . 2010 , a&a , 510 , 26 dove , j. , shull , j.m . , & ferrara , a. 2000 , apj , 531 , 846 dunkley , j. , et al . 2009 , apjs , 180 , 306 erb , d , pettini , m. , shapley , a. , steidel , c. , law , d. , reddy , n. 2010 , apj , 719 , 1168 fan , x. et al . 2006 aj , 32 , 1117 fernandez - soto , a. , lanzetta , k , & chen , h .- w . 2003 , mnras , 342 , 1215 froning , c. & green , j. 2009 , apss , 320 , 181 fujita , a. , martin , c. , maclow , m. , & abel , t. 2003 , apj , 599 , 50 genzel , r , et al . 2010 , mnras , 407 , 2091 giallongo , e. , cristiani , s. , dodorico , s. , & fontana , a. 2002 , apj , 568 , l9 gnedin , n. , kravtsov , a. , & chen , h .- w . 2008 , apj , 672 , 763 grimes , j. , heckman , t. , hoopes , c. , strickland , d. , aloisi , a. , meurer , g. , & ptak , a. 2006 , apj , 648 , 310 grimes , j. , heckman , t. , strickland , d. , dixon , w. v. , sembach , k. , overzier , r. , hoopes , c. , aloisi , a. , & ptak , a. 2007 , apj , 668 , 891 grimes , j. , heckman , t. , aloisi , a. , calzetti , d. , leitherer , c. , martin , c. l. , meurer , g. , sembach , k. , & strickland , d. 2009 , apjs , 181 , 272 hansen , m. & oh , s .- 2006 , mnras , 367 , 979 heckman . t. , armus , l. , & miley , g. 1990 , apjs , 74 , 833 heckman , t. , robert , c. , leitherer , c. , garnett , d. , & van der rydt , f. 1998 , apj , 503 , 646 heckman , t. , lehnert , m. , strickland , d. , & armus , l. 2000 , apj , 129 , 493 heckman , t. , sembach , k. , meurer , g. , leitherer , c. , calzetti , d. , & martin , c. l 2001 , 558 , 56 heckman , t. et al . 2005 , apj , 619 , l35 hoopes , c. et al . 2007 , apjs , 173 , 441 hopkins , p. , hernquist , l. , cox , t. , keres , d. , & wuyts , s. 2009 , 691 , 1424 hurwitz , m. , jelinsky , p. , & dixon , w. v. 1997 , 481 , l31 inoue , a. , iwata , i. , deharveng , j .- m . , buat , v. , & burgarella , d. 2005 , a&a , 435 , 471 iwata , i. et al . 2009 , apj , 692 , 1287 jia , j .- j . , ptak , a. , heckman , t. , overzier , r. , hornschemeier , a. , & lamassa , s. 2010 , submitted to apj kennicutt , r. 1998 , apj , 498 , 541 kornei , k. , shapley , a. , erb , d. , steidel , c. , reddy , n. , pettini , m. , & bogosavljevic , m. 2010 , apj , 711 , 693 krug , h. , rupke , d. , & veilleux , s. 2010 , apj , 708 , 1143 lehnert , m. & heckman , t. 1996 apj , 462 , 651 leitherer , c. , ferguson , h. , heckman , t. , & lowenthal , j. 1995 , apj , 454 , l19 leitherer , c. , schaerer , d. , goldader , j. , gonzlez delgado , r. , robert , c. , kune , d. , de mello , d. , devost , d. , & heckman , t. 1999 , apjs , 123 , 3 leitherer , c. , li , i .- h . , calzetti , d. , heckman , t. 2002 , apjs , 104 , 303 leitherer , c. , tremonti , c. , heckman , t. , & calzetti , d. 2010 , arxiv1011.0385 malkan , m , webb , w. , & konopacky , q. 2003 , 598 , 878 marcolini , a. , strickland , d. , dercole , a. , heckman , t. , & hoopes , c. 2005 , mnras , 362 , 626 massa , d. 1989 , a&a , 224 , 131 martin , c. 2005 , apj , 621 , 227 moos , h .- w . 2000 , 538 , l1 morton , d. 2003 , apjs , 149 , 205 murray , n. , quataert , e. , & thompson , t. 2005 , apj , 618 , 569 murray , n. , menard , b. , & thompson , t. 2010 , arxiv1005.4419 pettini , m. , rix , s. , steidel , c. , adelberger , k. , hunt , m. , & shapley , a. 2002 , apj , 569 , 742 overzier , r. et al . 2008 , apj , 677 , 37 overzier , r. et al . 2009 , apj , 706 , 203 overzier , r. , heckman , t. , schiminovich , d. , basu - zych , a. , goncalves , t. , martin , d. c. , & rich , r. m. 2010 , apj , 710 , 979 overzier , r. et al . 2011 , apj , 726 , l7 quider , a. , pettini , m. , shapley , a. , & steidel , c. 2009 , mnras , 398 , 1263 quider , a. , shapley , a. , pettini , m. , steidel , c. , & stark , d. 2010 , mnras , 402 , 1467 razoumov , a. , & somer - larson , j. 2010 , apj , 710 , 1239 robert , c. leitherer , c. , & heckman , t. 1993 , apj , 418 , 749 robert , c. pellerin , a. , aloisi , a. , leitherer , c. , hoopes , c. , & heckman , t. 2003 , apjs , 144 , 21 rupke , d. , veilleux , s. , & sanders , d. 2002 , apj , 570 , 588 rupke , d. , veilleux , s. , & sanders , d. 2005 , apj , 632 , 751 schwartz , c. , martin , c. , chandar , r. , leitherer , c. , heckman , t. , & oey , m.s . 2006 , apj , 646 , 858 shapley , a. , steidel , c. , pettini , m. , & adelberger , k. 2003 , apj , 588 , 65 shapley , a. , steidel , c. , pettini , m. , adelberger , k. , & erb , d. 2006 , apj , 651 , 688 siana et al . 2007 , apj . 668 , 62 siana , b. et al . 2010 , apj , 723 , 241 steidel , c. , pettini , m. , & adelberger , k. 2001 , apj , 546 , 665 steidel , c. , erb , d. , shapley , a. , pettini , m. , reddy , n. , bogosavljevic , m. , rudie , g. , & rakic , o. 2010 , 717 , 289 tremonti , c. , moustakas , j. , & diamond - stanic , a. 2007 , apj , 663 , 77 veilleux , s. , cecil , g. , & bland - hawthorn , j. 2005 , ara&a , 43 , 769 vanzella , e. et al . 2010 , arxiv1009.1140 verhamme , a. , schaerer , d. , & maselli , a. 2006 , 460 , 397 verhamme , a. , schaerer , d. , atek , h. , & tapken , c. 2008 , a&a , 491 , 89 walborn , n. , parker , j. , & nichols , j. 1995 , vizier on - line data catalog : iii/188 wang , b. 1995 , apj , 444 , 590a wise , j. , & cen , r. 2009 , apj , 693 , 984 yajima , h. , choi , j .- h . , & nagamine , k. 2010 , arxiv1002.33346

since the work of kaligosi and sanders ( 2006 ) , it is well - known that quicksort which is commonly considered as one of the fastest in - place sorting algorithms suffers in an essential way from branch mispredictions . we present a novel approach to address this problem by partially decoupling control from data flow : in order to perform the partitioning , we split the input in blocks of constant size ( we propose 128 data elements ) ; then , all elements in one block are compared with the pivot and the outcomes of the comparisons are stored in a buffer . in a second pass , the respective elements are rearranged . by doing so , we avoid conditional branches based on outcomes of comparisons at all ( except for the final insertionsort ) . moreover , we prove that for a static branch predictor the average total number of branch mispredictions is at most for some small depending on the block size when sorting elements . our experimental results are promising : when sorting random integer data , we achieve an increase in speed ( number of elements sorted per second ) of more than 80% over the gcc implementation of c++ ` std:\!:sort ` . also for many other types of data and non - random inputs , there is still a significant speedup over ` std:\!:sort ` . only in few special cases like sorted or almost sorted inputs , ` std:\!:sort`can beat our implementation . moreover , even on random input permutations , our implementation is even slightly faster than an implementation of the highly tuned super scalar sample sort , which uses a linear amount of additional space .

1604.06697

sorting a sequence of elements of some totally ordered universe remains one of the most fascinating and well - studied topics in computer science . moreover , it is an essential part of many practical applications . thus , efficient sorting algorithms directly transfer to a performance gain for many applications . one of the most widely used sorting algorithms is quicksort , which has been introduced by hoare in 1962 @xcite and is considered to be one of the most efficient sorting algorithms . for sorting an array , it works as follows : first , it chooses an arbitrary pivot element and then rearranges the array such that all elements smaller than the pivot are moved to the left side and all elements larger than the pivot are moved to the right side of the array this is called _ partitioning_. then , the left and right side are both sorted recursively . although its average number of comparisons is not optimal @xmath3 vs. @xmath4 for mergesort , its over - all instruction count is very low . moreover , by choosing the pivot element as median of some larger sample , the leading term @xmath5 for the average number of comparisons can be made smaller even down to @xmath6 when choosing the pivot as median of some sample of growing size @xcite . other advantages of quicksort are that it is easy to implement and that it does not need extra memory except the recursion stack of logarithmic size ( even in the worst case if properly implemented ) . a major drawback of quicksort is its quadratic worst - case running time . nevertheless , there are efficient ways to circumvent a really bad worst - case . the most prominent is introsort ( introduced by musser @xcite ) which is applied in gcc implementation of ` std:\!:sort ` : as soon as the recursion depth exceeds a certain limit , the algorithm switches to heapsort . another deficiency of quicksort is that it suffers from branch mispredictions ( or branch misses ) in an essential way . on modern processors with long pipelines ( 14 stages for intel haswell , broadwell , skylake processors for the older pentium 4 processors even more than twice as many ) every branch misprediction causes a rather long interruption of the execution since the pipeline has to be filled anew . in @xcite , kaligosi and sanders analyzed the number of branch mispredictions incurred by quicksort . they examined different simple branch prediction schemes ( static prediction and 1-bit , 2-bit predictors ) and showed that with all of them , quicksort with a random element as pivot causes on average @xmath7 branch mispredictions for some constant @xmath8 ( resp . @xmath9 , @xmath10 ) . in particular , in quicksort with random pivot element , every fourth comparison is followed by a mispredicted branch . the reason is that for partitioning , each element is compared with the pivot and depending on the outcome either it is swapped with some other element or not . since for an optimal pivot ( the median ) , the probability of being smaller the pivot is @xmath11 , there is no way to predict these branches . kaligosi and sanders also established that choosing skewed pivot elements ( far off the median ) might even decrease the running time because it makes branches more predictable . this also explains why , although theoretically larger samples for pivot selection were shown to be superior , in practice the median - of three variant turned out to be the best . in @xcite , the skewed pivot phenomenon is confirmed experimentally . moreover , in @xcite , precise theoretical bounds on the number of branch misses for quicksort are given establishing also theoretical superiority of skewed pivots under the assumption that branch mispredictions are expensive . in @xcite brodal and moruz proved a general lower bound on the number of branch mispredictions given that every comparison is followed by a conditional branch which depends on the outcome of the comparison . in this case there are @xmath12 branch mispredictions for a sorting algorithm which performs @xmath13 comparisons . as elmasry and katajainen remarked in @xcite , this theorem does not hold anymore if the results of comparisons are not used for conditional branches . indeed , they showed that every program can be transformed into a program which induces only a constant number of branch misses and whose running time is linear in the running time of the original program . however , this general transformation introduces a huge constant factor overhead . still , in @xcite and @xcite elmasry , katajainen and stenmark showed how to efficiently implement many algorithms related to sorting with only few branch mispredictions . they call such programs _ lean_. in particular , they present variants of mergesort and quicksort suffering only very little from branch misses . their quicksort variant ( called tuned quicksort , for details on the implementation , see @xcite ) is very fast for random permutations however , it does not behave well with duplicate elements because it applies lomuto s uni - directional partitioner ( see e.g. @xcite ) . another development in recent years is multi - pivot quicksort ( i.e.several pivots in each partitioning stage @xcite ) . it started with the introduction of yaroslavskiy s dual - pivot quicksort @xcite which , surprisingly , was faster than known quicksort variants and , thus , became the standard sorting implementation in oracle java 7 and java 8 . concerning branch mispredictions all these multi - pivot variants behave essentially like ordinary quicksort @xcite ; however , they have one advantage : every data element is accessed only a few times ( this is also referred to as the number of _ scans _ ) . as outlined in @xcite , increasing the number of pivot elements further ( up to 127 or 255 ) , leads to super scalar sample sort , which has been introduced by sanders and winkel @xcite . super scalar sample sort not only has the advantage of few scans , but also is based on the idea of avoiding conditional branches . indeed , the correct bucket ( the position between two pivot elements ) can be found by converting the results of comparisons to integers and then simply performing integer arithmetic . in their experiments sanders and winkel show that super scalar sample sort is approximately twice as fast as quicksort ( ` std:\!:sort ` ) when sorting random integer data . however , super scalar sample sort has one major draw - back : it uses a linear amount of extra space ( for sorting @xmath2 data elements , it requires space for another @xmath2 data elements and additionally for more than @xmath2 integers ) . in the conclusion of @xcite , kaligosi and sander raised the question : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ however , an in - place sorting algorithm that is better than quicksort with skewed pivots is an open problem . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ( here , in - place means that it needs only a constant or logarithmic amount of extra space . ) in this work , we solve the problem by presenting our block partition algorithm , which allows to implement quicksort without any branch mispredictions incurred by conditional branches based on results of comparisons ( except for the final insertionsort also there are still conditional branches based on the control - flow , but their amount is relatively small ) . we call the resulting algorithm blockquicksort . our work is inspired by tuned quicksort from @xcite , from where we also borrow parts of our implementation . the difference is that by doing the partitioning block - wise , we can use hoare s partitioner , which is far better with duplicate elements than lomuto s partitioner ( although tuned quicksort can be made working with duplicates by applying a check for duplicates similar to what we propose for blockquicksort as one of the further improvements in section [ sec : improvments ] ) . moreover , blockquicksort is also superior to tuned quicksort for random permutations of integers . [ [ our - contributions ] ] our contributions * we present a variant of the partition procedure that only incurs few branch mispredictions by storing results of comparisons in constant size buffers . * we prove an upper bound of @xmath14 branch mispredictions on average , where @xmath15 for our proposed block size ( theorem [ thm : few_branches ] ) . * we propose some improvements over the basic version . * we implemented our algorithm with an ` stl`-style interface . * we conduct experiments and compare blockquicksort with ` std:\!:sort ` , yaroslavskiy s dual - pivot quicksort and super scalar sample sort on random integer data it is faster than all of these and also katajainen et al.s tuned quicksort . [ [ outline ] ] outline section [ sec : prelims ] introduces some general facts on branch predictors and mispredictions , and gives a short account of standard improvements of quicksort . in section [ sec : partition ] , we give a precise description of our block partition method and establish our main theoretical result the bound on the number of branch mispredictions . finally , in section [ sec : experiments ] , we experimentally evaluate different block sizes , different pivot selection strategies and compare our implementation with other state of the art implementations of quicksort and super scalar sample sort . logarithms denoted by @xmath16 are always base 2 . the term _ average case _ refers to a uniform distribution of all input permutations assuming all elements are different . in the following ` std:\!:sort`always refers to its gcc implementation . [ [ branch - misses ] ] branch misses branch mispredictions can occur when the code contains conditional jumps ( i.e. _ if _ statements , loops , etc . ) . whenever the execution flow reaches such a statement , the processor has to decide in advance which branch to follow and decode the subsequent instructions of that branch . because of the length of the pipeline of modern microprocessors , a wrong predicted branch causes a large delay since , before continuing the execution , the instructions for the other branch have to be decoded . [ [ branch - prediction - schemes ] ] branch prediction schemes precise branch prediction schemes of most modern processors are not disclosed to the public . however , the simplest schemes suffice to make blockquicksort induce only few mispredictions . the easiest branch prediction scheme is the _ static predictor _ : for every conditional jump the compiler marks the more likely branch . in particular , that means that for every _ if _ statement , we can assume that there is a misprediction if and only if the _ if _ branch is not taken ; for every _ loop _ statement , there is precisely one misprediction for every time the execution flow reaches that loop : when the execution leaves the loop . for more information about branch prediction schemes , we refer to ( * ? ? ? * section 3.3 ) . [ [ how - to - avoid - conditional - branches ] ] how to avoid conditional branches the usual implementation of sorting algorithms performs conditional jumps based on the outcome of comparisons of data elements . there are at least two methods how these conditional jumps can be avoided both are supported by the hardware of modern processors : * conditional moves ( ` cmovcc ` instructions on x86 processors ) or , more general , conditional execution . in c++ compilation to a conditional move can be ( often ) triggered by + .... i = ( x < y ) ? j : i ; .... * cast boolean variables to integer ( ` setcc ` instructions x86 processors ) . in c++ : + .... int i = ( x < y ) ; .... also many other instruction sets support these methods ( e.g. arm @xcite , mips @xcite ) . still , the intel architecture optimization reference manual @xcite advises only to use these instructions to avoid unpredictable branches ( as it is the case for sorting ) since correctly predicted branches are still faster . for more examples how to apply these methods to sorting , see @xcite . [ [ quicksort - and - improvements ] ] quicksort and improvements the central part of quicksort is the partitioning procedure . given some pivot element , it returns a pointer @xmath17 to an element in the array and rearranges the array such that all elements left of the @xmath17 are smaller or equal the pivot and all elements on the right are greater or equal the pivot . quicksort first chooses some pivot element , then performs the partitioning , and , finally , recurses on the elements smaller and larger the pivot see algorithm [ alg : qs ] . we call the procedure which organizes the calls to the partitioner the _ quicksort main loop_. pivot @xmath18 choosepivot(@xmath19 $ ] ) cut @xmath18 partition(@xmath19 $ ] , pivot ) quicksort(@xmath20 $ ] ) quicksort(@xmath21 $ ] ) there are many standard improvements for quicksort . for our optimized quicksort main loop ( which is a modified version of tuned quicksort @xcite ) , we implemented the following : * a very basic optimization due to sedgewick @xcite avoids recursion partially ( e.g. ` std:\!:sort ` ) or totally ( here this requires the introduction of an explicit stack ) . * introsort @xcite : there is an additional counter for the number of recursion levels . as soon as it exceeds some bound ( ` std:\!:sort`uses @xmath22 we use @xmath23 ) , the algorithms stops quicksort and switches to heapsort @xcite ( only for the respective sub - array ) . by doing so , a worst - case running time of @xmath24 is guaranteed . * sedgewick @xcite also proposed to switch to insertionsort ( see e.g. ( * ? ? ? * section 5.2.1 ) ) as soon as the array size is less than some fixed small constant ( 16 for ` std:\!:sort`and our implementation ) . there are two possibilities when to apply insertionsort : either during the recursion , when the array size becomes too small , or at the very end after quicksort has finished . we implemented the first possibility ( in contrast to ` std:\!:sort ` ) because for sorting integers , it hardly made a difference , but for larger data elements there was a slight speedup ( in @xcite this was proposed as _ memory - tuned quicksort _ ) . * after partitioning , the pivot is moved to its correct position and not included in the recursive calls ( not applied in ` std:\!:sort ` ) . * the basic version of quicksort uses a random or fixed element as pivot . a slight improvement is to choose the pivot as median of three elements typically the first , in the middle and the last . this is applied in ` std:\!:sort`and many other quicksort implementations . sedgewick @xcite already remarked that choosing the pivots from an even larger sample does not provide a significant increase of speed . in view of the experiments with skewed pivots @xcite , this is no surprise . for blockquicksort , a pivot closer to the median turns out to be beneficial ( figure [ fig : skewed_pivots ] in section [ sec : skewed ] ) . thus , it makes sense to invest more time to find a better pivot element . in @xcite , martinez and roura show that the number of comparisons incurred by quicksort is minimal if the pivot element is selected as median of @xmath25 elements . another variant is to choose the pivot as median of three ( resp . five ) elements which themselves are medians of of three ( resp . five ) elements . we implemented all these variants for our experiments see section [ sec : pivot_experiments ] . our main contribution is the block partitioner , which we describe in the next section . the idea of block partitioning is quite simple . recall how hoare s original partition procedure works ( algorithm [ alg : hoare ] ) : @xmath26 < \mathrm{pivot}$ ] @xmath27 @xmath28 > \mathrm{pivot}$ ] @xmath29 @xmath30 swap(@xmath26 , a[r]$ ] ) ; @xmath27 ; @xmath29 @xmath31 two pointers start at the leftmost and rightmost elements of the array and move towards the middle . in every step the current element is compared to the pivot ( line 3 and 4 ) . if some element on the right side is less or equal the pivot ( resp . some element on the left side is greater or equal ) , the respective pointer stops and the two elements found this way are swapped ( line 5 ) . then the pointers continue moving towards the middle . the idea of blockquicksort ( algorithm [ alg : partition ] ) is to separate lines 3 and 4 of algorithm [ alg : hoare ] from line 5 : fix some block size @xmath32 ; we introduce two buffers @xmath33 $ ] and @xmath34 $ ] for storing pointers to elements ( @xmath35 will store pointers to elements on the left side of the array which are greater or equal than the pivot element likewise @xmath36 for the right side ) . the main loop of algorithm [ alg : partition ] consists of two stages : the scanning phase ( lines 5 to 18 ) and the rearrangement phase ( lines 19 to 26 ) . * integer*@xmath37 , \mathrm{offsets}_r[0 , \dots , { \ensuremath{b}\xspace}- 1]$ ] * integer*@xmath38 @xmath39 @xmath40 \gets i$ ] @xmath41 ) $ ] @xmath42 @xmath43 \gets i$ ] @xmath44 ) $ ] * integer*num @xmath45 swap(@xmath46\bigr ] , a\bigl[r- \mathrm{offsets}_r[\mathrm{start}_r + j]\bigr]$ ] ) @xmath47 ; @xmath48 @xmath49 @xmath50 @xmath51 @xmath52 compare and rearrange remaining elements like for classical hoare partition , we also start with two pointers ( or indices as in the pseudocode ) to the leftmost and rightmost element of the array . first , the scanning phase takes place : the buffers which are empty are refilled . in order to do so , we move the respective pointer towards the middle and compare each element with the pivot . however , instead of stopping at the first element which should be swapped , only a pointer to the element is stored in the respective buffer ( lines 8 and 9 resp . 15 and 16 actually the pointer is always stored , but depending on the outcome of the comparison a counter holding the number of pointers in the buffer is increased or not ) and the pointer continues moving towards the middle . after an entire block of @xmath32 elements has been scanned ( either on both sides of the array or only on one side ) , the rearranging phase begins : it starts with the first positions of the two buffers and swaps the data elements they point to ( line 21 ) ; then it continues until one of the buffers contains no more pointers to elements which should be swapped . now the scanning phase is restarted and the buffer that has run empty is filled again . the algorithm continues this way until fewer elements than two times the block size remain . now , the simplest variant is to switch to the usual hoare partition method for the remaining elements ( in the experiments with suffix ` hoare finish ` ) . but , we also can continue with the idea of block partitioning : the algorithm scans the remaining elements as one or two final blocks ( of smaller size ) and performs a last rearrangement phase . after that , some elements to swap in one of the two buffers might still remain , while the other buffer is empty . with one run through the buffer , all these elements can be moved to the left resp . right ( similar as it is done in the lomuto partitioning method , but without performing actual comparisons ) . we do not present the details for this final rearranging here because on one hand it gets a little tedious and on the other hand it does neither provide a lot of insight into the algorithm nor is it necessary to prove our result on branch mispredictions . the c++ code of this basic variant can be found in appendix [ app : code ] . if the input consists of random permutations ( all data elements different ) , the average numbers of comparisons and swaps are the same as for usual quicksort with median - of - three . this is because both hoare s partitioner and the block partitioner preserve randomness of the array . the number of scanned elements ( total number of elements loaded to the registers ) is increased by two times the number of swaps , because for every swap , the data elements have to be loaded again . however , the idea is that due to the small block size , the data elements still remain in l1 cache when being swapped so the additional scan has no negative effect on the running time . in section [ sec : blocksize_experiments ] we see that for larger data types and from a certain threshold on , an increasing size of the blocks has a negative effect on the running time . therefore , the block size should not be chosen too large we propose @xmath53 and fix this constant throughout ( thus , already for inputs of moderate size , the buffers also do not require much more space than the stack for quicksort ) . [ [ branch - mispredictions ] ] branch mispredictions the next theorem is our main theoretical result . for simplicity we assume here that blockquicksort is implemented without the worst - case - stopper heapsort ( i.e. there is no limit on the recursion depth ) . since there is only a low probability that a high recursion depth is reached while the array is still large , this assumption is not a real restriction . we analyze a static branch predictor : there is a misprediction every time a loop is left and a misprediction every time the _ if _ branch of an _ if _ statement is not taken . [ thm : few_branches ] let @xmath54 be the average number of comparisons of quicksort with constant size pivot sample . then blockquicksort ( without limit to the recursion depth and with the same pivot selection method ) with blocksize @xmath32 induces at most @xmath55 branch mispredictions on average . in particular , blockquicksort with median - of - three induces less then @xmath56 branch mispredictions on average . theorem [ thm : few_branches ] shows that when choosing the block size sufficiently large , the @xmath57-term becomes very small and for real - world inputs we can basically assume a linear number of branch mispredictions . moreover , theorem [ thm : few_branches ] can be generalized to samples of non - constant size for pivot selection . since the proof might become tedious , we stick to the basic variant here . the constant 6 in theorem [ thm : few_branches ] can be replaced by 4 when implementing lines 19 , 24 , and 25 of algorithm [ alg : partition ] with conditional moves . first , we show that every execution of the block partitioner algorithm [ alg : partition ] on an array of length @xmath2 induces at most @xmath58 branch mispredictions for some constant @xmath59 . in order to do so , we only need to look at the main loop ( line 4 to 27 ) of algorithm [ alg : partition ] because the final scanning and rearrangement phases consider only a constant ( at most @xmath60 ) number of elements . inside the main loop there are three _ for _ loops ( starting lines 7 , 14 , 20 ) , four _ if _ statements ( starting lines 5 , 12 , 24 , 25 ) and the min calculation ( whose straightforward implementation is an _ if _ statement line 19 ) . we know that in every execution of the main loop at least one of the conditions of the _ if _ statements in line 5 and 12 is true because in every rearrangement phase at least one buffer runs empty . the same holds for the two _ if _ statements in line 24 and 25 . therefore , we obtain at most two branch mispredictions for the _ _ if__s , three for the _ for _ loops and one for the min in every execution of the main loop . in every execution of the main loop , there are at least @xmath32 comparisons of elements with the pivot . thus , the number of branch misses in the main loop is at most @xmath61 times the number of comparisons . hence , for every input permutation the total number of branch mispredictions of blockquicksort is at most @xmath62 where @xmath63 it the number of branch mispredictions of one execution of the main loop of quicksort ( including pivot selection , which only needs a constant number of instructions ) and the @xmath64 term comes from the final insertionsort . the number of calls to partition is bounded by @xmath2 because each element can be chosen as pivot only once ( since the pivots are not contained in the arrays for the recursive calls ) . thus , by taking the average over all input permutations , the first statement follows . the second statement follows because quicksort with median - of - three incurs @xmath65 comparisons on average @xcite . [ rem : constant ] the @xmath64-term in theorem [ thm : few_branches ] can be bounded by @xmath66 by taking a closer look to the final rearranging phase . we give a rough heuristic estimate : it is save to assume that the average length of arrays on which insertionsort is called is at least 8 ( recall that we switch to insertionsort as soon as the array size is less than 17 ) . for insertionsort there is one branch miss for each element ( when exiting the loop for finding the position ) plus one for each call of insertionsort ( when exiting the loop over all elements to insert ) . furthermore , there are at most two branch misses in the main quicksort loop ( lines 177 and 196 in appendix [ app : code ] ) for every call to insertionsort . hence , we have approximately @xmath67 branch misses due to insertionsort . it remains to count the constant number of branch misprediction incurred during every call of partitioning : after exiting the main loop of block partition , there is one more scan and rearrangement phase with a smaller block size . this leads to at most @xmath68 branch mispredictions ( one extra because there is an additional case that both buffers are empty ) . the final rearranging incurs at most three branch misses ( lines 118 , 136 , 140 ) . selecting the pivot as median - of - three ( line 11 ) induces no branch misses since all conditional statements are compiled to conditional moves . finally , there is at most one branch miss in the main quicksort loop for every call to partition ( line 180 ) . this sums up to at most @xmath69 branch misses per call to partition . because the average size of arrays treated by insertionsort is at least 8 , the number of calls to partition is less than @xmath70 . thus , in total the @xmath64-term in theorem [ thm : few_branches ] consists of at most @xmath71 branch mispredictions . we propose and implemented further tunings for our block partitioner : 1 . loop unrolling : since the block size is a power of two , the loops of the scanning phase can be unrolled four or even eight times without causing additional overhead . 2 . cyclic permutations instead of swaps : we replace + swap(@xmath46\bigr ] , a\bigl[r- \mathrm{offsets}_r[\mathrm{start}_r + j]\bigr]$ ] ) + by the following code , which does not perform exactly the same data movements , but still in the end all elements less than the pivot are on the left and all elements greater are on the right : + temp @xmath72\bigr]$ ] @xmath73\bigr ] \gets a\bigl[r- \mathrm{offsets}_r[\mathrm{start}_r]\bigr]$ ] @xmath74\bigr ] \gets a\bigl[\ell + \mathrm{offsets}_l[\mathrm{start}_l + j]\bigr]$ ] @xmath46\bigr ] \gets a\bigl[r - \mathrm{offsets}_r[\mathrm{start}_r + j ] \bigr]$ ] @xmath75\bigr ] \gets$ ] temp + note that this is also a standard improvement for partitioning see e.g. @xcite . in the following , we always assume these two improvements since they are of very basic nature ( plus one more small change in the final rearrangement phase ) . we call the variant without them ` block_partition_simple ` its c++ code can be found in appendix [ app : code ] . the next improvement is a slight change of the algorithm : in our experiments we noticed that for small arrays with many duplicates the recursion depth becomes often higher than the threshold for switching to heapsort a way to circumvent this is an additional check for duplicates equal to the pivot if one of the following two conditions applies : * the pivot occurs twice in the sample for pivot selection ( in the case of median - of - three ) , * the partitioning results very unbalanced for an array of small size . the check for duplicates takes place after the partitioning is completed . only the larger half of the array is searched for elements equal to the pivot . this check works similar to lomuto s partitioner ( indeed , we used the implementation from @xcite ) : starting from the position of the pivot , the respective half of the array is scanned for elements equal to the pivot ( this can be done by one _ less than _ comparison since elements are already known to be greater or equal ( resp . less or equal ) the pivot ) ) . elements which are equal to the pivot are moved to the side of the pivot . the scan continues as long as at least every fourth element is equal to the pivot ( instead every fourth one could take any other ratio this guarantees that the check stops soon if there are only few duplicates ) . after this check , all elements which are identified as being equal to the pivot remain in the middle of the array ( between the elements larger and the elements smaller than the pivot ) ; thus , they can be excluded from further recursive calls . we denote this version with the suffix ` duplicate check ` ( dc ) . we ran thorough experiments with implementations in c++ on different machines with different types of data and different kinds of input sequences . if not specified explicitly , the experiments are run on an intel core i5 - 2500k cpu ( 3.30ghz , 4 cores , 32 kb l1 instruction and data cache , 256 kb l2 cache per core and 6 mb l3 shared cache ) with 16 gb ram and operating system ubuntu linux 64bit version 14.04.4 . we used gnu s ` g++ ` ( 4.8.4 ) ; optimized with flags ` -o3 -march = native ` . for time measurements , we used ` std:\!:chrono:\ ! : high_resolution_clock ` , for generating random inputs , the mersenne twister pseudo - random generator ` std:\!:mt19937 ` . all time measurements were repeated with the same 20 deterministically chosen seeds the displayed numbers are the average of these 20 runs . moreover , for each time measurement , at least 128 mb of data were sorted if the array size is smaller , then for this time measurement several arrays have been sorted and the total elapsed time measured . our running time plots all display the actual time divided by the number of elements to sort on the ` y`-axis . we performed our running time experiments with three different data types : * ` int ` : signed 32-bit integers . * ` vector ` : 10-dimensional array of 64-bit floating - point numbers ( ` double ` ) . the order is defined via the euclidean norm for every comparison the sums of the squares of the components are computed and then compared . * ` record ` : 21-dimensional array of 32-bit integers . only the first component is compared . the code of our implementation of blockquicksort as well as the other algorithms and our running time experiments is available at https://github.com/weissan/blockquicksort . [ [ sec : blocksize_experiments ] ] different block sizes figure [ fig : blocksizes ] shows experimental results on random permutations for different data types and block sizes ranging from 4 up to @xmath76 . we see that for integers only at the end there is a slight negative effect when increasing the block size . presumably this is because up to a block size of @xmath77 , still two blocks fit entirely into the l3 cache of the cpu . on the other hand for ` vector ` a block size of 64 and for ` record ` of 8 seem to be optimal with a considerably increasing running time for larger block sizes . as a compromise we chose to fix the block size to 128 elements for all further experiments . an alternative approach would be to choose a fixed number of bytes for one block and adapt the block size according to the size of the data elements . [ [ sec : skewed ] ] skewed pivot experiments [ sec : pivot_experiments ] means that @xmath78-th element is chosen as pivot of an array of length @xmath2 . ] we repeated the experiments from @xcite with skewed pivot for both the usual hoare partitioner ( ` std:\!:__unguarded_partition ` , from the gcc implementation of ` std:\!:sort ` ) and our block partition method . for both partitioners we used our tuned quicksort loop . the results can be seen in figure [ fig : skewed_pivots ] : classic quicksort benefits from skewed pivot , whereas blockquicksort works best with the exact median . therefore , for blockquicksort it makes sense to invest more effort to find a good pivot . [ [ sec : pivot_selection ] ] different pivot selection methods we implemented several strategies for pivot selection : * median - of - three , median - of - five , median - of - twenty - three , * median - of - three - medians - of - three , median - of - three - medians - of - five , median - of - five - medians - of - five : first calculate three ( resp . five ) times the median of three ( resp . five ) elements , then take the pivot as median of these three ( resp . five ) medians , * median - of-@xmath79 . all pivot selection strategies switch to median - of - three for small arrays . moreover , the median - of-@xmath79 variant switches to median - of - five - medians - of - five for arrays of length below 20000 ( for smaller @xmath2 even the number of comparisons was better with median - of - five - medians - of - five ) . the medians of larger samples are computed with ` std:\!:nth_element ` . despite the results on skewed pivots figure [ fig : skewed_pivots ] , there was no big difference between the different pivot selection strategies as it can be seen in figure [ fig : pivot_method ] . . ] as expected , median - of - three was always the slowest for larger arrays . median - of - five - medians - of - five was the fastest for ` int ` and median - of-@xmath79 for ` vector ` . we think that the small difference between all strategies is due to the large overhead for the calculation of the median of a large sample and maybe because the array is rearranged in a way that is not favorable for the next recursive calls . we compare variants of blockquicksort with the gcc implementation of ` std:\!:sort ` ( which is known to be one of the most efficient quicksort implementations see e.g. @xcite ) , yaroslavskiy s dual - pivot quicksort @xcite ( we converted the java code of @xcite to c++ ) and an implementation of super scalar sample sort @xcite by hbschle - schneider . for random permutations and random values modulo @xmath79 , we also test tuned quicksort @xcite and three - pivot quicksort implemented by aumller and bingmann from @xcite ( which is based on @xcite ) for other types of inputs we omit these algorithms because of their poor behavior with duplicate elements . [ [ branch - mispredictions-1 ] ] branch mispredictions we experimentally determined the number of branch mispredictions of blockquicksort and the other algorithms with the _ chachegrind _ branch prediction profiler , which is part of the profiling tool _ _ valgrind__. the results of these experiments on random ` int ` data can be seen in figure [ fig : branch_misses ] the ` y`-axis shows the number of branch misprediction divided the the array size . we only display the median - of - three variant of blockquicksort since all the variants are very much alike . we also added plots of blockquicksort and tuned quicksort skipping final insertionsort ( i.e. the arrays remain partially unsorted ) . we see that both ` std:\!:sort`and yaroslavskiy s dual - pivot quicksort incur @xmath80 branch mispredictions . the up and down for super scalar sample sort presumably is because of the variation in the size of the arrays where the base case sorting algorithm ` std:\!:sort`is applied to . for blockquicksort there is an almost non - visible @xmath6 term for the number of branch mispredictions . indeed , we computed an approximation of @xmath81 branch mispredictions . thus , the actual number of branch mispredictions is still better then our bounds in theorem [ thm : few_branches ] . there are two factors which contribute to this discrepancy : our rough estimates in the mentioned results , and that the actual branch predictor of a modern cpu might be much better than a static branch predictor . also note that approximately one half of the branch mispredictions are incurred by insertionsort only the other half by the actual block partitioning and main quicksort loop . finally , figure [ fig : branch_misses ] shows that katajainen et al.s tuned quicksort is still more efficient with respect to branch mispredictions ( only @xmath64 ) . this is no surprise since it does not need any checks whether buffers are empty etc . moreover , we see that over @xmath82 of the branch misses of tuned quicksort come from the final insertionsort . [ [ running - time - experiments ] ] running time experiments in figure [ fig : random_int ] we present running times on random ` int ` permutations of different blockquicksort variants and the other algorithms including katajainen s tuned quicksort and aumller and bingmann s three - pivot quicksort . the optimized blockquicksort variants need around 45ns per element when sorting @xmath83 elements , whereas ` std:\!:sort`needs 85ns per element thus , there is a speed increase of 88% ( i.e. the number of elements sorted per second is increased by 88% ) . and @xmath79 . ] the same algorithms are displayed in figure [ fig : modulo_int ] for sorting random ` int`s between @xmath84 and @xmath79 . here , we observe that tuned quicksort is much worse than all the other algorithms ( already for @xmath85 it moves outside the displayed range ) . all variants of blockquicksort are faster than ` std:\!:sort ` the ` duplicate check ` ( dc ) version is almost twice as fast . figure [ fig : powers ] presents experiments with data containing a lot of duplicates and having specific structures thus , maybe coming closer to `` real - world '' inputs ( although it is not clear what that means ) . since here tuned quicksort and three - pivot quicksort are much slower than all the other algorithms , we exclude these two algorithms from the plots . the array for the left plot contains long already sorted runs . this is most likely the reason that ` std:\!:sort`and yaroslavskiy s dual - pivot quicksort have similar running times to blockquicksort ( for sorted sequences the conditional branches can be easily predicted what explains the fast running time ) . the arrays for the middle and right plot start with sorted runs and become more and more erratic ; the array for the right one also contains a extremely high number of duplicates . here the advantage of blockquicksort avoiding conditional branches can be observed again . in all three plots the check for duplicates ( dc ) established a considerable improvement . of ` int ` with duplicates : _ left _ : @xmath86 = i \mod { \left\lfloor\mathinner{\sqrt{n } } \right\rfloor}$ ] ; _ middle _ : @xmath86 = i^2 + n/2\mod n$ ] ; _ right _ : @xmath86 = i^8 + n/2 \mod n$ ] . since @xmath2 is always a power of two , the value @xmath87 occurs approximately @xmath88 times in the last case . ] in figure [ fig : recordvector ] , we show the results of selected algorithms for random permutations of ` vector ` and ` record ` . we conjecture that the good results of super scalar sample sort on ` record`s are because of its better cache behavior ( since ` record ` are large data elements with very cheap comparisons ) . more running time experiments also on other machines and compiler flags can be found in appendix [ app : more_experiments ] . [ [ more - statistics ] ] more statistics table [ tab : more_stats ] shows the number of branches taken / branch mispredicted as well as the instruction count and cache misses . although ` std:\!:sort`has a much lower instruction count than the other algorithms , it induces most branch misses and ( except tuned quicksort ) most l1 cache misses (= l3 refs since no l2 cache is simulated ) . blockquicksort does not only have a low number of branch mispredictions , but also a good cache behavior one reason for this is that insertionsort is applied during the recursion and not at the very end . .instruction count , branch and cache misses when sorting random ` int ` permutations of size @xmath89 . all displayed numbers are divided by the number of elements . [ cols="^,^,^,^,^,^,^",options="header " , ] we have established an efficient in - place general purpose sorting algorithm , which avoids branch predictions by converting results of comparisons to integers . in the experiments we have seen that it is competitive on different kinds of data . moreover , in several benchmarks it is almost twice as fast as ` std:\!:sort ` . future research might address the following issues : * we used insertionsort as recursion stopper inducing a linear number of branch misses . is there a more efficient recursion stopper that induces fewer branch mispredictions ? * more efficient usage of the buffers : in our implementation the buffers on average are not even filled half . to use the space more efficiently one could address the buffers cyclically and scan until one buffer is filled . by doing so , also both buffers could be filled in the same loop however , with the cost of introducing additional overhead . * the final rearrangement of the block partitioner is not optimal : for small arrays the similar problems with duplicates arise as for lomuto s partitioner . * pivot selection strategy : though theoretically optimal , median - of-@xmath79 pivot selection is not best in practice . also we want to emphasize that not only the sample size but also the selection method is important ( compare the different behavior of the two versions of ` std:\!:sort`for sorted and reversed permutations ) . it might be even beneficial to use a fast pseudo - random generator ( e.g. a linear congruence generator ) for selecting samples for pivot selection . * parallel versions : the block structure is very well suited for parallelism . * a three - pivot version might be interesting , but efficient multi - pivot variants are not trivial : our first attempt was much slower . [ [ acknowledgments ] ] acknowledgments thanks to jyrki katajainen and max stenmark for allowing us to use their python scripts for measuring branch mispredictions and cache missses and to lorenz hbschle - schneider for his implementation of super scalar sample sort . we are also indebted to jan philipp wchter for all his help with creating the plots , to daniel bahrdt for answering many c++ questions , and to christoph greulich for his help with the experiments . 10 v8 instruction set overview , 2011 . document number : prd03-genc-010197 15.0 . intel 64 and ia-32 architecture optimization reference manual , 2016 . order number : 248966 - 032 . d. abhyankar and m. ingle . engineering of a quicksort partitioning algorithm . , 2(2):1723 , 2011 . martin aumller and martin dietzfelbinger . optimal partitioning for dual pivot quicksort - ( extended abstract ) . in _ icalp _ , pages 3344 , 2013 . martin aumller , martin dietzfelbinger , and pascal klaue . how good is multi - pivot quicksort ? , abs/1510.04676 , 2015 . paul biggar , nicholas nash , kevin williams , and david gregg . an experimental study of sorting and branch prediction . , 12:1.8:139 , 2008 . gerth stlting brodal , rolf fagerberg , and kristoffer vinther . engineering a cache - oblivious sorting algorithm . , 12:2.2:123 , 2008 . gerth stlting brodal and gabriel moruz . tradeoffs between branch mispredictions and comparisons for sorting algorithms . in _ wads _ , volume 3608 of _ lncs _ , pages 385395 . springer , 2005 . thomas h. cormen , charles e. leiserson , ronald l. rivest , and clifford stein . . the mit press , 3nd edition , 2009 . amr elmasry and jyrki katajainen . lean programs , branch mispredictions , and sorting . in _ fun _ , volume 7288 of _ lncs _ , pages 119130 . springer , 2012 . amr elmasry , jyrki katajainen , and max stenmark . branch mispredictions do nt affect mergesort . in _ sea _ , pages 160171 , 2012 . robert w. floyd . algorithm 245 : treesort 3 . , 7(12):701 , 1964 . john l. hennessy and david a. patterson . . morgan kaufmann , 5th edition , 2011 . charles a. r. hoare . , 5(1):1016 , 1962 . kanela kaligosi and peter sanders . how branch mispredictions affect quicksort . in _ esa _ , pages 780791 , 2006 . jyrki katajainen . sorting programs executing fewer branches . cph stl report 2263887503 , department of computer science , university of copenhagen , 2014 . donald e. knuth . , volume 3 of _ the art of computer programming_. addison wesley longman , 2nd edition , 1998 . shrinu kushagra , alejandro lpez - ortiz , aurick qiao , and j. ian munro . multi - pivot quicksort : theory and experiments . in _ alenex _ , pages 4760 , 2014 . anthony lamarca and richard e ladner . the influence of caches on the performance of sorting . , 31(1):66104 , 1999 . conrado martnez , markus e. nebel , and sebastian wild . analysis of branch misses in quicksort . in _ workshop on analytic algorithmics and combinatorics , analco 2015 , san diego , ca , usa , january 4 , 2015 _ , pages 114128 , 2015 . conrado martnez and salvador roura . optimal sampling strategies in quicksort and quickselect . , 31(3):683705 , 2001 . david r. musser . introspective sorting and selection algorithms . , 27(8):983993 , 1997 . charles price . nstruction set , 1995 . peter sanders and sebastian winkel . uper scalar sample sort . in _ esa _ , pages 784796 , 2004 . robert sedgewick . the analysis of quicksort programs . , 7(4):327355 , 1977 . robert sedgewick . implementing quicksort programs . , 21(10):847857 , 1978 . sebastian wild and markus e. nebel . average case analysis of java 7 s dual pivot quicksort . in _ esa _ , pages 825836 , 2012 . sebastian wild , markus e. nebel , and ralph neininger . average case and distributional analysis of dual - pivot quicksort . 11(3):22:142 , 2015 . j. w. j. williams . algorithm 232 : heapsort . , 7(6):347348 , 1964 . vladimir yaroslavskiy . ual - pivot quicksort algorithm , 2009 . url : http://codeblab.com/wp-content/uploads/2009/09/dualpivotquicksort.pdf . in figure [ fig : inversions ] and figure [ fig : basic_permutations ] we also included the new gcc implementation of ` std:\!:sort`(gcc version 4.8.4 ) marked as ` std:\!:sort`(new ) . the very small difference in the implementation of choosing the second instead of the first element as part of the sample for pivot selection makes a enormous difference when sorting special permutations like decreasingly sorted arrays . this shows how important not only the size of the pivot sample but also the proper selection is . in the other benchmarks both implementations were relatively close , so we do not show both of them . = i + n/2 \mod n$ ] . ] inversions ( @xmath90 random swaps of neighboring elements ) : _ left _ : @xmath91 ; _ middle _ : @xmath92 ; _ right _ : @xmath93 . ] of ` int ` with many duplicates : _ left : _ constant ; _ middle : _ @xmath86 = 0 $ ] for @xmath94 and @xmath86 = 1 $ ] otherwise ; _ right : _ random 0 - 1 values . ] and @xmath79 ; _ right : _ sorted . ] and @xmath79 ; _ right : _ sorted . ] and@xmath79 ; _ right _ : sorted . ] and@xmath79 ; _ right _ : sorted . ] here , we give the c++ code of the basic blockquicksort variant ( the final rearranging is also in block style , but there is no loop unrolling etc . applied ) . .... template < typename iter , typename compare > inline void sort_pair(iter i1 , iter i2 , compare less ) { typedef typename std::iterator_traits < iter>::value_type t ; bool smaller = less(*i2 , * i1 ) ; t temp = std::move(smaller ? * i1 : temp ) ; * i1 = std::move(smaller ? * i2 : * i1 ) ; * i2 = std::move(smaller ? temp : * i2 ) ; } template < typename iter , typename compare > inline iter median_of_3(iter i1 , iter i2 , iter i3 , compare less ) { sort_pair(i1 , i2 , less ) ; sort_pair(i2 , i3 , less ) ; sort_pair(i1 , i2 , less ) ; return i2 ; } template < typename iter , typename compare > inline iter hoare_block_partition_simple(iter begin , iter end , iter pivot_position , compare less ) { typedef typename std::iterator_traits < iter>::difference_type index ; index indexl[blocksize ] , indexr[blocksize ] ; iter last = end - 1 ; std::iter_swap(pivot_position , last ) ; const typename std::iterator_traits < iter>::value_type & pivot = * last ; pivot_position = last ; last-- ; int num_left = 0 ; int num_right = 0 ; int start_left = 0 ; int start_right = 0 ; int num ; //main loop while ( last - begin + 1 > 2 * blocksize ) { //compare and store in buffers if ( num_left = = 0 ) { start_left = 0 ; for ( index j = 0 ; j < blocksize ; j++ ) { indexl[num_left ] = j ; num_left + = ( ! ( less(begin[j ] , pivot ) ) ) ; } } if ( num_right = = 0 ) { start_right = 0 ; for ( index j = 0 ; j < blocksize ; j++ ) { indexr[num_right ] = j ; num_right + = ! ( less(pivot , * ( last - j ) ) ) ; } } //rearrange elements num = std::min(num_left , num_right ) ; for ( int j = 0 ; j < num ; j++ ) std::iter_swap(begin + indexl[start_left + j ] , last - indexr[start_right + j ] ) ; num_left -= num ; num_right -= num ; start_left + = num ; start_right + = num ; begin + = ( num_left = = 0 ) ? blocksize : 0 ; last -= ( num_right = = 0 ) ? blocksize : 0 ; } //end main loop //compare and store in buffers final iteration index shiftr = 0 , shiftl = 0 ; if ( num_right = = 0 & & num_left = = 0 ) { //for small arrays or in the unlikely case that both buffers are empty shiftl = ( ( last - begin ) + 1 ) / 2 ; shiftr = ( last - begin ) + 1 - shiftl ; start_left = 0 ; start_right = 0 ; for ( index j = 0 ; j < shiftl ; j++ ) { indexl[num_left ] = j ; num_left + = ( ! less(begin[j ] , pivot ) ) ; indexr[num_right ] = j ; num_right + = ! less(pivot , * ( last - j ) ) ; } if ( shiftl < shiftr ) { indexr[num_right ] = shiftr - 1 ; num_right + = ! less(pivot , * ( last - shiftr + 1 ) ) ; } } else if ( num_right ! = 0 ) { shiftl = ( last - begin ) - blocksize + 1 ; shiftr = blocksize ; start_left = 0 ; for ( index j = 0 ; j < shiftl ; j++ ) { indexl[num_left ] = j ; num_left + = ( ! less(begin[j ] , pivot ) ) ; } } else { shiftl = blocksize ; shiftr = ( last - begin ) - blocksize + 1 ; start_right = 0 ; for ( index j = 0 ; j < shiftr ; j++ ) { indexr[num_right ] = j ; num_right + = ! ( less(pivot , * ( last - j ) ) ) ; } } //rearrange final iteration num = std::min(num_left , num_right ) ; for ( int j = 0 ; j < num ; j++ ) std::iter_swap(begin + indexl[start_left + j ] , last - indexr[start_right + j ] ) ; num_left -= num ; num_right -= num ; start_left + = num ; start_right + = num ; begin + = ( num_left shiftl : 0 ; last -= ( num_right shiftr : 0 ; //end final iteration //rearrange elements remaining in buffer if ( num_left ! = 0 ) { int loweri = start_left + num_left - 1 ; index upper = last - begin ; //search first element to be swapped while ( loweri > = start_left & & indexl[loweri ] = = upper ) { upper-- ; loweri-- ; } while ( loweri > = start_left ) std::iter_swap(begin + upper-- , begin + indexl[loweri-- ] ) ; std::iter_swap(pivot_position , begin + upper + 1 ) ; // fetch the pivot return begin + upper + 1 ; } else if ( num_right ! = 0 ) { int loweri = start_right + num_right - 1 ; index upper = last - begin ; //search first element to be swapped while ( loweri > = start_right & & indexr[loweri ] = = upper ) { upper-- ; loweri-- ; } while ( loweri > = start_right ) std::iter_swap(last - upper-- , last - indexr[loweri-- ] ) ; std::iter_swap(pivot_position , last - upper);// fetch the pivot return last - upper ; } else { //no remaining elements std::iter_swap(pivot_position , begin);// fetch the pivot return begin ; } } template < typename iter , typename compare > struct hoare_block_partition_simple { static inline iter partition(iter begin , iter end , compare less ) { //choose pivot iter mid = median_of_3(begin , begin + ( end - begin ) / 2 , end , less ) ; //partition return hoare_block_partition_simple(begin + 1 , end - 1 , mid , less ) ; } } ; //quicksort main loop . implementation based on tuned quicksort ( elmasry , katajainen , stenmark ) template < template < class , class > class partitioner , typename iter , typename compare > inline void qsort(iter begin , iter end , compare less ) { const int depth_limit = 2 * ilogb((double)(end - begin ) ) + 3 ; iter stack[80 ] ; iter * s = stack ; int depth_stack[40 ] ; int depth = 0 ; int * d_s_top = depth_stack ; * s = begin ; * ( s + 1 ) = end ; s + = 2 ; * d_s_top = 0 ; + + d_s_top ; do { if ( depth < depth_limit & & end - begin > is_thresh ) { iter pivot = partitioner < iter , compare>::partition(begin , end , less ) ; //push large side to stack and continue on small side if ( pivot - begin > end - pivot ) { * s = begin ; * ( s + 1 ) = pivot ; begin = pivot + 1 ; } else { * s = pivot + 1 ; * ( s + 1 ) = end ; end = pivot ; } s + = 2 ; depth++ ; * d_s_top = depth ; + + d_s_top ; } else { if ( end - begin > is_thresh ) // if recursion depth limit exceeded std::partial_sort(begin , end , end ) ; else insertionsort::insertion_sort(begin , end , less ) ; //copy of std::__insertion_sort ( gcc 4.7.2 ) //pop new subarray from stack s -= 2 ; begin = * s ; end = * ( s + 1 ) ; --d_s_top ; depth = * d_s_top ; } } while ( s ! = stack ) ; }

the procyon ab binary system ( orbital period 40.838 years , a newly - refined determination ) , is near and bright enough that the component radii , effective temperatures and luminosities are very well determined , although more than one possible solution to the masses has limited the claimed accuracy . preliminary mass determinations for each component are available from hst imaging , supported by ground - based astrometry and an excellent _ hipparcos _ parallax ; we use these for our preferred solution for the binary system . other values for the masses are also considered . we have employed the tycho stellar evolution code to match the radius and luminosity of the f5 iv - v primary star to determine the system s most likely age as 1.87.13 gyr . since prior studies of procyon a found its abundance indistinguishable from solar , the solar composition of asplund , grevesse & sauval (=0.014 ) is assumed for the hr diagram fitting . an unsuccessful attempt to fit using the older solar abundance scale of grevesse & sauval (=0.019 ) is also reported . for procyon b , eleven new sequences for the cooling of non - da white dwarfs have been calculated , to investigate the dependences of the cooling age on ( 1 ) the mass , ( 2 ) the core composition , ( 3 ) the helium layer mass , and ( 4 ) heavy - element opacities in the helium envelope . our calculations indicate a cooling age of 1.19.11 gyr , which implies that the progenitor mass of procyon b was 2.59 . in a plot of initial vs final mass of white dwarfs in astrometric binaries or star clusters ( all with age determinations ) , the procyon b final mass lies several below a straight line fit .

1305.0587

procyon a is the 14th nearest star and stellar system to the sun , according to henry ( 2011 ) , at a distance of 3.51 parsecs or 11.46 light years see van leeuwen ( 2007 ) . its spectral type is f5 iv - v . bessel ( 1844 ) discovered that the ( large ) proper motion of procyon on the sky was perturbed by what he recognized to be an unseen companion . procyon b was first detected visually at the end of the 19th century by schaeberle ( 1896 ) . the first recognition that the companion was like those in the sirius and 40 eri systems , later to be called white dwarfs , may be the work of the irish amateur astronomer john ellard gore . however , he did not properly publish this result in the astronomical literature see the discussions of his work by fitzgerald ( 1966 ) and holberg ( 2009 ) . the first orbital elements were determined by auwers ( 1862 ) , who showed that the orbital period would be about 40 years . we now know that the more exact period is 40.838 years ( h. e. bond & g.h . schaefer , 2012 private communication ) . solar - like oscillations in procyon a were first reported by brown et al . ( 1991 ) , and several authors have since investigated the structure and evolution of that star through a seismic approach ( see , e.g. , dogan et al . 2010 and references therein ) . we come back to the asteroseismological properties of procyon a in 5 . like sirius a studied in liebert et al . ( 2005 , hereafter paper i ) , procyon a has an extremely accurate interferometric radius measurement , and an extremely accurate trigonometric parallax ; the luminosity is also very accurate . sirius a is certainly a main sequence star , while procyon a could be entering or about to enter the subgiant branch . in contast to the sirius system , however , the astrometric mass solution for the procyon system has been a controversial issue , as we shall discuss in 2 . in comparison with the white dwarf sirius b , procyon b is very different from other ( mostly da ) white dwarfs previously used for the problem of the star s initial - to - final mass relation ( ifmr ) . mainly it has a helium - dominated atmosphere polluted by heavy elements ( dqz spectral type ) , a lower mass , and a cooler temperature . the mass , radius and @xmath5 determined for the b component will be discussed in 3 . in 4 we address the atmospheric abundances of procyon a and the interior abundances to be used for modelling in 5 . here we employ the tycho stellar evolution code ( young & arnett 2005 ) to fit the position of procyon a in an hr diagram , and determine the age . in 6 eleven calculations of the cooling time of procyon b are presented , for different assumptions about the mass , the helium envelope mass , and its composition . the cooling time is the length of time since the degenerate entered the white dwarf sequence ; this has been used for the coolest white dwarfs to estimate ages of the galactic disk ( liebert , dahn , & monet 1988 ) and halo ( hansen et al . 2007 using the globular cluster ngc 6397 ) . the difference between the systemic age and the white dwarf cooling time yields the best theoretical estimate of the initial mass of the white dwarf . moreover , as discussed in 6 , procyon b presents a complicated case for the calculation of the cooling age . further remarks pertinent to this more challenging physical situation are discussed in that section . this paper adds a valuable new object with accurately determined stellar parameters ( masses , radii , luminosities ) to the initial - to - final mass relation ( ifmr ) for progenitors and white dwarfs . the vast majority of other data points have white dwarfs with hydrogen - rich atmospheres , while procyon b has a complicated helium - rich envelope and atmosphere . it is possible that the ifmr for helium - rich white dwarfs may differ from the hydrogen - rich distribution . we show in 7 how procyon b compares with sirius b and white dwarfs in star clusters for which the ifmr of low to intermediate mass stars has been determined . as noted in the introduction , a precise radius measurement is available for procyon a , from measurements of the angular diameter with the eso very large telescope interferometer . the radius is @xmath6 = [email protected]@xmath7 or @xmath00.65% ( aufdenberg , ludwig , & kervella 2005 ) . this radius value is based on the authors `` best estimate '' angular diameter of [email protected] mas at 2.2@xmath8 . note that the authors constructed model atmospheres for procyon in three ways : ( 1 ) stand - alone one - dimensional ( 1d ) phoenix structures , radiation fields , and spectra ( hauschildt et al . 1999 ) ; ( 2 ) co@xmath9bold 3d structures temporarily and spatially averaged to 1d , then read by phoenix for computation of the corresponding radiation fields ( freytag et al . 2002 ) ; and ( 3 ) stand - alone 1d atlas 12 ( kurucz 1992 ) structures , radiation fields and spectra . note that the quantitative assessment of the limb darkening , on which the radius value is determined , depends on the model atmospheres analysis . the authors discuss in detail their assessment of these different models , leading to their `` best estimate '' . these authors also cite the earlier measurement on mt . wilson , california using the mark iii interferometer at 500 and 800 nm ( mozurkewich et al 1991 ) , whose limb - darkened value of the angular diameter ( [email protected] mas ) agrees marginally within the errors . their new measurement compares to an older , slightly less accurate , but consistent value of [email protected]@xmath7 by the same group ( kervella et al . 2004 ) , which corresponds to an adopted angular diameter of [email protected] mas . casagrande et al . ( 2010 ) got a lower , less accurate , but consistent [email protected] mas . for the radius ( @xmath10 ) of the primary star in this paper , we adopt the aufbenberg et al . ( 2005 ) best value . note that the kervella et al . and aufdenberg et al . results differ by less than 1@xmath11 . adopting either one results in an essentially identical fit . the _ hipparcos _ satellite mission provided the precise measurement of the trigonometric parallax of [email protected] " ( or @xmath00.3% ) . using the bolometric energy distribution and the effective temperature determination ( 6,543@xmath084 k ) , the luminosity of procyon a is determined in aufdenberg et al . ( 2005 ) to be log @xmath12/@xmath13 = + [email protected] . note that chiavassa et al . estimate @xmath14 = 6591 k or 6556 k depending on how the bolometric flux is calculated , within the error bars of aufdenberg et al . the former value differs the most from aufdenberg et al.s preferred 6543 k , resulting in log @xmath12/@xmath13 = + 0.843 ( with one more significant digit than is warranted ) . the difference between these two values hardly matters in figure 1 . two decades ago the mass of the `` a '' component was believed to be near 1.75 @xmath3 ( irwin et al . . however , main sequence evolutionary models at that mass were too luminous by a factor of two compared with the observed luminosity ( guenter & demarque 1993 ) . long - term imaging by hst , combined with the interferometric radius and trigonometric parallax measurements cited above led to a great improvement in the stellar parameters . first , girard et al.(2000 ) produced an improved astrometric solution based on analysis of 250 photographic plates spanning 83 years , augmented with early hubble data . they found m@xmath15 = [email protected] @xmath3 , almost exactly the 1.50 @xmath3 predicted by guenther & demarque ( 1993 ) . the problem is that hst imaging has only been possible for only about half of the orbital period . the combining of space - based imaging and the much longer series of ground - based observations has often been fraught with difficulties . apparent systematic offsets often occur in the respective measurements of the separations and position angles . girard et al . ( 2000 ) found that , if they restricted the astrometric analysis to the excellent images obtained with the wfpc2 camera i.e. , with no ground - based data a substantially lower mass ( 1.465 @xmath3 ) for procyon a was the result . at that time hst imagery covered only about a decade . allende prieto et al . ( 2002 ) revisited the procyon a solution using the _ hipparcos _ parallax and the stellar angular diameter by mozurkevich et al . ( 1991 ) discussed previously . they derived the primary s mass to be a much lower [email protected] @xmath3 . this is consistent with the ground - based gatewood and han ( 2006 ) value of the primary s mass of [email protected] @xmath3 , determined from their multichannel astrometric photometer * map * camera , built at the university of pittsburg ( gatewood 1987 ) . the * map * parallax is [email protected] mas , which agrees well with the _ hipparcos _ value . schaefer et al . ( 2006 ) reported a preliminary solution resulting from a space telescope science institute - led program to extend the imaging program on the procyon binary orbit . their result was in excellent agreement with the girard et al . ( 2000 ) analysis including ground - based data . several more years of hst data since 2006 now result in the following masses : [email protected] @xmath3 for procyon a and [email protected] @xmath3 for procyon b ( g. h. schaefer and h. e. bond , 2012 private communication ) . this solution uses available hst data from 1995 through 2012 plus , appropriately weighted , the archival ground - based data . it is not difficult to see how the total mass of the system is specified from accurate hst astrometry . using newton s gravitational formula for kepler s third law , ( @xmath16 + @xmath17 ) = @xmath18 / @xmath19 @xmath20 , with the semimajor axis ( @xmath21 ) and period ( @xmath22 ) well known , the sum of the masses is fixed at 2.052 @xmath3 . ground - based observations of the wobble of the a component about the center - of - mass on the sky then fix @xmath23 , @xmath23 + @xmath24 = @xmath21 , and @xmath16/@xmath17 = @xmath24/@xmath23 . thus the solution provides separate values for @xmath16 and @xmath17 . the mass determinations remain a matter of dispute in the literature . we begin with 1.499 @xmath3 as the hst - based , preferred mass of the primary for fitting in 5 . we also consider 1.42 @xmath3 as a lower bound to the mass of the primary . note that , in their recent analysis , chiavassa et al . ( 2012 ) give arguments in favor of the 1.43 @xmath3 gatewood and han value . procyon b was first detected visually at the end of the 19th century by schaeberle ( 1896 ) , and was historically one of the first white dwarfs to be discovered . even early estimates of the mass ( @xmath17 ) of the white dwarf component , such as that by strand ( 1951 ) , showed that it is much closer to the mean value of field white dwarfs than is sirius b. the updated schaefer et al . hst solution gives [email protected] @xmath3 for the mass of the secondary , and we consider this the benchmark value for @xmath17 . to be sure this is much smaller than previous estimates for the secondary , such as 0.622 @xmath3 ( irwin et al . 1992 ) and [email protected] @xmath3 ( girard et al . 2000 ) . values in between these have been published . the kervella et al . ( 2004 ) lower mass solution of 1.42 @xmath3 for the primary led to @xmath17 = [email protected] @xmath3 for b. the gatewood and han ( 2006 ) primary mass of [email protected] @xmath3 resulted in [email protected] @xmath3 for the secondary . to illustrate the dependence on mass , the cooling time for a mass of 0.602 @xmath3 ( girard et al . 2000 ) is also discussed in 6 . as mentioned in the introduction , the nature of the `` b '' white dwarf poses additional problems . the atmosphere is helium - rich , with trace abundances of carbon , calcium , magnesium and iron , but no detection of hydrogen . this is an unusually diverse abundance pattern for a white dwarf atmosphere , and complicated greatly the determination of the atmospheric parameters . not knowing of the complication , in the absence of an ultraviolet spectrum , provencal et al . ( 1997 ) used pure hydrogen and helium atmospheres to model the spectral energy distribution . preferring the helium fit due to its optical `` dc '' spectrum , they found @xmath14 = 8,688 k. this implied a stellar radius of [email protected] @xmath7 . given the mass , this small radius implied that the interior composition of the white dwarf could not be carbon - oxygen , but rather something very dense like iron ( shipman & provencal 1999 ) . this conclusion was unsettling to stellar theorists . once the complicated ultraviolet spectrum was revealed with the space telescope imaging spectrograph on the hubble space telescope , provencal et al . ( 2002 ) got themselves out of this `` iron box '' . atmospheres with realistic trace adundances , obtained from fitting the spectrum and the 1,80010,000 spectral energy distribution were employed to obtain @xmath14 = 7,740@xmath050 k. they now estimated the radius to be [email protected] @xmath7 , consistent with a carbon - oxygen interior . note that procyon b is the only white dwarf in a binary system or star cluster which does not have an atmosphere amenable to a mass determination from either the fitting of hydrogen or he i lines . it is also the coolest such white dwarf . there is a long literature trail indicating that the procyon a atmospheric abundance can not be distinguished from solar . we cite five studies : heiter & luck ( 2003 ) , and luck & heiter ( 2005 ) , as part of a study of all bright , nearby stars in the northern hemisphere , analyzed procyon a using high resolution spectra . they found an iron abundance [ fe / h ] = @[email protected] . valenti & fischer ( 2005 ) , as part of a systematic study of nearby stars using echelle spectra , obtained [ fe / h ] = @[email protected] . in allende prieto et al.s ( 2004 ) study of stars more luminous than m@xmath26 = + 6.5 within 14.5 pc of the sun , a value of + 0.08 was determined for procyon a. finally , in the most comprehensive geneva - copenhagen study of solar neighborhood stars ( nordstrm et al . 2004 ) , a value of [ fe / h ] = + 0.05 was reported . both of the last two author groups discuss the possible systematic error due to the coupling of the [ fe / h ] determination to the error in the log @xmath27 value . note that procyon a is cool enough that peculiar diffusion / gravitational support mechanisms should not distort the surface abundances from the interior abundance . there could , nevertheless , be a small diffusion of heavy elements out of the thin convective envelope . however , recently there has been controversy about what the solar abundance actually is . inevitably , the abundance parameters to be run for the stellar evolution models of procyon a , to be described next in 5 , must therefore be linked to this controversy , so a few comments are in order . the newer solar abundance scale determined from 3d , non - lte calculations of the solar atmosphere ( asplund et al . 2004 , 2005 , 2009 the latter hereafter a09 ) , and now lodders 2010 ) have significantly lower abundances of oxygen , carbon and nitrogen than previous estimates ( grevesse & sauval 1998 ) . the abundance of neon is a particular problem , since it is impossible to measure this element in the sun ( or procyon a ) , but is likely to be comparable numerically to the three elements mentioned previously . the overall heavy element abundance parameter for the sun decreases to a primordial value of @xmath1=0.0153 . the asplund a09 scale has been challenged because it breaks the accordance of the `` standard solar '' interior model with helioseismological measurements ( bahcall et al . a solar model with these abundances incorrectly predicts the depth of the convection zone , the depth profiles of sound speed and density , and the helium abundance ( basu & antia 2004 ) . as a check on the magnitude of likely systematic errors due to uncertainty in the actual solar ( and procyon a ) abundance , we retain consideration of the older grevesse & sauval ( 1998 ) solar abundance which corresponds to @xmath1=0.019 . the tycho code http://chandra.as.arizona.edu/@xmath28dave/tycho-intro.html incorporates the most current microphysics for opacities , equation - of - state , and nuclear reactions . in particular , the code has been upgraded with an algorithm based on a physical analysis of 3d hydrodynamic simulations of convection , not an astronomically - calibrated version of the mixing length theory . it includes non - locality and time dependence of flow , dynamical acceleration , turbulent dissipation , kolmogorov heating , compositional effects and dynamically - defined boundary conditions ( instead of parameterized overshooting schemes ) , all in a single , self - consistent formulation . the code is regularly tested against observations of double - lined eclipsing binaries and cluster isochrones to ensure consistency and accuracy . it produces superior fits to observational test cases without adjustment of parameters . these changes make possible fits to nuclear - burning stars on the hr diagram , especially those on or near the main sequence ( arnett , meakin , & young 2009 , 2010 ) . robust , consistent ages for ( nondegenerate ) binary stars with well - determined luminosities , masses , radii and temperatures have been determined by young et al . ( 2001 ) , young & arnett ( 2005 ) . in additon to the mass , we have to input the abundance parameters @xmath29 and @xmath1 , where @xmath1 scales by the solar ratios of individual elements . since we can not regard the solar atmospheric abundance issue as fully resolved , we therefore have computed sequences using the lodders ( 2010 ) value of @xmath1 = 0.0153 and also evaluated the older solar abundance of grevesse and sauval ( 1988 ) which imply @xmath1 = 0.019 . in figure 1 we show evolutionary tracks for @xmath1 = 0.0153 of ( 1 ) 1.499 @xmath3 ( in black ) , the best value from preliminary hst astrometry ( 2 ) , fitting the correct luminosity and radius at age 1.74 gyr ; ( 2 ) 1.48 @xmath3 ( red ) at 1.87 gyr , the best fit to the luminosity ( to three significant digits ) ; ( 3 ) 1.463 @xmath3 ( green ) at 1.99 gyr ; and ( 4 ) the lower 1.42 @xmath3 ( blue ) at 2.51 gyr favored by largely ground - based studies cited in 2 . the highest three masses fit within 1@xmath4 late on the main sequence . for 1.42 @xmath3 the fit misses by more than 2@xmath4 late on the main sequence . note that best - fit ages include the pre - ms evolution , without pre - ms accretion , but only the tracks from the beginning of the main sequence are shown here . we also tried a fit with the grevesse & sauval ( 1998 ) @xmath1 = 0.019 abundance at 1.48 @xmath3 , but it is 1.45@xmath4 underluminous when it reaches the correct radius on the main sequence at age 1.96 gyr . given the evidence from a09 and lodders ( 2010 ) that this solar metallicity value is too large , this marginal attempt at a fit is not shown in the figure . we therefore believe that [email protected] gyr from the 1.48 @xmath3 fit is the best estimate of the age of the binary system , with the uncertainty given by the bracketing age values for the 1.499 and 1.463 @xmath3 fits . note that provencal et al . ( 2002 ) estimated the nuclear lifetime of 1.3 gyr for component b , estimating @xmath282.1 @xmath3 as its progenitor mass . these authors use the cooling age from models of wood ( 1995 ) for pure carbon cores the only calculations available then . we now explore a range of alternative compositions for the white dwarf in 6 . since our best - fit mass using the tycho models lies well within the error bars of the preliminary hst astrometric mass of 1.499 @xmath3 , we adopt 1.48 @xmath3 as the preferred or `` benchmark '' mass of procyon a. note that the fitting of a star with mass near 1.5 @xmath3 at this stage of evolution presents uncertainties due to having both a small convective core and a small convective envelope . due to the former , the main sequence lifetime is affected by the uncertainty in the amount of hydrogen fuel available for reactions in the core . note that remaining uncertainties in the treatment of mixing beyond the convective core boundary , in the primordial solar abundance mix , or in any of the other model physics probably add larger systematic errors , by amounts which are difficult to quantify . the convective envelope leads to implied small differences between the interior and atmospheric @xmath1 abundances , as noted earlier . for kervella et al.s ( 2004 ) similar analysis , the fit to the hrd parameters also occurs in a similar way . if 1.50 @xmath3 is assumed , their fit is late on the main sequence at the much younger age of 1.3 gyr . however , these authors assumed _ no overshooting _ of the convective core , i.e. , less fuel , vs our calculations . for their preferred mass of 1.42 @xmath3 , the corresponding fit is early in the h - shell burning phase at ages of 2.312.71 gyr . this age range is close to what we achieve at 1.42@xmath3 for which we determine an age of 2.51 gyr ( stated three paragraphs earlier ) , plotted as the blue track in figure 1 . as indicated in the introduction , stochastically excited @xmath30-modes have been observed in procyon a since first reported by brown et al . ( 1991 ) , and several authors have attempted to exploit the seismic potential of that star using different data sets and models ( e.g. , guenther & demarque 1993 ; barban et al . 1999 ; chaboyer et al . 1999 ; di mauro & christensen - dalsgaard 2001 ; eggenberger et al . 2005 ; provost et al . 2006 ; bonanno et al . however , their investigations have been limited by the accuracy of the detected pulsation frequencies . more recently , better data became available ( bedding et al . 2010 ) and have been modelled in a preliminary way by dogan et al . we note that dogan et al . ( 2010 ) report two possible seismic model for procyon a , one of which characterized by a total mass of 1.50 @xmath3 and an age of 1.83 gyr ( no uncertainties provided ) , which is remarkably close to our solution . we find this result most encouraging . the importance of the cooling age of a white dwarf was introduced in the introduction . for procyon b we first calculate the cooling time for a non - da white dwarf of @xmath17 = [email protected] @xmath3 , the benchmark mass , but also evaluate the earlier alternative of @xmath17 = 0.602 @xmath3 from kervella et al . the fitted @xmath31 is 7,740@xmath050 k , as discussed in 3 . these error bars are the published , internal errors of the fit to the spectral energy distribution and absorption features , but are uncomfortably less than 1% of the value itself . for this object with a complicated dqz spectrum , we feel more comfortable assuming quadruple the formal @xmath14 error or @xmath0200 k. there have unfortunately been a shortage of available evolutionary sequences for white dwarfs lacking thick hydrogen envelopes . most cluster white dwarf studies in the last few decades have used the evolutionary models of wood ( 1992 , 1995 ) , but always with an outer hydrogen layer . these calculations do include one sequence with no hydrogen , a `` thin '' helium envelope of 10@xmath32 , atop a pure carbon core but with solar abundances of elements heavier than helium ( wood 2012 , private communication ) . hansen ( 1999 ) showed that pure helium envelope cooling models for 0.6 @xmath3 by himself ( for c , o and c / o cores ) and by salaris et al . ( 1997 ) cool considerably faster than the wood sequence ( see hansen s fig . 7 ) . note that if a trace of heavy elements were assumed , the envelope opacity would be considerably increased . hence , there has been some confusion as to what the cooling rates of white dwarfs lacking hydrogen layers should be . this issue is addressed in 6.2 . for a cool , non - da white dwarf with a simple , featureless spectrum , one could argue that the use of a pure - he envelope sequence such as from hansen ( 1999 ) would yield the more robust cooling time . however , the spectrum of procyon b shows the effects of ( 1 ) dredge - up of carbon from the diffusion tail of the core , with an atmospheric abundance log [ c / he ] = @[email protected] ( provencal et al . 2002 ) ; and ( 2 ) accretion of heavy elements log mg / he = @xmath2510.4 , log ca / he = @xmath2512 , log fe / he = @xmath2510.7 with log h / he @xmath33 @xmath254 . provencal et al . speculate that the secondary could be accreting a stellar wind from the primary at a rate of about 2@xmath34 10@xmath35 yr@xmath36 . thus , a calculation including additional opacity in the envelope layer may be more appropriate . moreover , systematic errors need to be considered also due to the uncertain mix of carbon and oxygen in the core , and the thickness of the helium envelope . for this purpose one of us ( gf ) has calculated a new set of cooling tracks to estimate these errors and parameter dependences , and hopefully thus bound the possible cooling time of this white dwarf . salaris et al . ( 1997 ) explored the dependence of cooling rates on the internal chemical distribution of carbon and oxygen in the core . for their best choices of the combined effect of convective mixing and the @xmath37c(@xmath38,@xmath39)@xmath40o reaction rate , carbon - oxygen profiles showing an enhancement of oxygen in the central regions are obtained for all white dwarf masses . mass fractions of @xmath40o as high as 0.8 are found near 0.6 @xmath3 . this fraction declines with increasing core mass because the above reaction rate is highest at the lower core temperatures characteristic of lower masses . thus , a core composition dominated by oxygen , or at least a mixture of carbon and oxygen , is more likely than one of pure carbon for procyon b. sequences have been calculated for pure oxygen , pure carbon , and for mixtures of carbon and oxygen according to the mass - dependent and depth - dependent results of salaris et al . we pick the last of these three types of sequences as the benchmark core composition . the cooling time also depends on the thickness of the helium layer . the helium - rich atmospheric composition indicates that some of the helium layer mass could have been lost in a late helium - shell flash that disposed of all of the hydrogen ( iben 1984 ) . values in the literature that have been considered include masses between 10@xmath41 and 10@xmath42 . the atmospheric carbon abundance is a strong constraint , since this results from dredgeup of the diffusion tail of carbon from the core . the study ( fontaine & brassard 2005 ) fitting dredgeup models to match the spectra of procyon b and a sequence of field dq white dwarfs by dufour , bergeron , & fontaine ( 2005 ) shows that the atmospheric carbon abundance of procyon b favors a layer mass thickness of log @[email protected]@xmath00.5 ( see their fig . we therefore consider this value as the benchmark , with sequences also calculated for a reasonable range of @xmath252.0 and @xmath253.0 . finally , it is very difficult to estimate the opacities due to elements heavier than helium in the partially - degenerate helium layer . this will depend on how deeply the heavy elements have diffused into the envelope below the convection zone . in principle the carbon abundance profile can be calculated for the helium envelope from diffusion theory , with the atmospheric abundance as a surface boundary . to consider this dependence in an approximate way , sequences with @xmath1=0.001 and @xmath1=0 have been calculated , with the former as the benchmark while the latter is contradicted by the observation of carbon and heavier elements in the spectrum . in table 1 , the cooling ages for the parameters of procyon b are listed for 8 sequences at the favored mass of 0.553 @xmath3 , two for values at @xmath00.022@xmath3 , and one for the alternative 0.602 @xmath3 ( girard et al . also listed for each sequence are the core composition ( c , o , or a more realistic salaris c / o mix ) , the helium layer mass ( @[email protected] as discussed earlier ) , and the heavy element abundance @xmath1=0.001 or 0 . all assume zero hydrogen abundance . the benchmark cooling age of procyon b is given by the model for 0.553 @xmath3 , with c / o core , log m(he)/m = @xmath252.5 , and @xmath1 = 0.001 in the envelope ( benchmark parameters ) . the cooling value is 1.187 gyr , with a formal , internal error of @xmath00.104 gyr due to uncertainties in the assumed effective temperature and the assumed mass . according to table 1 , a range of @xmath0200 k about the estimated effective temperature of @xmath14 = 7740 k translates into an age uncertainty of @xmath00.085 gyr , while a range of @xmath00.022 @xmath3 about the benchmark mass of 0.533 @xmath3 translates into an age interval of @xmath00.061 gyr . added in quadrature , these uncertainties lead to @xmath00.104 gyr . in addition , two sources of systematic error should be considered here . the age difference between a mixed ( c / o ) and a pure oxygen core is relatively small at 0.010 gyr . for its part , uncertainty due to variations in the helium layer mass is @xmath00.043 gyr . added again in quadrature with the above internal error , these systematic effects lead to a total uncertainty of @xmath00.113 gyr on the age . so we reach the value of [email protected] gyr for the white dwarf cooling time of procyon b. on the other hand , the @xmath1 = 0 envelope sequence with otherwise benchmark assumptions yields a considerably larger age of 1.391 gyr . since this is inconsistent with ultraviolet observations of carbon and heavy elements in the ultraviolet spectrum , we consider a metal - free envelope an unrealistic assumption . note that the atmospheric values determined by provencal et al . ( 2002 ) already have a heavy element abundance ( dominated by carbon ) of [email protected] . this parameter increases greatly below the convection zone throughout the partially - degenerate helium envelope ( following profile of carbon dredge - up ) , until it reaches the c / o core . likewise , we do not consider the assumption of a pure carbon core realistic . the combination of a pure carbon core with zero metallicity envelope yields an even larger cooling age of 1.557 gyr . the cooling age listed for the alternative mass of 0.602 @xmath3 , assuming benchmark values of the other parameters above , is 1.406 gyr . a mass this large for the secondary disagrees by more than 2@xmath4 from that derived from hst astrometry and we no longer consider it in the rest of this paper . these several possibilities in this paragraph are not factored into the error bars in the preceding paragraph . the difference between the age of the procyon system and the cooling time of the white dwarf component , [email protected] gyr , corresponds to the sum of the pre - main sequence lifetime and the main sequence lifetime ( plus the much shorter red giant phases ) of the procyon b progenitor . using again our tycho evolutionary code , we thus find that the initial mass ( @xmath44 ) of the secondary is 2.59@xmath45 @xmath3 , if the envelope z value is fixed at 0.001 . however , if one allows for the extreme upper limit to the cooling time from the assumption of z(env ) = 0 , the error bars for the progenitor mass increase to @xmath2 @xmath3 . the quoted errors on @xmath44 are the quadrature sums of @xmath250.21,+0.40 due to the incertainty in the cooling time of the white dwarf , and @xmath250.16,+0.20 due to systematic error in the age of the binary system ( table 2 ) . in figure 2 , discussed in detail in the next section , the solid horizontal error bars for procyon b are for the case of z(env ) fixed at 0.001 , the extended light error bars reflect the increase in cooling time for the limiting case z(env ) = 0 . it may be worthwhile at this point to attract the attention of the reader on some subtleties of white dwarf cooling which leave their signature on the data presented in table 1 . for instance , as is well known ( and revealed again in the table ) , pure c core models systematically cool slower than models having cores made of heavier elements , all other things being the same . that a pure carbon core takes longer to cool is well understood , because the thermal energy in a given mass is locked up in more nucleons than for pure oxygen or a mixture . however , that the @xmath1 = 0 envelope takes longer to cool than one with heavy elements is not so intuitive , because one might think that the more opaque envelope with heavy elements would let the interior energy out more slowly . our calculations show that cooling proceeds in two phases . at first the less opaque model does let the thermal energy escape more quickly , and core temperature decreases more rapidly . despite this , however , the decrease in luminosity has to be slower for the less opaque model because of that extra energy that is made available to fuel the luminosity . hence , in a first phase , the less opaque model actually cools _ more slowly _ to a given luminosity or @xmath14 . however then , after the less opaque star has let more of its thermal energy escape , there is no turning back and the price to pay , in a second phase , is that it will now cool _ faster _ the rest of the way than its more opaque counterpart . in the case of procyon b at @xmath14 = 7,740 k , it is still warm emough to be in the first phase of that `` relative '' cooling and , hence , this is why the cooling age is larger for the less opaque model . if the star had a @xmath14 of @xmath285,000 k , then indeed the cooling age based on the less opaque envelope sequence would be shorter than than of the more opaque , as the conventional wisdom would expect . these arguments may be discerned from a careful perusal of section iii(a ) , subsection ( ii ) of tassoul , fontaine and winget ( 1990 ) , although this discussion compares an opaque envelope of hydrogen with less opaque counterparts . as discussed in the last paragraph of the introduction , the initial - to - final mass relation ( ifmr ) for progenitors and white dwarfs is fundamental in understanding a stellar population . this study of procyon adds a data point near the low mass end with well determined stellar parameters ; it may be valuable for testing whether differences exist in the ifmr for the distributions of hydrogen and helium - rich atmosphere white dwarfs . table 2 lists final ( @xmath46 ) and initial ( @xmath44 ) masses in @xmath3 , cooling times ( log in years ) , and estimated errors , first for procyon b and sirius b. the remaining data points come from white dwarfs found in galactic star clusters , in order of increasing age ( see footnotes ) . the collective data set is shown in figure 2 . again , the age of the population is obtained generally from fitting the main sequence turnoff of the cluster color - magnitude diagram . as given in ferrario et al . ( 2005 ) , the errors in @xmath44 fall into two categories observational ( obs ) or random , and systematic ( sys ) , as listed in table 2 . the observational errors devolve from uncertainties in the @xmath46 from fitting the balmer lines ( or , in one case , he i lines ) ; the systematic errors come from uncertainties in the cluster or binary system age . since the nuclear lifetime of the progenitor is the difference between the systemic age and the white dwarf cooling time , @xmath44 depends on the age as well . note that the uncertainties in the age determinations for the young clusters lead to large error bars in the estimates of the initial masses @xmath44 . we add gd 50 to the pleiades `` moving group '' , since dobbie et al . ( 2006 ) make a strong case based on astrometric and spectroscopic data that this ultramassive white dwarf is associated with the star formation event that created the pleiades cluster . they argue that it evolved as a single star from a progenitor of 6.3 @xmath3 , and that this may represent the first observational evidence that single - star evolution can produce white dwarfs with , in this case , a mass of 1.27 @xmath3 ( bergeron , saffer , & liebert 1992 ) or 1.2(+0.07,@xmath250.08 ) @xmath3 ( bergeron et al . they make a weaker case that the massive white dwarf pg 0136 + 251 , [email protected] @xmath3 ( bergeron , saffer , & liebert 1992 ) , may also be related to the pleiades . however , information is lacking on the total ( @xmath47 ) space motion of this object , so we do not include it here . we adopt an age of 125@xmath025 myr for the pleiades , which depends on the treatment of the convective core . this excludes a systematic uncertainty to the age which will be true for all clusters with main sequence turnoff stars of mass above @xmath28 1.3 @xmath3 . williams , bolte , & koester ( 2004 , 2009 ) present studies of white dwarfs in the cluster ngc 2168 ( m35 ) ; this work supercedes the data listed in ferrario et al . ( 2005 ) . in the updated listing , 2168 - 22 is excluded since it is likely magnetic , and the weak balmer lines could not be fit by the authors . the age of the cluster is believed to be 175@xmath050 myr ( sung and bessell 1999 , von hippel 2005 ) . for ngc 3532 ( age 300@xmath025 gyr ) we take the view that the analysis with one of the 8.2 m vlt telescopes by dobbie et al . ( 2009b ) supercedes the earlier eso observations of koester & reimers ( 1993 ) . for ngc 2099 ( age 490@xmath070 myr ) , we note that star 2099 - 14 at [email protected] @xmath3 likely has a helium core , implying formation from binary star evolution . a new white dwarf in the 500@xmath0100 myr coma berenices open cluster ( melotte 111 ) , called 1216 + 260 , has been studied by dobbie et al . ( 2009a ) . based on assuming a co core and a thick h layer for this da , they calculate the parameters listed in table 2 . for the praesepe and hyades clusters , likely members of the hyades `` moving group '' ( eggen 1958 , zuckerman & song 2004 ) , we adopt as the age 625@xmath050 myr . the white dwarfs from the hyades were studied by claver et al . ( 2001 ) and ferrario et al . the only db star among the cluster white dwarfs used here is 0437 + 138 in the hyades . the praesepe cluster stars were studied by claver et al . ( 2001 ) , ferrario et al . ( 2005 ) , dobbie et al . ( 2004 ) , and casewell et al . ( 2009 ) . for objects in common between these last two papers with many common authors , we adopt the values of the last paper , since the results were obtained with the uves spectrograph on the vlt ut2 . we omit 0837 + 218 since casewell et al . ( 2009 ) make a good case that this is not a cluster member . white dwarfs in older clusters overlapping and exceeding the likely age of procyon have been analyzed and published since the ferrario et al . work . for the older clusters ( as well as procyon b and sirius b ) , the errors for @xmath44 are relatively smaller . kalirai et al . ( 2008 ) identified four likely members of the open cluster ngc 7789 , with age [email protected] gyr , and found two likely member white dwarfs in ngc 6819 with age [email protected] gyr . we note that the signal - to - noise ratio of the spectra with fitted balmer lines are low compared to p. bergeron s usual standards ; we quote the listed errors here . kalirai et al . ( 2007 ) analyzed a number of the brightest white dwarfs in the very old galactic cluster ngc 6791 ( [email protected] gyr ) ; however , only 6791 - 7 was the only likely member not having a very low mass determination and likely helium core . the authors note that enhanced mass loss in the red giant phase , which is the likely origin of the helium white dwarfs , probably also occurred in this object , despite the fact that it retained a hydrogen atmosphere . finally , kalirai et al . ( 2009 ) analyze a number of white dwarfs at the tip of the observed sequence in the globular cluster messier 4 . this is of course a population ii system with [ fe / h ] = @[email protected] ( mucciarelli et al . 2011 ) , and likely age of [email protected] gyr ( hansen et al . the basic conclusion is that the @[email protected] @xmath3 stars now producing white dwarfs which have evolved to approximately [email protected] @xmath3 ( but we list the individual determination for each of the latter ) . procyon b has a final mass below nearly all others of similar initial mass . it is one of only two white dwarfs here with a helium atmosphere . the other is the db in the hyades mentioned earlier , which has a mass similar to the other da stars in that cluster and the praesepe . the dashed line plots the simple linear relation from ferrario et al . ( 2005 ) based on a fit to white dwarfs in more massive clusters available at that time . the procyon b point lies more than 0.1 @xmath3 or several @xmath4 below this line in the sense of having a lower final mass . we should note that the close , cool double - dc binary g 107 - 70ab ( harrington , christy , & strand 1981 ) appears to be a similar situation at first glance . it was barely resolved from ground - based imaging at the u.s . naval observatory , flagstaff station . it is no surprise that this old result is superceded by 13 years of hst imaging ( 1995.8 - 2008.8 ) , as of the latest fit schaefer et al . ( 2006 , and private communication ) . the preliminary period is [email protected] years , @xmath21 = 0.666 `` , and @xmath48 = [email protected] @xmath3 ( using the hipparcos parallax of [email protected] '' ) . since only a small difference in magnitude between the two components is indicated , this suggests two components each of not quite 0.6 @xmath3 . using the integrated light from the two stars , bergeron , ruiz , & leggett ( 1997 ) estimate 4900 k , log @xmath27 = 7.35 , and their broad - band @xmath49 energy distribution fits an h - rich atmosphere . that is , the spectrum would have been type da if the temperatures of the two stars were warmer . the hst astrometry above suggests two white dwarfs each of mass near the well known peak of the da mass distribution ( koester , schulz , & weidemann 1979 ) . note that the log @xmath27 value inferred above is incorrect , since the method used by bergeron et al . ( 1997 ) to estimate a surface gravity assumes a single star . thus g 107 - 70ab are not cooler analogs of procyon b. we have no explanation for why the procyon b remnant appears so undermassive vs the initial mass we have derived . suppose that virtually all of the helium envelope of the procyon b nuclear progenitor could have been been lost in the event ( late helium - shell flash ? ) that also removed all of the hydrogen . however , stellar models of the asymptotic giant branch phase show that the maximum helium envelope mass remaining would have been only @xmath280.01@xmath3 if left intact . thus the loss of the entire helium envelope fails to account for the total mass shortfall by an order of magnitude . rather , the evolution of the ( presumably ) carbon - oxygen core was evidently truncated before it could build up to a more normal mass for a single star with the indicated initial mass . the dilemma suggests that close binary evolution may have been involved , but the existing binary consists of two stars in an elliptical orbit of long period . hence , we are left with an interesting problem in ( binary ) stellar evolution . note that there are still rather few stars analyzed from population i clusters of age older than or similar to procyon . in the future the numerous white dwarfs in the old disk clusters m 67 and ngc 188 will make useful additions to round out the low mass end of such a diagram . we gratefully acknowledge gail schaefer of the chara array at georgia state university , and howard e. bond of stsci , for allowing us to quote preliminary hst results which fix the masses of procyon a and b , and for results on g 107 - 70ab . this work was originally supported by the national science foundation through grant ast-0307321 ( jl and kaw ) . gf wishes to acknowledge the contribution of the canada research chair program . cccccccc & & & & & 7.9676 & 0.012777 & 1.1069 + 1 & c / o & @xmath252.5 & 0.001 & 0.553 & 7.9681 & 0.012768 & 1.1870 + & & & & & 7.9687 & 0.012760 & 1.2772 + & & & & & 7.9297 & 0.013077 & 1.0516 + 2 & c / o & @xmath252.5 & 0.001 & 0.531 & 7.9303 & 0.013068 & 1.1274 + & & & & & 7.9309 & 0.013059 & 1.2122 + & & & & & 8.0045 & 0.012485 & 1.1650 + 3 & c / o & @xmath252.5 & 0.001 & 0.575 & 8.0050 & 0.012479 & 1.2501 + & & & & & 8.0055 & 0.012471 & 1.3438 + & & & & & 7.9637 & 0.012834 & 1.2336 + 4 & c & @xmath252.5 & 0.001 & 0.553 & 7.9642 & 0.012826 & 1.3236 + & & & & & 7.9648 & 0.012817 & 1.4251 + & & & & & 7.9709 & 0.012728 & 1.0992 + 5 & o & @xmath252.5 & 0.001 & 0.553 & 7.9714 & 0.012720 & 1.1770 + & & & & & 7.9720 & 0.012712 & 1.2650 + & & & & & 7.9730 & 0.012696 & 1.2962 + 6 & c / o & @xmath252.5 & 0.0 & 0.553 & 7.9736 & 0.012688 & 1.3914 + & & & & & 7.9742 & 0.012680 & 1.4990 + & & & & & 7.9693 & 0.012752 & 1.4500 + 7 & c & @xmath252.5 & 0.0 & 0.553 & 7.9699 & 0.012743 & 1.5573 + & & & & & 7.9705 & 0.012734 & 1.6774 + & & & & & 7.9762 & 0.012651 & 1.2634 + 8 & o & @xmath252.5 & 0.0 & 0.553 & 7.9767 & 0.012643 & 1.3544 + & & & & & 7.9772 & 0.012635 & 1.4565 + & & & & & 7.9661 & 0.012798 & 1.0705 + 9 & c / o & @xmath252.0 & 0.001 & 0.553 & 7.9667 & 0.012790 & 1.1488 + & & & & & 7.9673 & 0.012781 & 1.2374 + & & & & & 7.9687 & 0.012756 & 1.1521 + 10 & c / o & @xmath253.0 & 0.001 & 0.553 & 7.9692 & 0.012747 & 1.2347 + & & & & & 7.9698 & 0.012739 & 1.3266 + & & & & & 8.0485 & 0.012145 & 1.3120 + 11 & c / o & @xmath252.5 & 0.001 & 0.602 & 8.0489 & 0.012138 & 1.4063 + & & & & & 8.0494 & 0.012132 & 1.5112 procyon b & 0.553 & 0.022 & 9.075 & 0.074 & 2.59 & 0.21,0.40 & 0.16,0.20 & 1 + sirius b & 1.000 & 0.020 & 8.091 & 0.020 & 5.056 & 0.171,0.262 & 0.213,0.273 & 2 + lb 1497 & 1.023 & 0.026 & 7.781 & 0.062 & 6.542 & 0.346,0.458 & 0.866,1.614 & 3 + gd 50 & 1.264 & 0.017 & 7.785 & 0.045 & 6.3 & 2.5,0.8 & , & 4,5 + ngc 2516 - 1 & 0.931 & 0.098 & 7.760 & 0.230 & 5.411 & 1.017,0.500 & 0.524,0.386 & 6 + ngc 2516 - 2 & 1.004 & 0.058 & 7.621 & 0.179 & 5.096 & 0.406,0.239 & 0.415,0.311 & 6 + ngc 2516 - 3 & 0.969 & 0.039 & 7.922 & 0.079 & 6.141 & 0.692,0.427 & 0.924,0.587 & 6 + ngc 2516 - 5 & 1.054 & 0.063 & 7.883 & 0.136 & 5.902 & 0.986,0.524 & 0.786,0.514 & 6 + ngc 2168 - 1 & 0.873 & 0.091 & 7.228 & 0.260 & 4.39 & 0.23,0.09 & 0.35,0.27 & 7 + ngc 2168 - 2 & 1.015 & 0.067 & 7.657 & 0.202 & 4.79 & 0.47,0.26 & 0.46,0.36 & 7 + ngc 2168 - 5 & 0.916 & 0.075 & 6.103 & 0.088 & 4.21 & 0.00,0.00 & 0.29,0.23 & 7 + ngc 2168 - 6 & 0.877 & 0.096 & 6.077 & 0.082 & 4.21 & 0.00,0.00 & 0.29,0.23 & 7 + ngc 2168 - 11 & 0.802 & 0.096 & 8.047 & 0.173 & 6.63 & 0.00,1.38 & 1.89,0.96 & 7 + ngc 2168 - 12 & 0.922 & 0.092 & 7.314 & 0.281 & 4.43 & 0.30,0.13 & 0.36,0.27 & 7 + ngc 2168 - 14 & 1.010 & 0.072 & 7.865 & 0.167 & 5.32 & 0.95,0.44 & 0.73,0.48 & 7 + ngc 2168 - 15 & 0.888 & 0.088 & 7.551 & 0.236 & 4.63 & 0.38,0.22 & 0.41,0.32 & 7 + ngc 2168 - 27 & 1.022 & 0.072 & 7.840 & 0.170 & 5.22 & 0.71,0.42 & 0.68,0.45 & 7 + ngc 2168 - 29 & 0.882 & 0.078 & 7.331 & 0.235 & 4.44 & 0.23,0.12 & 0.36,0.28 & 7 + ngc 2168 - 30 & 1.011 & 0.122 & 7.808 & 0.289 & 5.12 & 1.49,0.55 & 0.63,0.43 & 7 + ngc 2287 - 2 & 0.910 & 0.040 & 7.908 & 0.075 & 4.450 & 0.580,0.380 & , & 8 + ngc 2287 - 5 & 0.910 & 0.040 & 7.964 & 0.071 & 4.570 & 0.640,0.430 & , & 8 + ngc 3532 - 1 & 0.86 & 0.04 & 7.778 & 0.095 & 3.830 & 0.18,0.15 & , & 9 + ngc 3532 - 5 & 0.820 & 0.04 & 7.580 & 0.115 & 3.71 & 0.150,0.130 & , & 9 + ngc 3532 - 9 & 0.760 & 0.04 & 7.00 & 0.113 & 3.57 & 0.12,0.11 & , & 9 + ngc 3532 - 10 & 0.96 & 0.04 & 8.173 & 0.060 & 4.58 & 0.47,0.33 & , & 9 + ngc 2099 - 2 & 0.69 & 0.10 & 7.961 & 0.165 & 2.76 & 0.08,0.05 & 0.162,0.201 & 10 + ngc 2099 - 3 & 0.76 & 0.13 & 8.190 & 0.190 & 2.92 & 0.16,0.14 & 0.197,0.258 & 10 + ngc 2099 - 4 & 0.87 & 0.15 & 8.428 & 0.217 & 3.02 & 0.26,0.19 & 0.324,0.495 & 10 + ngc 2099 - 5 & 0.83 & 0.14 & 8.271 & 0.188 & 3.02 & 0.26,0.19 & 0.219,0.305 & 10 + ngc 2099 - 7 & 0.88 & 0.19 & 8.378 & 0.289 & 3.26 & 0.78,0.40 & 0.278,0.405 & 10 + ngc 2099 - 9 & 0.61 & 0.05 & 8.268 & 0.067 & 2.97 & 0.06,0.07 & 0.218,0.302 & 10 + ngc 2099 - 10 & 0.74 & 0.04 & 8.090 & 0.068 & 2.81 & 0.04,0.02 & 0.178,0.225 & 10 + ngc 2099 - 11 & 0.96 & 0.06 & 8.149 & 0.095 & 2.86 & 0.07,0.07 & 0.188,0.242 & 10 + ngc 2099 - 12 & 0.55 & 0.07 & 8.394 & 0.126 & 3.30 & 0.30,0.22 & 0.290,0.429 & 10 + ngc 2099 - 13 & 0.79 & 0.05 & 8.229 & 0.074 & 2.93 & 0.06,0.07 & 0.207,0.278 & 10 + ngc 2099 - 14 & 0.45 & 0.08 & 8.487 & 0.085 & 3.31 & 0.20,0.26 & 0.400,0.674 & 10 + ngc 2099 - 16 & 0.83 & 0.06 & 8.689 & 0.082 & 5.20 & 7.0,1.20 & 3.737, & 10 + 1216 + 260 & 0.90 & 0.04 & 8.560 & 0.052 & 4.77 & 0.97,5.37 & , & 11 + 0352 + 098 & 0.719 & 0.030 & 8.270 & 0.046 & 3.094 & 0.045,0.052 & 0.114,0.134 & 12 + 0406 + 169 & 0.806 & 0.031 & 8.488 & 0.046 & 3.465 & 0.109,0.147 & 0.172,0.225 & 12 + 0421 + 162 & 0.680 & 0.031 & 7.970 & 0.055 & 2.892 & 0.020,0.023 & 0.090,0.103 & 12 + 0425 + 168 & 0.705 & 0.031 & 7.549 & 0.077 & 2.789 & 0.011,0.010 & 0.079,0.088 & 12 + 0431 + 125 & 0.652 & 0.032 & 7.752 & 0.068 & 2.825 & 0.014,0.017 & 0.083,0.093 & 12 + 0437 + 138 & 0.740 & 0.060 & 8.470 & 0.047 & 3.663 & 0.030,0.030 & , & 13 + 0438 + 108 & 0.684 & 0.031 & 7.203 & 0.060 & 2.757 & 0.003,0.004 & 0.076,0.084 & 12 + 0833 + 194 & 0.721 & 0.043 & 8.431 & 0.049 & 3.32 & 0.33,0.22 & , & 14 + 0836 + 197 & 0.909 & 0.030 & 8.136 & 0.049 & 2.981 & 0.031,0.035 & 0.101,0.115 & 15 + 0836 + 199 & 0.831 & 0.037 & 8.612 & 0.069 & 3.997 & 0.350,0.652 & 0.297,0.428 & 15 + 0836 + 199 & 0.752 & 0.044 & 8.489 & 0.049 & 3.46 & 0.43,0.27 & , & 14 + 0836 + 201 & 0.620 & 0.031 & 8.154 & 0.050 & 2.993 & 0.028,0.046 & 0.102,0.117 & 15 + 0837 + 185 & 0.804 & 0.044 & 8.504 & 0.050 & 3.50 & 0.48,0.29 & , & 14 + 0837 + 199 & 0.819 & 0.032 & 8.351 & 0.048 & 3.194 & 0.081,0.062 & 0.153,0.129 & 15 + 0837 + 199 & 0.737 & 0.043 & 8.253 & 0.052 & 3.07 & 0.20,0.15 & , & 14 + 0840 + 190 & 0.849 & 0.045 & 8.566 & 0.050 & 3.73 & 0.71,0.39 & , & 14 + 0840 + 200 & 0.761 & 0.033 & 8.522 & 0.045 & 3.572 & 0.186,0.129 & 0.255,0.196 & 15 + 0840 + 200 & 0.721 & 0.043 & 8.420 & 0.048 & 3.30 & 0.32,0.21 & & 14 + 0843 + 184 & 0.823 & 0.045 & 8.530 & 0.051 & 3.59 & 0.55,0.33 & , & 14 + ngc 7789 - 4 & 0.560 & 0.020 & 8.061 & 0.032 & 2.080 & 0.080,0.080 & , & 16 + ngc 7789 - 5 & 0.600 & 0.030 & 6.903 & 0.055 & 2.020 & 0.070,0.140 & , & 16 + ngc 7789 - 6 & 0.720 & 0.030 & 8.204 & 0.045 & 2.100 & 0.090,0.090 & , & 16 + ngc 7789 - 8 & 0.640 & 0.040 & 7.462 & 0.080 & 2.020 & 0.090,0.110 & , & 16 + ngc 6819 - 6 & 0.530 & 0.020 & 7.591 & 0.035 & 1.600 & 0.060,0.050 & , & 17 + ngc 6819 - 7 & 0.560 & 0.020 & 8.155 & 0.034 & 1.620 & 0.070,0.050 & , & 17 + ngc 6791 - 7 & 0.530 & 0.020 & 8.176 & 0.060 & 1.160 & 0.040,0.030 & , & 18 + m 4 - 0 & 0.520 & 0.040 & 7.602 & 0.070 & 0.800 & 0.050,0.050 & , & 19 + m 4 - 4 & 0.500 & 0.030 & 7.279 & 0.024 & 0.800 & 0.050,0.050 & , & 19 + m 4 - 6 & 0.590 & 0.040 & 7.255 & 0.050 & 0.800 & 0.050,0.050 & , & 19 + m 4 - 15 & 0.550 & 0.040 & 7.322 & 0.043 & 0.800 & 0.050,0.050 & , & 19 + m 4 - 20 & 0.510 & 0.050 & 7.690 & 0.075 & 0.800 & 0.050,0.050 & , & 19 + m 4 - 24 & 0.510 & 0.030 & 7.204 & 0.028 & 0.800 & 0.050,0.050 & , & 19 + asplund , m. , grevesse , n. , & sauval , a.j . 2005 , in _ cosmic abundances as records of stellar evolution and nucleosynthesis in honor of david l. lambert _ , asp . series , v. 336 , eds . t. g. barnes iii & f. n. bash ( san francisco : astronomical society of the pacific ) p. 25 lodders , k. 2010 , `` solar system abundances of the elements '' , in _ principles and perspectives in cosmochemistry , astrophysics & space sciences proceedings _ , vol . isbn 978 - 3 - 642 - 10351 - 3 ( berlin heidelberg : springer - verlag ) , p. 379 fig . 1 tycho evolutionary tracks in a log luminosity vs. log radius diagram beginning including the pre - main sequence at the asplund , grevesse , & sauval ( 2005 , 2009 = a09 ) primordial solar abundance of z=0.0122 for masses of ( in descending order of luminosity ) 1.499 @xmath3 ( black ) , the favored preliminary hst astrometric value , 1.48 @xmath3 ( red ) , the best fit to the observed luminosity , 1.463 @xmath3 ( green ) , and 1.42 @xmath3 ( blue ) . fig . 2 the initial to final mass relation ( ifmr ) for procyon b , sirius b and the white dwarfs in a number of star clusters discussed in the text . procyon b and sirius b ( from paper i ) are plotted as black , filled circles . the total data set for older open clusters from kalirai et al . ( 2007 , 2008 , 2009 ) are also plotted with filled symbols ngc 7789 ( red ) , ngc 6819 ( green ) , ngc 6791 ( blue ) and messier 4 ( magenta ) . shown with open circles taken from ferrario et al . ( 2005 ) and paper i are the pleiades ( red ) , ngc 2516 ( cyan ) , ngc 2168 ( magenta ) , the hyades ( blue ) and praesepe ( cyan ) , ngc 3532 ( green ) , and ngc 2099 ( yellow ) . the error bars include the best estimates of the cluster ages . the magenta line plots the simple linear relation from ferrario et al . the procyon b remnant mass appears low relative to the others , as discussed in the text .

we present a semi - analytical model of high redshift galaxy formation . in our model the star formation inside a galaxy is regulated by the feedback from supernova ( sne ) driven outflows . we derive a closed analytical form for star formation rate in a single galaxy taking account of the sne feedback in a self - consistent manner . we show that our model can explain the observed correlation between the stellar mass and the circular velocity of galaxies from dwarf galaxies to massive galaxies of . for small mass dwarf galaxies additional feedback other than supernova feedback is needed to explain the spread in the observational data . our models reproduce the observed 3-d fundamental correlation between the stellar mass , gas phase metallicity and star formation rate in galaxies establishing that the sne feedback plays a major role in building this relation . further , the observed uv luminosity functions of lyman - break galaxies ( lbgs ) are well explained by our feedback induced star formation model for a vast redshift range of . in particular , the flattening of the luminosity functions at the low luminosity end naturally arises due to our explicit sne feedback treatment . galaxies : high - redshift ; galaxies : star formation ; stars : winds , outflows ; galaxies : luminosity function ;

1401.7675

presently we possess a wealth of observations regarding the high redshift universe , thanks to present day technology . the galaxies are regularly being detected till redshift @xmath2 using lyman break technique ( steidel et al . , 2003 ) . the uv luminosity functions of lyman break galaxy ( lbg ) are well constrained upto redshift @xmath3 using hubble ultra deep field observations ( bouwens et al . , 2007 ; bouwens et al . 2008 ; reddy & steidel , 2009 ; oesch et al . , 2010 ; bouwens et al . , 2011 ) . the faint end slope of the uv luminosity functions is well established for @xmath4 and it shows flattening of the luminosity function at low luminosity end . the presence of gunn - peterson ( gunn & peterson , 1965 ) absorption in the spectrum of high redshift quasars tells us a transition from highly ionised inter galactic medium ( igm ) to partially ionised igm around redshift @xmath5 ( wyithe , loeb & carilli , 2005 ; fan et al . , 2006 ; mortlock et al . , 2011 ) . also the electron scattering optical depth ( @xmath6 ) measured in the cosmic microwave background radiation ( cmbr ) by wilkinson microwave anisotropy probe ( wmap ) constrains the reionization redshift to be @xmath7 ( komatsu et al . 2011 ) for a step reionization scenario . further , the high resolution absorption spectra of quasars show the presence of metals in very low density igm far away from galaxies . metals are produced inside galaxies and believed to be transported by outflows produced by the supernova ( sne ) explosions in the galaxy . the outflows are routinely being observed in low redshift galaxies as well as in high redshift galaxies ( martin , 1999 ; pettini et al . , 2001 ; martin , 2005 ) . these outflows are likely to expel metals along with a large amount of inter stellar medium ( ism ) . this , in turn , reduces the star formation in galaxies by reducing the available gas to form new generation of stars . thus the supernovae give a negative feedback to the star formation by throwing out gas from galaxies in the form of galactic winds . not only that , outflows also regulate the amount of metals in galaxies . starting from very early days , a tight correlation has been observed between luminosity and metallicity of galaxies ( lequeux et al . later , a more fundamental correlation has been found between the stellar mass and gas phase metallicity of galaxies in the local universe ( garnett , 2002 ; tremonti et al . , 2004 ; lee et al . , 2006 ; kewley & ellison , 2008 ) as well as in high redshift galaxies ( savaglio et al . , 2005 ; erb et al . , 2006 ; mannucci et al . , 2009 ; mannucci et al . , it has been observed that galaxies with higher stellar mass trend to have more metals compared to their lower stellar mass counter parts . further observations of local as well as high redshift universe show that the luminosity - metallicity or mass metallicity relation observed in galaxies is due to a more general relationship between stellar mass , metallicity of the gas and star formation rate ( mannucci et al . 2010 ; cullen et al . 2013 ) . since , the outflows from galaxies throw metal enrich gas it is most likely that outflows play an important role in building up this correlation ( kobayashi et al . , 2007 ; scannapieco et al . , 2008 ) . it has been seen from observations that the amount of gas / ism expelled from a galaxy due to outflows is inversely proportional to the mass of the galaxy ( martin , 1999 ; martin 2005 ) . this is expected if the hot gas produced in sne explosions drives the outflow . roughly 10% of total sne energy would be available for driving the outflow when the sne remnants started to overlap with each other ( cox , 1972 ) . the conservation of sne energy available to drive the outflow to the kinetic energy of outflowing gas would lead to a mass outflow rate inversely proportional to the square of the circular velocity ( @xmath8 ) of the galaxy . even if the hot gas loses its thermal energy due to radiative cooling the momentum of the gas and/or the cosmic rays produced in the sne shocks can still drive the outflow ( samui et al . , 2010 ; ostriker & mckee 1988 ) . in such cases as well the inverse relation between outflowing mass and the circular velocity of the galaxy still holds with a different scaling . thus due to supernova explosions small mass galaxies would lose more gas and experience a strong negative feedback to the star formation compare to higher mass galaxies . hence , it is important to build a complete model of high redshift galaxy formation taking account of all the observational evidences , particularly the sne feedback driven star formation in high redshift galaxies and their metal transport to the igm . numerical simulations are the best way to study all these together and a tremendous effort is going on ( for example , scannapieco et al . , 2005 & 2006 ; dave , oppenheimer & sivanandam , 2008 ; dave , oppenheimer , & finlator , 2011 ; scannapieco et al . , however , present state of art hydrodynamic simulations are far from reality . they are constrained by the resolutions as well as the amount of physical processes that they can take account together ( scannapieco et al . , 2012 ; stringer et al . 2012 ) . here , we build an analytical model of star formation in the high redshift galaxies regulated by the feedback from sne driven winds and try to explain the amount stellar mass and metals detected inside galaxies and the high redshift uv luminosity functions of lyman - break galaxies ( lbgs ) . in past , several authors have proposed semi - analytical models in order to understand the high redshift as well as low redshift galaxy formation process ( white & frenk , 1991 ; kauffmann , white & guiderdoni , 1993 ; cole et al . , 1994 ; baugh et al , 1998 ; somerville & primack , 1999 ; chiu & ostriker , 2000 ; granato et al 2000 ; choudhury & srianand , 2002 ; baugh et al . , 2005 ; shankar et al 2006 ) . these works clearly demonstrated the power of such semi - analytical modeling by predicting various observations regarding high redshift universe . however , they have not derived a universal closed analytical form for the time evolution of star formation rate ( sfr ) in a single galaxy . chiu & ostriker ( 2000 ) , choudhury & srianand ( 2002 ) and some of our earlier works ( samui et al . , 2007 , 2008 ; jose et al . , 2011 ) have used such a closed form without considering the sne feedback and also not deriving their star formation model from first principle . others have just considered star formation rate to be proportional to the available cold gas and not derived a single evolution equation for the evolution of the star formation rate including feedback . here , we solve for the star formation rate in a closed form starting from very basic physics governing the star formation and taking account of the negative feedback from galactic outflows on the star formation in a self - consistent manner . this closed form will be very useful while fitting the photometric observations of high redshift galaxies in order to find their star formation history , stellar mass etc . moreover , the new data are extended to much lower in stellar mass / luminosity and higher redshift where the process of reionisation is still going on . in one hand these low mass systems are the dominating sources of reionization . on the other hand they are much likely to prone to sne feedback . hence , it is timely to revisit feedback induced star formation in high redshift galaxies in the light of new improved data sets . further , semi - analytical models are always useful as they are computationally inexpensive and help to understand the average universe very well . also it is important to explore vast range of parameters that regulates the physical processes happening inside a galaxy . the paper is organised as follows . in section [ sec_sfig ] we clearly state our feedback induced star formation model in galaxies and how well it explains the stellar mass detected in dwarf galaxies to high mass galaxies . the mass - metallicity - sfr relation of galaxies is discussed in section [ sec_mass_metal ] . we present our model of uv luminosity function in section [ sec_lf ] . we show our model predictions of uv luminosity functions of lbgs and compare that with observations in section [ sec_fit ] . finally we draw our conclusions with some discussions in section [ sec_cd ] . through out this paper we assume a @xmath9cold dark matter ( @xmath9cdm ) cosmology with the cosmological parameter as obtained by recent wmap observation , i.e. @xmath10 , @xmath11 , @xmath12 and hubble parameter @xmath13 km / s / mpc . we model the star formation rate including the feedback from sne driven outflows in a galaxy as follows . we assume that the instantaneous star formation rate at a given time is proportional to the amount of cold gas present in the galaxy . once the dark matter halo virialises it accretes baryonic matter and a fraction , @xmath14 , of that becomes cold and available for star formation . the @xmath14 can be thought of as star formation efficiency and is a free parameter in our model . the baryon accretion rate in a galaxy of total mass @xmath15 at time @xmath16 after the formation of dark matter halo is taken as @xmath17 } \label{eqn_accretion}\ ] ] where @xmath18 is the gas mass and @xmath19 is the total baryonic mass in the halo . further , @xmath20 is the dynamical time of the galaxy ( barkana & loeb , 2001 ) . we assume that the total gas mass in the galaxy is equal to the dark matter mass times the universal dark matter to baryon mass ratio . note that integrating eq . [ eqn_accretion ] from @xmath21 to @xmath22 results @xmath23 . the exponential form of the baryon accretion rate can be understood as follows . once the dark matter halo virialises , the baryons are captured in the potential well and heated to virial temperature of the dark matter potential . in order to form stars the gas needs to cool and fall into the centre of the galaxy . if one assumes the rate of hot gas becoming cold is proportional to the amount of hot gas present , the increase in cold gas mass would follow an exponential form with time scale governed by the cooling time ( @xmath24 ) . however , as already mentioned , this cold gas has to collapse into the centre of the galaxy in order to form stars . the collapse of the gas into the centre of dark matter halo is governed by the dynamical time scale of the gravitational potential . for most of the galaxy masses that we are interested , @xmath25 ( silk , 1993 ) . this is true even at the mean overdensity of the collapsed halo which is @xmath26 180 times the background density . hence the effective cold gas accretion rate can be taken as an exponential form with time scale of the order of dynamical time scale for the system . further note that the accretions of cold gas in galaxies has been found to be important observationally and also considered in previous semi - analytical models ( i.e. kauffmann et al . 1993 , shankar et al . our exponential cold gas accretion rate tries to model that . such an exponential form for the cold gas accretion rate has been also used by other semi - analytical works ( for example see appendix of shankar et al . 2006 ) . at a given time some gas would already be locked inside stars . further , the massive stars are short lived and would explode as supernova after few times @xmath27 yrs . these supernovae can drive the cold gas out of galaxy as galactic wind . we assume that the amount of gas mass driven out by the supernova in the form of wind is proportional to the instantaneous star formation rate , neglecting the time delay of @xmath28 yrs between the star formation and subsequent explosion of supernova , i.e. @xmath29 here , @xmath30 is the wind mass ; @xmath31 is the star mass and over dot represents the time derivative . the proportionality constant @xmath32 depends on the driving mechanism of outflows . for example , if the outflows are driven by the hot gas produced by the sne , then @xmath33 , @xmath8 being the circular velocity of the galaxy . such outflows are referred to as energy driven outflows . if the hot gas loses its energy by radiative cooling , then outflows can be potentially driven by the pressure of cosmic rays produced in the sne shocks . it was shown analytically in samui et al . ( 2010 ) that in such cases @xmath33 as well . however , if the momentum of the cooled gas helps in driving the outflow , @xmath34 . these are called momentum driven outflows . detailed models of such supernova driven outflows can be found in weaver et al . ( 1977 ) , ostriker & mckee ( 1988 ) , scannapieco et al . ( 2002 ) , veilleux , cecil , & bland - hawthorn ( 2005 ) , samui et al . ( 2008 ) . finally , taking account of the mass in stars , mass in outflows and baryonic mass accreted we can write down the star formation rate at time @xmath16 after the formation of dark matter halo as @xmath35 , \label{eqn_dmdt}\ ] ] with @xmath36 is some proportionality constant that governs the duration of star formation . taking the time derivative of eq . [ eqn_dmdt ] and putting values of @xmath37 and @xmath38 from eq . [ eqn_accretion ] and [ eqn_eta ] respectively , we get @xmath39 \nonumber \\ & = & \frac{f_t}{\tau}\left[f_*\frac{m_b}{\tau}e^{-\frac{t}{\tau } } - \frac{d m_*}{dt } - \eta_w\frac{dm_*}{dt}\right].\end{aligned}\ ] ] integrating this with the boundary condition that @xmath40 and @xmath41 at @xmath21 we have @xmath42}\bigg [ e^{-\frac{t}{\tau } } - e^{-f_t(1+\eta_w)\frac{t}{\tau } } \bigg ] . \label{eqn_dm_dt_2}\ ] ] eq . [ eqn_dm_dt_2 ] gives the analytical form of star formation rate in a galaxy at age @xmath16 in presence of supernova feedback and is the backbone of our feedback induced star formation model . note that putting @xmath32 equals to zero would lead to star formation rate in a without feedback scenario and that form of star formation rate has been widely used in semi - analytical galaxy formation model in past by various authors ( see chiu & ostriker , 2000 ; choudhury & srianand , 2002 ; samui et al . , 2007 ; jose et al . , 2011 ; jose et al . , 2013 ) . it is interesting to note that the total baryonic mass that will be eventually converted to stars is @xmath43 which is independent of @xmath36 and inversely proportional to @xmath44 . we have already mentioned that @xmath45 where @xmath46 is the circular velocity of the halo for which @xmath47 and @xmath48 or @xmath49 depending on whether outflows are energy driven / cosmic rays driven or momentum driven . hence , for a small mass galaxy that has a lower circular velocity , @xmath32 is large compare to a high mass galaxy with higher @xmath8 . thus from eq . [ eqn_m_total ] it is clear that our model predicts a higher stellar mass fraction ( @xmath50 or equivalently @xmath51 ) in high mass galaxies and lower stellar mass fraction for small mass dwarf galaxies . the sne feedback regulates the total amount of star formed in a galaxy . for small mass dwarf galaxies the sne feedback is strong leading to a large decrease in the star formation in those galaxies . large mass galaxies due to their large gravitational potential are less prone to the sne feedback and suppression of star formation due to such feedback is small . indeed , observationally we found such a correlation between the stellar mass and circular velocity / mass of galaxy . in fig . [ fig_stellar_mass ] we show our model predictions of stellar masses as a function of both circular velocity and mass of a galaxy along with the observational data . the observed data points are taken from mcgaugh ( 2005 ) , stark et al . ( 2009 ) , walker et al . ( 2009 ) and gerhard et al . ( 2001 ) . note that in absence of any feedback , one expects @xmath52 which would be a straight line as shown by the dotted lines in fig . [ fig_stellar_mass ] . from observational data itself it is obvious that small mass galaxies have very small amount of stars . at @xmath53 the suppression is almost two orders of magnitude compared to the universal dark matter to baryon ratio whereas at @xmath54 the suppression is almost zero . thus star formation models that do not take account of feedback fail to reproduce the observation . one must consider the negative feedback of sne on subsequent star formation in a consistent way like in our model . we show our model predictions for @xmath48 ( left panel ) i.e. energy driven sne feedback and @xmath55 i.e. momentum driven sne feedback . the solid and dashed lines in both the panels are for @xmath56 km / s and 75 km / s respectively . we choose such normalisation as they provide a reasonable good fit to the data and also in samui et al . ( 2010 ) such a normalisation naturally arises for cosmic ray driven outflows ( see table 1 of samui et al . , we have assumed @xmath57 for all the models . note that with this choice of @xmath14 , less than 20% of baryon mass is converted to star for @xmath58 . it is clear from the figure that our model predictions match quite well with the observed data . for high mass galaxies with @xmath59 km / s the errors in the measurements as well as the scatter in the data are small and we obtain a good fit to the observed data for both @xmath60 and 1 . the spread in the data is within the statistical uncertainty of the model parameters . it is interesting to note that the observed data sample consists of different types of galaxies . while the stark et al . ( 2009 ) galaxy samples are gas dominated spiral galaxies , the mcgaugh et al . data consists of both gas dominated as well as gas poor star mass dominated galaxies . on the other hand , the gerhard et al . ( 2001 ) samples are compendium of the early type galaxies . hence our model reproduces the observed stellar mass fraction for different galaxy population with masses @xmath61 . for small mass dwarf galaxies with @xmath62 km / s the observational points are quite scatter although our model predictions agree reasonably well especially for @xmath60 . for @xmath55 it seems that additional feedbacks other than the sne feedback are needed to suppress star formation even more in those galaxies in order o explain majority of the data points . however in any model the spread in the observational data is much more than the statistical uncertainty of the model parameters . interestingly these dwarf galaxies are mostly satellite galaxies and they are likely to be affected by tidal striping and radiative feedback from reionization that we discuss later while considering the luminosity function of lbgs . hence , it is not surprising that our simple model of star formation including only the sne feedback fails to explain the spread in the observed data for dwarf galaxies . also note that for higher values of @xmath32 one may expect the radiative cooling of the outflowing material to be important and the hot thermal gas that drives the outflows cools efficiently making the outflow a momentum driven one ( see samui et al . 2009 for details example of such outflows ) . however , our models show that @xmath60 is more favourable even for low mass dwarf galaxies which points to the fact that the outflows from dwarf galaxies are most likely not momentum driven outflows . therefore , one must consider some alternative driving force for the outflows in these galaxies . in samui et . al ( 2010 ) we showed that cosmic rays can provide such alternative with correct scaling of @xmath60 . in order to investigate this more we also consider the ratio of stellar mass to halo mass ( shm ) as a function of halo mass . in fig . [ fig_shm ] we show our model predictions along with observationally derived shm ratio by leauthaud et al . ( 2012 ) ( green shaded region ) and moster et al . ( 2010 ) ( cyan hatched area ) with measurement uncertainty . leauthaud et al . used deep cosmos data along with halo occupation distribution model to constrain shm ratio . moster et al . derived the shm ratio from n - body simulation with recent sdss clustering data and galaxy galaxy lensing data . even though there are slight mismatch in their derived shm ratio , our model predictions match well considering the uncertainty in measurement . here also we see that momentum driven cases ( @xmath55 ) provide a poorer fit to the derived shm ratio compared to @xmath60 models in low mass region . we wish to point out that the decrease in shm ratio for @xmath63 is due to agn feedback that we have not considered here . also the observed galaxies have a limiting stellar mass of @xmath64 corresponds to @xmath65 that has been used to derived the shm . hence one should not compare our model predictions with this particular observation beyond this mass limit . in this section we focus on the gas phase metallicity of the high redshift galaxies in order to understand the observed mass - metallicity relation . the metallicity of a galaxy in our model can be calculated as follows . we assume that the amount of metals ejected by the supernova per unit of star formation is @xmath66 and these metals are mixed with the ism instantaneously . suppose at a given time the metallicity of the ism is @xmath67 . if at that instance @xmath68 amount of stars are formed then total gas mass reduced from the ism is @xmath69 . the second term arises due to the loss of gas through outflows . note that we neglect the recycled gas that is returned by the supernova into the ism as it is very small . for example , in a salpeter imf from @xmath70 the return fraction is less than 15% . given @xmath68 of star formation , the amount of metal lost from the ism is @xmath71 . however , supernova will produce @xmath72 metals . hence total change of metal mass ( @xmath73 ) in the ism is @xmath74\delta m_*.\ ] ] recall that @xmath75 . differentiating this and using above relations we obtain @xmath76 \delta m _ * \end{aligned}\ ] ] integrating this with boundary condition that the initial metallicity of the collapsed gas is @xmath77 , we obtain @xmath78 \label{eqn_mass_metal}\ ] ] in fig . [ fig_mass_metal ] we show our model prediction for stellar mass metallicity relation . note that the stellar mass is obtained from eq . [ eqn_m_total ] and metallicity is obtained from eq . [ eqn_mass_metal ] . the observed data points are taken from erb et al . ( 2006 ) for the sample of galaxies at @xmath79 . we take @xmath80 which means one supernova will form per @xmath81 of star formation and it will produce @xmath82 of oxygen ( starburst99 : leitherer et al . , 1999 ) . further , we assume @xmath83 . it is clear from the figure that our models reproduce the observed correlation very well . especially , for the case of @xmath60 models , the two normalisation circular velocities namely @xmath84 and 125 km / s nicely bracket the observed correlation function . predicted mass metallicity correlation by @xmath55 models is flatter than the observed values . therefore , this correlation also favours the @xmath60 models compared to @xmath55 models which produce poorer fit to the observation . and we conclude that the wind feedback is the main driver to determine the amount of metals in galaxies . as we have discussed early , recent observations have pointed out that the correlation of stellar mass and gas phase metallicity actually comes from a projection of more fundamental 3-d relationship between stellar mass , metallicity and star formation rate of galaxies . the offset seen in the mass metallicity relation between local universe and high redshift universe is due to selection bias of observing only higher star forming galaxies at high redshift ( mannucci et al . 2010 ) . hence any model of galaxy formation should be able to reproduce this fundamental 3-d relation . indeed our models do predict such a correlation between stellar mass , metallicity and sfr . in fig . [ fig_fmr ] we show both projections of this relationship as predicted by our models along with the observational data taken from mannucci et al . the top panels show metallicity as a function of stellar mass ( the color shaded area ) as obtained from all the four models discussed above ( the parameters of each model is indicated in each panel ) . the color bar represent the star formation rate of the galaxy . in each panel the color codded curves are the observational data with same color indicator for the star formation rate . in the bottom panels we show the metallicity as a function of star formation rate for a range of stellar masses . it is obvious from the figure that our models produce the 3-d correlation observed in galaxies . all our models produces the trend , i.e. for a given stellar mass , higher star forming galaxies have lower metallicity compared to low star forming counterparts or for a given star formation higher metallicity is obtained for higher stellar mass . however , it is interesting to point out that the model with @xmath60 and @xmath84 km / s provides the best fit with the observation . note that in this case the model has to reproduce both the projections simultaneously . like , the model with @xmath60 and @xmath56 km / s provide good fit with mass - metallicity relation but does not reproduce spread in sfr - metallicity relation . both the models with @xmath55 unable to explain the correctly the observational data . we therefore conclude like in previous section that the observations are well reproduced by @xmath60 models again indicating the importance of cosmic ray driven winds . after successfully explaining the stellar and metal mass detected in galaxies , we turn to the high redshift uv luminosity functions of lbgs . we broadly follow samui et al . ( 2007 & 2009 ) to calculate high redshift luminosity functions . below we briefly describe our model . we obtain luminosity evolution of a galaxy undergoing a burst of one unit of star formation at a specific wavelength ( say @xmath85 ) from population synthesis code starburst99 " ( leitherer et al . , 1999 ) assuming some initial mass function throughout this paper . ] for the stars formed . we convolve that with star formation rate of individual galaxy in our model ( i.e. eq . [ eqn_dm_dt_2 ] ) to obtain the luminosity as a function of galaxy age ( see eq . 6 of samui et al . , 2007 and also fig . 1 there ) . note that the difference with samui et al . ( 2007 ) model is that here we explicitly use the sne feedback to the star formation . further , all the light produced inside a galaxy by the stars do not reach to us due to presence of dust . a fraction , @xmath86 , of total luminosity of the galaxy can be observed from earth . we assume each luminous galaxy is formed inside a virialised dark matter halo provided the gas can cool and host star formation . the formation rate of dark matter halos per unit volume , @xmath87 , at a given redshift is calculated by taking the redshift derivative of sheth - tormen mass function ( sheth & tormen , 1999 ) . the sheth - tormen ( st ) mass function provides a good fit to the numerical simulation data ( brandbyge et al . , 2010 ) and hence we use it to calculate the formation rate of dark matter halos . note that the time derivative of mass function provides total change in the number of halos ; not just formation rate . one can extend sasaki formalism ( sasaki , 1994 ) for press - schechter mass function in order to get formation rate of halos from st mass function . however , it is well know that naive extension of this formalism does not work for other mass functions ( samui et al . , 2009 ; mitra et al , 2011 ) . hence we use the simple time derivative to get the formation rate of halos with the assumption that it closely follows the formation rate . however , to show the effect of mass function on the predicted luminosity functions , we also consider press - schechter ( ps ) mass function with sasaki formalism to calculate the formation rate of dark matter halos . in particular , we want to see if the conclusion of samui et al . ( 2009 ) is still valid with the new feedback induced star formation model . the cumulative luminosity function , @xmath88 for luminosity @xmath89 at a given redshift @xmath90 can be obtained from @xmath91 . \label{eqn_lf_cum}\ ] ] here , @xmath92 is the number of dark matter halos per unit volume between mass @xmath15 and @xmath93 and collapsed between redshift @xmath94 and @xmath95 . the heaviside theta function @xmath96 $ ] ensures that the integral is contributed only by galaxies that are formed at redshift @xmath94 and having luminosity , @xmath97 , greater than @xmath89 ( after correcting for dust reddening ) at observe redshift @xmath90 , with @xmath98 . taking derivative of eq . [ eqn_lf_cum ] with respect to @xmath89 and multiplying by the jacobi @xmath99 ( @xmath100 is the magnitude in ab system ( oke & gunn , 1983 ) ) we get the luminosity functions , @xmath101 , at a given redshift . in our model we calculate the reionization history and radiative feedback to the star formation in a self consistent way ( for detail see samui et al . , 2007 ) . a galaxy forming in the neutral region can cool with the help of atomic cooling if its virial temperature of the dark matter halo is greater than @xmath102 k. below this temperature a galaxy can cool and host star formation only in presence of molecular hydrogen . in this paper we consider only the atomic cooled halos as they are the main contributor to the luminosity functions in the observable range . in the ionised region of universe , due to the increase of jeans mass , a galaxy can host star formation only if its virial velocity is greater than 35 km / s . we assume a complete suppression of star formation for @xmath103 km / s and no suppression above @xmath104 km / s . for intermediate mass range we adopt a linear fit from 1 to 0 ( broom & loeb , 2002 ; benson et al . , 2002 , dijikstra et al . , we also assume a suppression of star formation in high mass halos by a factor of @xmath105^{-1}$ ] due to possible feedback from agn ( bower et al . , 2006 ; best et al . , 2006 ) in this section we show our model predictions for the uv luminosity functions of lbgs and compare them with observations in the redshift range @xmath106 . note that two free parameters of our model , the star formation efficiency , @xmath14 and dust reddening correction factor , @xmath107 come as a product like @xmath108 while calculating the luminosity of a galaxy and hence the luminosity function at a given redshift ( we have assumed @xmath109 ) . we use @xmath110-square minimization technique to fit observed luminosity functions at different redshifts taking @xmath108 as a free parameter . taking this product ( @xmath108 ) as independent of mass of the galaxy may seem simplistic . however , note that both @xmath14 and @xmath107 can depend on the mass of the galaxy ( bouwens et al . , 2012 ) . since we can not disentangle one from other while calculating the luminosity we do not consider mass dependent @xmath14 and/or @xmath107 here further , @xmath107 measures the amount of reddening due to dust . observationally it has been found that the high redshift galaxies exhibit smaller amount of dust compared to their low redshift counter part by measuring the uv continuum slope ( hopkins & beacom , 2006 ; bouwens et al . , 2012 ) . from physical point of view , high redshift galaxies should have less amount of metals and hence little dust . thus , we expect that @xmath108 should have different values at different redshift even if the efficiency remains same and we try to constrain this by fitting the observed uv luminosity function . before showing the luminosity functions at a given redshift , we show the star formation rate as obtained from eq . [ eqn_dm_dt_2 ] and resulting luminosity evolution of a single galaxy as a function of galaxy age . in fig . [ fig_sfr_lf ] , we show star formation rate ( sfr ) ( left panels ) and uv luminosity at 1500 ( right panels ) for galaxies with masses @xmath111 , @xmath112 , @xmath113 , @xmath114 and @xmath115 from top to bottom assuming collapsed redshifts of @xmath116 , 6 , 8 , 10 and 12 respectively . note that @xmath94 fixes the dynamical time scale @xmath20 ( barkana & loeb , 2001 ) and we choose @xmath94 such that the halo can collapse from a @xmath117 density fluctuation at @xmath94 . we also take @xmath109 for the rest of the paper . note that changing @xmath36 would change the duration of star formation . larger @xmath36 means a shorter duration of star formation representing a burst mode of star formation and smaller @xmath36 corresponds to a prolong star formation scenario in the galaxy . in fig . [ fig_sfr_lf ] we show models with @xmath60 & @xmath56 km / s ( dark green dotted curves ) , @xmath60 & @xmath84 km / s ( magenta long dashed curves ) , @xmath55 & @xmath56 km / s ( blue short dashed curves ) , and @xmath55 & @xmath84 km / s ( cyan dotted dashed curves ) . for comparison we also show the without feedback model ( @xmath118 ) with black solid line . [ fig_sfr_lf ] nicely demonstrates that our sne feedback models affect the star formation most in the low mass galaxies as @xmath32 is higher in those galaxies . the maximum suppression of star formation happens for models with @xmath60 and @xmath56 km / s and the effect is minimum for models with @xmath55 and @xmath84 km / s . for @xmath119 galaxy the suppression is almost two orders of magnitude in star formation rate that results two orders of magnitude decrease in the maximum luminosity compare to no feedback model ( bottom panels of fig . [ fig_sfr_lf ] ) . note that peak of the star formation hence the luminosity also happens early in time with @xmath120 . the suppression in star formation decreases with increasing galaxy mass ; we see one order of magnitude suppression in peak star formation rate and also in maximum luminosity for galaxy of @xmath121 ( middle panels of fig . [ fig_sfr_lf ] ) . for even higher mass galaxies sne feedback has very little effect on the star formation and hence on the luminosity of the galaxy . one hardly notices the difference ( less than factor 2 ) in star formation rate / luminosity for various feedback models with no feedback model for @xmath0 galaxy ( top panels of fig . [ fig_sfr_lf ] ) . interestingly such kind of feedback effect due to the sne on the star formation history of individual galaxies are also seen in full hydrodynamic simulations ( scannapieco et al . , 2006 ) hence our sne feedback model affects and regulates the star formation / luminosity of galaxies differentially , low mass galaxies get affected most and high mass galaxies least . such a feedback would change the slope of the luminosity functions at different redshifts making them flatter . indeed we see such flattening in the observational data of uv luminosity functions of lbgs . in figs . [ fig_lf_fit ] and [ fig_lf_fit_ps ] , we show our model predictions of luminosity functions in the redshift range @xmath1 along with the observational data for st and ps mass function respectively . the observed data points are taken from oesch et al . ( 2010 ) ( @xmath122 , 1.9 and 2.5 ) , reddy & steidel ( 2009 ) ( @xmath123 ) , bouwens et al . ( 2007 ) ( @xmath124 , 5 and 6 ) and mannucci et al . ( 2007 ) , bouwens et al . ( 2008 ) and bouwens et al . ( 2011 ) for @xmath125 and 8 . note that the open triangles at @xmath124 , 5 and 6 suffer from incompleteness problem ( see bouwens et al . , 2007 ) and hence we do not consider them while fitting . the fitted values of @xmath108 along with the best fit @xmath110-square per degree of freedom ( dof ) are tabulated in table [ tab_fit ] . note that all our models predict the reionization histories that are well within the available observational constraints . in particular , for all our models , we obtain redshift of reionization greater than 6 as inferred from the observations of lyman - alpha forest in distant quasars ( fan et al . , 2006 ) and electron scattering optical depth within one sigma of wmap 7yrs results ( @xmath126 ) ( komatsu et al . , 2011 ) . it is important to note that the feedback induced star formation model reduces the star formation severely in low mass galaxies . these galaxies are the main sources of the uv photon that causes the reionization of the igm and lower the star formation lesser the uv photon production . however , the fraction of uv photon that can escape from a galaxy and cause the ionisation of the igm is poorly constraint especially for low mass galaxies . we observe that using a escape fraction of 0.2 leads to a similar reionization history ( and electron scattering optical depth to the reionization ) for the feedback induced star formation models compared to the no feedback model with escape fraction of 0.1 . this value of escape fraction of uv photon is in good agreement with recent numerical results ( wise & cen 2009 ; pawlik et al . 2009 ; yajima et al . 2009 ) as well as observations ( bouwens et al . thus , our feedback models also produce the reionization history with reasonable physical parameters . -1 cm c|c|c|c|c|c|c|c|c @xmath90 & & & & + & @xmath108 & @xmath127/dof & @xmath108 & @xmath127/dof & @xmath108 & @xmath127/dof & @xmath108 & @xmath127/dof + + 1.5 & 0.048 @xmath128 0.003 & 0.81 & 0.037 @xmath128 0.002 & 0.28 & 0.053 @xmath128 0.003 & 0.56 & 0.044 @xmath128 0.003 & 0.40 + 1.9 & 0.049 @xmath128 0.002 & 0.89 & 0.038 @xmath128 0.001 & 0.24 & 0.054 @xmath128 0.002 & 0.49 & 0.045 @xmath128 0.001 & 0.22 + 2.5 & 0.079 @xmath128 0.003 & 4.12 & 0.064 @xmath128 0.002 & 2.44 & 0.087 @xmath128 0.003 & 3.05 & 0.074 @xmath128 0.002 & 2.46 + 3 & 0.059 @xmath128 0.001 & 6.40 & 0.056 @xmath128 0.001 & 1.51 & 0.067 @xmath128 0.001 & 4.13 & 0.062 @xmath128 0.001 & 1.80 + 4 & 0.062 @xmath128 0.002 & 1.11 & 0.052 @xmath128 0.001 & 0.40 & 0.068 @xmath128 0.002 & 0.46 & 0.060 @xmath128 0.001 & 0.38 + 5 & 0.050 @xmath128 0.001 & 0.98 & 0.041 @xmath128 0.001 & 1.36 & 0.052 @xmath128 0.001 & 1.14 & 0.045 @xmath128 0.001 & 1.51 + 6 & 0.062 @xmath128 0.002 & 0.68 & 0.052 @xmath128 0.002 & 0.55 & 0.066 @xmath128 0.002 & 0.56 & 0.057 @xmath128 0.002 & 0.51 + 7 & 0.120 @xmath128 0.006 & 0.38 & 0.083 @xmath128 0.004 & 0.38 & 0.086 @xmath128 0.004 & 0.40 & 0.072 @xmath128 0.003 & 0.50 + 8 & 0.114 @xmath128 0.007 & 0.26 & 0.080 @xmath128 0.004 & 0.45 & 0.099 @xmath128 0.005 & 0.55 & 0.083 @xmath128 0.004 & 0.65 + + 1.5 & 0.029 @xmath128 0.001 & 0.12 & 0.023 @xmath128 0.001 & 0.25 & 0.031 @xmath128 0.001 & 0.21 & 0.026 @xmath128 0.001 & 0.24 + 1.9 & 0.034 @xmath128 0.001 & 0.26 & 0.027 @xmath128 0.001 & 0.64 & 0.037 @xmath128 0.001 & 0.68 & 0.031 @xmath128 0.001 & 0.62 + 2.5 & 0.067 @xmath128 0.002 & 1.45 & 0.054 @xmath128 0.001 & 0.37 & 0.074 @xmath128 0.002 & 0.50 & 0.063 @xmath128 0.002 & 0.28 + 3 & 0.061 @xmath128 0.001 & 1.24 & 0.050 @xmath128 0.001 & 1.55 & 0.065 @xmath128 0.001 & 0.78 & 0.057 @xmath128 0.001 & 1.57 + 4 & 0.074 @xmath128 0.001 & 1.02 & 0.051 @xmath128 0.001 & 4.55 & 0.066 @xmath128 0.001 & 4.58 & 0.052 @xmath128 0.001 & 6.76 + 5 & 0.076 @xmath128 0.001 & 2.20 & 0.052 @xmath128 0.001 & 5.95 & 0.069 @xmath128 0.001 & 5.18 & 0.058 @xmath128 0.001 & 6.44 + 6 & 0.115 @xmath128 0.004 & 0.63 & 0.090 @xmath128 0.003 & 0.71 & 0.110 @xmath128 0.003 & 0.77 & 0.096 @xmath128 0.002 & 0.80 + 7 & 0.179 @xmath128 0.008 & 0.32 & 0.113 @xmath128 0.004 & 0.69 & 0.134 @xmath128 0.006 & 1.02 & 0.110 @xmath128 0.004 & 1.21 + 8 & 0.248 @xmath128 0.014 & 0.32 & 0.153 @xmath128 0.007 & 0.80 & 0.185 @xmath128 0.009 & 0.86 & 0.150 @xmath128 0.007 & 1.06 + we first consider the st mass function . it is clear from the fig . [ fig_lf_fit ] and also from table [ tab_fit ] that our models reasonably explain the uv luminosity functions of lbgs in a vast redshift range of @xmath129 . the flattening observed in the low end of luminosity functions ( @xmath130 ) particularly in the redshift range of @xmath131 are well explained by the sne feedback and the radiative feedback that we assume . except for @xmath132 and 3 the best fit @xmath110-square per degree of freedom is very close to unity or less than unity . this demonstrates how well our model predictions match with observations . in general , for all our models , values of @xmath108 show an increasing trend with redshift with some exceptions at @xmath132 and @xmath133 that we discuss later . this is in accord with the concordance model of structure formation . if one assumes a fixed @xmath57 which provides a reasonable fit to the observational data for the stellar mass in galaxies ( see fig . [ fig_stellar_mass ] ) one sees that the value of @xmath86 increases with increasing redshift . this means in the past the amount of dust present in a galaxy was less on average . as the universe became older the amount of metals in galaxies increased making them more dusty and hence more difficult to detect . also the mean values of @xmath107 obtained assuming @xmath57 is in good agreement with the observation of reddy et al . they found a mean value of @xmath134 at @xmath79 . further we notice that the models with @xmath84 km / s provide a better fit to the observational data compare to the models with @xmath56 km / s for both @xmath60 and 1 in all redshifts except for @xmath133 and 8 . in particular , for @xmath135 the models with @xmath84 km / s fit the faint end slope of the luminosity functions better than models with @xmath56 km / s . for @xmath136 both the models over fit the observational data as best fit values of @xmath127 per degree of freedom are less than unity and hence it can not identify the best fit model . at @xmath132 none of our models provide a good fit to the observational data . the bright end of the luminosity function is well explained by our models . however , they fail to reproduce the last two data points at the low luminosity end . interestingly we found the faint end slope of the observed luminosity function at @xmath132 is quite steep compared to @xmath123 luminosity function . it is very unlikely to change the slope within such a short time . we expect future improve observations would clarify this dispute . this also leads to a higher value of @xmath108 at @xmath132 compared to values that are obtained for @xmath137 . the observed luminosity function at @xmath123 clearly favours the models with @xmath84 km / s for both @xmath60 and 1 . the models with @xmath56 km / s fail to reproduce the faint end slope of the observed luminosity function . such a signature is also seen in case of @xmath122 and @xmath138 . the feedback in those models ( i.e. models with @xmath56 km / s ) are very strong to suppress the star formation in low mass galaxies and hence fail to produce enough number of low luminosity galaxy to explain the observed luminosity functions at @xmath139 . at @xmath124 we see that our models provide a reasonable fit to the observed data . the model with @xmath60 and @xmath56 km / s does not fit the last four observational points in the low luminosity end and hence has maximum value for the best fit @xmath127 . however , none of our models able to fit the last two data points in the low luminosity end ( the open triangles ) . as already mentioned above these two data points suffer major uncertainty in measurements and we do not consider them while fitting . if future observations confirm these data points then we have to change our model in order to explain the low luminosity end of the luminosity function at @xmath124 . all our models well explain the observed luminosity function at @xmath133 . however , we see from table [ tab_fit ] that the best fit values of @xmath127 per dof at this redshift are quite higher than unity except for model with @xmath60 and @xmath56 km / s for which best fit @xmath127/dof @xmath140 . we observe that the data point at @xmath141 has unusually low value of @xmath142 compare to its neighbouring data points and contribute significantly to the best fit @xmath127 values . it also leads to a lower value of @xmath108 compared to @xmath124 . ignoring this data point all our models fit the observed luminosity function with best fit @xmath127 per dof values close to unity . for redshift @xmath143 our model predictions match very well with the observed data . note that the errors in the measurements are very high especially at @xmath125 and 8 . thus all models provide good fit to the data with best fit @xmath127 per degree of freedom much less than unity . we expect that future observations would reduce the error bars and extend the measurements to even lower luminosity allowing us to put further constraints on star formation models . in passing we note that we can not distinguish two feedback models namely energy driven / cosmic ray driven sne winds ( i.e. @xmath60 ) and momentum driven sne winds ( @xmath55 ) with the present observational data of uv luminosity functions of lbgs at various redshifts . the reason behind this is that the two models differ maximum in low mass dwarf galaxies where the @xmath32 can be different by more than factor 10 . it was shown in samui et al . ( 2007 ) that the present observable range of luminosity functions are contributed mostly by the galaxies with mass @xmath61 . at this mass the @xmath32 for models with energy driven sne wind is larger only by factor 3 ( 2 ) in case of @xmath56 km / s ( @xmath84 km / s ) compared to models with momentum driven wind and hence can not be distinguished by the present observational data due to the large errors in the measurements . we now turn to the luminosity function as obtained using ps halo mass function and compare with st mass function . [ fig_lf_fit_ps ] shows the predicted luminosity functions and the fitted parameters are tabulated in table [ tab_fit ] . it is obvious from the figure that ps mass function also produces the shape and evolution of luminosity function correctly . however , table . [ tab_fit ] shows that in most of the redshift we considered ( i.e. except @xmath132 & 3 ) , ps mass function provide a poorer fit compared to st mass function . thus the flattening of luminosity functions at low luminosity end can only be explain due to sne feedback and can not be explain with different form of the halo mass function . further , our previous conclusions regarding the halo mass function remain same i.e. st mass function provides a better understanding of high redshift luminosity functions compared to ps mass function . we have built improved semi - analytical models of high redshift galaxy formation where star formation is regulated by the feedback due to sne driven outflows . we consider two models of feedback ; one in which the outflows are driven by the thermal energy of the sne remnants and in second the hot gas loses its thermal energy due to radiation and the momentum of the gas helps in driving the outflow . the effect of sne feedback to the star formation is calculated in a self - regulated manner . we derive an analytical form for the star formation rate in a galaxy that is regulated by the sne feedback . given the star formation rate in individual galaxy we calculate high redshift galaxy luminosity functions . our feedback induced star formation models are successful in explaining the observed stellar mass in galaxies of different types with mass range @xmath144 . for low mass dwarf galaxies our simple model of star formation including sne feedback produces the trend in correlation between the observed stellar mass and circular velocity but fails to explain the spread in it . other feedback mechanisms are expected to operate on such small mass galaxies in order to understand the observations . our models also produces the observed ratio of stellar mass to halo mass as a function of halo masses as obtained from recent sdss data . further , our models emphasize the importance of alternative driver of galactic outflows such as cosmic rays . the amount of metals detected in high redshift galaxies is well estimated by our models which demonstrates that the metal mass in a galaxy is vastly determined by the outflows . the 3-d correlation between gas phase metallicity , stellar mass and star formation rate is well explained by our feedback dominated star formation model showing the importance of sne feedback in building this observed relationship in galaxies . the observed uv luminosity functions of lbgs at @xmath1 are well explained with our feedback models . especially the flattening observed at low end of the luminosity functions arises naturally with our feedback model . in absence of such feedback one would produce more low mass galaxies and fail to reproduce the observed data points . the models with @xmath84 km / s provide a better fit to the observational data compared to @xmath56 km / s especially at low redshift ( the characteristic circular velocity , @xmath46 , fixes the normalisation of our feedback model ) . this implies that the supernova feedback affects the star formation in galaxies more compare to ionisation feedback for high mass galaxies . for galaxies with @xmath145 km / s one expects to have no effect of ionization feedback to the star formation where as sne feedback reduces the star formation by 40% in the same galaxies ( taking @xmath84 km / s ) . the present observational data of high redshift luminosity functions is not good enough to distinguish a thermal energy driven outflow from a momentum driven outflow . we need to reduce the error in the measurements and extend the data to even lower luminosity end where these two models behave differently . moreover , the flattening of the luminosity functions can not be explained as the difference between the form of halo mass function that one uses to calculate the formation rate of halos ( like ps or st mass function ) . it can only be understood due to sne feedback . hence , our feedback regulated star formation model can explain a vast range of observations of high redshift universe . having such kind of simple analytical form for physically motivated star formation of individual galaxy is very helpful for determining the star formation rate of high redshift galaxies by fitting spectral energy distribution ( sed ) with several broad band photometry . in such case the star formation history is very important to determine several physical quantities such as the star formation rate , stellar mass , age of the galaxy etc . in general one considers a constant star formation model or an exponentially decaying one in such fits . however , reddy et al . ( 2012 ) showed that high redshift galaxies ( @xmath146 ) are more expected to have a rising star formation history and very young age when they are detected . our successful star formation models actually predict such kind of rising star formation history at the young age and then gradually decreasing star formation history after a fraction of dynamical time . we expect that using our model in such sed fitting would lead to a better fit to the observational data and also provide more clearer picture of high redshift galaxies . since sne feedback is one of the main improvement of our model over past , it is important to investigate its effect on the global metal pollution . especially by the fact that in low mass galaxies that dominate in the metal production at high redshift should have a cosmic ray pressure driven outflows , one should see how far this model can explain the global metal budget . we hope to address that in future . the author thanks kandaswamy subramanian and raghunathan srianand for very useful discussions . the author also thanks anonymous referee for his / her useful comments that helped a lot to improve the quality of the paper . barkana , r. , loeb , a. 2001 , phr , 349 , 125 baugh , c. m. , cole , s. , frenk , c. s. , lacey , c. g. , 1998 , apj , 498 , 504 baugh c. m. , lacey c. g. , frenk c. s. , granato g. l. , silva l. , bressan a. , benson a. j. , cole s. , 2005 , mnras , 356 , 1191 benson , a. j. , lacey , c. g. , baugh , c m. , cole , s. , frenk , c. s. , 2002 , mnras , 333 , 156 best , p. n. , kaiser , c. r. , heckman , t. m. , kauffmann , g. , 2006 , mnras , 368 , l67 bouwens , r. j. , illingworth , g. d. , franx , m. , ford , h. , 2007 , apj , 670 , 928 bouwens , r. j. , illingworth , g. d. , franx , m. , ford , h. , 2008 , apj , 686 , 230 bouwens , r. j. , et al . , 2010 , apj , 708l , 69 bouwens , r. j. et al . , 2011 , apj , 737 , 90 bouwens , r. j. et al , 2012 , apj , 754 , 83 bower , r. g. , benson , a. j. , malbon , r. , helly , j. c. , frenk , c. s. , baugh , c. m. , cole , s. , lacey , c. g. , 2006 , mnras , 370 , 645 brandbyge , j. , hannestad , s. , haugblle , t. , wong , y. y. y. , 2010 , jcap , 09 , 014 bromm , v. , loeb a. , 2002 , apj , 575 , 111 chiu w. a. , ostriker j. p. , 2000 , apj , 534 , 507 choudhury , t. r. , srianand , r. , 2002 , mnras , 336 , l27 cole , s. , aragon - salamanca , a. , frenk , c. s. , navarro , j. f. , zepf , s. e. , 1994 , mnras , 271 , 781 cox , donald p. , 1972 , apj , 178 , 159 cullen , f. , cirasuolo , m. , mclure , r. j. , dunlop , j. s. , 2013 , arxiv1310.0816 dave , r. , oppenheimer , b. d. , finlator , k. , 2011 , mnras , 415 , 11 dave , r. , oppenheimer , b. d. , sivanandam , s. , 2008 , mnras , 391 , 110 dijkstra , m. , haiman , z. , rees , m. , weinberg , d. h. , 2004 , apj , 601 , 666 erb , d. k. , shapley , a. e. , pettini , m. , steidel , c. c. , reddy , n. a. , adelberger , k. l. , 2006 , apj , 644 , 813 fan , x. , strauss , m. a. , richards , g. t. et al . , 2006 , aj , 131 , 1203 granato , g. l. , lacey , c. g. , silva , l. , bressan , a. , baugh , c. m. , cole , s. , frenk , c. s. , 2000 , apj , 542 , 710 garnett , d. r. , 2002 , apj , 581 , 1019 gerhard , o. , kronawitter , a. , saglia , r. p. , bender , r. , 2001 , aj , 121 , 1936 j. e. , peterson b. a. , 1965 , apj , 142 , 1633 hopkins , a. m. , beacom , j. f. , 2006 , apj , 651 , 142 jose , c. , samui , s. , subramanian , k. , srianand , r. , 2011 , phrvd , 83 , 3518 jose , c. , subramanian , k. , srianand , r. , samui , s. , 2013 , mnras , 429 , 2333 kauffmann , g. , white , s. d. m. , guiderdoni , b. , 1993 , mnras , 264 , 201 kewley , l. j. , ellison , s. l. , 2008 , apj , 681 , 1183 kobayashi , c. , springel , v. , white , s. d. m. , 2007 , mnras , 376 , 1465 komatsu , e. et al . , 2011 , apjs , 192 , 18 leauthaud a. et al . , 2012 , apj , 744 , 159 lee , h. , skillman , e. d. , cannon , j. m. , jackson , d. c. , gehrz , r. d. , polomski , e. f. , woodward , c. e. , 2006 , apj , 647 , 970 leitherer . c. , et al . , 1999 , apjs , 123 , 3 lequeux , j. , peimbert , m. , rayo , j. f. , serrano , a. , torres - peimbert , s. , 1979 , a&a , 80 , 155 mannucci , f. , buttery , h. , maiolino , r. , marconi , a. , pozzetti , l. , 2007 , a&a , 461 , 423 mannucci , f. , cresci , g. , maiolino , r. , marconi , a. , gnerucci , a. , 2010 , mnras , 408 , 2115 mannucci , f. , et al . , 2009 , mnras , 398 , 1915 martin , c. l. , 1999 , apj , 513 , 156 martin , c. l. , 2005 , apj , 621 , 227 mcgaugh , s. s. , 2005 , apj , 632 , 859 mitra , s. , kulkarni , g. , bagla , j. s. , yadav , j. k. , 2011 , basi , 39 , 563 mortlock et al . , 2011 , nature , 474 , 616 moster , b. p. et al . , 2010 , apj , 710 , 903 oesch , p. a et al . , 2010 , apj , 725 , 150 oke , j. b. and gunn , j. e. , 1983 , apj , 266 , 713 ostriker , j. p. , mckee , c. f. , 1988 , rev . modern physics , 60 , 1 pettini , m. , shapley , a. , steidel , c.c . , et al . 2001 , apj , 554 , 981 pawlik , a. h. , schaye , j. , van scherpenzeel , e. , 2009 , mnras , 394 , 1812 reddy , n. a. , steidel , c. c. , fadda , d. , yan , l. , pettini , m. , shapley , a. e. , erb , d. k. , adelberger , k. l. , 2006 , apj , 644 , 792 reddy , n. a. , steidel , c. c. , 2009 , apj , 692 , 778 reddy , n. et al . , 2012 , apj , 744 , 154 samui , s. , srianand , r. , subramanian , k. , 2007 , mnras , 377 , 285 samui , s. , subramanian , k. , srianand , r. , 2008 , mnras , 385 , 783 samui , s. , subramanian , k. , srianand , r. , 2009 , newa , 14 , 591 samui , s. , subramanian , k. , srianand , r. , 2010 , mnras , 402 , 2778 s. , 1994 , pasj , 46 , 427 savaglio , s. , et al . , 2005 , apj , 635 , 260 scannapieco , c. et al , 2012 , mnras , tmp , 2970 scannapieco , c. , tissera , p. b. , white , s. d. m. , springel , v. , 2005 , mnras , 364 , 552 scannapieco , c. , tissera , p. b. , white , s. d. m. , springel , v. , 2006 , mnras , 371 , 1125 scannapieco , c. ; tissera , p. b. ; white , s. d. m. ; springel , v. , 2008 , mnras , 389 , 1137 scannapieco , e. , ferrara , a. , madau , p. , 2002 , apj , 574 , 590 shankar , f. , lapi , a. , salucci , p. , de zotti , g. , danese , l. , 2006 , apj , 643 , 14 sheth r. k. , tormen g. , 1999 , mnras , 308 , 119 silk , j. , 1993 , pnas , 90 , 4835 somerville , r. s. , primack , j. r. , 1999 , mnras , 310 , 1087 stark , d. v. , mcgaugh , s. s. , swaters , r. a. , 2009 , aj , 138 , 392 steidel , c. c. , adelberger , k. l. , shapley , a. e. , pettini , m. , dickinson , m. ; giavalisco , m. , 2003 , apj , 592 , 728 stringer , m. j. , bower , r. g. , cole , s. , frenk , c. s. , theuns , t. , 2012 , mnras , tmp , 2923 tremonti , c. a. et al , 2004 , apj , 613 , 898 veilleux , s. , cecil , g. , bland - hawthorn , j. , 2005 , ara&a , 43 , 769 walker , m. g. , mateo , m. , olszewski , e. w. , pearrubia , j. , wyn e. n. , gilmore , g. , 2009 , apj , 704 , 1274 weaver , r. , mccray , r. , castor , j. , shapiro , p. , moore , r. , 1977 , apj , 218 , 377 white s. d. , frenk c. s. , 1991 , apj , 379 , 52 wise , j. h. , cen , r. 2009 , apj , 693 , 984 wyithe , j. s. b. , loeb , a. , carilli , c. , 2005 , apj , 628 , 575 yajima , h. , umemura , m . , mori , m . , nakamoto , t. , 2009 , mnras , 398 , 715

the general antiparticle spectrometer experiment ( gaps ) is foreseen to carry out a dark matter search using low - energy cosmic ray antideuterons at stratospheric altitudes with a novel detection approach . a prototype flight from taiki , japan was carried out in june 2012 to prove the performance of the gaps instrument subsystems ( lithium - drifted silicon tracker and time - of - flight ) and the thermal cooling concept as well as to measure background levels . the flight was a success and the stable flight operation of the gaps detector concept was proven . during the flight about charged particle triggers were recorded , extensive x - ray calibrations of the individual tracker modules were performed by using an onboard x - ray tube , and the background level of atmospheric and cosmic x - rays was measured . the behavior of the tracker performance as a function of temperature was investigated . the tracks of charged particle events were reconstructed and used to study the tracking resolution , the detection efficiency of the tracker , and coherent x - ray backgrounds . a timing calibration of the time - of - flight subsystem was performed to measure the particle velocity . the flux as a function of flight altitude and as a function of velocity was extracted taking into account systematic instrumental effects . the developed analysis techniques will form the basis for future flights . dark matter , antideuteron , indirect detection , balloon flight

1307.3538

the existence of dark matter and its nature play a key role in understanding structure formation after the big bang and the energy density of the universe @xcite . dark matter can not be explained with known types of matter ; therefore , we are at the dawn of something significantly new . the importance of this problem becomes obvious by recalling that dark matter is approximately five times more abundant than regular matter . little is known about the nature of dark matter particles , but that they are relatively heavy , interacting gravitationally with regular matter , but otherwise , interacting only weakly if at all . if dark matter was in thermal equilibrium with the rest of the matter in the early universe and froze out when the temperature dropped due to expansion , it is a natural assumption in many models that dark matter particles are able to annihilate with each other and produce known standard model particles in this way . these particles would contribute to the known cosmic ray fluxes and , as the kinematic characteristics of these processes are different from the production mechanisms of the conventional cosmic rays , it could be possible to observe the imprint of dark matter in the diffuse cosmic ray spectra in the form of an excess . well - motivated theories beyond the standard model of particle physics contain candidates with exactly these properties . cosmic ray antiparticles without primary sources are ideal candidates for such a search . however recent results show that accomplishing this task with positrons and antiprotons appears to be difficult @xcite . antideuterons would also be generated in dark matter annihilations and are a potential breakthrough approach . secondary antideuterons , like antiprotons , are produced when cosmic ray protons or antiprotons interact with the interstellar medium , but the production threshold for this reaction is higher for antideuterons than antiprotons . collision kinematics also disfavor the formation of low - energy antideuterons in these interactions . moreover the steep energy spectrum of cosmic rays means there are fewer particles with sufficient energy to produce secondary antideuterons , and those that are produced will have relatively large kinetic energy . as a consequence , a low - energy search for primary antideuterons has very low background @xcite . this feature of antideuteron searches , along with the growing realization that such searches probe supersymmetric models in a broad way , has attracted considerable attention . many theoretical papers discuss aspects of antideuteron dark matter searches @xcite . in this regard , supersymmetric and universal extra dimension theories provide the most popular and theoretically well - motivated dark matter candidates @xcite . the absolute flux expected for antideuterons is very low , and therefore any attempt to measure it needs an exceptionally strong particle identification . predicted antideuteron fluxes from different dark matter models updated by more recent coalescence momentum value ( purple , red , green lines ) @xcite and secondary / tertiary background flux from cosmic ray interactions with the interstellar medium ( blue line ) @xcite . antideuteron limits from bess @xcite and sensitivities for the running ams @xcite and the planned gaps experiments @xcite are also shown . ] in the near future the challenging search for antideuterons will exclusively rely on the general antiparticle spectrometer ( section [ s - gaps ] ) and on the alpha magnetic spectrometer ( ams ) , a multi - purpose cosmic ray detector on the international space station @xcite . figure [ f - fig1_jap_pgaps ] shows the theoretically expected antideuteron fluxes from different dark matter models in comparison to the secondary background . the different boxes demonstrate the antideuteron flux limits of bess @xcite and the sensitivity reaches of gaps and ams @xcite , which reach for the first time the sensitivity to probe predictions of well - motivated models . both experiments have mostly complementary kinetic energy ranges but also some overlap in the interesting low - energy region . in addition , another very important virtue comes from the different detection techniques . ams identifies particles by analyzing the event signatures of different subsequent subdetectors and a strong magnetic field and gaps by slowing down the antideuteron and creating an exotic atom inside the target material and analyzing the decay . this allows the study of both a large energy range and independent experimental confirmation , which is crucial for a rare event search like the hunt for cosmic ray antideuterons . gaps detector concept with antiproton and antideuteron signatures . ] the general antiparticle spectrometer is designed to measure low - energy cosmic antideuterons . as mentioned above , the expected antideuteron flux is very low and therefore a large acceptance and long flight time are indispensable . figure [ f - fig1_jap_pgaps ] demonstrates that the dark matter signal above the background is expected to be the largest at low kinetic energies of 100500mev . it is therefore planned for gaps to carry out a series of long duration balloon flights from antarctica , where the deflection of low - energy charged cosmic rays in the geomagnetic field is the smallest . another important effect for the low - energy cosmic ray detection comes from the interaction of the solar wind with cosmic rays , which effectively decreases the observable interstellar flux . the strength of the solar modulation depends on the 11 year long solar cycle , which will approximately reach the next minimum in 2019 @xcite . the first science gaps flight will be possible from 2017 and will therefore only feel a small solar influence . the core of the detector will be a track reconstruction device consisting of 10 layers of lithium - drifted silicon ( si(li ) ) modules ( figure [ f - fig2_jap_pgaps ] ) that will be enclosed by a hermetically sealed time - of - flight system ( tof ) made of plastic scintillators with photomultiplier tube ( pmt ) readout . this box will be surrounded by another half - cube of plastic scintillators . the inner tracker core ( trk ) will be a cube of 2 m edge length and the outer tof half - cube will have a width of 4 m . these detector components will be used for a novel detection approach to clearly identify low - energy antideuterons . the idea is to stop low - energy antideuterons in the tracker material , to replace a shell electron of the target material with this antideuteron , and to form an excited exotic atom . the hydrogen - like atom will deexcite by autoionization followed by characteristic x - ray ladder transitions . at the end of the ladder transitions the antideuteron will annihilate in a hadronic interaction with the nucleus and produce pions and protons . the detector will be able to measure the velocity and the charge of the incoming particle in the tof as well as the stopping depth of a particle in the tracker and the development of the energy loss per layer throughout the slowing process . moreover , the tracker will resolve the characteristic x - ray energies and track the pions and protons . the main source of background for the antideuteron signal comes from antiprotons . therefore , good depth sensing and x - ray energy resolution along with a reliable tracking and counting of pions / protons are essential for a high background rejection . * left : * flight ready pgaps payload without insulation foam . * right : * event display showing the position of the time - of - flight and si(li ) tracker subsystems and a typical clean cosmic ray track reconstructed from flight data . ] a gaps prototype ( pgaps ) was constructed for a balloon flight to demonstrate a stable and low noise operation of all relevant detector components , to verify the thermal model and the si(li ) detector module cooling approach , and to study the incoherent background level of charged cosmic rays and x - rays . the small acceptance and short flight time compared to the full - scale experiment did not allow a study of antiprotons or antideuterons . however , the prototype consisted of all major components that will be part of the full gaps experiment ( figure [ f - fig3_jap_pgaps ] ) . the carbon fiber structure of the balloon gondola frame had a height of 1.2 m , the mass of the science part of the gondola was 308 kg , and the total power consumption was 430w . a full description of the instrument can be found here @xcite . in total six circular modules were arranged in two stacks of three with a vertical spacing of 20 cm and mounted into a watertight plastic vessel . the individual circular si(li ) modules were manufactured by semikon detektor gmbh of jlich , germany . for the full payload the detectors will be manufactured by the gaps collaboration @xcite . five modules had a thickness of 2.5 mm and one of 4.0 mm . the active area of each module had a diameter of 9.4 cm and was divided into eight strips . the strips on the p+ side were contacted by implanted boron and on the n+ side by lithium contacts . for full depletion the detectors had to be cooled down to at least -25 and were operated with reverse bias voltages of 185 - 240v . the cooling system used a closed - loop coolant pipe with fluorinert fluid to transport the heat from the detectors to a radiator using a pump . an attitude control system was designed to point the radiator to the anti - sun side during the flight . on the ground , the radiator was thermally coupled to a heat exchanger cooled by liquid nitrogen and the detectors were operated in a dry nitrogen atmosphere . for the science payload it will be important to resolve x - rays in the range of 10100kev while simultaneously measuring charged energy depositions up to 50mev . therefore , the pgaps si(li ) electronics had a dual readout scheme , with separate high and low gain processing chains for each detector strip , which consisted of two card cages with three analog readout boards each . the card cages , boards , and digital signal processing units had been flown on the nuclear compton telescope prototype in 2005 @xcite and were modified to meet the requirements for pgaps . the electronics were designed to work at low pressures and temperatures and were housed in a watertight plastic vessel . the electronics also recorded various scalers important for monitoring the quality of the signal processing of the different detector strips , temperatures , electronic status , and livetime . for in - flight calibration the payload was equipped with a silver target x - ray tube and provided , together with a gold filter , a characteristic spectrum with peaks at 26 and 36kev . the tube was mounted under the bottom time - of - flight layer and the position was optimized to illuminate the detectors as uniformly as possible . in addition , the 59.5kev line of an americium-241 ( am-241 ) radioactive source was used for ground calibration during the qualification , integration , and flight preparation stages . the time - of - flight system ( tof ) consisted of three layers of crossed plastic scintillator paddles , two above and one below the tracker . the top and the bottom layers were composed of two individual layers with three paddles each and the middle layer was composed of two layers of two paddles each . the paddles were made of bicron bc-408 material and were 50 cm long , 15 cm wide , and 3 mm thick . each paddle end was attached to a curved , acrylic light guide coupled to a hamamatsu r-7600 photomultiplier tube , which was operated at about 800 - 900v . prior to the flight , all flight ( and flight spare ) tof pmt assemblies were operated at full hv for a minimum of 5hours each in a low pressure environment . the test was not a full thermal - vacuum test since the temperature was not controlled . however , the pressure was varied over a range of pressures expected in flight ( 150torr ) , with most of the test taking place at pressures between 5 and 10torr . no failures or degraded performance were observed for any pmts in the tests . the spacing between the top and bottom layer was 0.94 m and the spacing between the top and middle layer was 0.38 m . the tof data were processed in a vme rack with modules . the vme system also made the tof trigger decision that will be further explained in the next section . in addition to the trigger decision , the electronics digitized the scintillator light output as measured by the pmts and generated time - to - digital converter ( tdc ) values with 50ps resolution based on a discriminator threshold . housekeeping data were also recorded . the tof electronics , flight computer , attitude control system electronics , and a fiber optic gyroscope were mounted together in a pressurized vessel since the tof electronics modules were not designed to operate in vacuum . pgaps was operated in several different data taking modes . the goal was both to collect tracks of charged particles traversing the tof and the tracker and to measure the stability of the x - ray performance of the tracker modules . therefore two different trigger schemes were deployed . the tof trigger mode was based on the coincident signals of two crossed paddles of the tof middle layer and at least one additional signal in the top or bottom tof layer , which triggered the simultaneous readout of all pmts and also of all tracker modules . in this way particle tracks with coincident signals in different detector layers could be recorded . during the tracker trigger mode each si(li ) module was able to self - trigger its own readout based on discriminator thresholds . eventually it will be crucial for gaps to detect x - rays in coincidence with charged particle tracks , but for calibration purposes no coincidence with other tracker modules or the tof was required . during flight , the tracker trigger mode was used for x - ray tube calibration and for incoherent x - ray flux measurements . in the tracker trigger mode the tof continued regular data taking as described above , but without triggering the readout of the tracker . it was also possible to run the payload only with the tof or tracker turned on . * left : * pgaps on the launcher and inflated balloon right before launch . * right : * pgaps during take off . ] * left : * pgaps balloon flight trajectory ( google maps ) . * right : * altitude as a function of time during flight . ] the scientific part of the pgaps payload was assembled at the space sciences laboratory of uc berkeley starting in summer 2011 and was shipped to the institute of space and astronautical science / japan aerospace exploration agency ( isas / jaxa ) facilities in sagamihara , japan in april 2012 where it underwent thermal vacuum testing and was matched to the gondola bus . the bus gondola supplied power by lithium batteries , telecommunication to ground , and ballast hoppers . the launch site was the taiki aerospace research field in taiki at the southern tip of the east coast of hokkaido , japan and was preceded by compatibility testing , e.g. , electromagnetic interference and communication , final rigging procedures , and a dress rehearsal @xcite . the launch took place at 4:55am jst on june 3rd , 2012 using a fb-100 helium balloon with a volume of 100,000m@xmath1 ( figure [ f - fig4_jap_pgaps ] ) . after the initial ascent the balloon drifted eastward for about three hours at 1015 km altitude ( boomerang altitude ) before dropping more ballast and further ascending to a maximum of 33 km ( float altitude ) @xcite . at that time of the season the winds at high altitude blew the balloon westward back to the coast of hokkaido . the balloon was released at 11:05am and the payload landed in the water at 11:36am where it was recovered by boat within 9min ( figure [ f - fig5_jap_pgaps ] ) . before launch the si(li ) detectors were cooled down and calibration data with atmospheric muons and x - rays were recorded . pgaps took science data , i.e. , energy depositions in the tracker and energy depositions and time values from the tof , from launch until the balloon was released , while housekeeping data were also recorded throughout descent . the 6:10h of science data taking were split into cadences of tof trigger mode ( @xmath2min ) , x - ray tube calibration ( @xmath3min ) , tracker trigger mode ( @xmath4min ) , and at the very end tof trigger mode with tracker turned off for 13min . these different modes made up 91% of the flight and consist of data with nominal values for currents and voltages , temperatures as well as electronics status . the number of tof triggers recorded during all well - defined data taking modes was @xmath5 , the number of tof events also triggering the tracker was @xmath6 , and the number of si(li ) detector triggers during x - ray tube calibration was @xmath7 . the tof took data continuously throughout the flight even when the tracker was operated in tracker trigger mode . the profile of the raw tof trigger rate as a function of altitude and atmospheric depth was generated by sampling the tof trigger rate over 2min intervals and filling these values together with the mean altitude of this sample into a two dimensional histogram ( figure [ f - fig6_jap_pgaps ] ) . the error bars reflect the error on the mean trigger rate for each altitude bin . the error bar length is affected by the number of available samples for each altitude bin , which is defined by the altitude change velocity , and the spread of trigger rate samples at the specific altitude . a maximum at 18 km is seen and is in good agreement with other measurements and air shower simulations @xcite . tof trigger rate during flight as a function of altitude and atmospheric depth . ] all electronics ( power distribution devices , tracker and tof electronics ) worked very reliably and the flight computer rebooted only once during ascent from boomerang altitude to float altitude due to the changing grounding environment . the livetime of the tracker electronics for processing events during tof trigger mode was on ground very close to 100% , at boomerang altitude 99.8% , and at float 99.5% . only during the high rate calibration with the x - ray tube did the livetime for the detector with the largest exposure drop to 95% . calibration of low gain channel with the help of the overlap region between low and high gain channel . ] calibration of low gain channel with the help of the overlap region between low and high gain channel . ] the high gain readout channels were calibrated using the x - ray tube lines at 26 and 36kev and the am-241 line at 59.5kev . figure [ f - fig7_jap_pgaps ] shows the results for one typical channel during the preparation period a few hours before the launch . the histogram for the measurement with the am-241 source shows a tail towards lower energies ( 4555kev ) that is assumed to be formed from scattered 59.5kev photons . the source could only be placed on top of the uppermost tof layer before launch and scattering could occur in the detector material between the source and the detector . the three line positions with their corresponding widths were used as input to a straight line fit . the measurements with am-241 could only be performed on ground as flying a radioactive source was not permitted . laboratory measurements of the high gain readout channels with test pulses during the electronics development phase showed a linear behavior of the adc response for the full 13bit range . therefore , the low gain channel for the charged particle measurement was calibrated by studying the overlap region of the high and low gain channel connected to the same detector strip . figure [ f - fig8_jap_pgaps ] shows one example for the correlation of the digitization of the same energy deposition between the already calibrated high gain channel and the analog - to - digital converter value ( adc ) measured in the low gain branch . a clear linear correlation is visible and straight line fits for every channel were used to map the low gain adc value to a calibrated energy value . active channels as a function of time during flight for individual si(li ) detectors ( blue , magenta , green , yellow , light blue , purple ) and integrated over all detectors ( black ) . in addition , the geometrical acceptance for tof trigger events with at least two tracker hits in active channels is shown ( red ) . ] active channels as a function of time during flight for individual si(li ) detectors ( blue , magenta , green , yellow , light blue , purple ) and integrated over all detectors ( black ) . in addition , the geometrical acceptance for tof trigger events with at least two tracker hits in active channels is shown ( red ) . ] as mentioned above , a critical parameter for the operation of si(li ) detectors is the temperature . ground testing showed that most detector channels started to deplete from about -15 to -20 . the coldest temperature at launch inside the tracker vessel measured on one of the detectors of the bottom layer was about -40 and the average temperature was -34 . figure [ f - fig9_jap_pgaps ] shows the temperature evolution of some of the relevant si(li ) detector components during the flight . the radiator was directly exposed to the ambient medium . during ascent it was not planned to have attitude control as the atmosphere is too dense and therefore the gondola was spinning randomly . upgoing spikes in the temperature distribution of the radiator are correlated with facing the sun side . after reaching float altitude the attitude control system was supposed to point the radiator towards the anti - sun side of the gondola to act as a heat dump for the tracker modules . unfortunately , due to an operational mistake this control was not possible . as a result the gondola was spinning at an approximately constant rate of one rotation every 5min , as measured by the fiber optical gyroscope , causing the tracker vessel to warm up by 0.02/min . one of the main goals of the pgaps flight was to verify the thermal model for the cooling approach of the si(li ) detectors . although the rotator failed , this goal was fully accomplished and the thermal analysis of all recorded temperature sensors led to a complete understanding of the thermal system @xcite . for flights of the full gaps instrument from antarctica the well established rotator from the columbia scientific balloon facility would be used . however , the effect of the increasing tracker temperature needs to be accounted for in the science flight data analysis . figure [ f - fig10_jap_pgaps ] shows the time evolution of the total number of active channels for the individual detectors over the course of the flight . under nominal operating conditions , out of 48 total channels 44 channels were operating while the other four channels were known not to be working since the qualification phase . at the time the tracker was turned off ( 10:30am ) the mean temperature was about -17 and 28 channels were still operational while the other 16 channels were either non - functional due to high leakage currents saturating the preamplifier or being no longer depleted . laboratory measurement after the flight at nominal temperatures showed that these channels were fully functional . the individual detectors and also the strips within a detector showed a variation with temperature , which is explainable by two different effects . some of the wire bonds between the si(li ) detector printed circuit boards to the detector surfaces were of visibly poor quality , which increased the noise level and therefore required lower operating temperatures compared to neighboring strips with better bonding quality . another effect was related to the surface condition of the grooves between the strips causing channels to deplete at different voltages due to inter - electrode capacitance differences . these effects were carefully investigated as for the full - scale payload the si(li ) modules will be fabricated in house by the gaps collaboration . for instance , one advantage of the gaps si(li ) modules over the semikon modules is the use of a robust pressure contact instead of wire bonding . another advantage is that the detectors will be structured on the n+ side instead of the p+ side . in this way , strips can be more easily separated as the depletion starts from the n+ side . * left : * x - ray tube ( preparation / red , right after launch / blue , float / green ) and am-241 source ( preparation / black ) measurements for one detector ( stack 0 , layer 0 , as defined in figure [ f - fig3_jap_pgaps ] ) integrated over all eight channels . energy values printed in the legend denote the mean and the fwhm of the gaussian fit . * right : * time evolution of the peak position of the gaussian fit for the 26kev x - ray tube peak . the error bars denote the widths of the gaussians . ] the left side of figure [ f - fig11_jap_pgaps ] shows the spectra for the x - ray energy deposition calibration for the detector closest to the x - ray tube . the spectra are integrated over all eight channels . the x - ray tube spectra were recorded during the launch preparation , right after launch , and after reaching float altitude . in comparison , and as already introduced in section [ s - trkcali ] , the am-241 spectrum taken a few hours before launch is shown . compared to the measurement shown for a different detector in figure [ f - fig7_jap_pgaps ] , a weaker am-241 line with respect to the x - ray tube lines is visible because less material is in front of the x - ray tube and more material in front of the am-241 source . the right side of figure [ f - fig11_jap_pgaps ] depicts the evolution of the position and width of the gaussian fit for the dominant 26kev x - ray tube peak over time . a small broadening is visible , which can be explained by a temperature increase of 11.5 over the course of the flight , which translates into an increase of the full width half maximum value ( fwhm ) of 0.08kev/. a similar measurement was carried out during the integration period using the am-241 source where an average of ( @xmath8)kev/ in the temperature range from -40 to -20 was observed . in conclusion , the observed effect of energy resolution change with temperature is small . once a strip was cold enough ( -20 ) to be depleted and have a small enough leakage current the x - ray resolution improved only by about 10 - 15% until it reached the nominal operational temperature of -35 . therefore , the temperature dependence would not significantly affect the ability to separate x - rays from antiprotonic exotic atoms from antideuteronic exotic atoms . the energy depositions of charged particles are much higher ( 1mev ) and behaved very stably . however , a good temperature regulation is critical to keep all detector channels at low leakage currents so as to not have holes in the acceptance . the effect of the increasing number of non - operational channels on the geometrical acceptance for pgaps for charged particle tracks with at least two tracker hits in tof trigger mode over the course of the flight is shown in figure [ f - fig10_jap_pgaps ] and will be further discussed in section [ s - accep ] . energy depositions normalized to charge @xmath9 for on ( black ) and off ( red ) track hits . ] energy depositions normalized to charge @xmath9 for on ( black ) and off ( red ) track hits . ] as outlined before , the tof trigger was set up in such a way to either trigger on a combination of top and middle layers or on a combination of middle and bottom layers . for pgaps about 10% of the charged particle triggers included at least one hit in the si(li ) modules . a track fit incorporating both tof and tracker hits was performed and is explained in detail in [ s - track ] . figure [ f - fig12_jap_pgaps ] illustrates the energy deposition @xmath10 distributions in the si(li ) detectors for different distances to the actual hit position for all reconstructed events during the flight integrated over all detectors . the distributions are normalized to the path length in the material:=,where @xmath11 is the zenith angle of the track and @xmath12 the thickness of the individual si(li ) module . the black distribution shows the behavior for strips on the track with a clear peak at ( @xmath13)kev / mm , a steep decline towards smaller energy depositions , and a long tail , as expected from a landau distribution for charged particle energy depositions . the red distribution represents the energy depositions for adjacent strips to the strip position calculated from the track parameters . it is very similar in shape to the on - track distribution starting from the most probable value ( mop ) . the resolution of the track fit is about 1.3 si(li ) strips wide and therefore allows hits in the neighboring strip of the calculated position . a smaller fraction of this distribution can also be attributed to cross - talk between adjacent strips due to capacitive coupling ( section [ s - trk_stab ] ) . below the mop more low energy entries are visible than for the on - track distribution . contributing effects are increased detector noise from the non - optimal tracker operating temperature ( section [ s - trk_stab ] ) , coherent atmospheric shower particles and lower energy depositions from interaction products of the primary charged particle in the detector material moving at a different angle , or to a very small degree incoherent atmospheric and cosmic x - rays . the distribution shapes for strips further than one strip away from the calculated track is radically different and only show a low energy contribution , supporting the reliability of the track fit . in the following , strips adjacent to the calculated track position with non - zero energy depositions are considered as on - track hits and farther away strips as off - track . as the mop for a minimum ionizing charged particle energy distribution scales as the square of its charge @xmath14 and the majority of all charged particles during the flight were expected to have an absolute charge value of @xmath15 , the @xmath16 values are transformed to absolute charge values @xmath9 by:|z|= and the corresponding distribution for all on and off track hits in the si(li ) during the flight is shown in figure [ f - fig13_jap_pgaps ] . the on - track distribution demonstrates a clear peak at @xmath15 and an additional shoulder at @xmath17 due to @xmath18 particles . the particle composition will be further discussed in section [ s - charge ] . in addition , the track fit information allows the extraction of the detection efficiency . using only the cleanest subsample of tracks with no off - track hits in tof or si(li ) , requiring at least two si(li ) hits in different layers , tracks going through the inner part of a module up to a radius of 3.7 cm , and neglecting non - operational si(li ) strips , the tracker showed an average detection efficiency of ( [email protected])% . this value should be interpreted as a lower bound as the si(li ) module operating temperature was not optimal . as the gaps antideuteron search will rely on a good x - ray identification , it is important to understand the level of coherent and incoherent x - ray backgrounds in coincidence with charged particles , e.g. , from particle interactions or atmospherically produced x - rays . for this study only the detector with the weakest temperature dependence was used . the critical energy range for the antideuteron analysis is between 20 and 100kev and both the on and off track distributions show a very similar behavior in this range ( figure [ f - fig14_jap_pgaps ] ) . the statistics for this study in tof trigger mode are low , but the data from the tracker trigger mode without the x - ray calibration tube can be used to better understand the atmospheric and cosmic x - ray component . in this regard , figure [ f - fig15_jap_pgaps ] illustrates the x - ray energy deposition flux at an average float altitude of about 32 km using a geometrical detector acceptance of 436@xmath20sr ( both sides ) and 27min of livetime . the coincidence flux of x - rays and charged particles will be presented after the charged particle flux discussion in section [ s - chargex ] . x - ray flux during tracker trigger mode . ] x - ray flux during tracker trigger mode . ] time difference between tof triggers of successive events during tof trigger mode . ] the livetime of the tof system is defined by the electronics deadtime of 280s . figure [ f - fig16_jap_pgaps ] shows the distribution of time differences between successive raw tof trigger events . as expected an exponential behavior is observed with a gap of 280s length in the beginning . the livetime is estimated by fitting an exponential to the distribution and calculating how many events were missed during the electronics processing period and results in @xmath21% . the energy deposition in each paddle was measured by two pmts and digitized with charge sensitive adcs . the raw adc spectrum of one pmt shows a constant offset ( pedestal ) followed by a landau distribution for the actual energy deposition ( figure [ f - fig17_jap_pgaps ] left ) . for each pmt , a search to determine the pedestal and most probable value of the landau distribution was performed . as discussed above , the mop position scales as @xmath22 for @xmath14 being the charge of the incident particle . the search was sampled over the flight to study the stability of the pedestal and mop positions ( figure [ f - fig17_jap_pgaps ] right ) . while the pedestal value @xmath23 was very stable , the mop value @xmath24 showed a slight increase over the course of the flight and was fitted with a straight line . each measured pmt adc energy deposition was calibrated to @xmath9 by performing : * left : * pedestal and most probable value positions for one typical tof pmt . the vertical red lines indicate the pedestal and mop value position , respectively . * right : * behavior of the most probable ( red ) and pedestal ( black ) value over the course of the flight for the same pmt . ] occupancy of tof pmts . bottom layer : pmt 0 - 11 , middle layer : pmt 12 - 19 , top layer : pmt 20 - 31 . ] similar to the previous tracker discussion , the following analysis will also make use of the path length corrected energy deposition : @xmath25 . despite the necessity of a slightly time dependent adc calibration , the tof operation was stable over the course of the flight for 31 out of 32 pmts . figure [ f - fig18_jap_pgaps ] depicts the pmt occupancy for the whole flight for energy depositions clearly above the pedestal ( @xmath26 ) . as it was a trigger requirement that the middle layer was always part of an event , the middle pmts show about twice as many entries as the top and bottom layer pmts . the distance between the top and middle layer was 18 cm shorter than the distance between middle and bottom , and therefore the top pmts show more entries than the bottom pmts . all tof pmts operated for the full duration of the pgaps flight , except for one tube in the middle tof layer . this pmt showed normal operation on the ground and in the flight while the ambient pressure stayed above 40torr . as soon as the pressure dropped below this value , the tube hv became unstable . this is consistent with corona discharge from exposure of part of the hv circuit to the low pressure ambient environment . since the behavior manifested as soon as the pressure was low enough ( and not after a period of outgassing ) , it suggests that the base assembly and potting was compromised some time after pmt vacuum testing . the pmt assemblies were integrated into the instrument about one year before the flight , and the paddle that this particular tube was mounted on was removed ( along with the other three middle layer paddles ) and re - installed every time the si(li ) detector vessel was accessed . therefore , it was subjected to some handling stress on a somewhat frequent basis . obviously in an experiment such as a gaps science flight , a more rigorous testing regime will be needed . gaps will have hundreds instead of dozens of pmts , and the flight time will be months . pmt assemblies for gaps will be tested in a low pressure environment for at least a week prior to acceptance . thermal - vacuum testing for at least a subset of assemblies will be considered as well . finally , some strengthening and enhanced strain relief of the pmt assembly is possible , especially for the cable feedthroughs . tdc time information as a function of energy deposition for the center slice of the paddle ( @xmath27 cm ) . the black and red data points / hyperbolic fits show the behavior of both pmts attached to the same paddle . ] tdc time information as a function of energy deposition for the center slice of the paddle ( @xmath27 cm ) . the black and red data points / hyperbolic fits show the behavior of both pmts attached to the same paddle . ] in addition to providing the main trigger for pgaps , the time - of - flight was also designed to measure charged particle velocities . a tdc value was recorded for a pmt when the energy deposition pulse crossed a preset discriminator threshold . the digitization had a precision of one count per 50ps . the velocity measurement requires an inter - calibration of all tdcs to reliably measure time of flights . important effects that need to be accounted for are the pulse propagation from the hit position to the pmt and the dependence of the tdc hit measurement on the pulse size at the pmt . all figures shown in this section depict the pmts from the same paddle . the steps described in the following were carried out for each paddle . to deconvolve the different effects the pgaps tracking information is used . figure [ f - fig19_jap_pgaps ] shows for both pmts the path length corrected energy deposition calibrated to charge @xmath28 as a function of the hit position calculated from the track parametrization for the coordinate along the paddle . the data points were fitted with straight lines and the signal is about 5 - 10% larger directly in front of the pmt compared to the center of the paddle . as expected , the lines for pmts on opposite sides cross in the center . as a first step in the timing calibration , all mean @xmath29 values in the center of each paddle were scaled to the same value ( mean of all uncalibrated paddle @xmath28 values in the center ) . using the same scaling factor @xmath30 but not the path length correction , the measured time as a function of @xmath9 for the @xmath27 cm slice in the center of a paddle for both pmts is shown in figure [ f - fig20_jap_pgaps ] . the data points are well fitted with a hyperbolic function and reveal that the tdc value dependence on the energy deposition is rather mild for signals above 0.8@xmath9 and nearly constant starting from about 1.2@xmath9 . for smaller signals the slope becomes much steeper . to decouple the measured tdc value @xmath31 from the pulse height the tdc hits are corrected to the same @xmath9 value using the hyperbolic fit : t@xmath32=t - a_i(-)with @xmath33 , @xmath34 being the fit values from the hyperbolic fit and @xmath35 the energy deposition scaling factor for each pmt @xmath36 . tdc time difference distribution of both pmts attached to a paddle for a 2.5 cm wide slice in the middle . ] tdc time difference distribution of both pmts attached to a paddle for a 2.5 cm wide slice in the middle . ] the next correction that needs to be applied is the timing offset between pmts at opposite ends . therefore , one of the pmts gets assigned a constant time shift to ensure that the @xmath37 behavior as a function of the coordinate along the paddle is at 0 in the center of the paddle ( figure [ f - fig21_jap_pgaps ] ) . the data points were fitted with a straight line . the slope of the fit equals @xmath38 where @xmath39 is the effective pulse velocity inside the paddle from the hit position to the pmt . an average value of @xmath40 over all paddles was measured and is about 20% slower than the velocity naively calculated from the refractive index itself ( @xmath41 ) because of photon reflections inside the scintillator material . gaussian fits to the @xmath37 distributions for the 2.5 cm slice in the center of the paddles give an average width of @xmath42ns that translates into an individual pmt timing resolution of @xmath43ns ( figure [ f - fig22_jap_pgaps ] ) . the timing resolution improvement using the outlined procedure is about 5% . the tdc value measurement behavior can therefore be interpreted as stable and is not prone to big systematic corrections ( figure [ f - fig20_jap_pgaps ] ) . for the actual velocity measurement , the individual paddle calibration of the last section has to be followed by the inter - paddle timing calibration . therefore , a mean time for each paddle using both pmts has to be calculated as well as the flight distance between them using the track fit . the flight distance between paddles with indices @xmath44 and @xmath45 , @xmath46 , is defined as : s_j , k = with @xmath47 being the three - dimensional hit coordinates on the paddles . the mean time value for one paddle is calculated by |t=12(t_+t_-)where @xmath48 are the corrected times of the pmts connected to the same paddle , @xmath49 the paddle length , and @xmath50 the effective pulse velocity inside the paddle . the particle velocity @xmath51 is then calculated by = .as the trigger was set up to always include the tof middle layer , the velocity calculation uses only combinations of top - to - middle and middle - to - bottom paddles . furthermore , downward going particles are defined as having a positive @xmath51 value . constant time offsets between paddles in different layers distort the velocity measurement and need to be corrected for . if @xmath51 were known the timing offset @xmath52 between two paddles @xmath44 and @xmath45 could be calculated by _ j , k = |t_k-|t_j+[e - off].as there is no reason to assume that the @xmath51 distributions for different paddle - to - paddle combinations should have different mean values , the offset finding algorithm takes the mean @xmath53 value as a free input parameter . the choice of @xmath53 is discussed after outlining the algorithm . only tof paddles on the track with both pmts having energy depositions above @xmath26 and tdc hits were used for the offset calibration . for each event , all offsets between the hit paddles and a prechosen reference paddle were calculated by using the input @xmath53 value , the distance known from the points where the track penetrates the tof paddles , and the mean paddle time values . these offsets were histogrammed per paddle where the center value of a gaussian fit to the distribution reflects the mean offset value for the choice of @xmath53 value and reference paddle ( figure [ f - fig23_jap_pgaps ] ) . to take into account all geometrically allowed combinations , all six paddles in the top layer were used subsequently as reference paddles . the resulting offsets were combined and averaged . histograms for different choices of @xmath53 value ( green : 0.5 , black : 0.6 , red : 0.7 ) for the paddle - to - paddle offset calculation with gaussian fits of the distribution above @xmath54 . ] histograms for different choices of @xmath53 value ( green : 0.5 , black : 0.6 , red : 0.7 ) for the paddle - to - paddle offset calculation with gaussian fits of the distribution above @xmath54 . ] the velocity was determined by all possible paddle - to - paddle combinations per event . the root mean square error @xmath55 of the different velocity measurements was used to justify the choice of @xmath53 in the following way : as a result of the finite tof timing resolution , events with @xmath56 are expected . therefore , a gaussian centered at @xmath54 with the average of the velocity root mean square error of all events @xmath57 as standard deviation should be able to explain the @xmath51 distribution above the speed of light . the @xmath57 value for each choice of @xmath53 value was calculated by averaging over the individual @xmath55 values per event . figure [ f - fig24_jap_pgaps ] shows the velocity distributions for three different choices of @xmath53 with corresponding gaussian fits with widths of @xmath58 where the only free parameter of the fit was the normalization factor . these fits underestimate the number of events above the speed of light . however , a systematic uncertainty band for the choice of @xmath53 from 0.5 to 0.7 will be used in the following to illustrate the effect on the velocity measurement ( section [ s - charge ] ) . to avoid the difficulty of the absolute velocity measurement calibration in the future , it is foreseen to calibrate the full gaps experiment with coincident led test pulses and with test beam measurements . furthermore , the redundant beta measurements produced an improvement ( by a factor of 2 ) in the timing resolution over single beta measurements ( to 0.7ns ) . summary of correction factors for flux analysis as a function of flight time . the total efficiency ( black ) is the result of multiplying the assembly efficiency ( red ) , the tracking efficiency ( blue ) , the relative tof livetime ( magenta ) , and the trk / tof interference factor ( green ) . ] summary of correction factors for flux analysis as a function of flight time . the total efficiency ( black ) is the result of multiplying the assembly efficiency ( red ) , the tracking efficiency ( blue ) , the relative tof livetime ( magenta ) , and the trk / tof interference factor ( green ) . ] the number of reconstructed events @xmath59 can be translated into a flux @xmath60 by taking into account measurement livetime @xmath61 geometrical acceptance @xmath62 , and detection efficiencies @xmath63:f=1.the directional distribution of clean charged particle tracks as a function of @xmath64 and the azimuth angle at float altitude is shown in figure [ f - fig25_jap_pgaps ] . the visible structure can be explained by the locations of active strips and tof paddles . figure [ f - fig26_jap_pgaps ] summarizes the different efficiencies and correction factors that are needed to transform the pgaps measurements into a non - instrument - specific flux and is explained in the following sections . the inverse of the black curve is the total systematic correction factor @xmath65 that needs to be applied to correct the measured number of clean events with at least two hit tracker strips to the number for the flux calculation . in tof trigger mode one tof data packet and six individual data packets from the different tracker modules were issued at the same time . these data packets contained synchronized time counter values with 100ns resolution to be able to assemble individual subsystem data packets to full events during the analysis . the details of the event assembly algorithm are described in [ s - assembly ] . unfortunately , not all events could be assembled to complete events due to time counter readout failures . however , as each tof trigger triggered the readout of the six tracker modules the event assembly efficiency can be defined as the number of complete events divided by the number of total individual tracker packets divided by six . the red curve in figure [ f - fig26_jap_pgaps ] shows the behavior as a function of time with assembly efficiency values of about 70% over the course of the flight . one dip shortly after reaching float altitude at 8:00am is visible and was caused by a flight computer reboot during the ascent from boomerang to float altitude and corresponding system adjustments . after this period the assembly efficiency showed the same value as before . the full gaps payload will follow a different approach not only relying on synchronized time counter values for the event building , but on onboard event building electronics . an interference between the tracker and tof electronics was present that manifested itself in a reduced rate when the tracker electronics were operated in tof trigger mode . the effect was studied by looking at the tof trigger rates when the tracker system was completely turned off and when the tracker system was fully operational in tof trigger mode . this measurement was carried out during the launch preparations on ground and at the end of the flight and shows a constant value for these two very different environments of ( @xmath66)% ( green curve ) . it was an unknown problem before the flight preparation in taiki . anyhow as mentioned above , the full gaps readout system will follow a different approach that will make this issue obsolete . as the tracking system electronics had a significantly shorter deadtime than the tof electronics ( about a factor of 20 ) the relative livetime for the tracking system during tof trigger mode was nearly 100% . the tof system showed a constant livetime of @xmath21% , as discussed in section [ s - tof - live ] . the geometrical acceptance of the pgaps detector using the tof trigger condition ( section [ s - trig ] ) and the clean track requirement ( [ s - track ] ) was calculated by applying the monte carlo approach from @xcite resulting in ( @xmath67)m@xmath68sr for particles coming from above and below , where the error on the acceptance reflects the uncertainty of the detector geometry . as previously discussed in section [ s - trk_stab ] , the increasing temperature of the tracking system caused a change of geometric acceptance over the course of the flight and led to a decrease of particle tracks with at least two tracker hits . a correction factor can be estimated by studying the number of events with two tracker hits in comparison to the total number of events . the blue graph in figure [ f - fig26_jap_pgaps ] shows this behavior normalized to the ratio of clean two tracker strip events over the number of total triggered events from the beginning of the flight when all detectors were active . total flux as a function of altitude . the gray shaded area depicts the systematic error from the flux correction explained in section [ s - eff ] . ] the charged particle flux as a function of altitude is shown in figure [ f - fig27_jap_pgaps ] . the efficiencies discussed in the last sections were treated as systematic effects and are shown as an error band . at boomerang altitude ( 1015 km ) the flux is about 30% higher than at float altitude . figure [ f - fig28_jap_pgaps ] shows the downward and upward fluxes measured at float altitude as a function of @xmath51 . the bin widths were chosen to reflect the pgaps tof timing resolution . for a reliable @xmath51 value it was required that a paddle used for the calculation had clean energy deposition in both pmts with @xmath26 and non - zero timing values . the flux was scaled accordingly for events not fulfilling this quality cut . the additional @xmath69 axis illustrates the corresponding kinetic energy using the proton mass . in addition to the systematic error bands due to detection efficiencies ( section [ s - eff ] ) and the choice of @xmath53 value ( section [ s - velo ] ) , the flux from air shower and geomagnetic simulations for taiki at 33 km altitude with planetocosmics is shown @xcite . the dashed blue histogram shows the simulated combined proton , @xmath18 particle , and muon flux assuming perfect timing resolution and exhibits a clear peak at @xmath54 . introducing a tof timing resolution of @xmath70ns and assuming an average distance between tof paddles of 0.45 m , entries from the sharp peak migrate to lower and higher velocity values ( solid blue histogram ) . as shown in @xcite , the atmospheric simulations are able to reproduce the fluxes in the atmosphere as measured by various experiments ( e.g. , bess ) . however , these experiments did not report the atmospheric fluxes below kinetic energies of 500mev . the pgaps measurements agree well with the simulations using 0.7ns tof timing resolution for velocities above @xmath71 , but show upward deviations at @xmath51 values of about 0.6 and 0.25 corresponding to kinetic energies for protons of 250mev and 30mev , respectively . below @xmath72 the simulations describe the spectrum very well and also the range from 0.4 to 0.5 is in good agreement within the statistical and systematic errors . the low energy range is especially prone to atmospheric interaction and geomagnetic deflection physics effects and the low - velocity pgaps results will be subject to further studies in the future . the total measured flux is mostly composed of downward going particles and contains only a small fraction ( 1% ) of upward going particles . in this regard , it is important to mention that the upward going flux was not corrected for the components and material under the science part of the gondola ( batteries , ballast hoppers , gondola bus , etc . ) . the upward coming flux is expected to be mostly composed of low - energy particles and therefore the material attenuates the actual flux significantly . downward ( black ) and upward ( red ) going charged particle flux as a function of velocity for the period at float . the gray area depicts the systematic error from the flux correction and the red shaded area from the choice of @xmath53 value between 0.5 and 0.7 . the dashed blue histogram shows the downward flux of protons , @xmath18 particles , and muons at taiki ( 33 km ) as predicted by air shower simulations @xcite . the solid blue histogram shows the results of the same simulations , but assuming 0.7ns tof timing resolution . the additional @xmath69 axis illustrates the kinetic energy range assuming that all particles are protons . ] the distribution of averaged energy depositions over all hit tof and tracker detectors per event for the time at float reveals a clear peak at @xmath15 and a second peak at @xmath17 coming from @xmath18 particles where the second peak was not observed on the ground ( figure [ f - fig29_jap_pgaps ] ) . the ratio of the integrals of the fitted landau distributions for the @xmath17 to @xmath15 populations is @xmath73% . for the same atmospheric depth of 9g/@xmath20 , measurements by the bess spectrometer for the 2001 flight from ft . sumner showed a ratio of the integral fluxes of @xmath18 particles to protons and muons of about 14% for the kinetic energy range of 0.510gev@xmath74 @xcite . the discrepancy can be explained by the geomagnetic cut - off , which is about twice as high at taiki ( 8gv ) than in ft . sumner ( 4gv ) @xcite . while protons with energies below the geomagnetic cut - off are abundantly produced in interactions of cosmic rays with the atmosphere , the probability of atmospheric production of @xmath18 particles is very low . therefore , the ratio of @xmath18 particles to protons and muons is expected to drop with increasing geomagnetic cut - off . average charge @xmath9 on track at float altitude . the black line shows a double peak landau fit and the red ( blue ) lines show the individual contributions of the @xmath15 ( @xmath17 ) distributions . the vertical lines indicate the 68.3% confidence interval around the peaks . ] the charged particle and x - ray results can be used to estimate the flux of events with coincident charged particles and atmospheric and cosmic x - rays in the range of antideuteron exotic atom ladder transitions . for the full payload , the geometric acceptance for charged particle tracks crossing all 10 tracker planes is about 3m@xmath68sr . pgaps measured an integrated charged particle flux of about 250m@xmath75sr@xmath76s@xmath76 at 33 km . assuming a rather pessimistic x - ray energy resolution , the interesting ranges for the antideuteron exotic atom ladder transitions are 2733kev , 4147kev , and 6470kev . the integrated x - ray flux of these ranges was measured to be @xmath77m@xmath75sr@xmath76s@xmath76 ( fig.[f - fig14_jap_pgaps ] ) . the x - rays have to be reasonably close to the charged particle track . therefore , this calculation assumes that the x - ray energy depositions have to occur in the detectors on the track , which translates into an x - ray acceptance of 0.44m@xmath68sr for the coincidence calculation . the exotic atom ladder transitions will happen within 50ns . as a result , the flux of charged particles through the whole experiment in coincidence with one x - ray in the antideuteron exotic atom ladder transition range is @xmath78m@xmath75sr@xmath76s@xmath76 and with two x - rays @xmath79m@xmath75sr@xmath76s@xmath76 , respectively . these background fluxes should be taken as conservative upper limits , as they are not taking into account other particle identification techniques like penetration depth in the tracker , the development of the energy loss throughout the detector , and for the case of two x - rays that they must come from two different ladder transitions . in conclusion , requiring more than one x - ray in the right energy ranges alone suppresses this type of background extremely well . the identification of dark matter is one of the most striking problems in physics and a low - energy cosmic ray antideuteron search has great potential in revealing deep insights . the gaps experiment is specifically designed to perform this task . a prototype gaps was successfully flown in june 2012 . the purpose of this flight was to test and thoroughly analyze the concepts that form the basis for future flights . all goals for the flight were met and it was shown that the si(li ) tracker detector modules and tof worked reliably under flight conditions , the thermal model was verified @xcite , and background particle and x - ray fluxes were measured . the detailed design work for the full gaps payload has been started already . we thank g. tajiri and d. stefanik for the mechanical engineering support , and also thank j. hoberman and b. mochizuki for the development of the gaps electronics . furthermore , we would like to thank c. hailey , k. kamdin , p. kaplan , m. lopez - thibodeaux , and t. zhang for their contributions to the project . we thank the scientific balloon office of isas / jaxa for the professional support of the pgaps flight . this work is partly supported in the us by nasa apra grants ( nnx09ac13 g , nnx09ac16 g ) and the ucla division of physical sciences and in japan by mext grants kakenhi ( 22340073 ) . k. perez s work is supported by the national science foundation under award no . the tof and the two tracker electronics signal processing units were connected to the clock board on the flight computer . the clock board generated a continuous rectangular 10mhz pulse , which was connected to discriminator circuits in the tof and tracker units and was used to count the number of transitions . the corresponding counters could be reset with a short individual pulse . the counter value was stored with 32bit precision and included in every event packet . in this way synchronized time counter values with 100ns resolution could be achieved between the different subsystem electronics . the 32bit counter was rolling over about every 7:15min and the first step in assembling full events of both tracker electronic units and the tof was to find the rollover positions . figure [ f - fig30_jap_pgaps ] shows the time counter values from the tof system for a data excerpt from the end of the flight as a function of incoming packet order . the rollovers are clearly visible and the vertical lines mark the positions found by the analysis algorithm . after adjusting for the number of rollovers by adding the corresponding number of seconds , the different electronics subsystems can be compared with each other . to find possible timing offsets between the subsystems , all rollover corrected time counter values of one subsystem were subtracted from the time counter values of another subsystem . the histogrammed differences are illustrated in figure [ f - fig31_jap_pgaps ] . as expected from a constant clock offset a very sharp peak is visible . in the first step the histogram spanned a wide time range of 2000s to also allow for finding missed rollovers . in the second iteration the same differences were filled into a histogram only as wide as the maximum bin of the coarse histogram before with a 100ns bin width ( figure [ f - fig32_jap_pgaps ] ) . a very sharp peak covering only one bin is visible and is used as the timing offset between the two subsystems under study . this procedure was carried out for each data taking block between clock resets and sampled over time to study if the timing offsets were constant . it was found that the offsets stayed absolutely constant after a clock reset , but could change after clock counter resets . figure [ f - fig33_jap_pgaps ] shows the result of the timing and offset calibration for all three electronics subsystems where the corrected time counter values for event data containing non - zero information follow a straight line compared to the also recorded unix computer clock time . data packets were merged into one event if the time counter values had a difference smaller than 1s . the track fitting procedure has the goal to fit a parametrized straight line to charged particle tracks inside the detector using the least squares @xmath80 approach . a typical reconstructed event using the procedure described in the following is shown in figure [ f - fig34_jap_pgaps ] . the tof paddles ran along the @xmath69 and @xmath81 directions of the instrument coordinate system . the si(li ) tracker is made up of two stacks with three circular modules each with a spacing of 8 cm between the layers . the centers of the circular si(li ) modules were offset from the central @xmath82 axis by 12 cm and one stack was rotated by @xmath83 and one by @xmath84 around the @xmath82 axis . the internal coordinate system of the si(li ) modules is described by the cartesian coordinates @xmath85 and needs to be rotated into the @xmath86 absolute instrument coordinate system : ) , + where @xmath87 are the rotation angles of the two stacks around the @xmath82 axis . also the internal si(li ) detector covariance matrix @xmath88 needs to be transformed to the absolute coordinate system and can be obtained from standard error propagation : the @xmath89 describe the errors in the internal coordinate system . the @xmath80 that needs to be minimized throughout the track fit is : ^2=_i^2_i=_i _ i^tu^-1 _ i , where @xmath90 denotes the difference vector between the position calculated from the straight line parametrization to the actual track point @xmath36 used for the fit . the @xmath91 calculation for the tof points being part of the track fit used the same approach , but with a rotational angle @xmath92 , which simplifies the calculation and removes the non - diagonal elements of the covariance matrix . an active tracker strip for the track fit was defined as having an energy deposition above @xmath94 while a tof paddle was used for the track fit if the energy deposition of at least one pmt of the paddle was above @xmath26 and also showed a non - zero timing value . each active detector needed to be associated with a three - dimensional track coordinate and corresponding error bars . the center of gravity of a tof paddle or si(li ) detector strip was used as the average track coordinate for the particular detector volume . as an equivalent to the @xmath95 range used for fits with error bars in only one dimension , the three - dimensional error bars for a track point were chosen such that they span a volume of 68.3% the size of the paddle or strip volume around the track point . the case for the rectangular box shaped tof paddles is trivial and the error bars are set to be @xmath96{68.3\,\%}/2 $ ] times the length of the paddle in its @xmath97 direction . the error boxes used for the case of the si(li ) modules are shown in figure [ f - fig35_jap_pgaps ] . the track fits were performed using the minuit fitting routines provided by root @xcite . a track is considered to be good if the requirement @xmath98 was fulfilled , where the @xmath93-value is defined as : p=_^2^f(t , n)twith the number of degrees of freedom @xmath99 and the @xmath80 probability density function @xmath100 . @xmath99 is the number of points used for the track fit subtracted by two . from purely statistical effects it is expected that the @xmath93-value distribution is uniform . a peak at @xmath101 corresponds to too many large @xmath80 values and is not in agreement with statistical fluctuations . for track quality reasons only the nearly uniform part above 5% of the distribution was taken into account ( figure [ f - fig36_jap_pgaps ] ) . clean tracks used for the analysis were required to have more than two hits in any active detector volume and in addition to show at least two tracker hits with energy depositions @xmath94 . figure [ f - fig36_jap_pgaps ] demonstrates the power of this requirement as the tighter tracker constraint ( red histogram ) makes the distribution much cleaner than the looser definition ( black histogram ) because the tof position resolution is much coarser than the tracker resolution . k. freese , europ . 36 , 113 ( 2009 ) . o. adriani et al . , nature 458 , 607 ( 2009 ) . o. adriani et al . lett . , 105 , 121101 ( 2010 ) . m. aguilar et al . , phys . b 646 , 145 ( 2007 ) . m. aguilar et al . , phys . rev . let .. 110 , 141102 ( 2013 ) . f. donato et al . , phys . d 62 , 043003 ( 2000 ) . r. duperray et al . , phys . rev . d 71 , 083013 ( 2005 ) . a. ibarra and s. wild . , phys . rev . d 88 , 023014 ( 2013 ) . n. fornengo et al . , j. cosmol . , 031 ( 2013 ) . a. barrau et al . , astron . 398 ( 2 ) , 403 ( 2003 ) . a. barrau et al . , phys . rev . d 69 , 105021 ( 2004 ) . h. baer et al . , j. cosmol . astropart . , 008 ( 2005 ) . f. donato et al . , phys . rev . d 78 , 043506 ( 2008 ) . m. kadastik et al . , phys . b 683 ( 45 ) , 248 ( 2010 ) . y. cui et al . , j. high energy phys . 1011 , 017 ( 2010 ) . m. grefe , j. phys . ser . 375 012035 ( 2012 ) . a. ibarra et al . , j. cosmol . , 021 ( 2013 ) . c. hailey , new j. phys . 11 , 105022 ( 2009 ) . v. choutko et al . . cosmic ray conf . 4 , 765 ( 2007 ) . t. aramaki , ( to be published ) ( 2013 ) . von doetinchem et al . 10th symposium on sources and detection of dark matter and dark energy in the universe ( ucla dark matter 2012 ) ( 2012 ) , to be published .

we present a mechanism to localize zero mode non - abelian gauge fields in a slice of ads . as in the u(1 ) case , bulk and boundary mass terms allow for a massless mode with an exponential profile that can be localized anywhere in the bulk . however in the non - abelian extension , the cubic and quartic zero mode gauge couplings do not match , implying a loss of 4d gauge invariance . we show that the symmetry can be restored at the nonlinear level by considering brane - localized interactions , which are added in a gauge invariant way using boundary kinetic terms . possible issues related to the scalar sector of the theory , such as strong coupling and ghosts , are also discussed . our approach is then compared with other localization mechanisms motivated by dilaton gravity and deconstruction . finally , we show how to localize the scalar component zero mode anywhere in the bulk which could be relevant in gauge - higgs unification models . umn - th-2527/06 + november 2006 * yang - mills localization in warped space * brian batell and tony gherghetta + +

hep-th0611305

gauge theories in a slice of anti - de sitter ( ads ) space have attracted a great deal of attention in recent years primarily because they provide geometrical explanations of particle physics mysteries , including the gauge hierarchy problem @xcite , and a window into strongly coupled gauge theories via the ads / cft correspondence @xcite . massless gauge fields propagating in the bulk of ads@xmath0 have been thoroughly examined in the literature @xcite and have many applications ( see for example @xcite ) . it is well - known that the zero mode of this field has a flat profile ( constant wavefunction ) in the extra - dimension , and that the dual theory is described by a ( mostly ) elementary source field interacting with a strongly coupled conformal field theory ( cft ) via a conserved cft current @xcite . not so familiar is the fact that gauge fields can be localized in the extra dimension , at least for the abelian case @xcite . in order to localize the zero mode , the bulk equation of motion and boundary conditions must be modified , which is accomplished through the introduction of mass terms in the bulk and on the boundaries . by tuning these masses , a zero mode solution with an exponential profile is allowed . the phenomenology and holographic description of localized u(1 ) gauge fields is described in @xcite . extending this mechanism to the non - abelian case is not so straightforward . in order to preserve four - dimensional ( 4d ) gauge invariance , the zero mode cubic and quartic gauge interactions must have the same coupling . these couplings are proportional to wavefunction overlap integrals , and for exponential zero mode wavefunctions they do not necessarily match . thus it seems the localization mechanism violates 4d gauge invariance . to have a viable low energy gauge theory , the equality of these couplings must be restored . in this paper we show that this can be achieved by considering interactions on the boundary , which provide the correct contributions to zero mode cubic and quartic terms to force the gauge couplings to be equal . these interactions can be added to the theory in a gauge invariant way via boundary kinetic terms . localizing both abelian and non - abelian gauge fields is important for both 5d model building and phenomenology as well as holographic physics . because the gauge field can be localized anywhere in the bulk , an entire new range of parameter space can now be explored . for example , if the standard model gauge bosons all propagate in the bulk , each field could in principle be localized at different places in the extra dimension . this may help to alleviate experimental bounds , as in the u(1 ) case , and could perhaps lead to novel model building scenarios . perhaps more interesting is the notion that if zero modes are localized near the infrared boundary , there may exist a dual theory in which the gauge fields are composite cft states . in this case , the dual gauge theory would be a valid effective field theory below the cutoff corresponding to the scale of conformal symmetry breaking . our approach is not without a few caveats that need to be addressed . first , a particular boundary kinetic term must be added to restore the gauge invariance at the nonlinear level . this represents an additional tuning beyond the tuning already required between bulk and boundary masses ( as well as the tuning between bulk and boundary cosmological constants in the original randall - sundrum model ! ) . more importantly , mass terms in the 5d theory must be generated through spontaneous symmetry breaking if we hope to preserve a 4d gauge symmetry . therefore , we must also account for the additional scalar fields present in the theory . these scalar fields complicate the analysis and may also introduce more severe problems , such as ghosts or strong coupling . although we are not able to offer a general analysis of these scalar issues , we present some simple examples illustrating the problems and speculate on possible resolutions . even if these issues can only be resolved in the uv completion of the model the nice features of the yang - mills localization will still be preserved at low energies and therefore makes the present study relevant . we also compare our approach to other methods of localizing gauge fields on the brane . as first shown in @xcite , gauge fields can be given a localized wavefunction through a dilatonic coupling . a similar mechanism is also motivated by deconstruction , in which the 4d gauge theories have nonuniversal gauge couplings at each site @xcite . these approaches share many of the features of our method , but there are also differences , which we point out . the scalar component @xmath1 is also utilized in models of electroweak symmetry breaking as an alternative to the higgs boson @xcite . the zero mode profile of this field is restricted by gauge invariance . however , we show that like the vector @xmath2 , it is possible to change the profile , and thus localize the @xmath1 zero mode anywhere in the bulk . this could affect the phenomenology of gauge - higgs unification models in novel ways . the organization of this paper is as follows . in section 2 we begin by reviewing how the addition of bulk and boundary mass terms allow for a localized zero mode gauge field . we then move on to the full non - abelian analysis , showing how boundary kinetic terms , if appropriately tuned , can restore 4d gauge invariance . in section 3 we discuss the additional scalar fields in the theory . we present a flat space analysis of the nambu - goldstone bosons associated with the massive vectors in the 4d theory . we also discuss the possibility of strong coupling and ghosts by considering a simple example containing only boundary masses ( no bulk masses ) . in section 4 we review other methods of gauge field localization , including localization with a dilaton and via deconstruction , and compare these methods with our approach . we discuss the localization of the scalar @xmath1 in section 5 . finally we conclude in section 6 , and suggest possible directions for future investigation . we will work with the following metric describing 5d anti - de sitter space : @xmath3 where @xmath4 is the extra coordinate and @xmath5 is the curvature scale related to the bulk cosmological constant . the fifth dimension @xmath4 is compactified on a @xmath6 orbifold and two three - branes are located at the orbifold fixed points @xmath7 and @xmath8 , referred to as the ultraviolet ( uv ) and infrared ( ir ) brane , respectively . the indices of 5d coordinates are labeled with latin letters ( @xmath9,@xmath10 , @xmath11 ) while 4d indices are labeled by greek letters ( @xmath12 , @xmath13 , @xmath11 ) . all 4d indices are raised and lowered with the minkowski metric @xmath14 . it is well - known that massless gauge fields propagating on the background ( [ adsmetric ] ) lead to a massless zero mode that is not localized in the extra dimension , but rather has a flat profile @xcite . instead to localize a zero mode , bulk and boundary mass terms must be added which modify both the bulk equation of motion and the boundary conditions for the gauge field @xcite . the massless mode then has an exponential profile and can be localized anywhere in the bulk . we will show how this is accomplished in section 2.1 . of course such mass terms must be generated in a gauge invariant way , so that a 4d gauge symmetry is preserved for the zero mode . the most obvious way to generate such terms is by the spontaneous breaking of gauge symmetry . this introduces extra scalar fields into the theory which may cause problems . for instance , if massless scalar modes persist in the 4d theory they may be undesirable for model building . other , more severe problems , such as the existence of ghosts or strongly coupled light scalar fields , may also occur . we will discuss these issues in section 3 . however , the gauge sector can be isolated from the scalar sector , and none of these more severe problems exist for the gauge field . a separate issue arises when we consider the localization of non - abelian gauge fields . as we will see in this section , the zero mode cubic and quartic couplings do not match , and thus the conventional 4d gauge symmetry does not exist in the simplest formulation . the effective couplings are proportional to the overlap integral of the wavefunctions , and are only equal when the zero mode is flat . to remedy this , we will add cubic and quartic interactions on the branes which will restore the equality of the zero mode gauge couplings . the only way this can be done while preserving gauge invariance is by adding a kinetic term to the brane , which we show in section 2.2 . we study only pure yang - mills theory in this paper . however , if fermions propagate in the bulk , we can generalize the mechanism and restore 4d gauge invariance by adding gauge covariant boundary kinetic terms for fermions . we begin by briefly reviewing how to localize the zero mode gauge field in the extra dimension . for simplicity we will consider a u(1 ) gauge field , since the generalization to the non - abelian case is straightforward at the quadratic level in the action . consider the following 5d equation of motion for a spin-1 field @xmath15 : @xmath16 the mass term arises from spontaneous symmetry breaking @xcite and contains both bulk and boundary terms : @xmath17 where we have defined @xmath18 . the bulk and boundary masses are written in terms of the ads curvature @xmath5 with dimensionless parameters @xmath19 and @xmath20 . as shown in @xcite , we must tune the parameters , @xmath21 to allow a zero mode solution . only real values for the mass parameters , @xmath22 , are considered and therefore @xmath20 can take any real value . let us choose the gauge @xmath23 for simplicity . the equation of motion ( [ eom1 ] ) becomes @xmath24a_\mu(x , y)=0~. \label{eom2}\ ] ] decomposing the vector @xmath2 as @xmath25 the solution of ( [ eom2 ] ) for the massless mode is @xmath26 since @xmath20 can be any real number ( [ tune ] ) , the zero mode can be localized anywhere in the fifth dimension . to see the necessity of tuning the parameters @xmath19 and @xmath20 , note the boundary condition derived by integrating the equation of motion ( [ eom2 ] ) across the boundary is @xmath27 the zero mode satisfies this equation only if the relation ( [ tune ] ) holds . otherwise , the boundary conditions project out the massless mode . it is possible that this tuning is enforced in a supersymmetric extension , as happens for bulk scalar fields @xcite . to summarize , the bulk mass term changes the profile of the zero mode from flat to exponential . clearly , an exponential wavefunction is incompatible with ordinary dirichlet or neumann conditions . the brane - localized mass modifies the boundary condition to be compatible with this exponential wavefunction . furthermore , this leads to interesting phenomenological possibilities and remarkably the localized u(1 ) gauge bosons can be given a holographic interpretation @xcite . let us now consider the following action for a non - abelian gauge field @xmath28 in a slice of ads@xmath0 : @xmath29~ , \label{a1}\end{aligned}\ ] ] where the field strength tensor is defined as @xmath30 with group structure constants @xmath31 and 5d coupling @xmath32 . again , the mass term is assumed to be generated by spontaneous symmetry breaking . the technique described in the previous section can also be applied to localizing non - abelian gauge fields . at the quadratic level , the equation of motion is the same as in the abelian case , and therefore the procedure is identical . however , when we examine the nonlinear terms of the yang - mills theory to compute the zero mode cubic and quartic interactions , we find that the couplings do not match . thus it seems that there is no longer any 4d gauge invariance . it is possible to restore the zero mode cubic and quartic couplings consistent with 4d gauge invariance if we consider brane - localized interactions . these terms play a similar role to the boundary mass terms , which cancels the bulk mass contribution to the zero mode action . we can not add these terms by hand to the boundary action without violating gauge invariance . to preserve the symmetry , we must add a boundary kinetic term , so that the action becomes @xmath33 . \label{a2}\ ] ] note the boundary term is suppressed by @xmath34 relative to the bulk term . it will be necessary to tune the dimensionless coefficient @xmath35 to match the zero mode couplings . for simplicity , we have chosen the same coupling on each brane , but it is possible to have them different , which we will discuss at the end of this section . also notice that one of the kinetic terms has a negative coefficient and therefore could introduce a ghost into the theory . however , it is the sum of the bulk and boundary kinetic energies which is important . when we perform the kaluza - klein reduction , we will require the wavefunction overlap integral , including the bulk and boundary terms , to be positive , which then implies the particle has positive energy . we will see that this requirement is satisfied for all values of @xmath35 that we will consider . in fact negative boundary kinetic terms have been previously considered in the literature . in ref . @xcite the effect of brane kinetic terms in the randall - sundrum model was studied and it was found that a ghost does indeed appear in the spectrum , but only if one of the boundary terms is made so strong and negative that it overwhelms the positive energy contributions from the bulk and other boundary . similar conclusions were drawn in ref . @xcite where negative kinetic terms in flat space were examined . thus , it seems from the 4d effective field theory vantage point , negative boundary kinetic terms in principle pose no problem to model building . however , note that this cancellation mechanism is nonlocal since locally there is always a negative kinetic term . with this modification to the action , the equation of motion at quadratic level is @xmath36a_\mu(x , y)=0~. \label{eom4}\ ] ] decomposing the field ( [ kk ] ) and demanding @xmath37 the equation of motion for the eigenfunction @xmath38 becomes @xmath39f^n(y)=-e^{2ky}m_n^2\left(1+\frac{\beta}{k^2}\sigma''(y)\right)~. \label{eom5}\ ] ] the bulk equation of motion is unaffected by the presence of the boundary term , but the boundary condition is now altered : @xmath40 the massive modes are influenced by the brane kinetic term and thus the spectrum will be modified . however , the boundary condition for the massless mode @xmath41 is the same as in the abelian case ( [ bc1 ] ) , and therefore the zero mode can be localized provided we demand the tuning condition ( [ tune ] ) . the normalized zero mode solution is given by @xmath42 note that for a flat zero mode ( @xmath43 ) , the normalization is unaffected by the presence of the brane - localized kinetic term since the wavefunction has the same value at each boundary . the normalization @xmath44 must be real , or equivalently the overlap integral ( [ normal ] ) must be positive so that the field is not ghostlike . thus , we must require that @xmath45 . we will see shortly that this limit is always satisfied if we require that a 4d gauge invariance be preserved for the zero mode . the interaction terms following from the action ( [ a2 ] ) are @xmath46~. \label{int1}\ ] ] localizing a massless mode comes at a price : the loss of _ 4d gauge invariance _ of the zero mode . without boundary interaction terms , the cubic and quartic vertices will have different strengths . the 5d symmetry may still be realized in the effective theory , but it will not take the standard form familiar in 4d theories . the gauge invariance can be restored if we add a boundary kinetic term with a precise coefficient ( [ a2 ] ) . let us investigate the effective 4d interactions and see the problem and its resolution explicitly . inserting the zero mode wavefunction ( [ wf ] ) into ( [ int1 ] ) , the couplings can be calculated by performing the overlap integrals . the cubic coupling is given by : @xmath47 while the quartic coupling is @xmath48 clearly for @xmath49 , @xmath50 and the 4d gauge symmetry is lost for general values of @xmath20 . thus we see the necessity of including a boundary kinetic term for nonzero @xmath20 . only when the mode is flat ( @xmath43 ) are the couplings identical . moreover , when the mode is flat arbitrary boundary kinetic terms ( with different coefficients ) may be added to the action without affecting the zero mode gauge invariance . we will demand that a 4d gauge symmetry be present in the theory for general values of @xmath20 , or in other words , we will require that @xmath51 . to find the value of @xmath35 that satisfies this requirement , we square ( [ cubic ] ) and equate it to ( [ quartic ] ) . solving for @xmath35 , we find @xmath52 , \label{beta}\ ] ] where we have defined @xmath53)^2}~.\ ] ] the exponential modulating function @xmath54 ranges from as low as @xmath55 when @xmath43 and as high as @xmath56 , when @xmath57 is very large . for any particular value of the localization parameter @xmath20 , there are two possible boundary kinetic terms that will preserve 4d gauge invariance , with strength @xmath58 or @xmath59 . the @xmath58 and @xmath59 branches are continuously connected at @xmath43 and this behavior is plotted in fig . [ fig1 ] . ( in units of @xmath60 ) as a function of @xmath61 . the @xmath58 branch is indicated by the solid line while the @xmath59 branch is indicated by the dashed line . the @xmath62 and @xmath63 branches are continuously connected at @xmath43 corresponding to a constant zero mode wavefunction . at @xmath43 , @xmath35 can take any value . ] it is instructive to study two limiting cases . for small values of @xmath20 , we find that the coefficient @xmath35 becomes @xmath64 this equation seems to imply that for the flat mode ( @xmath43 ) , we must also include boundary kinetic terms to maintain 4d gauge invariance . of course this is not true . the presence of boundary kinetic terms has no effect on the 4d gauge invariance when the mode is flat , which can easily be seen , for example , from the wavefunction overlap integrals that determine the 4d couplings , eqs . ( [ cubic ] ) and ( [ quartic ] ) . the boundary terms simply cancel when the wavefunction is constant . in the more generic case of different coefficients for each brane - localized kinetic term , the zero mode couplings are still equal . we can also study the limit where @xmath65 , in which case the coefficient @xmath35 reduces to @xmath66 we see that either a very small or very large boundary kinetic term must be added to restore gauge invariance in the low energy theory . now we turn to the effective 4d gauge couplings , which are computed by inserting ( [ beta ] ) back into ( [ quartic ] ) . the couplings can also take two possible values depending on which boundary kinetic term ( @xmath58 or @xmath59 ) is chosen . this is shown clearly in fig . [ fig2 ] with @xmath67 plotted as a function of @xmath20 in units of @xmath68 . plotted ( in units of @xmath68 ) as a function of @xmath20 for @xmath69 . for any given value of @xmath20 there are two characteristic behaviors , the @xmath59 branch ( dashed line ) , and the @xmath58 branch ( solid line ) . the @xmath62 and @xmath63 branches are continuously connected at @xmath43 . ] the two branches are again connected at @xmath43 , in which case the 4d gauge coupling becomes the familiar @xmath70 . for large @xmath57 the gauge coupling on the @xmath59 branch is approximately @xmath71 notice that a large value of @xmath72 is required to obtain a 4d gauge coupling of order one and therefore perturbativity of the 5d gauge theory does not allow for arbitrary localization on the @xmath59 branch . on the other hand for the @xmath58 branch ( shown as the solid line in fig . [ fig2 ] ) , the gauge coupling for highly localized modes ( @xmath73 ) becomes : @xmath74 in this case a 4d gauge coupling of order one for large @xmath57 can always be obtained for perturbative values of the 5d gauge coupling . thus , for extremely localized zero modes , only the @xmath58 branch is consistent with perturbativity for all large values of @xmath57 . for weak localization where @xmath75 , a 4d coupling of order one can be defined for both branches . as @xmath57 becomes of order unity and the mode is localized , only the @xmath58 branch would be consistent with perturbative values of @xmath68 . finally , let us verify that the zero mode is not a ghost for all values of @xmath35 . at the end of section 2.2 we found that @xmath76 , otherwise @xmath77 has negative kinetic energy and is ghostlike . from eq . ( [ beta - largealpha ] ) we see that the maximum value of @xmath78 is @xmath79 which occurs for very large values of @xmath57 on the @xmath58 branch . thus we see that the limit is always satisfied and the zero mode has positive energy . notice that in our approach , there are two relations that must be tuned : 1 ) the bulk and boundary masses must satisfy relation ( [ tune ] ) to permit a zero mode solution consistent with boundary conditions , and 2 ) a precise boundary kinetic term ( one on each brane ) with a coefficient given by ( [ beta ] ) , which is related to the boundary mass @xmath20 must be added to the action . of course , all of these parameters will receive quantum corrections , and therefore these relations are inherently radiatively unstable . this question will not be addressed in this paper and we simply assume that the tuning is enforced in the underlying theory to cancel any quantum effects that leads to deviations in the tunings . one might hope that there is some symmetry which could enforce the tuning relations . one possibility is supersymmetry . similar bulk and boundary mass relations for scalar fields are demanded by supersymmetry @xcite , and it is not unrealistic to think that the same thing could happen for gauge fields . however there are no known examples where boundary kinetic terms are demanded by supersymmetry , so it remains an unanswered question whether the second tuning will be enforced by any symmetry . let us briefly discuss the possibility of having different kinetic terms on the uv and ir brane , with strengths @xmath80 and @xmath81 . modifying the action ( [ a2 ] ) @xmath82 , \label{adiff}\ ] ] we can follow the procedure outlined in section 2.2 and compute the zero mode wavefunction and cubic and quartic couplings . demanding that the couplings are equal gives a constraint on the allowed @xmath35 values . for a given localization @xmath20 , we can choose , say , @xmath81 and then compute @xmath80 in terms of @xmath20 and @xmath81 . we must also demand that the normalization of the zero mode wavefunction is real ; otherwise the zero mode will be a ghost . this is illustrated in fig . [ fig3 ] for the case @xmath83 . and @xmath81 for the case @xmath83 , indicated by the solid line . the shaded region indicates the @xmath35 values which produce a zero mode ghost , and are thus not allowed . the line @xmath84 is also shown ( dotted line ) , and the @xmath58 and @xmath59 branches are also indicated . ] notice that it is not possible to have two positive kinetic terms ; at least one must be negative . this implies that for high energy processes , locally the kinetic term will appear to be negative , and the theory will have a ghost . at large distances , the nonlocal effects of both brane terms combine to form a perfectly sensible effective field theory , so long as the boundary terms are not so strong and negative that they overwhelm the positive bulk contribution . the shaded region indicates where this happens ; the negative ( and large ) brane kinetic terms produce a zero mode ghostlike gauge field . note also the case @xmath84 studied in the previous section lies in the ghost - free region . the structure of this plot is similar for different values of the localization parameter @xmath20 . as we localize the zero mode towards the ir brane ( increasing @xmath20 ) the asymptote moves towards , but never reaches , the @xmath4-axis . instead localizing the massless mode towards the uv brane ( decreasing @xmath20 ) , causes the asymptote to move further away towards large negative values . the qualitative behavior of the allowed values for @xmath80 and @xmath81 is the same in these cases . in particular , it is never possible to have both brane kinetic terms positive , and there is always a region where the brane kinetic terms produce a zero mode ghost . to localize a massless gauge field in the bulk , we found that it was necessary to add mass terms in the bulk and on the boundary . if we simply add a hard mass term which breaks the 5d gauge symmetry , there is no hope of recovering a 4d gauge symmetry in the effective theory . thus , it is necessary to generate the mass term through spontaneous symmetry breaking , i.e. a higgs mechanism . thus localizing a gauge field in this approach comes at the price of introducing additional scalar fields in the theory . these additional scalars complicate the analysis and obscure the 4d gauge symmetry compared to the case of a massless 5d bulk gauge field , in which the 4d symmetry is realized in a rather simple way . in that case , the scalar mode @xmath85 is the nambu - goldstone boson which is eaten by the massive vector @xmath86 @xcite . with additional scalar fields , the nambu - goldstone boson will be a particular linear combination of scalar fields , while another combination will be a physical scalar degree of freedom . although the analysis is complicated in warped space , we illustrate this feature with a flat space example in section 3.1 . more pressing is the possibility that these scalar modes may be massless , strongly coupled , or even ghostlike . we discuss these issues in section 3.2 using a simple example in warped space with only boundary masses . we find that natural boundary conditions allow a zero mode physical scalar field to exist . to preserve the gauge symmetry a boundary mass term for the vector @xmath2 necessarily contains a scalar kinetic term . when the bulk and boundary mass parameters are tuned to cancel the zero mode mass , we might then expect that the zero mode scalar kinetic term is canceled or even becomes negative , corresponding to a strong coupling problem or a ghostlike state , respectively . although we find no ghost in this example , the physical scalar zero modes are generically strongly coupled . it seems likely that a different choice of boundary conditions ( e.g. dirichlet ) will eliminate the scalar zero mode , but the general analysis is then complicated , as we will discuss . in this section our discussion is limited to the abelian case for simplicity , but of course similar issues will be faced in non - abelian extensions . let us revisit the abelian case , and consider the 5d action for the gauge field @xmath15 with bulk mass @xmath87 and the 5d nambu - goldstone boson @xmath88 which appears after spontaneous symmetry breaking : @xmath89~ , \label{a3}\ ] ] where we have only included quadratic terms in the lagrangian involving @xmath90 and @xmath91 . from a 5d point of view , @xmath91 is the nambu - goldstone boson which gives a mass to the vector @xmath90 . however when we compactify the extra dimension , the modes @xmath85 will play some role in giving the 4d vectors @xmath92 a mass @xmath93 . we might then expect that it is some combination of modes @xmath94 and @xmath95 which form the true nambu - goldstone boson for @xmath92 , while the orthogonal combination is a physical scalar particle . previously we did not take into account the nambu - goldstone boson which couples to @xmath2 at the quadratic level in the action . here we show that it is indeed possible to isolate @xmath2 by defining @xmath96 gauges . by separating the action ( [ a3 ] ) into components @xmath97 we obtain @xmath98~. \label{a4}\end{aligned}\ ] ] we can see that there is nontrivial mixing between @xmath99 , and @xmath91 , and it is not clear which combination of fields is the nambu - goldstone boson . we can remove the mixing in the @xmath2 sector by adding a gauge fixing term @xmath100 ^ 2~. \label{lgf}\ ] ] the bulk equation of motion ( [ eom2 ] ) then follows if we choose the unitary gauge given by @xmath101 . with boundary mass terms there will also be terms involving @xmath91 in the boundary action that mix with @xmath2 . this mixing can be removed in a similar manner by gauge fixing , except that in this case there will be terms proportional to delta - functions . still the vector @xmath2 can be isolated and its equation of motion can be analyzed straightforwardly as in section 2.1 . ideally , we would like to decompose the action ( [ a4 ] ) in order to see how the 5d symmetry manifests itself in the kaluza - klein action ( i.e. identify the nambu - goldstone boson ) , as is done with massless gauge fields @xcite . in warped space , however , it is unclear how to decompose the fields since they obey different equations of motion . the analysis can be performed in flat space , and we do find that the 4d nambu - goldstone boson is a linear combination of @xmath85 and @xmath102 . consider the flat space limit @xmath103 of the action ( [ a4 ] ) : @xmath104~. \label{flataction}\end{aligned}\ ] ] the structure of the action gives us a hint as to how to decompose the fields . consider the following expansion : @xmath105 where @xmath106 and the normalization is determined by @xmath107 the resulting 4d action is diagonal in kaluza - klein modes , but there is still mixing between @xmath108 , and @xmath102 : @xmath109~.\end{aligned}\ ] ] examining the action , it is natural to guess that the nambu - goldstone boson is proportional to the linear combination @xmath110 . however , we must also show that there is a scalar massless eigenstate . the mass mixing matrix @xmath111 has eigenvalues ( @xmath112 , @xmath113 ) and the eigenstates can be written as @xmath114 rewriting the action in terms of scalar mass eigenstates , we find @xmath115 , \label{bulk}\end{aligned}\ ] ] it is clear that @xmath116 is the nambu - goldstone boson . the coefficient of the mixing term between @xmath92 and @xmath116 is exactly the one needed to make the propagator for @xmath92 transverse . the unitary gauge is simply @xmath117 . notice that @xmath118 is a physical scalar degree of freedom , as we would expect from counting the degrees of freedom contained in a 5d massive vector field . if no additional boundary terms are considered , then the low energy theory contains no massless modes . it is possible to obtain massless modes by augmenting the bulk action with boundary potentials which modify the boundary conditions . in flat space the kaluza - klein decomposition is simple since the eigenfunctions are sinusoidal . in warped space , it is not clear that the eigenfunctions of the vector @xmath2 and the physical scalar and nambu - goldstone bosons will be related in a simple manner . in the case with no bulk and boundary masses @xcite , the decomposition proceeds as in ( [ aex ] ) and ( [ a5ex ] ) and the 4d action is simple . besides complicating the general analysis , scalar fields can in principle introduce more severe problems into the theory . to add boundary mass terms for the vector , we must also add kinetic terms for the scalar fields on each brane . the tuning condition ( [ tune ] ) implies that one of these kinetic terms has the wrong sign , associated with negative energy . at low energies this does not necessarily mean that the scalar has negative kinetic energy , since there are contributions from the bulk and the brane to the kinetic term of the scalar field . one must explicitly examine the low energy theory to see if the particle is a ghost . indeed , even though negative kinetic energy terms were added for the gauge field in section 1.2 to maintain 4d gauge invariance , the zero mode did have overall positive energy ( [ a2 ] ) . however this cancellation mechanism was nonlocal so that at high energies there is a ghost locally which eventually needs to be resolved in the uv completion of the theory . another potential problem is that the zero mode scalars might have a small or even vanishing kinetic term due to cancellations from bulk and brane contributions . the kinetic term must then be canonically normalized , leading to strong coupling in any interactions in which the scalar participates . these issues are difficult to examine in the general warped case with bulk and boundary masses because the action ( [ a4 ] ) does not allow for a straightforward kaluza - klein decomposition of the scalar fields . however , there is one warped special case , the `` specular '' solution in which the bulk mass is zero but boundary mass terms localize the gauge field , where we can analyze these scalar issues in detail . we find in this case that although there is no ghostlike particle , there is a strong coupling problem ( i.e. the kinetic term vanishes ) . we give suggestions as to how this problem might be eliminated . a massless gauge field in the bulk of ads@xmath0 is described by the action @xmath119~,\nonumber \\ & = & \int d^5x\left[-\frac{1}{4}f_{\mu\nu}f^{\mu\nu}-\frac{1}{2}e^{-2ky}(\partial_5 a_\mu -\partial_\mu a_5)^2 \right]~.\end{aligned}\ ] ] following the standard procedure @xcite , we expand the 5d fields in the following way : @xmath120 where the eigenfunctions @xmath38 satisfy the differential equation @xmath121 and the orthogonality condition @xmath122 we will focus on the massless mode , @xmath123 . normally the solution is taken to be @xmath124 , but since the differential equation is second order , there are in fact two linearly independent solutions . the general solution is given by @xmath125 we must impose boundary conditions that are consistent with the variational principle . the boundary terms that arise from variation of the action are @xmath126 \bigg\vert^{\pi r}_0~ , \label{b1}\ ] ] which forces us to impose dirichlet or neumann boundary conditions on @xmath127 . these choices force @xmath128 , and the exponential solution is removed . however , it is possible to keep the exponential solution if we add the following boundary mass term to the action : @xmath129~. \label{bmass}\ ] ] we will see that the boundary action is invariant under a 4d gauge symmetry . varying the boundary action yields @xmath130 \bigg\vert^{\pi r}_0~. \label{b2}\end{aligned}\ ] ] combining the terms in ( [ b1 ] ) and ( [ b2 ] ) and using the bulk equation of motion for @xmath1 : @xmath131 we find the natural boundary condition for @xmath38 to be @xmath132 this corresponds to @xmath133 in ( [ bc1 ] ) , which is consistent with ( [ tune ] ) . notice that there are other possibilities for boundary conditions consistent with the variational principle . for instance we can choose dirichlet conditions for @xmath1 . in this case the analysis is not as straightforward and the 4d gauge invariance is obscured , but there will be no @xmath1 zero mode . we will revisit this discussion shortly when we discuss strong coupling . for the zero mode , the exponential solution satisfies the boundary condition ( [ bc3 ] ) and the constant @xmath134 . the zero mode solution is given by @xmath135 note that the zero mode field @xmath136 is not eliminated in this case , but has the same bulk profile as @xmath137 in the fundamental domain @xmath138 . this is consistent because the wavefunction for @xmath139 is @xmath140 and is therefore odd with respect to orbifold symmetry . as mentioned above , the boundary action ( [ bmass ] ) respects a 4d gauge invariance : @xmath141 where , noting ( [ bc3 ] ) , the second relation is equivalent to @xmath142 on the boundary . it is instructive to examine how the _ abelian _ gauge symmetry , which involves no nambu - goldstone fields , arises for the zero mode . to this end , we will detune the condition ( [ bc3 ] ) to be @xmath143 where @xmath144 is now an arbitrary constant . the lowest lying mode becomes massive . the eigenfunctions are given by @xmath145~,\ ] ] where @xmath146 and @xmath147 are constants . the coefficient @xmath147 is found by applying the boundary conditions : @xmath148 and the masses are found by equating @xmath149 . we will now show that there is a smooth limit @xmath150 starting from the detuned case . performing the kaluza - klein reduction , the action becomes @xmath151~ , \nonumber \\ & = & \int d^4x\sum_n\left[-\frac{1}{4}(f^n_{\mu\nu})^2-\frac{1}{2}(\partial_\mu a^n_5)^2-\frac{1}{2}m_n^2(a^n_\mu)^2+m_n a^n_\mu \partial_\mu a^n_5 \right]~. \label{a8}\end{aligned}\ ] ] it is clear from the action that @xmath85 is the nambu - goldstone boson , which provides a correct contribution to the vacuum polarization amplitude of @xmath92 . expanding the gauge parameter @xmath152 , we see that the 4d gauge invariance is given by @xmath153 in the unitary gauge , @xmath23 , the nambu - goldstone @xmath85 is eaten by @xmath92 . now if we take the limit @xmath154 , the zero mode becomes massless @xmath155 . the zero mode action reduces to @xmath156~,\ ] ] and there is an abelian gauge invariance , as is evident from ( [ gs ] ) by taking @xmath123 . it is also clear from ( [ gs ] ) that when @xmath157 that we do not have freedom to set the zero mode @xmath158 . the expansion ( [ exp ] ) indicates that in the limit @xmath150 , the wavefunction for @xmath139 becomes infinite . we might argue that the expansion is not valid in this limit and instead use the following expansion : @xmath159 in this case we would find that although the wavefunction is finite , the kinetic term for @xmath139 vanishes . actually , this implies that as we tune the boundary mass parameter @xmath154 in ( [ bc4 ] ) to get a massless mode , we should properly canonically normalize the kinetic term . the smallness of the kinetic term then reappears as the large wavefunction in ( [ exp ] ) . the large wavefunction in ( [ exp ] ) means that any interactions will become strongly coupled , and therefore the theory is not well - defined if such interactions exist . indeed , any bulk interactions with @xmath90 would have this problem because of 5d gauge invariance . the problem is even more severe because the mode is massless , meaning that the strong coupling problem is present in the low energy theory . however there are no ghosts in the theory at low energies , as can be seen clearly by taking the limit @xmath150 in the action ( [ a8 ] ) . the kinetic term for @xmath1 is always properly normalized with the correct sign corresponding to positive energy . how can we avoid this strong coupling problem ? the most obvious possibility would be to impose a different boundary condition for @xmath1 consistent with the variational principle . a dirichlet condition , @xmath160 is our other option consistent with the variational principle , as can be seen from examining ( [ b2 ] ) . this implies @xmath161 on the boundary and will make the second term in ( [ b2 ] ) vanish . more importantly , this boundary condition will kill the zero mode @xmath162 . at the massless level , we would have an ordinary gauge theory with only the zero mode @xmath77 . however , for the massive modes , the action would no longer be diagonal with respect to kaluza - klein level @xmath163 because the wavefunctions associated with @xmath164 will be altered . in particular the zero mode would mix with these massive modes quadratically and the 4d gauge symmetry would be obscured at high energies . in principle there is no problem with this philosophy , and in fact this is akin to what happens in non - abelian theories when the zero mode interacts with the kaluza - klein modes via cubic and quartic terms , and the gauge invariance is realized in a non - trivial way ( each kaluza - klein mode transforms into other kaluza - klein modes ) . another way the strong coupling problem might be addressed is via quantum corrections . in the @xmath157 limit , the mass of the @xmath162 is not protected by a gauge symmetry in the 4d theory ( at least in the abelian case ) , and thus we would expect the mass to receive large corrections , quadratically dependent on the uv cutoff of the theory . in this way @xmath162 could decouple at low energies , relegating the problem to the uv . on the other hand , the masslessness of @xmath137 is protected by 4d gauge symmetry . we have seen that bulk and boundary masses alter the equations of motion and boundary conditions to permit a localized zero mode non - abelian gauge field . however , this is not the only way to achieve localization . another possibility is to couple the gauge field to a dilaton with a nontrivial background solution , as first illustrated in @xcite . interestingly , it has been suggested that this approach is related to localization via bulk and boundary mass terms , essentially through a field redefinition @xcite . this connection has also been alluded to via deconstruction , in which the wavefunction of the zero mode results from a position dependent gauge coupling varying from site to site @xcite . in this section , we wish to comment on this relation and point out the similarities and differences in each approach . in the end , these other approaches to gauge field localization may provide simpler models , in particular with regard to the scalar sector , which we have seen in our approach is quite difficult to analyze in a satisfactory way . let us begin by reviewing the basic idea considered in @xcite . the theory consists of 5d gravity coupled to a scalar field @xmath165 . the scalar @xmath165 acquires a bounce - like configuration , which can be idealized to a `` brane '' by taking a particular limit in the scalar potential . rs2-like solutions appear for the metric in this brane limit . if we introduce another scalar , the dilaton @xmath166 , it also acquires a nontrivial background , which in the brane limit turns out to be linear in @xmath4 , @xmath167 , where @xmath168 is a dimensionless constant . in fact if we simply begin with a brane , that is , without reference to the field @xmath165 , we still find a linear dilaton solution @xcite . in these cases , however , the dilaton @xmath166 back - reacts on the geometry , which is modified from pure ads@xmath0 . in our discussion below , we will assume the geometry is pure ads@xmath0 to compare with our model , even though we know of no 5d model with a linear dilaton solution and pure ads@xmath0 geometry . following @xcite , we can easily see how the gauge field is localized . consider the following action of a massless gauge field coupled to a dilaton @xmath166 , expanded about its background configuration : @xmath169~,\nonumber \\ & = & \int d^5x~ e^{-2 b k y}\left[-\frac{1}{4}f_{\mu\nu}f^{\mu\nu}-\frac{1}{2}e^{-2 k y } ( \partial_5 a_\mu-\partial_\mu a_5)^2 \right]~. \label{a5}\end{aligned}\ ] ] performing a kaluza - klein decomposition as in ( [ kk ] ) , the solution for the massless mode is a constant , @xmath170 . the zero mode action can then be written as @xmath171~.\ ] ] the profile of the gauge boson with respect to a flat metric is then seen to be @xmath172 . the coupling to the dilaton simply becomes the gauge boson wavefunction when the dilaton acquires a vacuum expectation value ( vev ) . this model can be related to localization with bulk and boundary masses by performing a field redefinition , @xmath173 . the action ( [ a5 ] ) transforms to @xmath174~. \label{a6}\ ] ] it appears that we have broken the 5d gauge symmetry by performing this @xmath4-dependent redefinition , but of course the symmetry is manifested in a different way . gauge transformations on the vector @xmath2 are the same while transformations on @xmath1 become rather unusual . under field redefinition , the gauge parameter transforms to @xmath175 , and we find @xmath176 we can still choose the gauge @xmath23 ) . ] , in which case the action ( [ a6 ] ) can be rewritten as @xmath177\nonumber \\ & & + \int d^4x~\frac{1}{2}b k\ , e^{-2 k y } ( a_\mu)^2 \bigg\vert_0~. \label{a7}\end{aligned}\ ] ] we see that the field redefinition has generated bulk and boundary mass terms . the relation of the coefficients of these mass terms is precisely the one necessary to force a localized massless mode , which can easily be seen by identifying @xmath178 and @xmath179 , in which case the tuning relation ( [ tune ] ) holds . again the wavefunction is given by @xmath180 . notice that in this model @xcite , there is only one brane , so that the wavefunction is normalizable only if @xmath168 is positive . why ca nt we simply start from the action ( [ a6 ] ) , and demand the strange gauge symmetry ( [ gauge ] ) without ever referring to the dilaton @xmath166 ? this would allow us to have a theory with a localized gauge field and _ no _ extra scalar fields . we can do this , but there are a couple of objections to this approach . the action ( [ a6 ] ) explicitly violates 5d lorentz invariance in the bulk . of course the dilatonic action ( [ a5 ] ) appears to violate the symmetry as well , but this is only because the vev of the dilaton spontaneously breaks lorentz invariance , and the symmetry is therefore realized non - linearly through fluctuations about the background . if we do nt have this extra scalar , then the theory breaks 5d lorentz invariance . on the other hand , 4d lorentz invariance is a symmetry of the action ( [ a6 ] ) , which is the only symmetry we need for phenomenological purposes . another objection , of course , would be that the unorthodox gauge symmetry in ( [ gauge ] ) would seem a bit unnatural if we did nt know about its origin from the dilaton coupling . it is a local symmetry nonetheless , indicating a redundancy in the description of the physical degrees of freedom in the system . the pure action ( [ a6 ] ) without a dilaton finds a stronger motivation through deconstruction as shown in @xcite . in this section we will review this approach and mention some of the advantages of the deconstruction viewpoint . consider @xmath44 sites each with an su@xmath181 gauge theory . the sites are connected by bifundamental scalar fields @xmath182 . the 4d lagrangian is then @xmath183 where the covariant derivative is defined as @xmath184 . the gauge symmetry is broken to a diagonal su@xmath185 by assuming that the scalar fields acquire vevs proportional to the unit matrix , @xmath186 . the mass lagrangian is then @xmath187 to reproduce the familiar case of a massless gauge field in ads@xmath0 in the continuum limit , the prescription is to take universal gauge couplings @xmath188 and @xmath4-dependent vevs @xmath189 , where the site `` coordinate '' is @xmath190 . this gives a flat zero mode gauge field . however , if we make different choices for the couplings and vevs , in particular assuming nonuniversal gauge couplings of the form @xmath191 and vevs of the form @xmath192 the continuum limit mass lagrangian becomes @xmath193~.\ ] ] this action can be rewritten identically to ( [ a7 ] ) , containing the precise bulk and boundary mass terms which will localize a massless mode . from the deconstructed point of view , this makes complete sense because the su@xmath185 gauge symmetry is still intact , implying the existence of a massless eigenstate which is a linear combination of the @xmath194 . what is appealing about the deconstruction approach to localizing the gauge field is that there is no reference to a dilaton . the fluctuations about the vacuum @xmath195 are related in the continuum to @xmath1 , and no other scalar field is present in the theory . indeed , after taking into account these fluctuations , the full action is given by ( [ a6 ] ) . the zero mode localization is simply a consequence of different gauge theories @xmath196 having different couplings @xmath197 , which is certainly allowed . the absence of the dilaton is much like the absence of the radion in generic deconstructed theories compared to their full five dimensional counterparts . the approaches to localizing gauge fields described in the previous section seem very similar to our approach . the localized wavefunction can be ascribed to a dilaton coupling in a nontrivial background , or to a position dependent gauge coupling . upon a field redefinition , bulk and boundary masses appear with precisely the right coefficients necessary to localize a zero mode , identical to ( [ mass ] ) . these observations were first made in @xcite . although the technical details of each approach are similar , the philosophy behind each is somewhat different . we are beginning with a pure gauge theory in a slice of ads@xmath0 , and attempting to generate the necessary mass terms through the spontaneous breaking of the symmetry . the 5d bulk lorentz symmetry is never broken . in the other approaches , the localized wavefunction is generated through an auxiliary mechanism , either the dilaton profile or a position dependent coupling , and is crucially dependent on breaking 5d lorentz invariance in the bulk . moreover , it remains to be shown whether there exists a solution to einstein s equations that produces a linear dilaton background with pure ads@xmath0 geometry . the number of additional physical scalar particles in each theory is different as well . in our approach , the number of scalar fields depends on the representation of the symmetry group in which scalar fields reside , and this number could be quite large . the analysis of these scalars is also complicated since there is mixing at the quadratic level in the action , and can in principle lead to strong coupling problems in the theory , as elaborated on in section 3 . on the other hand , in order to localize the field with a dilaton , only one scalar is needed per gauge field @xmath90 . fluctuations about the dilaton background in ( [ a5 ] ) need to be studied , but the analysis should be simpler because these perturbations will not mix with the gauge field at the quadratic level , but only appear as interactions . on the other hand , the deconstructed approach is even simpler because there are no additional scalar fields , although the continuum theory explicitly breaks 5d lorentz symmetry . at the non - abelian level , these approaches really diverge . we found it necessary to add brane - localized kinetic terms to restore the equality of the zero mode gauge couplings . the other approaches accomplish this by having position dependent couplings . to see this , consider the non - abelian generalization of ( [ a5 ] ) . to compare with our approach , we redefine the field @xmath198 to generate bulk and boundary mass terms . upon this redefinition , the cubic and quartic interactions become @xmath199~. \label{int2}\ ] ] we can view the 5d gauge couplings as position dependent : @xmath200 . inserting the zero mode wavefunction @xmath201 and computing the overlap integrals , we find that the 4d cubic and quartic couplings are equal , and thus the low energy theory is gauge invariant . in our approach , the 5d gauge couplings are constant in the bulk . the phenomenology of each localization mechanism would also be different . in our approach , brane kinetic terms must be included , which can drastically alter the mass spectrum , wavefunctions , and couplings to matter , as illustrated by a number of studies @xcite . brane kinetic terms are not necessary in the other approaches , but of course may be included for study . finally , the scalar content of each theory is much different , which would affect the phenomenology . although we have focused on the gauge field @xmath2 in this paper , many models utilize the @xmath1 field in one way or another . one well known example is the hosotani mechanism @xcite , or radiative symmetry breaking , where quantum corrections induce a mass for the zero - mode @xmath139 , which is then identified as the higgs boson . a recent twist on this idea , motivated by holography , are composite higgs models @xcite , where the dual of the @xmath1 is a composite pseudo - nambu - goldstone boson of a strongly coupled theory . the ability to localize the @xmath1 field in these contexts could have interesting effects on the phenomenology of these models . with a massless 5d gauge boson in the bulk , the profile of @xmath139 is fixed by gauge invariance , and in fact is localized near the ir brane . as is the case with the vector @xmath2 , it is possible to localize the scalar @xmath1 zero mode at different points in the bulk . using bulk and boundary masses generated by spontaneous symmetry breaking to localize this field goes against the philosophy of the gauge - higgs models where @xmath1 is to be used as an _ alternative _ to spontaneous symmetry breaking . hence , we will show how the @xmath1 can be localized either through a dilaton coupling , or through deconstruction , similar to the case for @xmath2 in section 4 . we reiterate that at this point it is unknown whether a 5d system with a linear dilaton solution in a pure ads@xmath0 background exists . starting from the action ( [ a5 ] ) , we add the following gauge fixing term to diagonalize the action : @xmath202 ^ 2~. \label{lgf2}\ ] ] the @xmath1 action becomes @xmath203~. \label{a9}\ ] ] decomposing the field , @xmath204 , we find that the kaluza - klein modes have a @xmath205-dependent mass , indicating that they are not physical particles . however for the zero mode this is not the case . the eigenfunction @xmath206 obeys the equation of motion @xmath207 and with the boundary condition @xmath208 has solution @xmath209 inserting the solution into the action ( [ a9 ] ) , we have @xmath210~,\ ] ] indicating that with respect to a flat metric , the wavefunction is @xmath211 . the physical zero mode is localized on the uv brane when @xmath212 while for @xmath213 it is localized on the ir brane . note that this agrees with the zero mode localization of a pure bulk scalar with bulk mass @xmath214 and boundary mass @xmath215 @xcite . zero mode gauge fields in a compact extra dimension can be localized through the addition of bulk and brane mass terms . the inclusion of these terms modifies the bulk equation of motion and the boundary conditions such that a zero mode with an exponential wavefunction exists in the theory . in the non - abelian case , however , the cubic and quartic couplings of this zero mode are not equal , meaning that the gauge symmetry of the 4d theory is lost . we have shown how to recover the 4d gauge symmetry by adding gauge interactions on the boundaries . in this way the sum of the bulk plus boundary contributions to the interactions of the zero mode restores the equality of the gauge couplings , and then the 4d gauge invariance appears in its usual form . we can add these brane interactions in a gauge invariant way via brane - localized kinetic terms . one caveat in this approach is that there are two relations in this theory that must be tuned , one between the bulk and boundary masses to produce a localized massless mode wavefunction , and another between kinetic terms and the boundary mass parameter . furthermore even though the zero mode had overall positive kinetic energy , via a nonlocal cancellation mechanism , negative kinetic energy terms were required on at least one brane . these terms would need to be resolved in any sensible uv completion of the model . there are also caveats related to the additional scalar fields that must be present in the theory . to add mass terms in the bulk and on the boundary in a gauge invariant way , the terms must be generated by breaking the 5d gauge symmetry , which necessitates introducing additional scalar fields in the theory . in the warped case , a complete analysis of these extra scalars is limited because of the nontrivial mixing . we have instead presented this analysis in flat space , where the eigenfunctions of the different fields are similar . if one could extend this analysis to the warped case , many of the scalar issues , such as strong coupling , could be addressed more generally . we have shown , however , that in the warped case with only boundary masses , a strong coupling problem may exist in the scalar sector . this problem may be resolved by a different choice of boundary conditions or by radiative corrections to the scalar zero mode mass . even if the problems of the scalar sector can not be addressed without a more fundamental uv description , the yang - mills localization will still be preserved at low energies . the phenomenology of localized gauge fields in the randall - sundrum model would therefore be interesting to explore . if we consider the standard model in the bulk , then each gauge boson can in principle be localized at different points in the extra dimension . gauge fields in the bulk were originally considered in @xcite , and it was found that the mass of the first kaluza - klein excitation must be pushed to a rather high scale to suppress four - fermion operators . however , localizing gauge fields can help lower the kaluza - klein mass scale , as illustrated in the abelian case @xcite . it is conceivable that localizing gauge fields could have other phenomenological benefits . also , the phenomenology would be different depending on the localization mechanism , either bulk and boundary masses , dilaton coupling , or deconstruction . perhaps the most interesting aspect of localized gauge fields is the holographic interpretation @xcite . if the zero mode is localized on the uv brane , we find that the massless particle in the dual theory is primarily an elementary field external to the cft . on the other hand , a zero mode localized on the ir brane corresponds in the dual theory to a composite massless gauge field built out of cft states . for example , if we consider the standard model gauge fields propagating in the bulk with zero modes localized towards the ir brane , then in the dual theory the forces that we experience - electromagnetic , weak , and strong - are really emergent phenomena , a result of strongly coupled ( unknown ) cft dynamics at low energy . this is similar to the emergent gravity scenario considered in ref . @xcite , and adding boundary kinetic terms for gravity as done for non - abelian gauge fields in this paper , suggests a similar nonlinear extension for gravity . finally , we discussed the localization of the @xmath1 field , the scalar component of the 5d gauge field . we have shown that it is possible to localize this field anywhere in the bulk with a dilaton coupling or via deconstruction . this opens up new phenomenological possibilities in models where @xmath1 is the pseudo - nambu - goldstone boson responsible for electroweak symmetry breaking . in particular , motivated by the dual interpretation , the dual higgs field can become even more composite as the zero mode is localized on the ir brane . we thank emilian dudas and erich poppitz for helpful discussions . t.g . acknowledges the cern theory division and the galileo galilei institute for theoretical physics in florence for their warm hospitality and the infn for partial support where portions of this work were done . this work was supported in part by a department of energy grant de - fg02 - 94er40823 at the university of minnesota , and an award from research corporation . l. randall and r. sundrum , phys . lett . * 83 * , 3370 ( 1999 ) [ arxiv : hep - ph/9905221 ] . j. m. maldacena , adv . * 2 * ( 1998 ) 231 [ int . j. theor . * 38 * ( 1999 ) 1113 ] [ arxiv : hep - th/9711200 ] . s. s. gubser , i. r. klebanov and a. m. polyakov , phys . b * 428 * ( 1998 ) 105 [ arxiv : hep - th/9802109 ] . e. witten , adv . * 2 * ( 1998 ) 253 [ arxiv : hep - th/9802150 ] . n. arkani - hamed , m. porrati and l. randall , jhep * 0108 * ( 2001 ) 017 [ arxiv : hep - th/0012148 ] . r. rattazzi and a. zaffaroni , jhep * 0104 * ( 2001 ) 021 [ arxiv : hep - th/0012248 ] . m. perez - victoria , jhep * 0105 * ( 2001 ) 064 [ arxiv : hep - th/0105048 ] . k. agashe and a. delgado , phys . d * 67 * , 046003 ( 2003 ) [ arxiv : hep - th/0209212 ] . h. davoudiasl , j. l. hewett and t. g. rizzo , phys . b * 473 * , 43 ( 2000 ) [ arxiv : hep - ph/9911262 ] . a. pomarol , phys . b * 486 * , 153 ( 2000 ) [ arxiv : hep - ph/9911294 ] . t. gherghetta , arxiv : hep - ph/0601213 . k. ghoroku and a. nakamura , phys . d * 65 * , 084017 ( 2002 ) [ arxiv : hep - th/0106145 ] . i. kogan , s. mouslopoulos , a. papazoglou , and g. ross , nucl . b * 615 * , ( 191 2001 ) [ arxiv : hep - ph/0107307 ] . b. batell and t. gherghetta , phys . d * 73 * , 045016 ( 2006 ) [ arxiv : hep - ph/0512356 ] . a. kehagias and k. tamvakis , phys . b * 504 * , 38 ( 2001 ) [ arxiv : hep - th/0010112 ] . h. abe , t. kobayashi , n. maru , and k. yoshioka phys . d * 67 * , 045019 ( 2003 ) [ arxiv : hep - ph/0205344 ] . y. hosotani , phys . b * 126 * , 309 ( 1983 ) ; phys . b * 129 * , 193 ( 1983 ) . s , j. huber and q. shafi , phys . d * 63 * , 045010 ( 2001 ) [ arxiv : hep - ph/0005286 ] . t. gherghetta and a. pomarol , nucl . b * 586 * , 141 ( 2000 ) [ arxiv : hep - ph/0003129 ] . h. davoudiasl , j.l . hewett , and t.g . rizzo phys . d * 68 * , 045002 ( 2003 ) [ arxiv : hep - ph/0212279 ] . f. del aguila , m. perez - victoria , and j. santiago , jhep * 0303 * ( 2003 ) 051 [ arxiv : hep - th/0302023 ] . l. randall and m. schwartz , jhep * 0111 * , 003 ( 2001 ) [ arxiv : hep - th/0108114 ] . m. tachibana , [ arxiv : hep - th/0108164 ] . m. carena , e. ponton , t. tait , and c.e.m . wagner phys . d * 67 * , 096006 ( 2003 ) [ arxiv : hep - ph/0212307 ] . r. contino , y. nomura and a. pomarol , nucl . b * 671 * , 148 ( 2003 ) [ arxiv : hep - ph/0306259 ] . k. agashe , r. contino , and a. pomarol , nucl . b * 719 * , 165 ( 2005 ) [ arxiv : hep - ph/0412089 ] . t. gherghetta , m. peloso and e. poppitz , phys . rev . d * 72 * , 104003 ( 2005 ) [ arxiv : hep - th/0507245 ] .

the rotation of the disk of the large magellanic cloud ( lmc ) is derived from the radial velocities of 422 carbon stars ( kunkel , irwin , & demers 1997 , a&as , 122 , 463 ) . new aspects of this analysis include the propagation of uncertainties in the lmc proper motion with a monte carlo , and a self - consistent modeling of the rotation curve and disk kinematics . the lmc rotation curve reaches a maximum circular velocity of 72 7 km s at kpc and then declines . the rotation curve is well fit by a truncated , finite - thickness exponential disk model with no dark halo , implying a total mass of 5.3 1.0 . the velocity dispersion in concentric radial bins from to 5.6 kpc decreases from 22 to 15 km s , then increases to km s at larger radii . constant - thickness disk models in virial equilibrium can not be reconciled with the data even if the effects of lmc or galactic dark halos are included . if the disk is virialized , the scale height rises from = 0.3 to 1.6 kpc over the range of = 0.5 to 5.6 kpc . thus the lmc disk is flared . we model the velocity dispersion at large radii ( 6 kpc ) as a maximal flared disk under the influence of the galactic dark halo , which favors a mean density for the latter of pc at the lmc distance . lmc stellar kinematics play an important role in elucidating the nature of machos , a dark population inferred from lmc microlensing . we have constructed a truncated and flared maximal disk model for the lmc which is kinematically based . our model does not include a nonvirialized component such as tidal debris . the instantaneous probability of microlensing from lmc stars in our model is , where is the disk inclination . our upper limit on the self - lensing optical depth is in good agreement with that obtained from less sophisticated models , and is an order of magnitude too small to account for the macho microlensing signal .

astro-ph0006018

if the galactic dark halo were composed entirely of massive compact halo objects ( machos ) , the instantaneous probablity of microlensing ( the `` optical depth '' ) towards the large magellanic cloud ( lmc ) would have been @xmath15 ( pacznski 1986 ) . the most recent observational result for machos with masses up to about 1 @xmath5 is @xmath16 , which comes from an analysis of 5.7 years of lmc microlensing survey data ( alcock et al . thus an all - macho galactic dark halo is ruled out . however , the observed optical depth is significantly higher than the estimate for known galactic and lmc stellar populations , @xmath17 ( alcock et al . the excess lmc microlensing signal may be telling us the macho fraction of the galactic dark halo or revealing gaps in our understanding of the essential structure of the lmc and galaxy . critical discussions of the microlensing optical depth from known stellar populations have developed along two lines . one debate has been about the possible existence of a `` new '' stellar population which would account for the lmc microlensing result . some example suggestions include an intervening dwarf galaxy or tidal stream ( zhao 1998 ; zaritsky & lin 1997 , zaritsky et al . 1999 ) , and a very warped galactic disk ( evans et al . 1998 ) . these hypotheses have been tested , and in some cases ruled out ( alcock et al . 1997 ; beaulieu & sackett 1998 ; bennett 1998 ; gould 1998 , 1999 ; gyuk , flynn , & evans 1999 ; ibata , germaint , & beaulieu 1999 ; johnston 1998 ) . the second debate has been about the importance of lmc star - star `` self - lensing '' ( sahu 1994 ; wu 1994 ) . in particular , although the lmc is well represented by an exponential disk model which often serves as the basis for the self - lensing optical depth calculation , it may exhibit some important detailed structures . our paper is motivated by this lmc self - lensing problem . the line - of - sight velocity dispersion of lmc stars yields a strong constraint on star - star self - lensing from the virialized lmc disk : @xmath18 ( gould 1995 ) . gould s elegant limit is probably uncertain by no more than a factor of @xmath72 . however , several lmc models have been devised which would increase the self - lensing optical depth over the gould limit . one example of a detailed structure that might increase the self - lensing optical depth is a highly - inclined and flared lmc disk ( zhao 1999 ) . it has also been suggested that the oldest lmc disk stars have a very large characteristic scale height , or that the lmc harbors an as yet unseen but massive stellar spheroid 2 - 4 for the former , and a very large m / l for the latter ) , and the density profiles . ] ( e.g. aubourg et al . 1999 ; salati et al . 1999 ; evans & kerins 2000 ) . finally , it is possible that a nonvirialized stellar component ( i.e. , a shroud of tidal debris ; weinberg 2000 ) acts to increase the self - lensing optical depth ( zhao 1998 ) . kinematic studies play a critical role in testing these various proposed lmc models . for example , the flare of the lmc disk may be inferred from its velocity dispersion at different radii , and the existence of a non - equilibrium stellar component lying near the lmc might be proven with a large kinematic survey ( zhao 1999 ; graff et al . 1999 ; see also zaritsky et al . 1999 ; ibata et al . in addition , one could search directly for an lmc stellar population with a velocity dispersion of @xmath750 km s@xmath1 , the prediction for a spheroid or very thick disk . these latter components must be identified before their importance to the self - lensing optical depth can be assessed . the lmc rotation curve provides a framework for detailed kinematic studies such as these . the decomposition of the lmc rotation curve into disk and dark halo components has several implications for microlensing . first , as emphasized by gyuk et al . ( 1999 ) , the total mass of the lmc is an important constraint on self - lensing optical depth calculations . a high - mass model will typically predict a high self - lensing optical depth , with an additional dependence on whether the mass lies mostly in the disk or in a halo . unfortunately , recent analyses of the lmc rotation curve have lead to mass estimates that range over a factor of @xmath74 , which can be attributed primarily to different assumptions about a dark halo ( kim et al . 1998 , kunkel et al . 1997 , schommer et al . second , a massive lmc dark halo might significantly affect the lmc disk kinematics ( e.g. bahcall 1984 ) , with possible consequence for constraints on self - lensing or determining the flare of the disk . the influence of an lmc dark halo on the lmc disk kinematics has not been investigated in detail . we note that the interplay between the disk and dark halo is also of general interest for studies of galaxy formation and evolution . finally , if the lmc has a dark halo with a physical makeup similar to that of the galactic dark halo ( i.e. composed partly of machos ) , the macho fraction of the galactic dark halo implied by the observed optical depth would be lowered ( alcock et al . 2000 ) . motivated by the above considerations , we present a new analysis of the lmc rotation curve . in this work we are particularly concerned with the flare of the lmc disk . however , other issues pertinent to microlensing are discussed as our calculations and analyses permit . we refer extensively to the rotation curve recently presented by kim et al . ( 1998 ) and the impressive radial velocity dataset for lmc carbon stars summarized by kunkel , demers , irwin & albert ( 1997 ; hereafter kdia ) . a significant subset of these latter data are public ( kunkel , irwin & demers 1997 ) , and archived electronically at the cds . we also refer extensively to the analysis of the lmc space motion presented by kroupa & bastian ( 1997 ) . our paper is organized as follows . in 2 , we present the impetus for this work , a comparison of constant - thickness exponential disk models ( with no dark halos ) to the lmc disk velocity dispersions reported by kdia . in 3 , we reanalyze the lmc rotation curve and velocity dispersion curve using archived carbon - star radial - velocity data . in 4 , we present a multi - mass component kinematic model for the lmc . in 5 , we compare our carbon - star rotating disk solution to our model . we discuss the microlensing implications of our analyses in 6 and conclude in 7 . disks in galaxies often have exponential surface brightness profiles from which we infer exponential mass surface density profiles of the form @xmath19 by assuming constant mass - to - light ratios ( @xmath20 ) . in equation ( 1 ) , @xmath21 is the central mass surface density , @xmath22 is the radial scale length , and @xmath9 is a cylindrical radial coordinate . for an infinitely thin disk described by equation ( 1 ) , freeman ( 1970 ) calculated that the rotation curve reaches a maximum circular velocity at @xmath23 @xmath24 where @xmath25 is the total mass of the disk , which is related to the central surface density by @xmath26 for the case of a disk with a modest finite thickness , @xmath27 will be decreased by @xmath28 ( van der kruit & searle 1982 ) . a useful model with a finite thickness is the exponential disk ( also known as a double - exponential disk ) , which has a density distribution @xmath29 in equation ( 4 ) , the variable @xmath8 is the vertical , or what we will sometimes refer to as the `` z '' scale height . if we choose the spatial density normalization @xmath30 , the exponential disk projects to the surface density of the infinitely thin freeman disk , i.e. equation ( 1 ) . the z - velocity dispersion , @xmath31 , for the exponential disk is given by wainscoat , freeman and hyland ( 1989 ) , @xmath32^{1/2}\ ] ] which assumes that the disk is virialized . equation ( 5 ) is strictly valid for a radially infinite disk . the radial dependence ( i.e. , the @xmath33 term ) in equation ( 5 ) scales the local density normalization , as prescribed by others ( van der kruit & searle 1982 ; van der kruit & freeman 1984 ) . a model extensively discussed by van der kruit & searle ( 1981 , 1981b , 1982 ) is the isothermal disk . the vertical distribution of stars in an isothermal disk is described by the sech@xmath34 function , as originally derived by spitzer ( 1942 ; see also 4.2 of this paper ) . the spatial density for this disk model is , @xmath35 which projects to the surface density of the infinitely thin freeman disk if @xmath36 . assuming virialization , the z - velocity dispersion of the isothermal disk is proportional to the z - scale height and independent of z : @xmath37^{1/2}\ ] ] where the radial dependence scales the local density normalization . equation ( 7 ) is strictly valid for a radially infinite disk . we note that the `` z - scale height '' in the isothermal disk is not strictly a scale height ( and not the same as @xmath8 in the double - exponential model ) , but it is a similar parameter . further comparisons of the exponential and isothermal disk models are found in wainscoat , freeman & hyland ( 1989 ) , and van der kruit ( 1988 ) . the systematic error in the velocity dispersion predicted by the radially infinite isothermal disk model has been shown to be only @xmath710% in the range @xmath38 using numerical studies of truncated isothermal disk models ( van der kruit & searle 1982 ) . we will use the isothermal disk model in our analyses , because these numerical calculations provide us with an estimate of the model accuracy , and will later allow us to approximately account for the finite radial extent of the lmc disk ( see 4.3 of this paper ) . observations of edge - on spirals indicate that both the exponential and isothermal models are viable representations of real disks ( de grijs , peletier , & van der kruit 1997 ) . the specific choice of model does not affect the main results of this work . it is reasonable to compare the model z - velocity dispersion to the observed velocity dispersion at different projected radii in the lmc for the case of uniform anisotropy ( @xmath39 constant at different radii ; van der kruit & freeman 1984 ) . although we do not know if this is true for the lmc , the assumption of uniform anisotropy is supported by kinematic studies of the old galactic disk ( lewis & freeman 1989 ) . gould ( 1995 ) notes that if @xmath40 1 as found elsewhere , the line of sight velocity dispersion measured for an inclined disk overestimates @xmath41 . therefore , our model z - velocity dispersion is likely an upper - limit on the true z - velocity dispersion . combining equations ( 2 ) , ( 3 ) and ( 7 ) yields @xmath42 which is most accurate for disks in the range @xmath38 . the z - scale height has been observed to be constant over a large range of @xmath9 in numerous disk galaxies ( van der kruit & freeman 1984 ; de grijs , peletier , & van der kruit 1997 ) , implying that @xmath43 decreases with increasing @xmath9 in these galaxies . indeed , this trend of decreasing velocity dispersion has been observed in over a dozen spiral / disk galaxies ( bottema 1993 ) , including our own . in summary , if the disk velocity dispersions are the same at different radii , the disk is likely flared ( e.g. zhao 1999 ) however , this inference assumes that ( 1 ) the disk is virialized , ( 2 ) the radial scale length is constant , ( 3 ) the disk velocity dispersions exhibit uniform anisotropy , and ( 4 ) the galaxy has no dark halo ( or that the dark halo has a negligible affect on the disk velocity dispersions ) . finally , one must be careful to compare the observed velocity dispersions to the simple model described above over a restricted range of radii , because real galactic disks are finite in radial extent . we compare to the lmc in the next section . in this section , we will adopt a plausible value for the product @xmath44 in equation ( 8) , and compare that model to lmc disk velocity dispersion data . following kdia and many others , we adopt a distance to the lmc of 50.1 kpc , and an inclination of @xmath14 = 33 degrees . for comparison , wide - field uv polarimetric image data have yielded a precise but model - dependent estimate of the inclination of the lmc disk : @xmath45 degrees ( cole , wood , & nordsieck 1999 ) . westerlund ( 1997 ) summarizes a number of other inclination measurements . we adopt @xmath46 kpc for the radial scale length of the lmc disk . the surface brightness profile of the lmc disk is well - fit by an exponential with a scale length @xmath47 kpc ( de vaucouleurs 1957 , bothun & thompson 1988 ) . moreover , the surface density profile of intermediate - age long period variables ( lpvs ) in the lmc is also fit by an exponential with a scale length of @xmath48 kpc ( hughes , wood & reid 1991 ) . this result directly associates our kinematic dataset with an lmc population that follows the @xmath22 = 1.6 kpc exponential profile , because some of the intermediate - age lpvs are also carbon stars . finally , the rr lyrae stars in the lmc also appear to lie in a @xmath47 kpc exponential disk ( alcock et al . since the rr lyraes are presumably much older than the carbon stars , the carbon stars were likely born into the same disk . we note that no stellar population in the lmc has yet been shown to be inconsistent with a @xmath46 kpc exponential profile . we assume that the carbon stars observed by kdia properly represent this lmc disk . the kdia dataset consists of radial velocity measurements for 759 carbon stars spanning a true lmc radius of 2 to 10 kpc . kdia summarize the results of 11 different zonal solutions for a rotating disk . in figure 1 , we plot the kdia velocity dispersions as a function of radial distance , which we take as the central value for each zonal solution given in kdia s table 1 . we show data only for @xmath9 = 3 to 6 kpc . this is the range of radii where equation ( 8) will most accurately predict the velocity dispersions . the dashed line in figure 1 is the mean of the plotted , observed velocity dispersions , @xmath49 13.7 km s@xmath1 . assuming a constant value of @xmath43 equal to the mean , we find @xmath50 = 0.4 , consistent with the data . the solid line plotted in figure 1 shows the prediction of equation ( 8) assuming values of @xmath51 km s@xmath1 and @xmath8 = 0.35 kpc at @xmath52 . this model predicts @xmath53 km s@xmath1 at @xmath54 kpc . the fit for this model distribution is @xmath50 = 8.7 , which can be rejected with high significance . models with larger scale heights intersecting the data at larger @xmath9 give similarly poor fits are not strictly apppropriate because the zonal solutions given by kdia are not all independent . however , subsets of the @xmath55 measurements which are independent also give poor fits to the distribution given by equation ( 7 ) . ] . one possible interpretation of the observed constancy of @xmath43 with @xmath9 is that the z - scale height increases as @xmath56 , i.e. _ the lmc disk is flared . _ an alternate interpretation of figure 1 might be a constant - thickness disk under the dynamical influence of a dark halo ( bahcall 1984 ) . indeed , kim et al . ( 1998 ) compared their rotation curve with the kdia carbon star rotation curve and , on this basis , argued that an lmc dark halo is dynamically significant at a radius of @xmath74 kpc . however , kdia favored a different interpretation ; they attributed the rising outer portion of their carbon star rotation curve to tidal effects , and not a dark halo curve in figure 1 , supporting this aspect of our analysis in 2 . ] . before concluding that the lmc disk is flared , it is worth testing disk plus dark halo kinematic models in a self - consistent manner . with this model comparison in mind , let us consider the kdia analysis in greater detail . we have several specific concerns with the kdia analysis . first , and perhaps most importantly , kdia excluded some carbon stars from their solutions for their association with a polar ring . this procedure might have artificially lowered the disk velocity dispersions . by retaining these stars in our solutions , we guarantee that our results will be properly comparable to our model ( which will not include a polar ring ) . kdia also corrected for the lmc transverse motion by forcing the position angle of the kinematic line of nodes to match the photometric line of nodes in each zonal solution . this is a questionable procedure . moreover , it is inconsistent with the analysis by kim et al . ( we will correct the carbon - star velocities for the transverse motion of the lmc in a manner consistent with the rotation curve analysis by kim et al . in fact , we will make our carbon star rotation curve analysis similar in other ways to the rotation curve analysis ( i.e. , by adopting the same kinematic center ) , in order to lend maximum credibility to detailed comparisons of the two rotation curves . we assemble positions ( right ascension and declination ; equinox 1950 ) and galactocentric radial velocities for magellanic cloud carbon stars from kunkel , irwin & demers ( 1997 ) . we discard all small magellanic cloud ( smc ) carbon stars , `` inter - cloud '' carbon stars ( their table 15 ) , and carbon stars located near the center of the lmc ( their table 17 ; see also 3.3 ) . we also discard one rogue carbon star ( c0433@xmath576607 ) for its highly discrepant velocity . the resulting homogeneous dataset contains 422 carbon stars , representing a significant subset of the data analyzed in kdia . ongoing observational campaigns to obtain carbon star radial velocities will likely increase the sample size in the coming years ( suntzeff 1998 ) . the archival dataset we have assembled is sufficient for the purposes of this study . feast , thackeray & wesselink ( 1961 ) have discussed the apparent rotation induced by the transverse motion of the lmc . each of our 422 carbon star velocities are corrected for this projected radial velocity gradient as follows . we adopt the lmc space motion calculated by kroupa & bastian ( 1997 ) , which is derived from an average of the lmc proper motion they measure with _ hipparcos _ data and the lmc proper motion measured by jones , klemola , & lin ( 1994 ) . the _ hipparcos _ and jones et al . ( 1994 ) measurements of the lmc proper motion agree within their respective error bars . we note that kim et al . ( 1998 ) make their transverse motion correction using the proper motion measurement by jones et al . ( 1994 ) . it is useful to establish a galactic coordinate system for the vector analysis that follows . following kroupa & bastian ( 1997 ) , the galactocentric coordinates used here are such that the galactic center and the sun are located at 0 and @xmath58 kpc along the @xmath59-axis , respectively . the positive @xmath60-axis points toward the north galactic pole , and the positive @xmath61-axis points in the direction of galactic rotation at the position of the sun . in these coordinates , the position vectors of the sun and lmc are @xmath62 @xmath63 in units of kpc . the velocity vector of the lmc is @xmath64 in units of km s@xmath1 ( kroupa & bastian 1997 ) . the galactocentric radial velocity of the lmc is the vector dot product @xmath65 km s@xmath1 , where @xmath66 is a normalized unit vector . the vector between the sun and lmc is @xmath67 , and the radial velocity component of the lmc seen from the local standard of rest ( accounting for the sun s peculiar velocity and galactic rotation ) at the position of the sun is @xmath68 km s@xmath1 . the vector @xmath69 points toward the `` center '' of the lmc , which is assumed to be at @xmath70 , @xmath71 ( kroupa & bastian 1997 ) . the unit vector connecting the sun to any position in the sky in galactocentric coordinates is @xmath72 the transverse motion velocity correction that we seek is the difference between the projected radial velocity toward the lmc center and the projected radial velocity toward each carbon star . if we define the vector @xmath73 , the apparent rotation induced by the lmc space motion is @xmath74 . we make our correction by subtracting @xmath75 from the observed carbon star velocities . small differences between the kroupa & bastian ( 1997 ) model used here for the transverse motion radial velocity correction and assumptions ( i.e. , for the lmc center and distance ) in subsequent sections have a negligible effect on our results . first , we convert each carbon star s right ascension and declination into spherical coordinates in units of radians . we designate these @xmath76 and @xmath77 , respectively . we adopt the center of the lmc at right ascension @xmath78 and declination @xmath79 ( equinox 2000 ) following kim et al . we designate the lmc center with @xmath80 and @xmath81 ( also in units of radians and precessed to equinox 1950 ) . the distribution of carbon stars is shown in figure 2 ( where north is up and east is to the right ) . the smc is located well outside of the boundaries of the figure in a south - westerly direction ( toward the lower left ) . we remind that @xmath76 and @xmath77 are spherical coordinates plotted on a rectilinear scale . therefore , the distribution of carbon stars shows a @xmath7cos(@xmath77 ) distortion . our solution proceeds as follows . we calculate the position angle ( @xmath82 ) and radial distance from the center of the lmc ( @xmath83 in units of radians ) for each carbon star using the formulae @xmath84\ ] ] @xmath85\ ] ] @xmath86 equation ( 15 ) assigns the correct sign and phase to the position angle depending on which quadrant each star resides in figure 2 . we have defined @xmath82 to increase west of north . we deproject each radial distance on the sky to true lmc radius ( @xmath9 in units of kpc ) using , @xmath87^{1/2 } \ \right ) \right|\ ] ] we take the absolute value of the inverse tangent in equation ( 16 ) to guarantee a positive value for @xmath9 . we accommodate a twisting disk model by introducing @xmath88 . the parameter @xmath88 allows for a changing kinematic line of nodes in different radial zone solutions . we note that @xmath88 is sensitive to the adopted lmc transverse velocity ( meatheringham et al . we define our different zones by making cuts in true radius . each zonal solution is derived by fitting the function @xmath89 to the distribution of galactocentric radial velocities , @xmath90^{1/2 } } \ + \ v_{sys } \right)\ ] ] we must choose the correct sign for @xmath89 at the appropriate position angle , which is the meaning of the ` @xmath0 ' in equation ( 17 ) . we choose ` @xmath91 ' for @xmath92 = @xmath93 to @xmath94 , and ` @xmath57 ' for @xmath92 = @xmath94 to @xmath95 . the fitted model could be generalized to account for a warped and decentered disk by letting @xmath14 and ( @xmath80 , @xmath81 ) vary in each zonal solution . however , we fix @xmath14 , @xmath80 , and @xmath81 to the values given above in all solutions . we derive solutions in four radial zones which span @xmath9 = 2.5 - 5 , 5 - 6 , 6 - 7 , and 7 - 13 kpc . these were chosen to provide an adequate number of stars and phase coverage in each zone . the inner - most and outer - most boundaries were chosen to be inclusive of the entire dataset when @xmath88 = 0 . we begin by solving for @xmath96 , @xmath88 , and @xmath97 that yield a minimum dispersion about @xmath89 in each of our zones . we average the values of @xmath96 , excluding the outermost zone ( which was sometimes poorly constrained ) . in a second round of fitting , we fix @xmath96 to the three - zone average value , and search only for values of @xmath88 and @xmath97 that minimize the velocity dispersion . this procedure was easily reproduced , and facilitated our bootstrap and monte carlo error analyses ( described below ) . the mean radii in each final zonal solution are @xmath98 = 4.0 , 5.6 , 6.5 , and 7.8 kpc . the minimum velocity dispersions occurred at @xmath88 = @xmath5737 , @xmath5722 , @xmath578 , and 3 degrees . we find circular velocities @xmath97 = 72 , 68 , 65 , and 65 km s@xmath1 and velocity dispersions @xmath43 = 17.7 , 14.9 , 19.3 and 20.6 km s@xmath1 , respectively ( in order of increasing radius ) . the four zonal solutions are shown in figure 3 . we indicate the carbon stars with small open circles . each carbon star is plotted twice at @xmath82 and ( @xmath82 + @xmath99 ) . the best fit curve is plotted as a solid line . the mean radius , circular velocity , and velocity dispersion are also labeled in each panel . our first estimate of the errors is made with a bootstrap resampling analysis ( barrow , sonada , & bhavsar 1984 ) . we create 250 artificial datasets by randomly removing 1 - 10% of the carbon stars from our full dataset , and then find new rotating disk solutions for each of these . in a bootstrap reasampling analysis , the variances in the mean values of interest averaged over all of the artificial datasets are taken to represent the internal variances of the full dataset . we find errors in @xmath97 of 1.3 , 1.7 , 2.3 , and 2.4 km s@xmath1 for the four radial zones ( in order of increasing radii ) . we find an error in @xmath43 of 0.3 , 0.2 , 0.4 , and 0.5 km s@xmath1 , and errors in @xmath88 of 1.1 , 1.0 , 2.4 , and 1.5 degrees , respectively . our second calculation is designed to estimate the errors associated with our choice of zonal boundaries and the movement of stars between zones due to the changing @xmath88 parameters . for each of the four zones , we find new solutions for 64 different variations in the zonal boundaries . we varied each boundary by no more than @xmath0 0.25 kpc , which yielded roughly comparable numbers of stars and mean radii in each zonal solution ( except for the outer boundary of the outer - most zone which was varied by 1.5 kpc in steps of 0.25 kpc ) . the variances of @xmath97 , @xmath43 , and @xmath88 in the zonal boundary solutions are similar to those found with the bootstrap analysis . moreover , we find no systematic trends with boundary choice , except for the outermost zone . in the outermost zone , @xmath97 decreases systematically ( from @xmath760 to 30 km s@xmath1 ) as the inner boundary is varied from @xmath76.7 to 7.2 kpc . finally , we perform a monte carlo calculation to propagate the error on the lmc space motion to the rotation curve . we create 500 artificial datasets , where the lmc space motions are drawn randomly from a sample that reproduces the best fit values and errors in equation ( 11 ) . we assume gaussian distributions for the space motion errors . this calculation yields errors in @xmath97 of 7.1 , 8.7 , 10.4 and 15.8 km s@xmath1 , errors in @xmath43 of 0.1 , 0.2 , 0.3 and 0.4 km s@xmath1 , and errors in @xmath88 of 6.6 , 4.6 , 7.6 and 8.4 degrees , in the four zones ( in order of increasing radius ) . similar values were found with as few as @xmath7250 artificial datasets , indicating that our calculation is robust . for most of the parameters in our rotating disk solution , the error from the lmc space motion dominates . our final solutions and adopted errors are summarized in table 1 . we note that there would be no stars in common between the different solutions if all zones had the same @xmath88 , but in our final solutions , @xmath75% of the stars are located in two zones ( see table 1 ) . however , since we have accounted for the movement of small numbers of stars across the zonal boundaries in our error analysis , we may consider each zonal solution as independent in subsequent analyses . we previously discarded 22 radial velocity measurements for carbon stars very near the center of the lmc ( kunkel , irwin & demers 1997 ; their table 17 ) because this sample of stars did not span an adequate range of @xmath82 to yield an interesting zonal rotation curve solution . however , the velocity dispersion of these carbon stars is of considerable interest . this sample of stars at @xmath100 kpc has a velocity dispersion of @xmath101 km s@xmath1 . guided by our error analyses above , we adopt an error of 1 km s@xmath1 . the average velocity dispersion of carbon stars in our five zones is @xmath43 = 18.9 km s@xmath1 . we defer further discussion and interpretation of the lmc rotation curve and disk velocity dispersions until 5 , after we present our theoretical kinematic model . we will consider two representations of the contribution of the luminous mass in the lmc disk to the rotation curve . the first is derived empirically by kim et al . ( 1998 ) from the @xmath9-band surface brightness data of de vaucouleurs ( 1958 ) . for two assumptions of the disk mass - to - light ratio , we present this stellar rotation curve in table 2 . the second representation of the disk rotation curve we will consider is the finite - thickness , truncated disk model of van der kruit and searle ( 1982 ) . we adopt their model no . 2 ( truncated at @xmath9 = 4@xmath22 ) from their appendix a.1 . table 3 summarizes this disk model rotation curve , which is given in units of the radial scale length ( @xmath22 ) and the equivalent maximum circular velocity for an infinitely thin exponential disk ( @xmath27 ; see 2.1 of this paper ) . this model will also be used to calculate a correction to the scale heights inferred from the disk velocity dispersions at small radii in the lmc ( see 4.3 of this paper ) . when comparing either of these two disk rotation curves to our carbon star data points , we will add in quadrature the small additional contribution of gas in the lmc disk to the rotational velocity . kim et al . ( 1998 ) calculated the gas rotation curve using the single - dish data of luks & rolfs ( 1992 ) , and assuming a 30% contribution from . the gas rotation curve is summarized in table 4 . finally , we will account for a pseudo - isothermal dark halo in the traditional manner , assuming a density profile , @xmath102 where the circular velocity is a function of the integrated total mass @xmath103\ ] ] the psuedo - isothermal dark halo is parameterized by a central density , @xmath104 , and core radius , @xmath105 . we add the circular velocities of the disk ( @xmath106 ) and psuedo - isothermal halo ( @xmath107 ) in quadrature in order to calculate the rotation curve of our model : @xmath108 . the effect of a spherical dark halo on the velocity dispersion of an embedded disk has been extensively discussed by bahcall ( 1984 ) , bahcall & casertano ( 1984 ) and bottema ( 1993 ) . the vertical motions of stars in a disk and halo system are governed by poisson s equation @xmath109 where @xmath110 and the first moment of the boltzmann equation @xmath111 the effective halo density , @xmath112 , includes the contribution from the radial deriviative of the total circular velocity , @xmath97 . if the rotation curve is fairly flat , we may discard the radial deriviative term in equation ( 22 ) and equate @xmath113 . this formulation of the problem ( bahcall 1984 ) neglects the @xmath114 cross terms in poisson s equation . with the following redefinitions ( bahcall 1984 ) @xmath115 and @xmath116 we rewrite the equations governing the vertical motions of the disk stars as @xmath117 with the boundary conditions @xmath118 the solution for @xmath119 = 0 was first published by spitzer ( 1942 ) : @xmath120 the solution to equation ( 25 ) for @xmath121 is @xmath122 the differential equation ( 25 ) must be solved numerically for intermediate values of @xmath119 . the @xmath119 parameter define in equation ( 24 ) is the ratio of effective halo density to the disk density in the plane of the disk . it is particularly useful parameter because it relates ratio of the velocity dispersion in a disk plus halo system , @xmath123 , to the velocity dispersion in a disk with no halo , @xmath124 . following bottema ( 1993 ) , and suppressing the subscript @xmath60 for clarity , @xmath125 where @xmath76 is the surface density @xmath126 in the limit of @xmath121 , @xmath127 we have solved for @xmath128 as a function of @xmath119 numerically ; these values are listed in table 5 . in the limit of large @xmath119 , these values confirm equation ( 31 ) . the error in @xmath76 is @xmath730% at @xmath129 and less than 20% at @xmath130 , in the sense that equation ( 31 ) overestimates the numerically calculated value of @xmath76 . finally , we note that the density profile of the dark halo may be derived from the observed velocity dispersion in the disk at large @xmath9 , if the disk scale height is constant . equating @xmath113 and @xmath131 , and combining equations ( 6 ) , ( 7 ) , ( 24 ) , ( 29 ) , and ( 31 ) gives @xmath132 where the approximate density profile of the dark halo ( at large @xmath9 ) follows from equation ( 18 ) . thus , if a galactic disk is embedded in a dark halo , one expects @xmath133 at large @xmath9 with the slope proportional to @xmath134 . the velocity dispersion in the lmc bar will yield an accurate estimate of the scale height only if we make a numerical correction to account for the finite extent of the lmc . we note that applying a pure disk model at small radii in the lmc is reasonable because the lmc has no bulge , and the surface brightness profile ( even in the bar ) is purely exponential ( bothun & thompson 1988 ) . we appeal to the numerical integrations by van der kruit & searle ( 1982 ) to estimate a `` correction factor '' for the disk velocity dispersions given by equation ( 7 ) . in table a1 of van der kruit & searle ( 1982 ) , model no . 2 , the ratio of the actual z - force ( @xmath135 ) to the model z - force is 0.55 , 0.82 , 0.95 , 1.07 , 1.06 , 0.64 , and 0.77 at @xmath136 = 0 , 1 , 2 , 3 , 4 , 5 and 6 , respectively . we have simply averaged over the tabulated values of @xmath137 ( various other weighted - average schemes give similar values ) . the correction to the model velocity dispersons is the square root of the z - force ratios ( van der kruit & searle 1982 ) , and is @xmath725% near the lmc center . the scale height inferred from the velocity dispersion in the lmc bar would be underestimated by a factor of @xmath72 without the correction . here we summarize the theoretical calculations presented thus far . in 2.1 , we introduced the basic formulae that describe an exponential disk with no dark halo , and a prediction for the velocity dispersion perpendicular to the plane of the disk . the latter calculation was made by assuming a radially infinite disk . for a disk truncated at @xmath136 = 4 or 5 , the error in this predicted velocity dispersion is less than 10% at radii of a few scale lengths ( van der kruit & searle 1982 ) . in 4.1 , we described the rotation curve of an exponential disk embedded in a psuedo - isothermal dark halo . we presented a stellar rotation curve derived from lmc surface brightness data , and a gas rotation curve , both from kim et al . ( 1998 ) . we presented a model rotation curve for a finite - thickness disk truncated at @xmath136 = 4 , from van der kruit & searle ( 1982 ) . in 4.2 , we calculated the velocity dispersion perpendicular to the plane of a disk that is embedded in a psuedo - isothermal dark halo . at large radii , the dynamical influence of the dark halo can dominate . in this limit , we showed that the disk velocity dispersion can be predicted by an analytic formula . we estimated the error of this approximation using numerical integrations . in 4.3 , we give a numerical correction to the velocity dispersion perpendicular to the plane of the disk ( the prediction from 2.1 ) that accounts for truncation of the disk at @xmath136 = 4 , which is adopted from van der kruit and searle ( 1982 ) . although the lmc disk may not be truncated at precisely four radial scale lengths , by using the same model throughout this work , our analysis is self - consistent . in figure 4 , we present the lmc rotation curve and a `` maximal '' disk decomposition . our four zonal solutions are indicated with bold dots and error bars . data points on the rotation curve from kim et al . ( 1998 ) are shown with bold cross symbols . for comparison , we plot the carbon star zonal solutions of kdia with open circles . also shown in figure 4 are the gas and stellar disk rotation curves . we assume m / l = 2.2 for the stellar rotation curve , which is similar to that found for the galactic disk ( bahcall , flynn & gould 1992 ) . disk mass - to - light ratios of @xmath72 are favored by dynamical stability arguments ( bottema 1993 ) . finally , we plot the sum of the gas and stellar disk rotation curves as a bold solid line . our four zonal solutions and the solutions of kdia agree within our respective errors . the model rotation curve and our data points show quite good agreement . in fact , although the overall scaling of the model is a free parameter ( the m / l ratio of the stellar disk ) , the variation with radius is reproduced remarkably well . the disagreement between the model and data points would appear less severe if error bars similar to those on the carbon star data points were appropriate . if the lower envelope of the carbon star error bars were more representative of the true lmc rotation curve , then one could adopt a slightly smaller m / l ratio and the agreement with the data points would improve . however , since the data points appear to deviate from a smooth curve , most notably the dip at @xmath138 3 kpc , it is possible that all of the data points lying below the model curve represent a significant failure of the model . we speculate that a bar in the lmc may be responsible , although we have not attempted any modeling of this effect . in figure 5 , we show the same carbon star and data points as in figure 4 . we now replace the stellar rotation curve derived from surface brightness data with the finite - thickness , truncated model disk rotation curve . the gas rotation curve , and the sum of the gas and model disk rotation curves are also plotted . the model curve assumes a scale length of 1.6 kpc and an infinitely thin disk equivalent maximum circular velocity . a finite - thickness disk has a maximum circular velocity approximately 5% smaller than an equivalent infinitely thin disk ( see table 3 ) . in this decomposition , we adopt the finite - thickness disk model , but account for the contribution from gas ( approximately 5% near maximum ) , and quote here the maximum circular velocity for an equivalent infinitely thin disk . ] @xmath27 = 71 km s@xmath1 . the model disk rotation curve is clearly consistent with the data , and fits the run of and carbon star data points even better than the ( semi - empirical ) stellar rotation curve shown in figure 4 . the small differences between the model disk rotation curve , the stellar disk rotation curve , and the data points as illustrated in figures 4 and 5 are not critical to our conclusions . we estimate the mass of the lmc disk using @xmath27 = 71 km s@xmath1 and the formulae in 2.1 , which yields @xmath139 @xmath3 @xmath140 . the error is estimated by adopting the uncertainty of the maximum observed circular velocity in our carbon star solutions ( @xmath07 km sec@xmath1 ; see table 1 ) for the uncertainty of @xmath27 . this maximal disk model has a surface density normalization of @xmath141 @xmath5 pc@xmath142 . kim et al . ( 1998 ) estimate the total mass of gas in the lmc to be 0.5 @xmath3 @xmath140 . thus , the total mass of the lmc is @xmath143 @xmath3 @xmath140 . in figure 6 , we present a `` minimal '' disk decomposition of the lmc rotation curve . we plot the same carbon star and data points as in figures 4 and 5 . we plot the stellar rotation curve assuming m / l = 1 , the gas rotation curve , and the sum of the stellar and gas rotation curves . we also plot the contribution of a pseudo - isothermal dark halo , and the sum of the disk and halo rotation curves . in this decomposition , we calculate @xmath144 @xmath3 @xmath140 and @xmath145 @xmath5 pc@xmath142 . the shallow and declining run of carbon star data points favors a small core radius for the lmc dark halo . the dark halo shown has @xmath105 = 1 kpc and @xmath104 = 0.10 @xmath5 pc@xmath12 . by assuming m l = 1 in the disk , this decomposition has a maximal halo , and yields a total lmc mass of @xmath75 @xmath3 @xmath140 ( corresponding to an upper limit on the lmc global mass - to - light ratio of @xmath74 ) . in figure 7 , we plot our lmc disk velocity dispersions as a function of true radius with bold dots and error bars . we plot the results of kdia with open circles ( see also fig . 1 ) . the prediction of equation ( 8) for our maximal disk model is shown as a dotted line and labeled . we plot this model prediction again , but now corrected for the truncation of the lmc disk ( the `` @xmath146 force correction '' ) as a bold solid line . we have assumed a constant scale height of @xmath147 kpc , which normalizes the latter curve to intersect the carbon star data point in the lmc bar . a constant - thickness disk model can not be reconciled with the data by any choice of @xmath8 . for our minimal disk decomposition , we must account for the effect of the lmc dark halo on the disk velocity dispersions . by combining equations ( 6 ) , ( 18 ) , and ( 24 ) , and equating @xmath148 , we may estimate the @xmath119 parameter , @xmath149 assuming a constant scale height of @xmath150 kpc , the spatial density normalization of the minimal disk is @xmath151 = 0.068 @xmath5 pc@xmath12 . adopting the parameters of the maximal lmc dark halo from above , we find @xmath152 1.1 , 1.0 , 1.2 , 1.9 and 3.4 at integer steps of @xmath136 = 1 to 5 . for these values of @xmath119 , the error associated with equation ( 31 ) is too large , and thus we appeal directly to our numerical integrations in table 5 . for this decomposition , we begin with the velocity dispersion according to equation ( 8) , and then make the @xmath153 force correction . we calculate @xmath154 using equation ( 33 ) , and use spline interpolations of the numerical data in table 5 to calculate @xmath76 . the model disk velocity dispersions are corrected according to equation ( 29 ) by a factor of @xmath128 ( equating @xmath155 ) . these various curves are plotted in figure 8 . our choice of @xmath8 = 0.4 kpc normalizes the dark halo - corrected curve to intersect the data point in the lmc bar . a constant - thickness disk in the presence of a maximal dark halo can not be reconciled with the observed velocity dispersions by any choice of @xmath8 . for comparison , a fit to the velocity dispersion data with the minimal disk and an arbitrary pseudo - isothermal dark halo implies a maximum circular velocity in the rotation curve of @xmath7500 km s@xmath1 , which is about 7 times higher than observed . our maximal disk model yields @xmath8 = 0.25 , 0.93 , and 1.62 kpc at @xmath9 = 0.5 , 4.0 , and 5.6 kpc , respectively . for comparison , the minimal disk yields @xmath8 = 0.55 , 1.62 , and 2.61 kpc at the same radii . at larger radii ( @xmath10 6 kpc ) , where the disk velocity dispersions begin to rise , the implied scale heights for the maximal disk model are @xmath8 = 5.1 and 28.4 kpc ( at @xmath9 = 6.5 and 8.2 kpc , respectively ) . the scale height inferred from the last data point is much larger than the tidal radius of the lmc ( weinberg 2000 ) , and obviously wrong . we will return to the interpretation of these high velocity dispersions at large radii in a moment . a linear regression on the three data points with @xmath156 kpc yields : @xmath157 where @xmath9 is in units of kpc , and @xmath55 is in units of km s@xmath1 . we adopt this regression to represent the flare of the lmc disk . the change of scale height is represented by a function of the form @xmath158 where @xmath8 and @xmath9 are in units of kpc . we find @xmath159 , and @xmath160 kpc . equation ( 35 ) predicts the scale height to within 0.1 kpc of the values estimated above . assuming a constant mass - to - light ratio , our flared disk model must project to a surface density that varies as @xmath33 in order to reproduce the observed surface brightness profile of the lmc ( de vaucouleurs 1957 , bothun & thompson 1988 ) . therefore , the spatial density of the flared disk , @xmath161 , must change with an effective radial scale length that compensates for the variation of scale height with @xmath9 . this effective radial scale length is easily calculated by considering @xmath162 where we require that @xmath163 . substituting @xmath164 = 1.4 yields @xmath165 = 0.583 ( note that @xmath165 and @xmath164 are dimensionless scale factors ) . it also follows that @xmath166 = @xmath167 = 1.064 @xmath5 pc@xmath12 . in summary , our flared disk model has the following spatial density , @xmath168 where @xmath169 is given in equation ( 35 ) . the dynamical influence of the galactic dark halo is calculated following the discussion in 4.2 of this paper ( see also aubourg et al . we designate the mean density of the galactic dark halo at the distance of the lmc as @xmath170 . if we substitute @xmath170 for @xmath171 in equation ( 20 ) , then @xmath119 from equation ( 24 ) yields @xmath76 from table 5 , and thus the correction to the disk velocity dispersions through equation ( 29 ) . we estimate @xmath170 following griest ( 1991 ; see his eqn . [ 3 ] ) . we assume that the galactic dark halo has a core radius of 3 kpc , and a solar neighborhood spatial density normalization of 0.0079 @xmath5 pc@xmath12 . we adopt the distance from the sun to the galactic center of 8.5 kpc , and the distance from the sun to the lmc of 50.1 kpc , which yields @xmath170 = 0.00025 @xmath5 pc@xmath12 . we estimate @xmath119 = 0.001 , 0.007 , 0.040 , 0.224 and 1.246 at integer steps of @xmath136 = 1 to 5 . in figure 9 , we plot the same data points as in figures 7 and 8 . we indicate the regression of equation ( 34 ) as a solid line , and this regression corrected for the effect of the galactic dark halo as a bold solid line . although the fit is not perfect , the agreement with the last data point is quite good . thus the influence of the galactic dark halo on our flared disk model is of the correct magnitude to account for the outer - disk velocity dispersions . in addition , the radius where the galactic dark halo begins to have a significant dynamical effect ( i.e. , where the velocity dispersions begin to rise in the disk ) is also reasonably reproduced . a more detailed modeling of the outer - disk velocity dispersions is beyond the scope of this paper . we find that @xmath119 is quite small at all radii of interest for the constant - thickness maximal disk model ( accounting for the galactic dark halo but no lmc dark halo ) , and for the constant - thickness minimal disk model ( accounting for both the galactic and lmc dark halos ) . in summary , constant - thickness exponential disk models can not be reconciled with the observed run of lmc disk velocity dispersions . as discussed above , we adopted the mean density of the galactic dark halo at a distance of 50 kpc of @xmath170 = 0.00025 @xmath5 pc@xmath12 , or @xmath172 ( griest 1991 ) . it is worth considering the range of allowed galactic dark halo profiles . assuming flat rotation curves at large radii , many authors have made galaxy models with pseudo - isothermal density profiles ( e.g. caldwell & ostriker 1981 ; bahcall , schmidt , & soneira 1982 ) . typical core radii range from 2 to 8 kpc . the average value of the density at 50 kpc predicted by these models is @xmath173 , with a standard deviation of 0.2 dex . as a final check , we compare the n - body cold dark matter simulation of the galactic dark halo presented by dubinski ( 1994 ) . the initial density profile for this dark halo model was similar to an ellipsoidal hernquist potential , which is like the universal dark halo profile found in dark matter - dominated galaxies ( kravstov et al . dubinski s ( 1994 ) initial density profile was then compressed by the galactic potential ( e.g. blumenthal et al . 1986 ) and in its final form predicts @xmath172 . the density profile of the galactic dark halo assumed for machos ( alcock et al . 2000 ) is also pseudo - isothermal ; thus our analysis of the lmc disk kinematics supports this dark halo profile over the range of interest ( @xmath174 to 50 kpc ) . it has been suggested that the lmc is embedded in a `` shroud '' of tidal debris ( weinberg 2000 ) . such debris is unlikely to contribute significantly to the lmc self - lensing optical depth because too little mass is involved , and it is likely to be located close to the disk ( weinberg 2000 ; see 6.2 ) . nonetheless , it is worth considering the possibility that our sample of true disk carbon stars is contaminated by tidal debris , which might reconcile a constant - thickness disk model with the disk kinematic data . in this case , tidal debris would also affect our interpretation of the outer - disk velocity dispersions . the notion of tidal debris surrounding the lmc is somewhat similar to the suggestion by kdia that the lmc harbors a polar ring , also of tidal origin . ( we note that kdia analysed a larger sample of carbon stars , in which they claim the kinematic signature of a polar ring is evident . ) we have thus far presented an interpretation of the radial velocity data for 422 carbon stars in the lmc that does not require a polar ring , or tidal debris . the nature of tidal debris ( in a polar ring or otherwise ) is investigated as follows . if we assume that a constant - thickness maximal disk model represents the true lmc disk , the `` excess '' velocity dispersion in each zonal solution may be calculated by subtracting the model contribution from the observed dispersions in quadrature , which yields @xmath175 = 15.0 , 13.7 , 18.8 and 20.5 km s@xmath1 at @xmath9 = 4.0 , 5.6 , 6.4 and 8.2 kpc , respectively . next , if we assume that the contaminating debris has a uniform velocity dispersion of @xmath750 km s@xmath1 , the ratio of contaminating stars to total stars in each zonal solution would be of order @xmath710 - 20% . ( this ratio is inversely proportional to the velocity dispersion assumed for the tidal debris . ) in each of our zones , we estimate the number of @xmath750 km s@xmath1 tidal debris stars would be 12 , 17 , 10 , and 8 ( in order of increasing radius ) , if the lmc disk were of constant thickness . we calculate that our four disk zones subtend relative sky areas of 0.40/0.16/0.16/1.00 , in order of increasing zone radius . a regression of the relative numbers of tidal debris stars with the relative zone areas shows an anti - correlation , significant at the @xmath71@xmath43 level . assuming that the true disk and tidal debris carbon stars are similarly affected by incompleteness in the data , this is not what we naively expect from the tidal debris scenario described by weinberg ( 2000 ) . the two outermost zones are illustrative . they have area ratios that differ by a factor of @xmath76 , yet the estimated numbers of contaminating tidal debris stars are 10 and 8 . therefore , the contaminating debris must have a non - uniform radial distribution of surface density , velocity dispersion , or both . it is also possible , but even more contrived , that a spatially - dependent incompleteness in the carbon star dataset has conspired with a non - uniform spatial distribution of tidal debris to yield the observed velocity dispersions . in light of this analysis , we prefer an interpretation of the lmc rotation curve and disk velocity dispersions that invokes a negligible contamination from tidal debris ( i.e. , our maximal flared disk model ) . the unambiguous detection of nonvirialized lmc stars and the accurate characterization of their spatial distribution are clearly desirable . observations of such a population / structure would be relevant to our interpretation of the lmc rotation curve and disk kinematics , with possible additional implications for lmc microlensing ( e.g. zhao 1999 ; graff et al . 1999 ; weinberg 2000 ) . here we calculate the lmc self - lensing optical depth of our maximal flared disk model by directly integrating the spatial density of stars given in equation ( 37 ) . we begin with an integral of the form @xmath176 which yields the optical depth to any source star in the lmc at position @xmath177 , a cylindrical coordinate system in the plane of the disk . the position angle @xmath178 is defined to be zero at the near side if the inclined disk . the integration variable is the line of sight distance between the source and lens . we use the subscripts `` @xmath179 '' and `` @xmath180 '' to denote the coordinates of the sources and lenses , respectively . the approximation in equation ( 38 ) is that @xmath61 is always much smaller than the distance to the source ( @xmath181 50.1 kpc ) , which is reasonable for lmc self - lensing . note that @xmath61 is simply related to @xmath182 , the distance from the observer to the lens , and @xmath183 , the distance from the observer to the source @xmath184 where @xmath14 is the inclination angle of the disk ( measured from the plane of the sky ) . for completeness , we provide the following geometric identity relating the source and lens positions to @xmath61 : @xmath185 it is necessary to integrate over the line of sight a second time , in order to calculate a density - weighted average of @xmath186 over the distribution of source stars @xmath187 which may be properly compared to the observed optical depth . the integration over @xmath183 is made in practice by transforming to a variable in our lmc cylindrical coordinate system ( i.e. , @xmath188 ) . we truncate the lmc disk at a conservative radius of 10 kpc . our results are insensitive to the choice of truncation radius at the few percent level . we calculate @xmath189 for multiple lines of sight , designated by ( @xmath190 ) in the plane of the disk . our calculation is the case of `` pure '' self - lensing ( gould 1995 ) , which yields an upper limit . the effects of dust obscuration and nonlensing mass in the plane of the disk will lower the real value of the optical depth for a fixed total mass of the lmc . we have calculated the self - lensing optical depths in each of the 30 survey fields in the 5.7-year lmc microlensing analysis by alcock et al . ( 2000 ; see also gyuk et al . 1999 ) . in the central - most fields , we find @xmath191 , while in the fields lying at the largest true radii , we find @xmath192 . the field - averaged optical depth for lmc disk self - lensing is : @xmath193 . the lmc inclination dominates the uncertainty in @xmath194 . first , we note that the inclination is probably not a serious concern for the rotation curve and disk kinematics because the shape of the @xmath89 function ( eqn . [ 15 ] ) is fairly insensitive to @xmath14 over the range of plausible values . however , gould ( 1995 ) and gyuk et al . ( 1999 ) show that , to first order , @xmath195 is proportional to the square of the mass - weighted velocity dispersion and an inclination factor of sec@xmath196 . therefore , we simply rescale our calculation of @xmath194 for different inclinations , @xmath197 which should be quite accurate for the lmc inclinations typically found . as noted in 2.2 , cole et al . ( 1999 ) find @xmath45 . thus their @xmath72@xmath43 upper limit , @xmath198 , yields @xmath199 . for comparison , bothun & thompson ( 1988 ) find @xmath200 , which corresponds to @xmath201 . the contribution to the uncertainty in @xmath194 from the velocity dispersion follows from gould s ( 1995 ) formula and our equation ( 34 ) ; it is of order @xmath202 , or a negligible 0.03% . finally , the @xmath710% uncertainty in @xmath203 yields a @xmath720% uncertainty on the total lmc mass , but this corresponds to a small uncertainty in the factor of @xmath204 in equation ( 42 ) . the self - lensing optical depth on the near and far sides of the minor axis at the distances spanned by the 30 fields in alcock et al . ( 2000 ) varies by @xmath75% due to the inclination of the disk ( e.g. gould 1994 ) . thus the flaring of the lmc disk has only a minor effect on the spatial distribution of the self - lensing optical depth , which is a potential diagnostic of the lens population ( e.g. alcock et al . the estimate of a @xmath75% minor axis asymmetry in the optical depth would increase for larger adopted values of the inclination . gyuk et al . ( 1999 ) recently reviewed different self - lensing calculations in the literature . to first order , self - lensing from the virialized lmc disk is proportional to the following combinations of derivable quantities ( gyuk et al . 1999 ) , @xmath205 our model accounts for `` second order '' effects , such as the finite radial extent and flare of the lmc disk . we have shown that uncertainties in the lmc proper motion contribute a @xmath71% uncertainty to @xmath43 and a @xmath720% uncertainty to @xmath25 ( via @xmath203 ) , assuming a truncated and flared maximal disk model . assuming this model , our numerical calculation of the self - lensing optical depth ( eqn . [ 42 ] ) is appropriate . the maximal disk model is consistent with lmc dynamics . however , if a model of the lmc with a dark halo is preferred , the halo will probably contribute to lmc self - lensing in addition to the disk . in this case , our equation ( 42 ) is not appropriate ( see gyuk et al . we have adopted a minimal lmc disk , and provided an example maximal dark halo which is consistent with the rotation curve . measurements of the lmc disk mass to light ratio would better constrain the minimal lmc disk model ( our model is extreme ) . for the more complicated case of an lmc with a dark halo , constraints on self - lensing are weaker . this is due partly to the uncertainty that the lmc proper motion contributes to the rotation curve , which constrains the total lmc mass . the lmc self - lensing optical depth may depend systematically on the age of the stars whose kinematics are studied . historically , it has been difficult to prove that old lmc stellar populations have different velocity dispersions . for example , the ch stars in the lmc were originally thought to trace the elusive lmc halo population , by analogy with ch stars in the galaxy ( hartwick & cowley 1988 , cowley & hartwick 1991 ) . however , the ch stars in the lmc are now believed to represent a much younger population ( suntzeff et al . we note that our average carbon star velocity dispersion ( 18.9 km s@xmath1 ) is similar to that found for the ch stars ( 20 km s@xmath1 ) . according to aubourg et al . ( 1999 ) and salati et al . ( 1999 ) , the lmc could harbor an old population with a large velocity dispersion . these authors invoke wielen s ( 1977 ) age - velocity dispersion calibration from the solar neighborhood . we note that weilen s calibration was derived for stars younger than @xmath73 gyr , and extrapolation beyond this age may not be appropriate . moreover , weinberg ( 2000 ) predicts that the velocity dispersions in the lmc disk , under the influence of the galactic tidal field , will remain constant ( or decrease ) over time . extant observational studies have not yielded a clear picture of the variation of velocity dispersion with age in the lmc . for example , the old globular clusters in the lmc , with ages of @xmath712 gyrs ( olsen et al . 1998 ) , exhibit a velocity dispersion of @xmath721 - 24 km s@xmath1 ( schommer et al . 1992 ; see also freeman , illingworth , & oemler 1982 ) . thus , the lmc carbon stars and ancient clusters support the hypothesis of a constant disk velocity dispersion ( for stellar populations older than @xmath73 gyr ) , in agreement with weinberg s ( 2000 ) prediction . however , the velocity dispersion of old lpvs in the lmc , with ages of @xmath78 gyr ( hughes , wood , & reid 1991 ; olszweski , sunzteff , & mateo 1996 ) , is @xmath735 km s@xmath1 ( hughes et al . further observational studies are needed , particularly given the provocative old lpv result . last , we suggest that a shroud of tidal debris or a polar ring , if they exist , would probably make a negligible contribution to the lmc self - lensing optical depth because the total amount of mass involved is small . weinberg ( 2000 ) reached a similar conclusion . the rotation of the disk of the lmc has been derived from the radial velocities of 422 carbon stars ( kunkel , irwin , & demers 1997 ) . we have propogated the uncertainty in the lmc space motion to the lmc rotation curve with a monte carlo calculation . the associated disk velocity dispersions found in each rotating disk zonal solution are less sensitive to systematic uncertainties from the lmc space motion than the circular velocities obtained . we note that our carbon star rotation curve is derived in a manner consistent with the rotation curve analysis by kim et al . ( 1998 ) ; thus our respective results are properly comparable . we have fit our carbon star rotation curve and the rotation curve with a truncated , maximal lmc disk model , yielding a total lmc mass of @xmath143 @xmath3 @xmath140 . we also conclude that the disk of the lmc is flared . extrapolating the flare inferred at small radii ( where tidal perturbations are small ) to the outer - disk , and accounting for the influence of the galactic dark halo , we are able to approximately reproduce the observed disk kinematics . this model favors an isothermal density profile for the galactic dark halo out to a distance of 50 kpc . our truncated and flared maximal disk model yields a limit on the spatially - averaged lmc self - lensing optical depth of @xmath206 . for plausible values of the lmc inclination , this low self - lensing rate compared to the measured microlensing rate allows for the existence of a dark lens population in the galactic halo ( alcock et al . 2000 ) . finally , we caution that we have not included in our analysis ( 1 ) the dynamical effects of an lmc bar , ( 2 ) large - scale non - circular motions , ( 3 ) non - uniform anisotropy of the disk velocity dispersions , or ( 4 ) arbitrary spatial distributions of tidal debris ( i.e. , a polar ring or stars out of virial equilibrium ) . despite these shortcomings , our truncated and flared maximal disk model successfully accounts for the general dynamical characteristics of the lmc , lending inferences from the model high weight . d.r.a . acknowledges howard e. bond for support of work performed at the space telescope science institute , and kem h. cook for support of work performed at the lawrence livermore national laboratory . nasa research grant nag5 - 6821 `` uv , visible , and gravitational astrophysics research and analysis '' and doe contract w7405-eng-48 are recognized . c.a.n . acknowledges support from national physical science consortium graduate student fellowship . we thank ken freeman , david bennett , tim axelrod , howard bond , kailash sahu , geza gyuk , greg bothun , and the anonymous referee for their helpful discussions and comments . lccccclccccc 2.5 - 5.0 & 91 & 4.0 & @xmath207 & 72 @xmath0 7.1 & 17.6 @xmath0 0.3 5.0 - 6.0 & 139 & 5.6 & @xmath208 & 68 @xmath0 8.7 & 14.9 @xmath0 0.2 6.0 - 7.0 & 122 & 6.4 & @xmath209 & 65 @xmath0 10.4 & 19.3 @xmath0 0.4 7.0 - 13.0 & 73 & 8.2 & 3 & 65 @xmath0 15.8 & 20.6 @xmath0 0.5 ccc|ccc 0.5 & 14.0 & 30.8 & 5.5 & 28.6 & 62.7 1.0 & 23.5 & 51.7 & 6.0 & 28.3 & 62.1 1.5 & 28.6 & 62.7 & 6.5 & 28.0 & 61.5 2.0 & 30.8 & 67.7 & 7.0 & 28.0 & 61.5 2.5 & 31.4 & 68.9 & 7.5 & 28.0 & 61.5 3.0 & 31.4 & 68.9 & 8.0 & 27.7 & 60.9 3.5 & 31.4 & 68.9 & 8.5 & 27.4 & 60.3 4.0 & 30.8 & 67.7 & 9.0 & 27.4 & 60.3 4.5 & 30.2 & 66.4 & 9.5 & 27.4 & 60.3 5.0 & 29.1 & 64.0 & 10.0 & 27.2 & 59.7 cc|cc 0.5 & 5.0 & 5.5 & 22.8 1.0 & 9.0 & 6.0 & 22.5 1.5 & 10.0 & 6.5 & 21.0 2.0 & 12.0 & 7.0 & 20.0 2.5 & 16.0 & 7.5 & 19.0 3.0 & 22.0 & 8.0 & 18.5 3.5 & 24.2 & 8.5 & 17.0 4.0 & 23.8 & 9.0 & 16.0 4.5 & 23.0 & 9.5 & 15.0 5.0 & 23.0 & 10.0 & 14.5 ll|ll 0.000 & 1.000 & 6.000 & 0.339 0.250 & 0.865 & 7.000 & 0.317 0.500 & 0.776 & 8.000 & 0.298 1.000 & 0.658 & 9.000 & 0.283 1.500 & 0.582 & 10.000 & 0.269 2.000 & 0.528 & 12.000 & 0.248 2.500 & 0.486 & 15.000 & 0.223 3.000 & 0.453 & 20.000 & 0.194 4.000 & 0.404 & 30.000 & 0.160 5.000 & 0.367 & 50.000 & 0.124

using the previously gained insight about the particle / field relation in conformal quantum field theories which required interactions to be related to the existence of particle - like states associated with fields of anomalous scaling dimensions , we set out to construct a classification theory for the spectra of anomalous dimensions . starting from the old observations on conformal superselection sectors related to the anomalous dimensions via the phases which appear in the spectral decomposition of the center of the conformal covering group we explore the possibility of a timelike braiding structure consistent with the timelike ordering which refines and explains the central decomposition . we regard this as a preparatory step in a new construction attempt of interacting conformal quantum field theories in d=4 spacetime dimensions . other ideas of constructions based on the- or the perturbative sym approach in their relation to the present idea are briefly mentioned .

hep-th0005134

it had been known for a long time that conformal quantum field theory exhibits in addition to the general spin - statistics theorem another more characteristic structural property which we will refer to as the `` anomalous dimension - central phase '' connection . it relates the anomalous scale dimension of fields modulo integers ( semi - integers in the case of fermion fields ) to the phase obtained by performing one complete timelike sweep around the compactified minkowski world @xcite and hence is analogous to the univalence superselection rule of the semi - integer spin which historically spells the beginning of the issue of superselection rules in the famous paper of wick , wightman and wigner @xcite . in the spin case one associates with a @xmath3 spatial rotation sweep the statistics phase @xmath4 of the spin - statistics connection @xcite and the question of whether there is a commutation relation behind the superselected coherent subspace corresponding to each phase factor of the timelike sweep poses itself naturally . the word `` central '' here refers to the center @xmath5(@xmath6 ) of the infinite sheeted covering group @xmath6 which has one abelian generator @xmath5 for spacetime dimensions @xmath7 in chiral conformal theories the sweep is lightlike and the spin and its timelike analogue coalesce , whereas in higher dimensions there is the problem of consistency with the well known dhr superselection theories of internal symmetries based on spacelike commutativity . it is our aim to show that the analogy is deeper than expected at first sight , namely the statistics aspect of the spin has an algebraic counterpart in form of a timelike braid group ( plektonic ) commutation relation . here the notion of global causality in the covering of the compactified minkowski spacetime is important because it was on the basis of this concept that the `` einstein causality paradox '' @xcite was solved @xcite . the conformal decomposition theory resulted from the attempt to avoid the covering formalism ( which is not natural from a particle physics viewpoint ) and to deal instead with projected fields which behave like sections over compactified minkowski spacetime rather than globally causal fields on the covering . whereas the latter are wightman fields , the former are not since they carry with them a source and a range projector , and hence are similar to the exchange algebra fields of chiral theories @xcite . as in the chiral case the spectrum of anomalous dimensions ( possibly modulo a common abelian contribution ) is determined in terms of the admissable braid group representations . the chiral observables on @xmath8 correspond to conformal observables on compactified minkowski space @xmath9 . for the latter one has space- and timelike commutativity ( validity of huygens principle ) , in fact this distinction and the notion of causality altogether becomes meaningless and only lightlike distances have an invariant meaning . in d=1 + 1 dimensions one has accumulated a good understanding of conformal theories and in particular of their associated superselected charge structure . one knows that they can be decomposed into the @xmath10 chiral light cone components . there is a systematic way to classify localizable representation of chiral observable algebras ( at least in principle ) and one finds charge - carrying fields which obey a lightlike exchange algebra @xcite@xcite in which those new objects satisfy braid group commutation relations either of the abelian kind ( anyonic ) or with plektonic ( which includes the nonabelian case ) r - matrices with quantized statistical phases . since the latter determine the spectrum of anomalous dimensions ( or spectrum of `` twists''= scale dimension minus spin ) modulo integers@xmath11 one has a theory of anomalous dimension as soon as one knows how to classify physically admissable representations of the infinite braid group or more precisely the ribbon braid group @xmath12 . the classification of the latter is done by the method of tracial states on @xmath13 which follow a combinatorial version of the field theoretic cluster decomposition property , the so - called markov property @xcite@xcite . this method was originally invented in the early 70@xmath14by doplicher haag and roberts ( dhr ) in order to classify the admissable permutation group statistics which is associated with the algebraic superselection theory of compactly localized charges in @xmath15 @xcite in a formulation without field multiplets . in subfactor theory @xcite this method was independently discovered in a vastly more general context and called it very appropriately the method of `` markov traces '' ( due to vaughn jones ) which in turn was gratefully re - adopted by the physicists . the name markov in this context reveals a lot about the conceptual scope of this theory because markov junior refers to the russian mathematician who made important contribution to the early study of the braid group , but at the same time one is invited to think about markov senior the probabilist since , while for a physicist the property of this tracial state ( which then allowed its iterative determination ) was a discrete version of the field theoretic cluster decomposition property , to a mathematician this procedure was more reminiscent of a discrete stochastic process . a field theoretic version of the classification of admissable braid group representations based on the markov trace formalism can be found in @xcite@xcite@xcite @xcite the reader will notice that in our enumeration of achievements in chiral theories we have omitted the better known representation theory of specific algebras as the energy - momentum tensor algebra or current algebras . this is not to ignore their important role in the modern development of chiral theories but rather a result of the fact that they have no direct counterpart in higher dimensions . so if we want to use chiral theory as a theoretical laboratory for higher dimensional conformal field theories we are forced to de - emphasized those aspects and highlight instead others , as space- and time - like commutation structures which are independent on spacetime dimensions . the weak point of the present approach to higher dimensional conformal theories is of course the total lack of nontrivial examples i.e. of higher dimensional conformal models with anomalous scaling dimensions . as already mentioned the new timelike superselection structure can unfortunately not be illustrated by the representation theory of any known algebra ( unlike chiral theory ) and it is also not possible to explore this timelike region by lagrangian methods ( which tend to favor the euclidean or spacelike regions ) . there is of course the folklore that the only lagrangian conformal 4-dimenional theories are a special kind of supersymmetric yang - mills theories which , if it could be made more rigorous in a clearer conceptional setting ( less computational recipes , more physical principles ) would be a remarkable observation . the present non - lagrangian approach suggests another picture : instead of a scarceness of models one should expect a similar wealth as in the case of nonperturbative chiral theories . i think that in the near future one will have new concepts and methods for their construction from timelike braid group data . more remarks can be found at the end . in lack of illustrative examples which could show that the present requirements allow for nontrivial realizations , we are limited to consistency checks . the main new problem which was absent in the chiral case is the consistency of timelike braiding with spacelike locality i.e. one has to show that the new timelike plektonic structure is in harmony with the standard boson / fermion local commutation relations for the spacelike region . this will be the subject of the third section . the next section contains a review of the geometric setting of conformal symmetry . although most of these results are known since the 70s , we find it convenient for setting notation and concepts to present an updated version . the isomorphism of conformal field theory with qft in 5-dim . anti de sitter spacetime belongs logically ( though not historically ) to that section which deals with issues of compactification , covering and global causality . some comments on the relation of conformal qft and particle physics are in order . in most of the recent literature standard notions of particle physics as particles and scattering theory have been used in the conformal setting without qualifications . but a closer examination shows that conformal field theory is not a theory of interacting particles , at least not if these concepts are still used in a way which is not completely void of their original physical meaning . there are no discrete zero mass shells ( light cone delta functions in momentum space ) in an _ interacting _ dilation invariant theory and there is a fortiori no lsz s - matrix ( the asymptotic lsz limits in fact vanish ) . any kind of interaction dissolves immediately the zero mass shell into the continuum with a continuous mass distribution of enhanced weight at @xmath16 . there remains of course the interesting question of what kind of residual particle physics information one can extract from the scale invariant limit of a massive particle theory . for a discussion of this and related points see @xcite . the structural problems of conformal qft to which we draw our attention in the sequel do not depend on particle interpretations . since massless particles in a conformal theory can not interact @xcite , the physically interesting interacting fields in a conformal theory are those with anomalous dimension . whereas conformal free fields are commutative in the huygens sense of time- and spacelike commutativity ( huygens principle in wave optics transposed into the setting of local quantum physics ) , conformal interactions lead to reverberations@xmath17 inside the light cone which are in correspondence with the appearance of anomalous dimension and produce a paradoxical violation @xcite of einstein causality if one uses the standard description of minkowski space and allows `` big '' special conformal transformations to act on spacelike vanishing commutators with the usual transformation rules . hence it was of interest to investigate the global spacetime interpretation of such fields in more detail in order to remove the paradox . for aspects of global localization in interacting conformal field theories one needs to introduce the covering of the conformally compactified minkowski space with its nontrivial topology . this is a well - studied old subject @xcite@xcite@xcite , which led to an important global decomposition theory which , after lying dormant for a number of years ( when gauge theories took the center of the stage ) , @xciteobtained an unexpected fresh push in the special case of d=1 + 1 @xcite where conforms observables decompose ( similar to free fields ) additively in to the two light cone components . the latter act on a tensor - factorized hilbert space whereas charge - carrying fields act on a tensor product of extended superselected hilbert spaces . the fastest way to obtain a first glance at the formalism and physical use of the conformal covering space is to notice that the wigner representation theory for the poincar group for zero mass particles allows an extension to the conformal symmetry ( without extending the hilbert space and the degree of freedoms ) : poincar group @xmath18 . besides scale transformations , this larger symmetry also incorporates the fractional transformations ( proper conformal transformations ) @xmath19 the conformal reflection @xmath20 itself is not a moebius transformation , but in free field theories it is known to be implemented by a unitary transformation @xcite . for fixed x and small b the formula ( [ sc ] ) is well defined , but globally it mixes finite spacetime points with infinity and hence requires a more precise definition ( in particular in view of the positivity energy - momentum spectral properties ) in its action on quantum fields . as preparatory step for the quantum field theory concepts one has to achieve a geometric compactification . this starts most conveniently from a linear representation of the conformal group so(d,2 ) in 6-dimensional auxiliary space @xmath21 ( i.e. without field theoretic significance ) with two negative ( time - like ) signatures @xmath22 and restricts this representation to the ( d+1)-dimensional forward light cone @xmath23 where @xmath24 denotes the d - dimensional minkowski length square . the compactified minkowski space is obtained by adopting a projective point of view ( stereographic projection ) @xmath25 it is then easy to verify that the linear transformation which keep the last two components invariant consist of the lorentz group and those transformations which only transform the last two coordinates yield the scaling formula @xmath26 leading to @xmath27 the remaining transformations , namely the translations and the fractional proper conformal transformations , are obtained by composing rotations in the @xmath28-@xmath29 and boosts in the @xmath30-@xmath31 planes . the so obtained spacetime is most suitably parametrized in terms of a `` conformal time '' @xmath32@xmath33 so that the conformally compactified minkowski space is a piece of a multi - dimensional cylinder carved out between two d-1 dimensional boundaries which lie symmetrically around @xmath34 where they touch each other @xcite ( fig.1 below ) . if one cuts the cylinder wall this region @xmath35 looks like a double cone subtended by two points at infinity @xmath36 @xmath37 the boundary region @xmath38 consists precisely of all points which are past / future light like from @xmath39 ( the light rays on the cylinder are continuous lines starting from @xmath40 and ending at @xmath41 which are points to be identified on @xmath9 ) . in this way the cylinder is equipped with a tiling into infinitely many ordinary minkowski spaces . @xmath42 whereas matter in @xmath9 is subject to the quantum version of the huygens principle namely observables commute if the light ray subtended by one localization region can not reach the other , the covering @xmath43 comes with a global causality structure . the relevance of this covering space for the notion of relativistic causality was first pointed out first by i. segal @xcite and the above parametrization as well as many other contributions which became standard in conformal qft appeared for the first time in the work of luescher and mack @xcite . formally this framework solves the `` einstein causality paradox of conformal quantum field theory '' @xcite which originated from `` would be '' conformal models ( locally conformal invariant ) of quantum field theory as e.g. the massless thirring model which violates huygens principle . the naive reason for this apparent violation turned out to be that there exist continuous curves of special conformal transformations which lead from timelike separations with one point at the origin via the lightlike infinity to spacelike separation . this obviously generates a contradiction with the locality structure of the thirring model whose timelike anti - commutator unlike the spacelike one does not vanish . the covering structure formally solves this causality paradox by emphasizing that in fact there are no covering transformation which violate global causality , a paradox only arises if points become projected out of @xmath43 into @xmath44 in other words , one must keep track of the global path which remains space / time- like and not just its end points . if one depicts as before the covering space as a cylinder ( fig.1 below ) , then it contains infinitely many copies of the original minkowski space which appear in the projection to ( @xmath45 subspace as a finite rhomboid region @xcite . using the above parametrization in terms of @xmath46 and the `` conformal time '' @xmath47 one can immediately globalize the notion of time like distance and one finds the following causality structure ( @xcite@xcite ) @xmath48 where @xmath49 now denotes the spacelike/@xmath50timelike separation in the global sense . since it is expressed in terms of the difference of two coordinates on the light cone in the 6-dim . ambient spacetime , a conformal transformation which is linear in the @xmath51-variables leaves it invariant . for the description of the dirac - weyl compactified minkowski space the use of the following simpler parametrization close to standard minkowski coordinates is more convenient @xmath52 similarly one may use the quadratic polynomial @xmath53 appearing in the denominator ( and the jacobian ) of the special conformal transformations in order to decide whether two points are globally timelike / spacelike ( connectable by timelike / spacelike geodesics ) without using ( [ glob ] ) @xcite@xcite . the formulation in terms of conformal covering space would be useful if the world ( including laboratories of experimentalists ) would also be conformal , which certainly is not the case . therefore it is helpful to know that there is a way of re - phrasing the physical content of local fields ( which violate the huygens principle and instead exhibit the phenomenon of `` reverberation '' @xcite inside the forward light cone ) in the minkowski world @xmath54 of ordinary particle physics without running into the trap of the causality paradox of the previous section ; in this way the use of the above @xmath55- parametrization would loose some of its importance and this may be considered as an alternative to the lscher - mack approach on covering space . this was first achieved in a joint paper involving one of the present authors @xcite whose main point was that the global causality structure could be encoded into a global decomposition theory of fields with respect to the center of the conformal covering ( conformal block decomposition ) . local fields , although behaving apparently irreducibly under infinitesimal conformal transformations , transform in general reducibly under the action of the global center of the covering @xmath56 this reduction was precisely the motivation in for the global decomposition theory of conformal fields in @xcite . @xmath57 these component fields behave analogous to trivializing sections in a fibre bundle ; the only memory of their origin from an operator on covering space is their quasiperiodicity @xmath58 ^{\delta -\zeta } \left [ \sigma _ { -}(b , x)\right ] ^{\zeta } } % f_{\delta } ( x)_{\alpha , \beta } \\ \zeta & = & \frac{1}{2}(\delta _ { f}+\theta _ { \beta } -\theta _ { \alpha } ) \notag\end{aligned}\ ] ] where the second line is the transformation law of special conformal transformation of the components of an operator @xmath59 with scale dimension @xmath60 sandwiched between superselected subspaces @xmath61 anf @xmath62 using the explicit form of the conformal 3-point function it is easy to see that phases are uniquely given in terms of the scaling dimensions @xmath63 which occur in the conformal model @xcite . @xmath64 a central projector projects onto the subspace of all vectors which have the same scaling phase i.e. onto a conformal block associated with the center , so the labelling refers to ( in case of bosons ) the anomalous dimensions @xmath65 these subspaces of operators are bigger than the chiral blocks , since the anomalous dimension e.g. does not distinguish between with charged fields and their anti - fields which carry the conjugate charge . having understood the physical interpretation of the central decomposition does not yet mean that we have a theory of the spectrum of admissable anomalous dimensions ( critical indices ) in higher dimensional conformal theories , but the close analogy to chiral theories gives sufficient incentive to look for such a theory . this will be the subject of the next section . the prize one has to pay for this return to the realm of particle physics on @xmath54 in terms of component fields ( [ comp ] ) is that these projected fields are not wightman fields . they depend on a source and range projector and if applied to a vector the source projector has to match the hilbert space i.e. @xmath66 annihilates the vacuum if @xmath67 is not the projector onto the vacuum sector . this is very different from the behavior of the original @xmath59 which , in case it was localized in a region with a nontrivial spacelike complement , can never annihilate the vacuum . this kind of projected fields are well known from the exchange algebra formalism of chiral qft @xcite but they appear in a rudimentary form already in @xcite . inside two and 3-point functions the projectors are unique and may be omitted but for @xmath68 there are several projected n - point functions and therefore they are needed . structural properties of the real time formulation as this timelike decomposition formula remain totally hidden in the euclidean formulation . they lead to cuts with multi - valuedness in an analyticity region which is beyond the standard bhw @xcite extended permuted tube region ( see next section ) of standard poincar invariant theories . as in the chiral case one defines conformal observable fields as those which commute with the center generator @xmath5 and , as a consequence are free of these cuts i.e. have meromorphic ( rational ) correlation functions on @xmath9 and its complex extensions as it is well known from the chiral conformal observables ( where the terminology `` holomorphic '' was unfortunately attributed to the observable fields instead of referring to a particular state their ground state correlation functions only ) the structure of the center in chiral conformal field theories is determined by the discrete spectrum of the rotation operators for the compactified @xmath69lightrays @xmath70 where the right hand side is the standard virasoro algebra notation . it is well - known that this operator shares with the light ray translations @xmath71 the positivity of its spectrum . this becomes in fact obvious if one represents it in terms of @xmath72@xmath73 where @xmath74 is the representer of the chiral conformal reflection @xmath75 ( in linear lightray coordinates @xmath76 ) and @xmath77 is the generator of the fractional special conformal transformation ( [ sc ] ) . however the two - dimensional inversion does not factorize since the chiral inversion rewritten in terms of 2-dim . vector notation corresponds to @xmath78 the `` wrong '' sign in the spatial part can be corrected by a parity transformation @xmath79 which mixes the two chiral components . in defining an object which transforms as a vector this has to be taken in consideration @xmath80 the vector formula ( [ vec ] ) is valid in any dimension i.e. does not require light ray factorization . it leads to a family of operators with discrete spectrum @xmath81 which are dependent on a timelike vector @xmath82 as in the chiral case one only needs to add to the poincar+scale transformations the ( timelike ) conformal rotation @xmath83 the other components of @xmath84 are generated by the action of the lorentz group . to understand the geometric action of @xmath85 it is helpful to depict the covering world @xmath43 with a copy of the minkowski world inside . from ( [ m ] ) one obtains the identification of the covering world with the surface of a d+1 dimensional cylinder @xcite . in fig.1 only two of the d-2 components of the d - dimensional * e*-vector have been drawn , the others have been set zero . for depicting the spacelike complement of a double cone @xmath86 in in @xmath43 it is more convenient to cut open the cylinder in @xmath32-direction and identify opposite sides as in fig.2 . ( 132,226 ) ( 34.0,178.0)(34.0,178.0)(35.88,177.38)(37.75,176.82)(39.59,176.25)(41.38,175.69)(43.15,175.13)(44.88,174.6)(46.56,174.08)(48.22,173.55 ) ( 48.22,173.55)(49.86,173.05)(51.45,172.55)(53.02,172.08)(54.54,171.58)(56.04,171.11)(57.52,170.67)(58.95,170.23)(60.36,169.79)(61.74,169.36 ) ( 61.74,169.36)(63.08,168.94)(64.4,168.51)(65.68,168.11)(66.94,167.73)(68.18,167.33)(69.37,166.94)(70.55,166.58)(71.69,166.19)(72.83,165.83 ) ( 72.83,165.83)(73.93,165.5)(75.0,165.14)(76.05,164.8)(77.05,164.48)(78.05,164.14)(79.01,163.8)(79.98,163.51)(80.9,163.19)(81.8,162.88 ) ( 81.8,162.88)(82.68,162.58)(83.51,162.29)(84.37,161.98)(85.16,161.69)(85.94,161.41)(86.73,161.13)(87.48,160.86)(88.19,160.58)(88.91,160.3 ) ( 88.91,160.3)(89.58,160.05)(90.26,159.76)(90.91,159.51)(91.54,159.26)(92.15,159.0)(92.73,158.75)(93.3,158.48)(93.87,158.23)(94.41,157.98 ) ( 94.41,157.98)(94.94,157.73)(95.44,157.48)(95.94,157.23)(96.43,156.98)(96.9,156.75)(97.33,156.5)(97.79,156.25)(98.19,156.0)(98.62,155.76 ) ( 98.62,155.76)(99.01,155.51)(99.4,155.26)(99.76,155.01)(100.12,154.76)(100.48,154.5)(100.8,154.25)(101.12,153.98)(101.44,153.73)(101.76,153.47 ) ( 101.76,153.47)(102.05,153.19)(102.33,152.94)(102.62,152.66)(102.9,152.39)(103.16,152.11)(103.41,151.83)(103.66,151.55)(103.91,151.26)(104.12,150.98 ) ( 104.12,150.98)(104.37,150.69)(104.58,150.38)(104.8,150.08)(105.01,149.76)(105.22,149.44)(105.41,149.14)(105.62,148.8)(105.8,148.48)(106.01,148.14 ) ( 106.01,148.14)(106.19,147.8)(106.37,147.44)(106.55,147.08)(106.73,146.73)(106.91,146.36)(107.08,145.98)(107.26,145.61)(107.44,145.22)(107.62,144.8 ) ( 107.62,144.8)(107.8,144.41)(107.98,144.0)(108.0,144.0 ) ( 34.0,178.0)(34.0,178.0)(33.34,177.75)(32.7,177.52)(32.04,177.27)(31.44,177.05)(30.81,176.83)(30.21,176.61)(29.62,176.38)(29.04,176.16 ) ( 29.04,176.16)(28.46,175.97)(27.92,175.75)(27.37,175.55)(26.81,175.35)(26.29,175.13)(25.77,174.94)(25.26,174.75)(24.76,174.57)(24.27,174.38 ) ( 24.27,174.38)(23.78,174.19)(23.29,174.0)(22.84,173.8)(22.37,173.63)(21.94,173.44)(21.5,173.26)(21.05,173.11)(20.63,172.92)(20.22,172.76 ) ( 20.22,172.76)(19.84,172.58)(19.43,172.41)(19.04,172.26)(18.68,172.08)(18.31,171.92)(17.93,171.76)(17.59,171.58)(17.25,171.44)(16.89,171.29 ) ( 16.89,171.29)(16.56,171.11)(16.26,170.97)(15.93,170.8)(15.64,170.66)(15.34,170.51)(15.05,170.33)(14.76,170.19)(14.47,170.05)(14.22,169.89 ) ( 14.22,169.89)(13.93,169.75)(13.68,169.58)(13.43,169.44)(13.21,169.3)(12.97,169.14)(12.75,169.0)(12.52,168.83)(12.31,168.69)(12.1,168.54 ) ( 12.1,168.54)(11.89,168.38)(11.69,168.23)(11.51,168.08)(11.31,167.94)(11.14,167.79)(10.97,167.63)(10.8,167.48)(10.64,167.33)(10.5,167.16 ) ( 10.5,167.16)(10.34,167.01)(10.19,166.86)(10.06,166.69)(9.92,166.55)(9.8,166.38)(9.67,166.23)(9.56,166.05)(9.44,165.89)(9.34,165.73 ) ( 9.34,165.73)(9.22,165.55)(9.14,165.39)(9.05,165.23)(8.96,165.05)(8.88,164.86)(8.8,164.69)(8.71,164.51)(8.64,164.33)(8.59,164.16 ) ( 8.59,164.16)(8.51,163.98)(8.46,163.8)(8.39,163.61)(8.36,163.41)(8.3,163.23)(8.26,163.04)(8.22,162.83)(8.19,162.63)(8.14,162.42 ) ( 8.14,162.42)(8.13,162.23)(8.09,162.01)(8.06,161.8)(8.05,161.58)(8.03,161.36)(8.03,161.14)(8.01,160.92)(8.01,160.69)(8.0,160.47 ) ( 8.0,160.47)(8.0,160.23)(8.0,160.0)(8.0,160.0 ) ( 34.0,78.0)(34.0,78.0)(36.02,78.52)(38.0,79.04)(39.93,79.55)(41.84,80.05)(43.72,80.55)(45.56,81.04)(47.36,81.52)(49.11,81.99 ) ( 49.11,81.99)(50.84,82.43)(52.54,82.88)(54.18,83.33)(55.81,83.75)(57.4,84.18)(58.95,84.58)(60.47,84.99)(61.97,85.38)(63.43,85.77 ) ( 63.43,85.77)(64.86,86.15)(66.25,86.54)(67.61,86.9)(68.94,87.27)(70.23,87.61)(71.51,87.97)(72.75,88.3)(73.94,88.65)(75.15,88.97 ) ( 75.15,88.97)(76.3,89.3)(77.41,89.63)(78.51,89.93)(79.58,90.25)(80.62,90.55)(81.65,90.86)(82.62,91.15)(83.58,91.44)(84.51,91.74 ) ( 84.51,91.74)(85.44,92.02)(86.33,92.29)(87.19,92.58)(88.01,92.86)(88.83,93.13)(89.62,93.4)(90.4,93.66)(91.12,93.93)(91.87,94.19 ) ( 91.87,94.19)(92.55,94.44)(93.23,94.72)(93.9,94.97)(94.51,95.22)(95.15,95.49)(95.73,95.74)(96.3,96.0)(96.87,96.25)(97.41,96.5 ) ( 97.41,96.5)(97.93,96.75)(98.43,97.02)(98.91,97.27)(99.37,97.52)(99.8,97.77)(100.25,98.04)(100.66,98.29)(101.05,98.55)(101.44,98.8 ) ( 101.44,98.8)(101.8,99.08)(102.16,99.33)(102.5,99.61)(102.8,99.88)(103.12,100.15)(103.43,100.43)(103.69,100.69)(103.98,101.0)(104.23,101.27 ) ( 104.23,101.27)(104.48,101.55)(104.72,101.86)(104.94,102.15)(105.15,102.44)(105.33,102.77)(105.54,103.08)(105.72,103.4)(105.9,103.72)(106.05,104.04 ) ( 106.05,104.04)(106.19,104.36)(106.36,104.69)(106.5,105.05)(106.62,105.4)(106.75,105.75)(106.87,106.11)(106.98,106.49)(107.08,106.86)(107.18,107.25 ) ( 107.18,107.25)(107.26,107.63)(107.36,108.04)(107.44,108.43)(107.51,108.86)(107.58,109.27)(107.66,109.69)(107.73,110.15)(107.8,110.58)(107.87,111.05 ) ( 107.87,111.05)(107.93,111.51)(108.0,111.98)(108.0,112.0 ) ( 34.0,78.0)(34.0,78.0)(33.34,78.22)(32.7,78.47)(32.05,78.68)(31.45,78.91)(30.84,79.13)(30.25,79.33)(29.64,79.55)(29.09,79.77 ) ( 29.09,79.77)(28.53,79.97)(27.96,80.16)(27.43,80.36)(26.88,80.55)(26.37,80.75)(25.87,80.93)(25.37,81.11)(24.87,81.29)(24.39,81.47 ) ( 24.39,81.47)(23.94,81.63)(23.46,81.8)(23.03,81.97)(22.6,82.13)(22.17,82.3)(21.75,82.44)(21.31,82.61)(20.93,82.77)(20.54,82.93 ) ( 20.54,82.93)(20.14,83.08)(19.77,83.22)(19.39,83.38)(19.06,83.52)(18.7,83.65)(18.35,83.8)(18.02,83.94)(17.68,84.08)(17.38,84.22 ) ( 17.38,84.22)(17.06,84.36)(16.77,84.5)(16.47,84.63)(16.18,84.77)(15.92,84.9)(15.64,85.04)(15.38,85.16)(15.1,85.3)(14.85,85.43 ) ( 14.85,85.43)(14.6,85.58)(14.38,85.69)(14.14,85.83)(13.93,85.97)(13.71,86.11)(13.5,86.24)(13.27,86.38)(13.09,86.52)(12.89,86.65 ) ( 12.89,86.65)(12.68,86.79)(12.52,86.93)(12.34,87.05)(12.17,87.19)(12.0,87.33)(11.81,87.5)(11.67,87.63)(11.51,87.79)(11.36,87.93 ) ( 11.36,87.93)(11.22,88.08)(11.09,88.24)(10.94,88.4)(10.81,88.55)(10.69,88.72)(10.56,88.88)(10.44,89.04)(10.34,89.19)(10.21,89.38 ) ( 10.21,89.38)(10.11,89.54)(10.01,89.72)(9.89,89.9)(9.8,90.08)(9.71,90.27)(9.61,90.44)(9.53,90.65)(9.44,90.83)(9.36,91.04 ) ( 9.36,91.04)(9.26,91.25)(9.19,91.44)(9.11,91.66)(9.03,91.88)(8.96,92.11)(8.88,92.33)(8.81,92.55)(8.75,92.79)(8.67,93.02 ) ( 8.67,93.02)(8.61,93.27)(8.55,93.52)(8.47,93.77)(8.42,94.04)(8.36,94.29)(8.3,94.55)(8.22,94.83)(8.17,95.11)(8.11,95.4 ) ( 8.11,95.4)(8.05,95.69)(8.0,95.99)(8.0,96.0 ) ( 160,282 ) on the other hand the living space of the observable algebra is the dirac - weyl compactification @xmath9 of @xmath54 which is depicted as fig.3 with opposite two sides @xmath87 and @xmath88 identified . vice versa the minkowski space results from puncturing the dirac - weyl compactification @xmath9 at @xmath89 and simultaneously removing the whole subtended lightlike d-1 dimensional subspace . note that as a result of the identification the union of the timelike and spacelike complements form a connected set in @xmath9 ; in fact that part of the spacelike complement of a double cone @xmath86 in the covering @xmath43 beyond @xmath54 becomes converted into the huygens timelike region with respect to @xmath90 the first use of these geometrical properties in the setting of algebraic qft is due to hislop and longo @xcite . the pictures are closely related to those used by penrose , except that penrose does not use them for compactification since he is dealing with a conformal class of spacetime metrics and not with conformal invariant observable matter fulfilling the huygens principle . ( 176,266 ) it is hard to resist mentioning that the d+2 dimensional setting for the compactification and subsequent covering of d - dimensional minkowski spacetime also lends itself to obtain a natural relation with the dd+1 dimensional @xmath91 ( anti de sitter ) spacetime by taking instead of the surface of the light cone a hyperbolic region inside this light cone @xmath92 it is immediately clear that asymptotically this @xmath93 dimensional lorentzian noncompact manifold in the associated @xmath51-parametrization has a d - dimensional conformal boundary which in the above picture corresponds to the asymptotic coalescence of the @xmath94 with the light cone directions . this asymptotic pointlike relation can of course not be continued inside the @xmath94 spacetime , but the action of the same @xmath95 group on the two manifolds suggests a relation between d - dimensional conformal double cone ( conformal transforms of rindler wedge regions ) and d+1 dimensional wedge regions on @xmath94 i.e. between those regions which result from projecting d+2 dimensional wedges in the ambient space ( on which there is a transitive linear action of @xmath95 ) onto @xmath96 and @xmath97using the setting of algebraic qft , rehren @xcite converted this geometric relation into an isomorphism between algebraic quantum field theories where the principle objects are localized algebras and not their coordinatizations in terms of pointlike field generators . in fact as one naively expects the isomorphism negates a relation between pointlike field theories ( see below ) . although from a logical point of view the above observation on ads spacetime belongs naturally to the conformal compactification setting of the 70s where most of the above observations were made , history ( as almost always and in particular in this case ) did not follow logic . rather part of the isomorphism namely the mapping @xmath98 for the corresponding qfts was first observed by string theorist at the end of the 90 s @xcite in connection with their speculative ideas on quantum gravity . although the underlying idea was that the asymptotic map characterizes a unique ads theory once the conformal asymptote has been prescribed , it was rehren s construction which supplied a constructive mathematical proof for the full isomorphism and at the same time also highlighted the conceptual scope of ( field - coordinate - free ) local quantum physics . the isomorphism can be brought closer to the realm of particle physics if one highlights the above analogy between the hamiltonian @xmath99 and the conformal `` hamiltonian '' @xmath100 by asking the following question : is there a qft which maintains the symmetries but for which the conformal hamiltonian of the @xmath101 becomes the true hamiltonian ? the answer is unique , it is precisely the same @xmath94 theory of the rehren isomorphism which changes the physical interpretation and the spacetime affiliation , but not the group theoretical and algebraic net structure . the particle physics nature of this isomorphism can be further clarified by studying concrete examples e.g. the fate of free conformal / ads fields under this isomorphism . the result is rather interesting @xcite . a pointlike ads free fields has too many degrees of freedom with the result that its conformal image is a special `` generalized free field '' with a homogenous mass distribution which destroys the primitive causality i.e. the requirement that the algebraic data in a time slice ( in order to avoid short distance problems of spacelike surfaces ) which covers a compact spatial region fix the data in the `` causal shadow '' region ( the double cone shaped causal envelop ) . using rehren s graphical illustration @xcitewhich depicts conformal qft on a cylindrical boundary of the ads world , one sees that there are more and more degrees of freedom from the inside of ads entering the causal shadow as one moves upward in time . vice versa if one starts from a free conformal theory than the ads image can not sustain pointlike fields but only configurations which are delocalized ( and hence diluted ) in one direction as some kind of nielsen - olsen string i.e. the memory from its d - dimensional pointlike origin is stored in the ads image . this means in practical terms that the nice idea to start with a lagrangian ads theory end re - process them with the witten prescription to the conformal side in order to enrich the set of conformal models does not work . one obtains conformal theories in this way but they are unphysical , a fact which remains concealed in the euclidean formulation . this also affects the conjectures string theory sym relation to the extend that it relies on a relation between @xmath1 and conformal sym lagrangian theories . the algebraic isomorphism itself is not affected limited by these deficiencies in pointlike ( lagrangian ) relations and their remains the intellectually challenging problem of understanding the algebraic conformal decomposition theory and its dhr origin directly for qft on ads . just because the isomorphism changes the interpretation , it is by no means obvious without looking carefully at details , what the observations in this article mean in a ads spacetime . since the spectrum of anomalous dimensions becomes encoded into the spectrum of a hamiltonian one even could be optimistic and hope for certain simplifications on the ads side . a rather trivial and unfortunately atypical case is @xmath102 whose bona fide hamiltonian is the virasoro @xmath103 and in fact it is the only system with a maximal sl(2,r ) symmetry having that compact ( discrete spectrum ) opaerator as its hamiltonian . the above analogies between the d=1 + 1 case and the higher dimensional conformal field theory should however not lead one into overlooking a remarkable difference . already on a purely classical level the characteristic value problem for the free wave equation is totally different from either its massive counterpart or from the d2 case . whereas in the latter cases the data on one lightray or lightfront is complete , the zero mass d=1 + 1 case needs both the lightray data in order to determine the d=1 + 1 theory . in the qft the manifestation of this is the tensor factorization into the chiral degrees of freedom which amounts to a doubling of degrees of freedom . in the next section we will see that this also leads to an exceptional behavior in the timelike huygens structure and the associated timelike braid group structure . so the chirally factorizing d=1 + 1 situation is a guide in certain higher dimensional aspects and stands in interesting contrast to others . in the previous section we have emphasized certain analogies between the timelike algebraic structure in higher dimensional and chiral conformal theories . for the latter the decomposition theory has a more fundamental explanation in terms of a local plektonic superselection structure . this in particular means that the components appearing in ( [ comp ] ) permit a local refinement ; they can be further reduces into dhr localized charge sectors of an observable local algebra which lives on @xmath8 and which in concrete models is generated by the energy momentum tensor , current algebras , w - algebras etc . the commutation structure of charge - carrying chiral fields obtained by the dhr method is an exchange algebra @xmath104g_{\alpha , \beta ^{\prime } } ( y)f_{\beta ^{\prime } , \gamma } ( x),\,\,x > y \label{time } \\ f_{a,\beta } g_{\beta , \gamma } & = & \sum_{\beta ^{\prime } } r_{\beta , \beta ^{\prime } } ^{(\alpha , \gamma ) } [ c(f),c(g)]g_{\alpha , \beta ^{\prime } } f_{\beta ^{\prime } , \gamma } , \,\,locf > locg\end{aligned}\ ] ] where in the second line we have used the more general operator formulation of aqft and the c s in the bracket @xmath105 $ ] denotes the dependence of the r - matrices on the superselected charges c of the participating operators . the localization is always meant relative to the observables @xcite . the artin relations are a consequence of the associativity of this algebra . here the indices @xmath106 refers to projectors on irreducible dhr representation spaces . although the latter are a refinement of the projectors appearing in the center @xmath5 , we will for simplicity in writing maintain the same notation . the validity of such an exchange algebra structure would supply a natural local explanation for the center superselection rules . we will therefore postulate this structure for the timelike region in higher dimensional conformal theories ( where now @xmath49 to @xmath107timelike ordering ) and , in lack of concrete examples , test its consistency . it is reasonable to start with consistency checks in the standard wightman framework of pointlike covariant fields . the most powerful tool of wightman s formulation is provided by the analytic properties of correlation functions . it is well known that the complexified lorentz group may be used to extend the tube analyticity associated with the physical positive energy - momentum spectrum . the famous bhw theorem @xcite insures that this extension remains univalued in a new complex domain and the jost theorem characterizes its real points . finally spacelike locality links the various permutations of the position field operators within the correlation function to one permutation ( anti)symmetric analytic master function which is still univalued . the various correlation functions on the physical boundary with different operator ordering can be obtained by different temporal @xmath108 prescriptions descending to the real boundary from within the tube . complexifying the scale transformations , the conformal correlations can be extended into a still bigger analyticity region which even incorporates `` timelike jost points '' . but trying to find a univalued master function which links the various orders together fails in the presence of fields with anomalous dimensions and only works for observable fields which are local fields on the compactification @xmath109 the latter are the analogs of chiral observables , except that apart from ( composite ) free fields one does not have algebraic examples since virasoro- and kac - moody algebras do not exist in higher dimensions . the timelike braid group structure suggest that the role of the permutation group in the analytic extension from timelike points should be replaced by the braid group . the resulting ramification takes place in a region which is obtained analytic extension by the complex dilation group , which leaves the old univalued bhw region free of branch cuts . therefore the timelike braid structure is consistent with the bhw analyticity structure . let us look at other arguments which test the consistency of the old spacelike locality with the new timelike localization structure because the problem of coexistence of these two regions is the main difference to the chiral case . this significant difference even remains if one glues together the two chiralities to a 2-dimensional conformal theory . as already mentioned before , it has its origin already in the classical wave propagation theory one needs two sets of characteristic light ray data to e.g. determine the amplitudes in a wedge region whereas for any higher dimensional theory ( and even massive d=1 + 1 propagation ) one light front is enough . a plektonic charge structure which is only visible in the timelike region would immediately explain the appearance of a nontrivial timelike center and the spectrum of anomalous dimension . it would sort of `` kinematize '' conformal interactions and reveal conformal qfts as basically free theories if it would not be for that part of interaction which sustains the timelike plektonic structure . of course the situation trivializes if the theory has no anomalous dimensions and nontrivial components . analogous to @xcite ( remarks at end of section v.4 ) we conjecture that this characterizes interaction free conformal theories which are generated by free fields . what makes this issue somewhat complicated is the fact that contrary to chiral theories we do not have a single nontrivial example because this issue is neither approachable from the representation theory of known infinite dimensional lie - algebras nor from the formal euclidean functional integral method . the remaining strategy is to show structural consistency of the spacelike local- with the conjectured timelike plektonic- structure and to find a new construction method ( non energy - momentum tensor- or current- algebra based , non - lagrangian ) . here we are mainly be concerned with consistency arguments and in the following we will comment how local / plektonic on - vacuum relations between two fields can be commuted through to a generic position . assume for simplicity , as we have already tacitly done in our decomposition formulas before , that we are in a `` minimalistic '' situation ( similar to minimal models or w - algebras in the chiral setting ) where the field theory has no internal symmetry group . actually the whole discussion can be carried out in the presence of nontrivial inner symmetries , but the additional complications do not essentially alter the following consistency considerations . so we assume that the fields can be given `` timelike '' charge indices @xmath110 and their conjugates @xmath111 resulting from projectors on charge spaces so that the decomposition is as in the chiral case where the charge projectors with the same phase factors @xmath112 constitute a refinement of a central projector . clearly @xmath113 and its conjugate @xmath114 contribute tu the same central projector . in fact we may take over a substantial part of the formalism and concepts of @xcite . if one replaces the chiral translation+dilation augmented by the circular rotation generator @xmath103 by the spacetime symmetry group which leaves the timelike infinite point fixed ( poincar+dilations ) augmented by the generator of conformal time @xmath100 instead of the chiral @xmath115 one would of course also have to change the title of the old paper from `` einstein causality and artin braids '' to `` huygens causality and artin braids '' referring to the timelike ordering for which the conformal observables fulfill the huygens principle of vanishing commutators . the `` on - vacuum '' structure of commutation relations follows from the structure of the conformal 3-point functions @xmath116 ^{\delta _ { 3}}}\frac{1}{\left [ -(x_{13})_{\varepsilon } ^{2}\right ] ^{\delta _ { 2}}}\frac{1}{\left [ -(x_{23})_{\varepsilon } ^{2}\right ] ^{\delta _ { 1 } } } \\ & & \delta _ { 1}=\frac{1}{2}(\delta _ { g}+\delta _ { h}-\delta _ { f}),\,\delta _ { 2}=% \frac{1}{2}(\delta _ { f}+\delta _ { h}-\delta _ { g}),\,\delta _ { 3}=\frac{1}{2}% ( \delta _ { f}+\delta _ { g}-\delta _ { h } ) \notag\end{aligned}\ ] ] where the @xmath117-prescription was explained in the introduction . for spacelike and timelike distances one concludes @xmath118 since this relation is valid on all quasiprimary composites @xmath119 they consist by definition of the equal point limit of the associated primary @xmath120 ( lowest scale dimension operator in a superselected charge class ) multiplied with a polynomial in the observable field . these composites applied to the vacuum form a dense set in the respective charge sector ] and hence the on - vacuum formula is a consequence of the structure of 3-point functions . the spacelike local commutativity off - vacuum is consistent with that on - vacuum since for @xmath121 timelike with respect to a spacelike pair @xmath122 we have ( here @xmath123 denote the superselected charges of the participating operators ) @xmath124 and therefore the off - vacuum vanishing of the @xmath59-@xmath125 commutator is consistent with the on - vacuum vanishing of this commutator if there holds a certain relation between @xmath126 and @xmath127 which is identically fulfilled for @xmath128 . similarly one does not run into inconsistencies if one tries to obtain a timelike off - vacuum @xmath59-@xmath125 situation from the on - vacuum placement by commuting through a @xmath99 which is spacelike to the timelike @xmath59-@xmath125 pair @xmath129 where in the second line we commuted @xmath59 through @xmath125 _ before _ trying to bring both to the vacuum . since their is no rule to commute the @xmath130 with @xmath131 for @xmath132 there is no way to get to the same @xmath133 order as in the first line and hence no consistency relation is to be checked . the absence of rules for spacelike commutations for projected fields protects the formalism to run into inconsistencies . if components for spacelike distances would commute than they belong to the vacuum . let us also briefly look at the compatibility of the timelike plektonic structure with the conformal structure of the 4-point function of 4 identical hermitian fields @xmath134 ^{\delta _ { f}}\sum_{\gamma } w_{\gamma } ( u , v),\,\ , \notag \\ \,\,\,u & = \frac{x_{43}^{2}x_{21}^{2}}{(x_{42})_{\varepsilon } ^{2}(x_{31})_{\varepsilon } ^{2}},\,\,v=\frac{x_{32}^{2}x_{41}^{2}}{% ( x_{42})_{\varepsilon } ^{2}(x_{31})_{\varepsilon } ^{2 } } \notag\end{aligned}\ ] ] whereas the spacelike commutations leads to functional relations for @xmath135 with the exchange of two fields causing a rational transformation of the @xmath136 ( apart from multiplying the @xmath137 by rational @xmath136 factors)@xmath11 the timelike commutation of the off - vacuum fields @xmath138 produces rational transformation together with r - matrix mixing of the @xmath139-components leading to nontrivial monodromies @xmath140 as in the chiral case @xcite these relations characterize equivalence classes of operators carrying the same superselected charges ( conformal blocks ) and the selection of individual correlation functions have to be made by using in addition their short distance behavior in terms of scaling dimensions . despite some similarities with the chiral case , the dependence of @xmath141 on two instead of one cross ratios requires the use of more elaborate techniques ( mellin - barnes techniques , generalized hypergeometric functions ) than the hypergeometric formalism ( which is sufficient for the chiral one variable cross ratio dependence ) . we find it conceivable that in higher dimensions even for minimalistic models there could be deformation parameters ( coupling constants ) which may not show up in continuously changing anyonic phases . here we will not pursue this matter . let us finally take notice of what the algebraic approach can add to these consistency consideration . one possible point of departure for the algebraic approach would be to start from a doplicher - roberts field net @xmath142 on @xmath54 and use its assumed local conformal invariance in order to construct a unique extension @xmath143 on the covering @xmath144 this has been done in the work @xcite where it was also shown that extended net fulfills the important property of haag duality which in the bosonic case is a maximalization of causality @xmath145 where we have used the standard notation of aqft : a dash on the algebra means the von neumann commutant in the hilbert space of the operator algebra , whereas on the localization region @xmath86 ( double cone , or any conformal transform thereof ) it means spacelike complement in the global sense of @xmath144 in order to obtain information about a timelike braid group structure we need to identify an observable subnet @xmath146 on @xmath54 consisting of operators which commute with @xmath5 and is haag dual for timelike distances such that the restriction of the field net to the observable net decomposes into irreducible representations of the latter which obey braid group fusion laws . the projectors onto those irreducible subspaces would than be the desired refinements of the central @xmath5 projectors by which one obtains the @xmath66 exchange algebra operators . a prerequisite for this idea to work would be the existence of an autonomous timelike haag dual net which is sufficiently nonlocal in the spacelike sense i.e. sufficiently different from a spacelike haag dual net . this is necessary in order to obtain sectors which are different from those of the spacelike based dhr theory . a theory of observables @xmath146 on the compactification @xmath9 is automatically haag dual , but the localization of the commutant of a double cone would consist of a spacelike and timelike part @xcite . we would need a dualization which involves only the timelike complement in @xmath54 of double cones and which should be sufficiently nonlocal with respect to a conformal dhr dualization based on double cone algebras defined in terms of intersections of infinitely many wedges . algebraically the haag dual net on @xmath9 results from conformal transformations from the wedge algebra , whereas the double cones of the timelike dual net are obtained from intersections of the forward- with the shifted backward light cone algebra in @xmath44 from a pointlike field point of view the difference in these algebras is related to the way in which smeared fields are used to generate algebra . if they are used on @xmath54 then the net at infinity is diluted because the test functions vanish on @xmath147 and in order to re - establish the duality balance on has to make the algebras in the net bigger which do not contain points at infinity . this and the ensuing loss of spacelike commutativity of these algebras has been studied in @xcite . a detailed investigation of `` cutting holes '' into the circle of chiral nets and then producing new conformal nets from the haag dualization of the punctured nets can be found in @xcite . although such studies of chiral theories can not answer the above consistency problems in higher dimensional theories , they do supply valuable concepts and mathematical methods for further studies . as in many other areas of field theory , consistency problems find their satisfactory solutions only through mathematically controllable model constructions and the structural analysis is only a preparatory step in the classification and construction of models outside the standard lagrangian realm . its main purpose is to prepare our intuition in an area where it is presently underdeveloped or beset by prejudices . thinking of what could be the right setting for such constructions , only one idea seems to offer sufficient conceptual depth and mathematical power . this the modular localization approach based on the tomita - takesaki modular theory and using new physical concepts as `` polarization - free generators '' of wedge algebras . there are two ways of using that theory and both are independent of spacetime dimensions @xcite@xcite , although special situations in low dimensions greately favor their analytical control . one approach is based on the fact that the s - matrix is a relative modular invariant of wedge algebras and is behind the bootstrap - formfactor constructions in d=1 + 1 factorizing models . conformal theories do not have an s - matrix , but there are indications that their tcp operators ( @xmath148modular conjugations for the modular theory of the forward light cone algebra ) can be related to a simpler reference situation by generalized `` twist operators '' which contain the braid information . this idea should be first tested for chiral theories since even there exists up to date no systematic method to construct a model associated with a given plektonic superselection structure . in contradistinction to the standard methods which start from concrete observable algebras ( virasoro- , current- w - algebras ) and introduce the charge - carrying fields as intertwiners between the irreducible representations , the modular method aims directly at the charged fields . it is analogous to the wigner poincar group representation approach to free fields where the charged fields came first and the group invariant observable algebras ( which are more complicated since they involve composite fields ) were later introduces only for purposes of a better structural understanding . the modular program is in a certain sense a continuation of the wigner approach in the presence of interactions . presently 4-dimensional conformal theories are very scarce and in fact do not yet exist on the same conceptual and mathematical level as chiral models . perturbative ideas about their construction have not yet given satisfactory results . one idea which was mentioned in the second section as not leading to physically viable conformal theories was the use of perturbative lagrangian ads models in the @xmath1-@xmath2 connection . recently there have also been direct attempts to obtain nontrivial conformal theories within the family of perturbatively renormalizable supersymmetric yang - mills models with vanishing callan - symanzik @xmath149-function ( which is a necessary condition for conformal invariance in perturbation theory ) @xcite . there are many lowest order perturbative calculations for gauge invariant quantities ( often involving additional approximations ) , but they presently did not reach a level where they could be used for a test with the braiding ideas in this paper . even the conjectures on the existence of certain `` protected '' quantities are not very clear . protected objects by definition do not receive any contributions from the interaction i.e. they retain their zero order free field values . a chiral illustration of a protected quantity would be a nontrivial chiral theory which has an energy momentum tensor with the free field value c=1 of which there exist plenty . `` nontrivial '' here simply means that there are other unprotected quantities in the model . what is conceivable is a protection of an entire subalgebra as in this analogy . a protection of only certain correlation functions on the other hand ( say the two - or three - point normalization constant of a certain operator ) which does not follow a ( supersymmetric ) charge rule which characterizes a subalgebra of operators is hardly reconcilable with one s understanding of the omnipresence vacuum polarization in interacting qft . so the question of the quality of arguments about protection remains open . the crucial question for a future comparison with the timelike braid group structure is whether the perturbation theory can be pushed far enough so that one can extract anomalous dimensions . if it works it is probably limited to models with an abelian braid group phase , which in the analogy to chiral models would mean something analogous to the massless thirring model but not to e.g. a minimal model without a coupling deformation parameter . since it is not unrealistic to expect that the first conceptually and mathematically controllable 4-dimensional nontrivial models will be conformal ( because they stay close to free field theories without being identical to them ) , this line of research has a certain urgency and importance for the future of qft . another fascinating but at the same time very speculative idea which emerges from the present setting on the structure of higher dimensional conformal theories is the suggestion that there may be charge superselection rules and inner symmetries which do not respect the inner / spacetime factorization pattern of the coleman - mandula theorem ( and a fortiori not its prerequisites ) even after adjusting the prerequisites in order to incorporate supersymmetry . braided charges and their fusions certainly have a very different structure than multiplets of fields on which compact group act . in fact a timelike braided structure would inexorably remain linked with spacetime and dynamics . we have gotten so much used to think in terms of nonabelian symmetry groups that it is helpful to remind oneself from time to time that there actually does not exist a single exact nonabelian continuous flavor symmetry in nature . so to look for explanations of the observed regularities outside group theory may not be that absurd as it appears at first sight . but the idea that the observed regularities may be remnants of timelike braidings remains a farfetched wild speculation as long as there is no understanding of how conformal theories can be naturally related to particle models in a more controllable way than assigning to them a short distance universality class .. _ acknowledgements _ : i am indebted to karl - henning rehren for making some suggestions based on an earlier smaller and still somewhat opaque version . i thank lucio fassarel for some mathematical suggestion and for helping me with the pictures . i owe also thanks to francesco toppan for discussions which were helpful in the formulation . last not least i would like to thank detlev buchholz for an interesting exchange of emails on the issue of possible non - locality of timelike - defined double cone algebras . through this exchange of mails it became also clear to me that as a result of his familiarity with zero mass problems the possibility of a timelike braided structure was probably on his mind some time ago . b. schroer , _ particle physics and _ _ qft at the turn of the century : old principles with new concepts , _ ( an essay on local quantum physics ) , invited contribution to the issue 2000 of jmp , j. math . phys . * 41 * , ( 2000 ) 3801 i. e. segal , _ causality and symmetry in cosmology and the conformal group _ , montreal 1976 , proceedings , group theoretical methods in physics , new york 1977 , 433 and references therein to ealier work of the same author .

this paper describes the development and testing of a general relativistic magnetohydrodynamic ( grmhd ) code to study ideal mhd in the fixed background of a kerr black hole . the code is a direct extension of the hydrodynamic code of hawley , smarr , and wilson , and uses evans and hawley constrained transport ( ct ) to evolve the magnetic fields . two categories of test cases were undertaken . a one dimensional version of the code ( minkowski metric ) was used to verify code performance in the special relativistic limit . the tests include alfvn wave propagation , fast and slow magnetosonic shocks , rarefaction waves , and both relativistic and non - relativistic shock tubes . a series of one- and two - dimensional tests were also carried out in the kerr metric : magnetized bondi inflow , a magnetized inflow test due to gammie , and two - dimensional magnetized constant- tori that are subject to the magnetorotational instability .

astro-ph0210518

accretion into black holes is believed to account for a wide variety of astrophysical phenomena , from solar - mass black holes in x - ray binaries , to supermassive black holes in active galactic nuclei . the basic theory of black hole accretion was laid out more than thirty years ago in a number of now - classic papers ( e.g. novikov & thorne 1972 ; shakura & sunyaev 1973 ; lynden - bell and pringle 1974 ) . since then , theoretical progress has been steady , but perhaps at a slower rate than might have been hoped . the problem is complex , involving time - dependent , three - dimensional dynamics of magnetized plasmas in the relativistic potential of kerr black holes . solutions are impossible to obtain analytically beyond simplified cases that rely upon time - stationarity and spatial symmetry . because of this , numerical experiments must play an increasingly important role in driving theory . the importance of numerical simulations in the investigation of such complex physics has long been recognized , but until recently the necessary computational hardware was not available . we are now in a position to pursue full 3d simulations with general relativistic mhd codes . again , the basics were laid out thirty years ago by wilson ( 1972 ) who pioneered both hydrodynamic and magnetohydrodynamic ( mhd ) two - dimensional simulations of black hole accretion ( wilson 1975 ; 1977 ; 1978 ) . since then there have been a significant number of efforts by a number of groups to develop accretion simulations for both newtonian and relativistic gravitational fields , and for both hydrodynamics and magnetohydrodynamics . this paper details our current effort to develop a fully general relativistic mhd accretion code . why focus on general relativistic mhd ? in fact , for many applications newtonian gravity ( or a pseudo - newtonian mock - up of a black hole ) is sufficient . in some important cases , however , it is not . the detailed physics of gas interacting with a black hole provides , in principle , a rare test of strong - field general relativity , but only if we understand the unique properties of accretion in the fully relativistic case . distinctive and astrophysically interesting effects are expected from the kerr metric , including those due to the lack of a stellar surface , the presence of an innermost stable circular orbit , and the dragging of inertial frames . the need to include mhd in an accretion simulation is also now quite clear . magnetic fields play an essential role in the outward transport of angular momentum in accretion disks through the action of the magnetorotational instability ( mri ; balbus & hawley 1998 ) . this , in turn , means the simulations must be three - dimensional , for only in three dimensions can there be a self - sustaining ( and accretion - sustaining ) mhd dynamo . in this paper we describe an algorithmic approach to the three - dimensional equations of general relativistic mhd that is based upon techniques first developed for axisymmetric hydrodynamics around black holes ( hawley , smarr & wilson 1984a , 1984b ; hereafter hswa and hswb ) . the hydrodynamic portion of the code was recently tested and employed in three - dimensional simulations in the kerr metric by de villiers & hawley ( 2002 ) . here we cast the equations of relativistic mhd in a form that is compatible with the hsw formulation . we will describe how those equations are evolved numerically . we will also discuss a suite of tests suitable for relativistic codes , and show how our code performs against those tests . it is important to recognize that the development of numerical algorithms in any area of physics with few analytic solutions , and even fewer experimental observations , is likely to be difficult . progress will be made only by exploring a variety of approaches . such a situation necessarily precludes the development of only one numerical algorithm , or even a single code implementation of a promising algorithm . some approaches will have strengths for certain applications , but it is unlikely that one numerical technique will be optimal for all problems . in any case , experience gained , both with what works and what does not , will be essential as we work toward developing a consensus within the community . with this in mind , our own efforts have concentrated on extending the established general relativistic hydrodynamic approach of hsw , since our main area of interest , that of accretion flows , has been well handled ( in the hydrodynamic case ) by this solver , and by similar algorithms designed to perform newtonian mhd simulations . our test suite has been developed in conjunction with gammie , mckinney , & tth ( 2002 ) , who use an alternative scheme for general relativistic mhd . the plan of this paper is as follows . in 2 we write the equations of general relativistic mhd in the form that they will be solved in the code . in 3 we describe the numerical implementation of our algorithm . in 4 we describe the one - dimensional minkowski ( special relativistic ) test problems . in 5 we describe the test problems in the kerr and schwarzschild metrics , and present our test results . the summary and conclusions are presented in 6 . we wish to study the evolution of a magnetized fluid in the background spacetime metric of a kerr black hole . we work in boyer - lindquist coordinates , for which the line element has the form , @xmath1 in keeping with misner , thorne , & wheeler ( 1973 ) , we use the metric signature @xmath2 , along with geometrodynamic units where @xmath3 ; the black hole mass is unity , @xmath4 . the determinant of the 4-metric is @xmath5 , and @xmath6 where @xmath7 is the lapse function , @xmath8 , and @xmath9 is the determinant of the spatial @xmath10-metric . the state of the relativistic test fluid at each point in the spacetime is described by its density , @xmath11 , internal energy , @xmath12 , @xmath13-velocity @xmath14 , and isotropic pressure , @xmath15 , which is related to the first two scalars via the equation of state of an ideal gas , @xmath16 , where @xmath9 is the adiabatic exponent . the equations of general relativistic mhd are the law of baryon conservation , @xmath17 where @xmath18 is the covariant derivative , the conservation of stress - energy , @xmath19 where @xmath20 is the energy - momentum tensor for the fluid , and maxwell s equations , @xmath21 where @xmath22 is the electromagnetic field strength tensor , @xmath23 is the current @xmath13-vector , and @xmath24 the charge density . the dual tensor is defined as @xmath25 where @xmath26 $ ] is the contra - variant form of the levi - civita tensor . maxwell s equations are supplemented by the equation of charge conservation @xmath27 . the energy - momentum tensor consists of perfect fluid and electromagnetic parts , @xmath28 , and has the form @xmath29 following lichnerowicz ( 1967 ) and anile ( 1989 ) , we define the magnetic induction and the electric field in the rest frame of the fluid , @xmath30 we adopt the ideal mhd limit and assume infinite conductivity ( the flux - freezing condition ) , wherein the electric field in the fluid rest frame is zero , i.e. , @xmath31 . we combine the definition of the magnetic induction ( [ bdef.2 ] ) with ( [ dual.2 ] ) and the condition for infinite conductivity to obtain @xmath32 where @xmath33 $ ] . the orthogonality condition @xmath34 follows directly from ( [ bdef.2 ] ) and the anti - symmetry of @xmath22 . using these results , it is possible to rewrite the electro - magnetic portion of the energy - momentum tensor as @xmath35 where @xmath36 , and @xmath37 . the induction equation ( [ maxwell.2 ] ) is solved using the equivalent form @xmath38 working directly with @xmath39 replaces covariant derivatives with simple coordinate derivatives . this is the basis for the constrained transport ( ct ) formulation of evans & hawley ( 1988 ; hereafter eh88 ) for the induction equation . in ct we make the identification @xmath40 where @xmath41 are the ct magnetic field variables . with @xmath42 for one of the indices in ( [ ctinduction ] ) we get the evolution equations for @xmath41 , while permuting over the spatial indices yields @xmath43 which is the familiar divergence - free condition . the ct magnetic field is considered the fundamental expression of the magnetic field for evolution in the induction equation . equation ( [ fmunudef.2 ] ) relates the ct field to the field @xmath44 ( or @xmath45 ) which is fundamental in the definition of @xmath46 . the induction equation ( [ maxwell.2 ] ) can also be rewritten by substituting definitions , @xmath47 by expanding this equation using the product rule and applying the orthogonality condition ( [ bnorm ] ) , we obtain the identity @xmath48 proofs of the results for ( [ fmunudef.2 ] ) , ( [ tmunuem ] ) , and ( [ max2.b ] ) can be found in appendix a. in preparation for discretizing these equations we list the specific variables that will be used in the code . first , the transport velocity ( also known as the coordinate velocity ) @xmath49 is defined @xmath50 where @xmath51 , and @xmath52 is the relativistic gamma - factor , which can be expressed as @xmath53^{-1/2},\ ] ] a result which follows directly from the definition of transport velocity and the normalization condition on the @xmath13-velocity , @xmath54 the magnetic field of the fluid is described by two sets of variables , the constrained transport ( ct ) magnetic field , @xmath55 , and magnetic field @xmath13-vector @xmath56 . the latter is fundamental to the definition of the total four momentum , @xmath57 and the normalization condition @xmath58}^2 , \ ] ] which is algebraically equivalent to ( [ vnorm ] ) . finally , we define auxiliary density and energy functions @xmath59 and @xmath60 . the set of variables @xmath61 , @xmath62 , @xmath63 , @xmath55 , @xmath64 , and @xmath65 will be those on which the numerical scheme is built . since we are working in a coordinate basis , we also record here the identities for the divergence of a four - vector and a tensor @xmath66 @xmath67 the equation of baryon conservation ( [ barcons ] ) is unaltered by the presence of magnetic fields , and can be expanded in terms of the code variables to read ( keeping in mind that the metric is stationary ) , @xmath68 the equation of energy conservation is derived by contracting ( [ tmncons ] ) with @xmath69 , @xmath70 by using identity ( [ bident ] ) and the law of baryon conservation ( [ barcons ] ) , we recover the local energy conservation law @xmath71 which is unaffected by the presence of magnetic fields in the ideal mhd limit and corresponds to an equation of entropy conservation . applying the definition for the auxiliary energy function @xmath62 , the energy equation is rewritten as follows : @xmath72 which is the same as in hsw . the momentum conservation equations follow from @xmath73 using the definition of momentum @xmath74 the first term in the preceding expression can be rewritten as @xmath75 and simplified to @xmath76 , and also using @xmath77 , which holds for any symmetric tensor , we rewrite the momentum equation as @xmath78 to obtain the final form of the equations , multiply ( [ mom.2 ] ) by the lapse @xmath7 , split the @xmath79 index into its space ( @xmath80 ) and time ( @xmath42 ) components , and restrict @xmath81 to the spatial indices ( @xmath82 ) only : @xmath83 the @xmath81 index can be restricted to the spatial indices because the equation that arises from @xmath84 for the time components of momentum and magnetic fields is redundant , corresponding to a total energy conservation equation . in our formalism , we solve equation ( [ enfinal ] ) separately for the internal energy . the time component of @xmath63 is obtained from the normalization condition on the momentum ( [ momnorm ] ) , and on the relations between components of the magnetic field . we note that there are alternate ways to write this equation that , while analytically equivalent , correspond to significant differences in numerical implementation . one is to evolve directly the total stress energy @xmath85 and solve algebraically for the primitive variables . this approach is taken by koide , shibata , & kudoh ( 1999 ) , komissarov ( 1999 ) , and gammie , mckinney & tth ( 2002 ) . yet another alternative is to write the divergence of the electromagnetic stress - energy tensor in the form @xmath86 ( the familiar @xmath87 force from newtonian mhd ) and regard it as a source term for the hydrodynamic momentum evolution ; this approach was used by wilson ( 1978 ) . the equations of energy and momentum conservation have been expressed in terms of the magnetic field @xmath13-vector @xmath45 since this yields expressions structurally compatible with the existing hydrodynamic equations . this has important advantages in the numerical implementation of these equations . however , the induction equation is evolved with the constrained transport ( ct ) approach of eh88 using the variables @xmath88 . the practical consequence is that we must translate between the two sets of variables . working with the second of maxwell s equations , ( [ maxwell.2 ] ) , we have previously identified the space - space components of @xmath39 with the ct field @xmath41 . it is easy to show that we obtain the same result using ( [ max2.b ] ) , @xmath89 by splitting the above equation into two pieces and defining the ct magnetic field as @xmath90 it is possible to write @xmath91 the starting point for the eh88 ct scheme . in addition , the time - space components of @xmath92 can be obtained by applying the condition of infinite conductivity , @xmath31 , to obtain @xmath93 these can be identified with @xmath94 , the ct electromotive forces ( emfs , eh88 ) . the divergence - free character of the magnetic field is maintained by evolving the induction equation using these ct variables . hence the magnetic induction @xmath56 is derived from the ct magnetic field . we obtain the spatial components using ( [ fmunudef.3 ] ) . to obtain @xmath95 , we use the definition of the magnetic induction ( [ bdef.2 ] ) , @xmath96 and we expand this definition and substitute for the faraday tensor . this allows us to obtain the time - component of the magnetic field , @xmath97 this relationship can also be obtained from the orthogonality condition ( [ bnorm ] ) . using this result and the relations in ( [ fmunudef.3 ] ) we obtain , after some algebra , a useful expression for @xmath98 , @xmath99 the advantage of this form is that it guarantees that @xmath98 is positive . the derivation is supplied in appendix a. the grmhd code evolves time - explicit , operator - split , finite difference forms of equations ( [ masscons ] ) , ( [ enfinal ] ) , ( [ mom.3 ] ) , and ( [ ct.3b ] ) . variables are placed on a fixed spatial grid using boyer - lindquist coordinates . they are advanced in time over a timestep size @xmath100 that remains fixed for the duration of the simulation , and is determined by the extremal light - crossing time for a grid zone , as described in hswb . the evolution algorithm is a three - dimensional generalization of the solver described in hswb extended to include the contribution of magnetic fields . the hydrodynamic portion of the solver was described in de villiers and hawley ( 2002 ) . one timestep consists of three sequential sub - steps : the induction step , the transport step , and finally the source step . the induction step updates the magnetic field vector @xmath55 using the constrained transport framework of eh88 . this is discussed in greater detail below . after the update , the ct field is transformed to the magnetic field in the fluid rest frame , @xmath101 . velocities @xmath64 are computed using the normalization condition during the source step . three versions of grmhd have been used in development and testing : ( 1 ) a 1d version with minkowski metric to do alfvn and shock wave tests , described here in 4 , ( 2 ) a 2d axisymmetric version in the kerr metric to perform the tests described in 5 , and ( 3 ) the full 3d production version of the code . the 3d version of the code uses message passing parallelism with domain decomposition , where the global grid is partitioned into subgrids , with each subgrid assigned to a processor . data on each subgrid are evolved independently during a timestep and data on subgrid boundaries are exchanged when required through message - passing calls . this results in a highly scalable code that exhibits good speedup over the full range of practically realizable subgrids . the transport step , as its name implies , handles the transport terms in the continuity , energy , and momentum equations , @xmath102}_{transport } = -{1 \over \sqrt{\gamma}}\ , \partial_i\left[\sqrt{\gamma}\,{\cal{x}}\,v^i\right ] \ , , \ , \left({\cal{x } } \equiv d , e , s_j\right).\ ] ] other than the modified definition of the inertia , these expressions are completely analogous to those in hswb . the source step handles the remaining terms in the energy and momentum equations . the energy terms @xmath103}_{source } = -p\,\partial_{t } w - { p\over\sqrt{\gamma}}\,\partial_{i}\left[\sqrt{\gamma}\,w\,v^i\right]\ ] ] are evolved as described in hswb . the momentum source terms contain new elements , namely @xmath104}_{source } = -\alpha\,\partial_t\left[b_j\,b^t\right]+ { \partial_i \left[\sqrt{\gamma}\,\alpha\,b_j\,b^i\right ] \over \sqrt{\gamma } } -\left[{s_\epsilon\,s_\mu \over s^t}- \alpha\,b_\mu\,b_\epsilon\right]\ , { \partial_j g^{\mu\,\epsilon } \over 2}- \alpha\,\partial_j\left[p+{{\|b\|}^2 \over 2}\right].\ ] ] in the following expressions , superscript indices for the current time step are understood . the spatial differencing of some value @xmath105 is shown in compact form through the shift operator @xmath106 ; any averaging of zone- and face - centered quantities that may be required to properly center calculations is done as described in hswb . the time derivative of the magnetic fields is evaluated from the newly - obtained values from the induction step , and stored values from the previous time step , @xmath107 .\ ] ] the magnetic gradient term discretizes readily , @xmath108 as does the pressure gradient term , @xmath109 finally , the accelerations due to gradients in the metric , @xmath110\ , \partial_j\,g^{\mu\,\epsilon}.\ ] ] are computed using @xmath111 and @xmath112 obtained from @xmath113 using momentum normalization ( [ momnorm ] ) ; the metric derivatives , @xmath114 , are evaluated analytically over the whole grid and stored in memory at the beginning of the simulation . in the source step , after the pressure gradient and acceleration terms have been computed , the transport velocity @xmath115 , and the lorentz factor @xmath116 are obtained using velocity normalization . to guarantee that the normalization is well - behaved in evacuated zones , we work with the momenta @xmath117 rather than applying the normalization condition directly to the velocities . although the two normalizations are analytically equivalent , use of the momentum protects against numerical rounding errors that cause spurious floating point exceptions when extremely tenuous material is moving at velocities close to that of light . the source step also includes an artificial viscosity in both the energy and momentum equations to provide a mechanism to increase the entropy of the fluid in shocks . the artificial viscosity is unchanged from hswb , except for modifying the definition of the inertia to include the magnetic energy . the ct framework of eh88 is based on two basic ideas : the use of @xmath39 as the fundamental variable so that the induction equation reduces to simple partial derivatives , and a staggered - grid centering of terms so that the numerical divergence of the ct magnetic field is constrained to be zero . the ct magnetic field component @xmath41 is located on the @xmath82 face of a zone ( a cube in three dimensions ) , and the ct emfs , @xmath118 , that are differenced to evolve @xmath41 are located on the zone edges that define the @xmath82 face . the ct formalism imposes no stringent conditions on exactly how the emfs are computed . eh88 noted that some of the magnetic field variables appear in the discretrized equations for the emfs in the guise of transport terms , requiring that they be upwinded for stability , while others appeared as shear terms , apparently requiring no special treatment . eh88 used simple space - centered differencing for those terms , but stone & norman ( 1992 ) found that this prescription is inadequate . the results from simple alfvn pulse propagation tests where the alfvnic velocity exceeds that of the background fluid found substantial undamped dispersion error , resulting in excessive zone - to - zone noise . stone & norman ( 1992 ) proposed to calculate the ct emfs by solving a one dimensional analytic linear alfvn wave equation to obtain improved values of @xmath41 and @xmath119 in the construction of the emfs . this approach became known as the method of characteristics constrained transport ( mocct ) . since then several variants have been proposed ( e.g. hawley & stone 1995 ; clarke 1996 ) to further improve the computation of newtonian mhd . while the traditional approach to moc does not carry over readily to gr mhd , it is nevertheless possible to address the problem by using the same motivation that underlies mocct . instead of using an analytic alfvn wave solution we use a numerical solution of the simplified one - dimensional induction and momentum equations to obtain the improved emfs . our implementation of ct for the grmhd code is accomplished as follows . following the example of hawley & stone ( 1995 ) , the starting point for constructing , for instance , the @xmath120-emf , @xmath121 lies in obtaining the discretized solutions to the 1d incompressible mhd equations . on a staggered mesh , @xmath122 is located on zone edges in such a way that it is constructed from @xmath123- and @xmath124-components relocated in the @xmath124- and @xmath123-directions respectively . to take a specific example , we would like obtain an edge - centered estimate @xmath125 to construct the above emf . we do this by solving the 1d linearized induction equation in the @xmath123-direction ( i.e. @xmath126 ) for small transverse ( @xmath124 ) perturbations . this simplified induction equation has the form , @xmath127 by applying the usual differencing we obtain for the left - hand side , @xmath128 where the overbar denotes the original eh88 upwinding . we can now readily obtain estimates of the starred quantity , @xmath129 , \ ] ] where the factor of @xmath130 ensures proper time - centering of the starred quantity ( the starred values are centered between the @xmath131 and @xmath132 time levels ) . a similar approach can now be used to generate @xmath133 , and produce an improved estimate of @xmath122 . the procedure is just as easily applied to the other two emfs to complete the process for all ct variables . note that because of the fully covariant nature of the ct equations , this process applies to any coordinate system . for consistency , we also need to evolve the linearized momentum equation for the transverse velocities . care must be taken in deriving these equations since , in contrast to the ct induction equation , metric terms are explicitly involved in the expressions . continuing with the example for @xmath122 , and regarding the @xmath123 direction to be either @xmath134 or @xmath135 , and @xmath124 a direction transverse to this , we obtain @xmath136 where , @xmath137 ( the details are found in appendix a. ) note that the above equations are similar to the newtonian results , but with the relativistic factors @xmath138 and @xmath139 . by simply applying the usual differencing , however , we readily obtain expressions for an improved estimate of the transverse velocity , @xmath140 where the overbar denotes the edge - centered average , from which , @xmath141.\ ] ] in the kerr metric , the calculation in the @xmath142 direction is somewhat different due to the @xmath143 metric term . in this case we obtain @xmath144\,v^y - \kappa\,\,{\cal{b}}^\phi\partial_\phi\,{\cal{b}}^y \approx 0.\ ] ] where @xmath138 and @xmath139 are unchanged and @xmath145 combined with the starred quantities for the ct variables , we now have a complete set of variables from which to construct the emfs . the first series of tests is designed to verify code performance in the special relativistic limit . for this purpose we use a one - dimensional version of the code and the minkowski metric . the tests include alfvn wave propagation , fast and slow magnetosonic shocks , rarefaction waves , and both relativistic and non - relativistic shock tubes . many of these tests have appeared previously in the literature in one form or another . in particular , komissarov ( 1999 ) ( hereafter referred to as k99 ) presented a series of challenging test problems for a special relativistic mhd godunov scheme , and stone _ et al . _ ( 1992 ) presented a test suite for newtonian mhd codes . our presentation will conform to the formats of these published results to facilitate comparison . the propagation of a linear alfvn wave is simple in conception , but , as demonstrated by stone & norman ( 1992 ) , can be quite revealing as a numerical test . the special relativistic alfvn wave includes the displacement current normally neglected in newtonian mhd . this is what limits wave speeds to less than @xmath146 . we consider a fixed background magnetic field @xmath147 in minkowski spacetime with constant fluid velocity @xmath148 and a small transverse perturbation with velocity @xmath149 and ct magnetic field @xmath150 . the induction and momentum equations for the perturbed quantities have the form given in equations ( [ alfind ] ) and ( [ alfvel ] ) . equation ( [ alfvel ] ) can be combined with the induction equation ( [ alfind ] ) by taking partial derivatives [ @xmath151 for the former and @xmath152 for the latter ] to get an equation for @xmath149 , @xmath153\,v^y = 0\ ] ] which describes the propagation of alfvn waves with speeds @xmath154 where @xmath155 . ( the equation for @xmath150 can be obtained in a similar manner . ) figure [ valfven ] plots @xmath156 as a function of the background velocity of the fluid ( @xmath148 ) for four values of @xmath157 and illustrates several key properties of the wave speeds : ( 1 ) for @xmath158 , waves move with equal and opposite speed ( for all @xmath157 ) ; ( 2 ) @xmath156 is constrained to the upper bound of @xmath159 ; ( 3 ) for a given @xmath157 , there is a @xmath160 that yields a stationary wave , @xmath161 ; ( 4 ) for @xmath162 , both waves move in the same direction , with @xmath163 lagging behind the main fluid flow , and @xmath164 leading ; ( 5 ) as @xmath165 , all wave speeds converge , @xmath166 . we describe four tests , listed in table [ 1dpulse ] , carried out in a periodic box of width @xmath167 . in each case the fluid has constant density @xmath168 and energy density @xmath169 . individual tests are distinguished by the amplitude of the @xmath123-component of the ct magnetic field @xmath147 , given here in terms of @xmath170 , and the amplitude of the background @xmath123-component fluid velocity @xmath148 . the waves are initialized with a square pulse in the transverse velocity @xmath149 with amplitude @xmath171 , namely @xmath172 the transverse magnetic field is initially zero , @xmath173 . table [ 1dpulse ] summarizes the results and figure [ v1pulse ] overlays @xmath149 for each test at time @xmath174 at three grid resolutions , @xmath175 , @xmath176 , and @xmath177 zones . in test alf2 the fast and slow pulses have wrapped around and re - entered the grid from the left , the fast pulse having done so twice . in alf4 , the fast pulse has also wrapped around while the slow pulse is stationary . ( in this linear regime the pulses overlap cleanly provided that the spatial resolution is sufficiently high to minimize losses due to numerical diffusion . ) measuring from the center of each pulse ( originally at @xmath178 ) we verify that the observed pulse speeds agree with ( [ alfven.5 ] ) . with a stationary background , alf1 , the pulses separate from the initial perturbation with equal and opposite velocities and also with equal amplitudes such that @xmath179 . in a moving background , the splitting of the velocity pulses is no longer symmetrical , as is especially obvious in alf2 ( although @xmath179 still holds ) . the slow velocity pulse has a greater amplitude than the fast pulse . we find that the ratio of the velocity amplitudes scales inversely with their lorentz factors , @xmath180 it is easy to verify that this holds for each model in table [ 1dpulse ] . the pulses in the magnetic field variable , @xmath181 , on the other hand , are equal in magnitude and opposite in phase for all @xmath148 . the numerical values given in table [ 1dpulse ] can be verified by direct computation using results presented in appendix b. .[1dpulse ] results of 1d alfvn pulse tests . [ cols= " < , > , > , > , > , > , > , > , > , > , > " , ] to validate the general relativistic aspects of the code we require code tests that involve magnetic fields and the kerr geometry . test problems with analytic solutions include one dimensional magnetized bondi inflow , and a two - dimensional bondi flow along a split monopole field ; in these tests the field does not affect the dynamics of the problem . we also perform a one - dimensional magnetized inflow test problem where the field does act dynamically ( gammie 1999 ) . finally , we have considered two - dimensional magnetized constant-@xmath0 tori that are subject to the magnetorotational instability . while there is no analytic solution to this problem , the results can be compared to those obtained in non - relativistic simulations . the analytic expressions for the hydrodynamic bondi solution of hswa are unchanged in the presence of a radial magnetic field ( see proof in appendix a ) . however , this result does not necessarily hold for a numerical solution to the bondi problem . a non - trivial test of the magnetic components of the code consists , therefore , in maintaining the equilibrium bondi solution for _ any _ magnitude of radial magnetic field . the magnetized bondi problem is set up in the schwarzschild metric ( @xmath182 ) . the radial grid extends from just outside the event horizon at @xmath183 to @xmath184 ; the critical point is at @xmath185 . grids of @xmath186 , @xmath187 , @xmath188 , @xmath175 , and @xmath176 points are used , and the magnetic field at the critical point is set to @xmath189 , and @xmath190 , where the @xmath157 parameter sets the grid - averaged ratio of gas to magnetic pressure . the bondi solution is generated in the initialization routine by a numerical root - finder on a logarithmically scaled grid of @xmath176 points . this solution is then sampled at the required resolution for the particular test run ( to ensure proper alignment of the coarser grids ) . to complete the test , the unmagnetic case is also presented . the solution is allowed to evolve to a time sufficient to allow numerical convergence on each grid ( @xmath191 ) . figure [ bondierror ] shows the results of spatial accuracy tests . the spatial accuracy of variable @xmath192 is obtained from the @xmath193 norm over an interval @xmath194 , @xmath195 where @xmath196 denotes the converged solution , and @xmath197 the original solution , as described above . different regions of the flow are dominated by numerical errors either from a particular routine in the numerical solver or a particular code variable . here , the bounds of integration were chosen to range from @xmath198 to @xmath199 , which represents a region of the flow where pressure error dominates . figure [ bondierror ] is for the variables @xmath200 , and is typical of error curves for all code variables . the code is clearly first - order accurate for these variables . in the split monopole test , we initialize a 2d @xmath201 grid with the radial bondi solution , but the radial magnetic field has opposite polarity above and below the equator , hence establishing a current sheet along the equator . this test is made into a numerical stability test by applying small enthalpy perturbations to the initial state ( without this , the code is simply a set of 1d radial tests ) . the test is repeated with increasingly large values of magnetic field , until the code breaks . there is no known analytic solution for the evolution of a current sheet , but it can be used as a probe of the code s stability in the presence of strong , fluctuating current sheets . the split monopole test was carried out with @xmath202 at the critical point , and a grid of @xmath203 scaled logarithmically in both the @xmath134 and @xmath135 coordinates , with an initial @xmath204 random enthalpy perturbation . the test was allowed to run to @xmath205 ( @xmath206 free - fall times ) . for @xmath207 , the code variables show no visible change over the course of the run ; the enthalpy perturbations do not grow , nor do they trigger a numerical instability . for @xmath208 , the late - time solution shows evidence of reconnection at the equator near the horizon , and erratic behavior . the effective @xmath157 where reconnection occurs is of order @xmath209 . given the moderate resolution , the results are consistent with the spatial accuracy tests in that the region near the horizon requires high resolution in order to minimize numerical error . gammie ( 1999 ) describes a simplified 1d model for inflow from the inner edge of an accretion disk in the kerr geometry that serves as an excellent test problem for relativistic mhd . this solution is similar to the weber & davis ( 1967 ) rotating magnetized wind which was used as a test problem stone _ the model assumes that the inflow is cold ( @xmath210 ) , time - steady , purely radial , confined to the equatorial plane , and initiated from the edge of an accretion disk located at the marginally stable orbit ( @xmath211 ) . at @xmath211 the solution is assumed to satisfy the conditions @xmath212 and @xmath213 . the flow is characterized by a fast magnetosonic critical point . the numerical procedure to find a particular solution to this inflow is described in appendix d. our test consists in replicating the solution described by gammie _ ( 2002 ) for a flow near a kerr black hole with @xmath214 ( @xmath215 , @xmath216 , @xmath217 ) , with @xmath218 , @xmath219 , @xmath220 , @xmath221 . the critical point for this test case is @xmath222 . figure [ g99m1a ] shows a plot of the solution for this test problem on a logarithmically scaled grid of @xmath176 points . the inner edge of the grid is at @xmath223 and the outer edge of the grid is at @xmath224 . in boyer - lindquist coordinates , the event horizon can not be part of the computational domain because of the coordinate singularity . in the gammie inflow test , the marginally stable orbit , which is the formal outer boundary of the problem , is also excluded since the density is divergent as @xmath134 approaches @xmath211 from below . the plot overlays the late - time solution ( after @xmath225 time steps ) on the initial state , but the two curves are indistinguishable at the chosen plot scale . figure [ g99err ] shows the results of convergence tests carried out analogously to those for the bondi problem , with the range of integration going from @xmath226 to @xmath224 . here we have an example of a highly - resolved near - horizon region , and the convergence rate is close to second order . the final set of tests involve constant-@xmath0 disks ( @xmath227 , the specific angular momentum ) , solutions of the axisymmetric gr hydrodynamic equations described in hswa . here , we add weak poloidal magnetic field loops that overlay the hydrodynamic solution ; this will trigger the mri . we quantify the strength of the magnetic field through the @xmath157-parameter , the ratio of the volume - averaged gas pressure to magnetic pressure . the initial magnetic field is obtained from the definition of @xmath92 in terms of the @xmath13-vector potential , @xmath228 , @xmath229 restricting the field to poloidal loops is done by using @xmath230 , where @xmath231 using ( [ avec.1 ] ) , it follows that @xmath232 and @xmath233 . this choice of @xmath234 produces poloidal field loops that coincide with isodensity contours . the constant @xmath235 is set by the input parameter @xmath157 . the constant @xmath236 sets a suitable minimum density within the disk , and it is chosen to keep the initial magnetic field away from the outer edge of the disk . an axisymmetric ( 2d ) magnetized constant-@xmath0 torus was set up in both the schwarzschild and kerr metrics . for the schwarzschild case , the disk has a specific angular momentum @xmath237 , an initial pressure maximum at @xmath238 , and an orbital period at the pressure maximum of @xmath239 . for the kerr case , the black hole is maximally rotating , @xmath240 , and the disk has a prograde specific angular momentum @xmath241 , an initial pressure maximum at @xmath242 , and an orbital period at the pressure maximum of @xmath243 . this particular choice of parameters yields a disk that is similar to the schwarzschild case , although the equipotentials which define the overall disk structure are notably different , as can be seen in the left column of figure [ mri ] . for both tests the average field strength is @xmath244 , the initial state is perturbed with random @xmath204 enthalpy fluctuations . the simulations are run for @xmath206 orbits . the mri develops after a few orbits ( center column , fig . [ mri ] ) , and is soon fully developed ( right column , fig . [ mri ] ) . by the tenth orbit , the mri - induced turbulence has settled down considerably , as is to be expected since the imposition of axisymmetry precludes the development of the azimuthal modes which sustain the mri . the observed qualitative differences between the schwarzschild and kerr cases can be attributed to the different shapes of the potential cusp near the horizon . the density plots bear a strong resemblance to the previous axisymmetric mri studies of hawley ( 2000 ) in a pseudo - newtonian potential ; this indicates that the numerical solver is well behaved in the moderately strong gravitational fields that exist near the pressure maximum , where the implied code comparison is taking place . since mri studies are the main area of application of this code , it is reassuring that we can both trigger the mri in a general relativistic treatment of mhd , and that we qualitatively reproduce established results . ( 2002 ) obtain similar results with their mri torus test . in this paper we document the development of a general relativistic mhd code that is a direct extension of the general relativistic hydrodynamic solver of hawley , smarr and wilson ( 1984b ) . as before , the equations of relativistic mhd are written in a nonconservative form , supplemented here by the relativistic induction equation . the magnetic field is added to the definition of the inertia for the four momentum and the magnetic field adds several source terms to its evolution . the equations are evolved on a staggered grid using the time - explicit model of `` transport @xmath245 source . '' the induction equation is evolved using the constrained transport structure of eh88 . we implement a version of ct that is inspired by the method of characteristics technique employed in the popular zeus code ( stone & norman 1992 ) , yet is simpler to implement in a general relativistic framework . we have performed a number of 1d and 2d tests to validate the code and to note its limitations . the tests include a suite of non - relativistic and relativistic mhd shock tube tests , the fast and slow rarefaction and slow magnetosonic shock tests of komissarov ( 1999 ) , and simple alfvn pulse propagation . the formulation for these test and their results are documented to establish a basis for comparison against other existing and future codes . the main shortcoming of our code revealed by these tests is due to the already - known limitations of the artificial viscosity algorithm of the underlying hydrodynamic solver . improvements to this algorithm are an ongoing area of research , but is important to stress that these limitations do not pose an undue restriction on our main area of study : dynamical , mhd - driven accretion flows around spinning black holes . although the set of general relativistic black hole accretion test problems is limited , we have also verified that the code reproduces the standard results for bondi flow in the presence of magnetic fields , and have observed the triggering and evolution of the mri in thick accretion tori for both the kerr and schwarzschild black holes . the mri evolution can be compared qualitatively with pseudo - newtonian simulations ( e.g. hawley 2000 ) . one new test problem with an analytic solution is the magnetized gammie ( 1999 ) inflow test . this solution examines the behavior of a dynamically important magnetic field in the strong - field region of the kerr metric . the results of these tests provide evidence that this new solver is correctly reproducing magnetized flows outside the event horizon of a kerr black hole . this new gr mhd code should provide a valuable tool with which to study accretion flows in the kerr metric . some of the following proofs use two identities for contractions of the levi - civita tensor , @xmath246 @xmath247 and identities for the @xmath248-symbol , @xmath249 @xmath250 to prove ( [ fmunudef.2 ] ) , it is easiest to work backwards from the result . expand the right - hand side of ( [ fmunudef.2 ] ) using ( [ bdef.2 ] ) and ( [ dual.2 ] ) ; the resulting expression simplifies through identities ( [ levi.1 ] ) and ( [ delta3 ] ) , followed by the velocity normalization condition , infinite conductivity , and antisymmetry of @xmath22 : @xmath251 to prove ( [ tmunuem ] ) , expand the em component of tensor ( [ tmndef ] ) using ( [ fmunudef.2 ] ) , and simplify using identities ( [ levi.1])-([delta3 ] ) , followed by the velocity normalization condition , and orthogonality condition ( [ bnorm ] ) : @xmath252\\ \nonumber & = & -{1 \over 4\,\pi } \left(g^{\mu \gamma}\ , \delta^{\rho\,\sigma\,\nu}_{\delta\,\epsilon\,\gamma}\,b^{\delta}\,u^\epsilon\,b_{\rho}\,u_\sigma- { g^{\mu \nu } \over 2}\,\delta^{\rho\,\sigma}_{\delta\,\epsilon}\,b^{\delta}\,u^\epsilon \,b_{\rho}\,u_\sigma\right)\\ \nonumber & = & -{1 \over 4\,\pi } \left(-{g^{\mu\,\nu}\,b^{\alpha}\,b_{\alpha } \over 2}+ b^{\mu}\,b^{\nu}-b^{\alpha}\,b_{\alpha}\,u^\mu\,u^\nu\right)\\ \nonumber & = & { 1 \over 2}\,g^{\mu\,\nu}\,{\|b\|}^2+u^\mu\,u^\nu\,{\|b\|}^2 -b^{\mu}\,b^{\nu}.\end{aligned}\ ] ] to prove ( [ max2.b ] ) , the second of maxwell s equations , @xmath253 , can be rewritten by substituting definitions , and using identities ( [ levi.2 ] ) and ( [ delta2 ] ) : @xmath254 which simplifies , using ( [ delta2 ] ) , to @xmath255 . to prove ( [ bsqdef ] ) , we begin with an identity derived from the expression for @xmath52 ( [ gred ] ) , @xmath256 or @xmath257 now expand @xmath258 , @xmath259 and substitute @xmath260 for the spatial components of the magnetic field . making use of the above identity , it follows that @xmath261\,{(b^t)}^2 + 2\,g_{t \phi}\left [ { { \cal{b}}^\phi \over \sqrt{4 \pi}\,w\,\sqrt{\gamma}}+b^{t}\,v^\phi \right]\,b^t\\ \nonumber & + & { 2 \over \sqrt{4 \pi}\,w\,\sqrt{\gamma}}\,\left [ g_{rr}\,{\cal{b}}^r\,v^r + g_{\theta \theta}\,{\cal{b}}^\theta\,v^\theta+ g_{\phi \phi}\,{\cal{b}}^\phi\,v^\phi \right]\,b^t\\ \nonumber & + & { 1 \over 4 \pi\,w^2\,\gamma } \left[g_{rr}\,{({\cal{b}}^r)}^2 + g_{\theta \theta}\,{({\cal{b}}^\theta)}^2 + g_{\phi \phi}\,{({\cal{b}}^\phi)}^2 \right ] .\end{aligned}\ ] ] now substitute the expression for @xmath95 ( [ bdef.5 ] ) and simplify , noting that @xmath262 to get , after a few steps , expression ( [ bsqdef ] ) . to derive ( [ alfvel ] ) we first obtain the components of the fluid - frame magnetic field , which are , to leading order , @xmath263 and also note the approximate value for @xmath52 , @xmath264 we then expand the momentum equation and substitute the expressions for @xmath265 and @xmath95 . the left - hand side of the momentum equation reads @xmath266 - { g_{xx}\,g_{yy}\,{\cal{b}}^x\,{\cal{b}}^y\,v^x \over \alpha\,{\left(\sqrt{4\,\pi\,\gamma}\right)}^2 } \right ) \\ \nonumber & \approx & { g_{yy}\,(\rho\,h\,w^2+{\|b\|}^2 ) \over \alpha } \,\partial_t\,\left(v^y\right ) - { g_{xx}\,g_{yy}\,{\cal{b}}^x\,v^x \over \alpha\,{\left(\sqrt{4\,\pi\,\gamma}\right)}^2}\ , \partial_t\,\left({\cal{b}}^y\right ) .\end{aligned}\ ] ] recall that @xmath123 stands for either @xmath134 or @xmath135 , and @xmath124 stands for a transverse direction , either @xmath134 , @xmath135 , or @xmath142 . as for the left hand side , we neglect the purely geometric terms and retain only @xmath267 combine these results and replace @xmath268 with the results for the induction equation , @xmath269 by regrouping these terms and substituting the approximate expressions for @xmath52 and @xmath258 given above , equation ( [ alfvel ] ) follows immediately . a similar procedure yields expression ( [ alfvel_phi ] ) for transverse velocities when @xmath270 . to prove that the bondi solution is unchanged by the presence of a radial magnetic field , we revisit the calculations presented in hswa we are looking for a time - independent purely radial solution to the equations of grmhd in the presence of a constant ct radial magnetic field @xmath271 in the schwarzschild metric . we will work with the more general formalism of equations ( [ barcons ] ) and ( [ tmncons ] ) . the equation of continuity ( [ barcons ] ) is not dependent on magnetic fields , so @xmath272 leads to the same conserved quantity as in the hydrodynamic case , @xmath273 the energy - flux conservation equation is the @xmath42-component of ( [ tmncons ] ) , @xmath274 and contains an explicit dependence on the magnetic field . however , by expanding the @xmath275 component of the energy momentum tensor , @xmath276 and using equations ( [ fmunudef.3 ] ) and ( [ bdef.5 ] ) to obtain the fluid - frame magnetic field , @xmath277 we obtain after simple substitutions @xmath278 clearly , the magnetic contributions cancel ( this is simply restating the general result that there is no net force due to magnetic fields parallel to the direction of motion ) . we therefore recover the hydrodynamic expression , @xmath279 which leads to the same conserved quantities as in the hydrodynamic case , @xmath280 and @xmath281 to prove ( [ alfgam ] ) , we used the method of characteristics . we write the momentum and induction equations as a matrix equation , @xmath282 where @xmath283 and @xmath284 \ , , \ , b = \left [ \matrix { 1 & \chi \cr 0 & 1 \cr } \right ] \ , , \,\chi = -{\eta^2 v^x\over { \cal b}^x\ , ( 1+\eta^2 ) } .\ ] ] subject to the initial state @xmath285 we obtain the eigenvalues from the roots of @xmath286 , which are simply the expressions for @xmath156 , ( [ alfven.5 ] ) . to solve this system , we seek a matrix @xmath15 such that @xmath287 where @xmath288 and @xmath289 . the rows of matrix @xmath15 satisfy @xmath290 . it is straightforward to show that the matrix @xmath15 has the form , @xmath291 \ , , \ , \zeta = { { \cal b}^x ( 1+\eta^2)\over \eta\,\sqrt{\eta^2 + w^{-2 } } } .\ ] ] we obtain the results @xmath292 which yield @xmath293 as discussed in the text , the velocity amplitudes are clearly asymmetrical , except for the case where @xmath158 , and the magnetic field amplitudes are always equal , but opposite in sign . it is easy to show that the ratio of the velocity amplitudes is indeed the ratio of the gamma factors , @xmath294 . it is also possible to verify by direct computation that the amplitudes of the pulses given in the text agree with the above formulas . relativistic mhd shocks have been discussed by anile ( 1989 ) and komissarov ( 1999 ) . their work builds upon the monograph of lichnerowicz ( 1967 ) who systematically developed the relations for gr mhd shock invariants . we summarize here the relevant results from lichnerowicz , translating symbols to our own notation . we use these results to construct our shock tests . mhd shocks are discontinuities in one or more of the mhd variables whose propagation through spacetime lies on a hypersurface @xmath295 . it is useful in the analysis of these shocks to construct a @xmath13-vector @xmath296 normal to @xmath295 . shocks can be characterized by invariant scalar and @xmath13-vector quantities the jump conditions which are derived readily from the equations of grmhd . mhd shocks can be assigned to one of two categories , tangential and non - tangential , and this classification hinges on whether the invariants have non - zero projections onto the normal @xmath296 . slow and fast magnetosonic shocks are examples of non - tangential shocks , so it is this type of shock that is of interest here . the analysis of non - tangential shocks was shown by lichnerowicz to reduce to the study of five scalar quantities , @xmath11 , @xmath297 , @xmath298 , @xmath299 , and @xmath98 which are related through five positive - definite scalar invariants ( derivable from the jump conditions ) , @xmath300 & = & 0 , \\ \label{shkinvarb } \left[{\cal{b}}\right ] & = & \left[h \ , \eta \right ] = 0,\\ \label{shkinvarh } \left[{\cal{h}}\right ] & = & \left[{\eta^2 \over { \cal{a}}^2}- { \|b^2\| \over \rho^2}\right ] = 0,\\ \label{shkinvarl } \left[{\cal{l}}\right ] & = & \left[\alpha + { q \over { \cal{a}}^2}\right ] = 0,\\ \label{shkinvark } \left[{\cal{k}}\right ] & = & \left [ \left(\rho^2+{\cal{a}}^2\right ) { h^2 \over \rho^2}+ k^2 \left ( 2\,{h \over \rho } - { \cal{h } } \right)\right ] = 0 , \end{aligned}\ ] ] where @xmath301 = { \cal{x}}_{+ } - { \cal{x}}_{-}$ ] indicates the change in a quantity @xmath302 across the shock , and @xmath303 magnetosonic shocks are characterized by the parameter @xmath7 . the determination as to whether a magnetosonic wave is fast or slow hinges on the sign of the pre- and post - shock values of parameter @xmath7 . physically realizable shocks are characterized by @xmath304 . a slow shock has @xmath305 and a fast shock has @xmath306 . to construct an initial state , we solve the scalar invariants using the mathematica findroot function . first , we postulate a left state in the rest frame of the shock , where the normal vector @xmath296 has an especially simple expression ( @xmath307 ) . next , we set up the scalar invariants as a system to be solved by the findroot function , supplying an initial guess for the right state . this is an iterative procedure , requiring adjustments to the initial guess . once a solution is found , it is possible to construct a moving shock by repeating the above process with a small non - zero shock speed , for which the normal vector now has the form @xmath308 . once a solution is found the process is repeated , gradually increasing the shock speed and using the previous converged right state as a starting point for the solution at the new shock speed . this is a pragmatic solution to a problem that has a high degree of complexity , and it allows us to set up a family of closely related shock solutions to aid in testing . given that the findroot function has converged on a right state , it is possible to compute the parameter @xmath7 , and classify the shock . it is also prudent to verify that the left and right states are true solutions by verifying that they satisfy the shock invariants by direct computation . once this is done , the left and right states can be transferred directly to a numerical grid . the magnetized gammie inflow is characterized by three conserved quantities , @xmath309,\\ f_e & = & -2\,\pi\,r^2\,{t^r}_{t } = -2\,\pi\,r^2\,\left [ \rho\,u^r\,u_t+{g^{rr}\,g^{\theta \theta } \over 4\,\pi}\ , f_{r \theta}\,f_{t \theta } \right],\end{aligned}\ ] ] which are readily derived from the equations of baryon and momentum conservation . note that gammie writes @xmath310 . from the induction equation , @xmath311 we obtain @xmath312 , and @xmath313 which follows from the boundary condition . infinite conductivity ( @xmath314 ) yields @xmath315 finally , the velocity normalization condition yields @xmath316 where the mix of upper and lower indices was made to coincide with code variables . the critical point is a saddle point in the two - dimensional parameter space of the energy function @xmath317 . it is apparent that @xmath318 as given above is not a simple function of these parameters , and it is necessary to substitute the other expressions in order to obtain the desired form for @xmath318 . since the resulting expression for @xmath318 is quite lengthy , we will only sketch the procedure . the process of substitution and elimination described here was carried out using mathematica . first , the conserved quantity @xmath319 is used to obtain @xmath320 . next , the conserved quantity @xmath321 is used to isolate @xmath322 , @xmath323\,{1 \over \rho u^r}.\ ] ] now substitute the expressions for @xmath324 and @xmath325 expressed in terms of @xmath326 and @xmath327 , to yield an expression of the form @xmath328 where @xmath329 and @xmath330 are expressions involving the constants @xmath321 , @xmath319 , @xmath331 , and @xmath332 , as well as @xmath333 and metric terms . now substitute this expression for @xmath322 in the velocity normalization condition and `` solve '' the resulting quadratic for @xmath327 , denote this solution @xmath334 to emphasize that there are two distinct roots . finally , take the conserved quantity @xmath318 and substitute the expression for @xmath325 ( expressed in terms of @xmath326 and @xmath327 ) and then perform a final substitution for @xmath326 and @xmath327 by @xmath334 . simplify the resulting expression , which has the desired functional form : @xmath335 . the critical point must satisfy the conditions @xmath336 and the conserved quantity @xmath318 must also satisfy the constraint at the outer boundary @xmath337 these three conditions allow us to solve for @xmath338 , @xmath339 , and @xmath321 . in order to find the critical point , we will eventually use the findroot function of mathematica , but we first need to obtain a rough location for the saddle point . to do this , we produce a coarse contour plot over the range @xmath340 and a large range of @xmath333 , say @xmath341 for a given choice of root of @xmath334 . if the contour plot does not show evidence of a saddle point , we choose the other root . figure [ fesurf ] illustrates this for our specific test problem . the saddle point is shown to lie near @xmath342 . this is then used as an initial guess for the mathematica findroot function , from which the parameters used in the test were obtained . to generate an initial state within our code , we apply a numerical root - finder to the energy function @xmath335 ( coded using the mathematica fortranform command ) , and generate a numerical solution for @xmath333 ( using the parameters in the text ) . once @xmath333 is found , we construct @xmath327 and @xmath322 using the algebraic expressions obtained from mathematica ( also coded using fortranform ) . once these variables have been initialized the remainder of the code variables can be obtained in a straightforward manner . anile , a. m. 1989 , relativistic fluids and magnetofluids ( cambridge : cambridge university press ) balbus , s. a. , & hawley , j. f. 1998 , rev . phys . , 70 , 1 brio , m. , & wu , c. c. 1988 , j. comput . phys . , 75 , 400 clarke , d. a. 1996 , apj , 457 , 291 de villiers , j. p. & hawley , j. f. 2002 , apj , 577 , 866 evans , c. r. & hawley , j. f. 1988 , apj , 332 , 659 ( eh88 ) gammie , c. f. 1999 , apj , 522 , l57 gammie , c. f. , mckinney , j. , & toth , g. 2002 , apj , submitted hawley , j. f. 2000 , apj , 528 , 462 hawley , j. f. & stone , j. m. 1995 , comput . comm . , 89 , 127 hawley , j. f. , smarr , l. l. , & wilson , j. r. , 1984a , apj , 277 , 296 ( hswa ) hawley , j. f. , smarr , l. l. , & wilson , j. r. , 1984b , apjs , 55 , 211 ( hswb ) koide , s. , shibata , k. , & kudoh , t. 1999 , apj , 522 , 727 komissarov , s. s. 1999 , mnras , 303 , 343 ( k99 ) komissarov , s. s. 2002 , astro - ph/0209213 . lichnerowicz , a. 1967 , relativistic magnetohydrodynamics ( new york : benjamin ) lynden - bell , d. & pringle , j.e . 1974 mnras 168 , 603 misner , c. w. , thorne , k. s. , & wheeler , j. a , 1973 , gravitation ( san francisco : w.h . freeman ) norman , m. l. & winkler , k .- h . a. 1986 in astrophysical radiation hydrodynamics , k .- h . a. winkler & m. l. norman eds . ( dordrecht : reidel ) , 449 novikov , i.d . & thorne , k . s. 1973 in black holes les astres occlus , c. de witt ed . ( new york : gordon and breach ) , 346 ryu , d. , & jones , t. w. 1995 , apj , 442 , 228 ( rj95 ) shakura , n. i. & sunyaev , r. a. 1973 , a&a 24 , 337 stone , j. m. , hawley , j. f. , evans , c. r. , & norman , m. l. 1992 , apj , 388 , 415 stone , j. m. & norman , m. l. 1992 , apjs , 80 , 753 weber , e. j. , & davis , l. 1967 , apj , 148 , 217 wilson , j. r. 1972 , apj , 173 , 431 wilson , j. r. 1975 , seventh texas symposium on relativistic astrophysics eds : p.g . bergmann , e.j . fenyves , and l. motz , ann . sci.,262 , 123 wilson , j. r. 1977 , in proceedings of the marcel grossman meeting ed . r. ruffini ( amsterdam : north holland , 1977 ) , 393 wilson , j. r. 1978 , proceedings of international school of physics fermi course lxv , ed . r. giacooni and r. ruffini ( amsterdam : north holland ) , 644

the description of the inspiral of a stellar - mass compact object into a massive black hole sitting at a galactic centre is a problem of major relevance for the future space - based gravitational - wave observatory lisa ( laser interferometer space antenna ) , as the signals from these systems will be buried in the data stream and accurate gravitational - wave templates will be needed to extract them . the main difficulty in describing these systems lies in the estimation of the gravitational effects of the stellar - mass compact object on his own trajectory around the massive black hole , which can be modeled as the action of a local force , the _ self - force_. in this paper , we present a new time - domain numerical method for the computation of the self - force in a simplified model consisting of a charged scalar particle orbiting a nonrotating black hole . we use a multi - domain framework in such a way that the particle is located at the interface between two domains so that the presence of the particle and its physical effects appear only through appropriate boundary conditions . in this way we eliminate completely the presence of a small length scale associated with the need of resolving the particle . this technique also avoids the problems associated with the impact of a low differentiability of the solution in the accuracy of the numerical computations . the spatial discretization of the field equations is done by using the pseudospectral collocation method and the time evolution , based on the method of lines , uses a runge - kutta solver . we show how this special framework can provide very efficient and accurate computations in the time domain , which makes the technique amenable for the intensive computations required in the astrophysically - relevant scenarios for lisa .

0903.0505

one of the main sources of gravitational radiation for the future esa - nasa gravitational wave observatory , the laser interferometer space antenna ( lisa ) @xcite , is the capture and inspiral of a stellar - mass compact object ( sco ) , with masses in the range @xmath0 , into a massive black hole ( mbh ) in a galactic centre , with masses in the range @xmath1 . these systems are usually known as extreme - mass - ratio inspirals ( emris ) since the mass ratios involved are in the range @xmath2 . during the inspiral phase the system is driven by the emission of gravitational radiation , and hence there is loss of energy and angular momentum that makes the orbit shrink until the sco plunges into the mbh . it is expected that lisa will be able to detect @xmath3 emri@xmath4 @xcite up to distances within @xmath5 @xcite ( see also @xcite for more details on the astrophysics of emris ) . recently , it has been suggested @xcite that inspirals of scos into intermediate - mass black holes , with masses in the range @xmath6 and presumably located in globular clusters , could be detected by future second - generation ground interferometers like advanced ligo @xcite and advanced virgo @xcite . the mass ratios are in the range @xmath7 , and in consequence they are called intermediate - mass - ratio inspirals ( imris ) . it is expected that techniques to describe emris may also been used for imris , at least to a certain degree of precision . during the long inspiral , an emri will spend many cycles inside the lisa band , of the order of @xmath8 during the last year before plunge into the mbh @xcite . however , the gravitational - wave signals from emris will be buried in the lisa data stream with the instrumental noise and the gravitational wave foreground ( produced by compact binaries in the lisa band ) . to extract these signals and the relevant physical parameters that characterize them , we need to have a very precise _ a priori _ theoretical knowledge of the gravitational waveforms . this means to describe the inspiral of the sco taking into account the gravitational backreaction , that is , the influence of the sco gravitational field on its own motion . this is an interesting but difficult theoretical problem , and different methods have been developed to solve it ( see @xcite ) . while techniques for constructing templates good enough for detection are getting ready , mainly based on the use of the adiabatic approximation @xcite , methods to build templates good enough for extraction of physical information are not yet fully developed . the main difficulty being that one requires a more precise treatment of the self - gravity of the sco and its impact on the gravitational waveform . in relation to this fact , there is currently a significant activity on the study of the accuracy of the different types of adiabatic approximations that have been introduced @xcite . on the other hand , the _ self - force _ approach to the equations of motion @xcite ( see also @xcite ) is an step forward to a precise estimation of the radiation - reaction effects and in consequence , towards the construction of accurate waveform templates . in this approach the backreaction effects on the sco are described as the action of a local force , the _ self - force _ , which can be computed in terms of the perturbations generated by the sco with respect the mbh background spacetime . in practice , to compute the self - force one needs to regularize the perturbations , similarly as it happens in electromagnetism @xcite . to that end , a _ mode sum _ regularization scheme has been designed ( in the gravitational case it has been formulated in the lorenz gauge ) @xcite . it tells us how to subtract , mode by mode ( for a schwarzschild mbh the modes correspond to a harmonic decomposition of the perturbations ) , the singular part of the perturbations that does not contribute to the self - force . therefore , what we need is a method of computing the full retarded solution of the perturbative equations for applying the mode - sum regularization scheme and obtain in this way the self - force . the perturbative field equations are a set of ten linear partial differential equations ( pdes ) for the metric perturbations @xmath9 ( @xmath10 at linear order , where @xmath11 is the spacetime metric and @xmath12 is the background metric describing the mbh ) , which only in certain gauges ( e.g. , the regge - wheeler gauge @xcite ) can be decoupled . in any case , to solve them completely we need to resort to numerical methods . following the initial studies of black hole quasinormal modes , frequency methods were used successfully @xcite , and it was found they provide accurate results for emris with moderate eccentricities . however , the frequency domain approach has more difficulties with highly eccentric orbits , which are of interest for lisa , since one has to sum over a large number of modes to obtain a good accuracy , and convergence may be an issue . this has opened the door to time - domain methods , which are not affected much by the eccentricity of the orbit and may be more efficient for the case of high - eccentricity emris . in the last years there has been an intense activity on this front , both for a nonrotating background @xcite , and for a rotating background @xcite . the main drawbacks of time - domain methods have mainly two origins : ( i ) the fact that one has to resolve very different physical scales ( both spatial and temporal ) present in the problem due to the extreme mass ratios involved ( see , e.g. @xcite ) . that is , using a standard numerical discretization of the problem we are led to resolve the typical gravitational wavelengths ( comparable to the size of the mbh ) and , at the same time , scales in the vicinity of the sco , which are crucial for evaluating the self - force . this translates in a demanding requirement of computational resources . ( ii ) the fact that the sco is described as a point - like object . this introduces dirac delta distributions in the sco energy - momentum distribution that lead to loss of differentiability in the solution of the perturbative field equations . this fact can degrade the convergence properties of the numerical algorithms used . moreover , such a localized distribution of matter can also introduce spurious high - frequency modes that contaminate the numerical solution and , in consequence , degrade its accuracy . recently , there have been different proposals to improve the performance of time domain methods . barack and goldbourn @xcite have introduced a new technique to compute the scalar field generated by a pointlike scalar charge orbiting a black hole . this technique consists in subtracting from each azimuthal mode ( in the kerr geometry the field equations are not fully separable in the time domain and one has to tackle them in 2 + 1 dimensions ) of the retarded field a piece that describes the singular behavior near the particle . this is done through a careful analytical study of the scalar field near the particle , using a _ puncture _ scheme which resembles the puncture model used for simulations in numerical relativity @xcite . this technique has been extended to the electromagnetic and gravitational cases by barack , goldbourn and sago @xcite . on the other hand , vega and detweiler @xcite have introduced another new method for regularizing the solution of the field equations . their approach , tested on a simplified model of a charged particle orbiting a nonrotating black hole , regularizes the retarded field itself by identifying and removing first , in an analytical way , the singular part of the retarded field . this alternative approach to the mode - sum regularization scheme yields a finite and differentiable remainder from which the self - force can be computed . this remainder is the solution to a field equation with a nonsingular source , which avoid the problem ( ii ) above . finally , lousto and nakano @xcite have also introduced an analytical technique to remove the particle singular behaviour . their method is global and also produces a well behaved source for the field equations . whereas these new techniques help in dealing with problem ( ii ) above , they do not completely solve the problem ( i ) since the regular source terms that these new schemes produce still have associated with them a length scale ( or , from the numerical point of view , there are still special spatial resolution requirements associated with those source terms ) . in this paper , we introduce a new time - domain scheme towards the computation of the self - force which , for the case of a nonrotating black hole , eliminates completely any length scale associated with the sco . this is done by using multiple subdomains and locating the particle in the interface between two of them ( this has similarities with what was done in @xcite using the finite element method , where in one of the numerical schemes proposed the particle was located between two elements ) . in this way , the dirac delta distributions do not appear in our equations and the presence of the sco enters through the boundary conditions that communicate the solutions at the different subdomains . as a consequence , we are solving wave - type equations with smooth solutions , which avoid the problems described in ( ii ) ( preliminary results have been reported in @xcite ) . regarding ( i ) , we just need to provide the numerical resolution to describe the field near the particle , but not the particle itself , which makes the computation much more efficient . our numerical algorithms are based on the pseudospectral collocation ( psc ) method ( see , e.g. @xcite ) , which has been applied to numerical relativity @xcite , and recently it has also been used in @xcite for one - dimensional head - on collisions of black - holes . and very recently , in @xcite , a discontinuous galerkin method has been introduced and some gravitational waveforms for extreme - mass - ratio binaries are computed . this work uses similar techniques to the ones introduced in @xcite and the ones that we present here . in this paper , we describe a set of techniques and methods to use the psc for the computation of the self - force on a charged scalar particle in circular orbits around a non - rotating black hole . we also show some results of the numerical implementation . the organization of the paper is as follows : in section [ modeldescription ] we introduce the basics of the model of a charged scalar particle orbiting a nonrotating black hole , including the basic formulae for the computation of the self - force via the mode - sum regularization scheme . in section [ timedomainframework ] we introduce all the ingredients of a new time - domain numerical framework for the computation of the self - force in such scenario , from the mathematical foundations to the practical implementation of the computations . in section [ results ] we show the performance of a numerical code we have designed to implement the new scheme , and results of the computation of the self - force , in particular for the innermost stable circular orbit . in section [ discussion ] we draw conclusions from the results shown and discuss possible future avenues in the development of these techniques for the simulations of emris in relevant physical situations . throughout this paper we use the metric signature @xmath13 and geometric units in which @xmath14 . in this paper we present a new technique for the computation of the self - force . in order to simplify things we focus in the particular case of a charged scalar particle in circular geodesics around a non - rotating mbh . in this simplified model , the spacetime metric is not dynamical , in the sense that it is not affected neither by the particle nor by the scalar field that it generates . in our case , since we are considering non - rotating bhs , the metric is the schwarzschild metric , which can be written as follows : @xmath15 where @xmath16 are the so - called schwarzschild coordinates , @xmath17 ( where @xmath18 is the bh mass ) , and @xmath19 is the _ tortoise _ coordinate , given by : @xmath20 in this geometry we assume there is a particle with scalar charge @xmath21 , associated to a scalar field @xmath22 . this particle is orbiting the bh , and in doing so generates scalar field @xmath23 , which in turn influences the particle trajectory . that is , the particle motion is affected by the field created by itself . in this way , this model contains all the ingredients of the gravitational case , in which the particle motion is influenced by its own gravitational field . the equation for the scalar field is then ( see , e.g. @xcite ) : @xmath24 that is , a wave - type equation with a source term that describes the particle energy density due to its scalar charge . in this equation , the spacetime metric @xmath11 is the schwarzschild metric ( [ schmetric ] ) , @xmath25 denotes the associated canonical connection ; @xmath26 denotes proper time associated with the particle along its timelike worldline @xmath27 , which we denote by @xmath28 ; @xmath29 is the invariant dirac functional in schwarzschild spacetime , which is defined by the relation @xmath30 and the equivalent one for the other argument . in this relation , @xmath31 denotes the metric determinant . taking into account the properties of @xmath29 , it follows that the source term in the scalar field equation ( [ geo ] ) only has support on the particle worldline @xmath27 . the equations of motion for the particle trajectory ( @xmath28 ) that one would obtain from energy - momentum conservation are : @xmath32 where @xmath33 and @xmath34 are the particle mass and 4-velocity , respectively . however , this expression is a formal one due to the fact that the force @xmath35 diverges on the particle worldline @xmath27 ( see @xcite for a derivation of the _ regularized _ equations of motion ) . an analysis of the solutions of ( [ geo ] ) and ( [ particlemotion ] ) reveals ( see for details @xcite ) that the gradient of the field , @xmath36 , can be split into two parts @xcite : a singular piece , @xmath37 , which contains the singular structure of the field and satisfies the same field equation , that is , @xmath38 and a regular part , @xmath39 , which satisfies an homogeneous wave equation @xmath40 and which is solely responsible of the deviation of the particle from geodesic motion around the bh , @xmath41 this part is associated with the tail part of the scalar field . this is the analogous equation to the _ misataquwa _ equation of the gravitational case ( see @xcite ) , and @xmath42 is called the _ self - force _ on the particle . in order to solve the equation for the scalar field [ eq . ( [ geo ] ) ] it is very convenient to take advantage of the spherical symmetry of the schwarzschild spacetime and decompose @xmath23 in scalar spherical harmonics , @xmath43 ( see appendix [ sphericalharmonics ] ) , @xmath44 where the harmonic numbers @xmath45 take the usual values : @xmath46 , and @xmath47 . since we want to compute the self - force on the particle for geodesics orbits , and since geodesic motion takes place on a plane in schwarzschild geometry , we will assume , without loss of generality , that the plane is given by @xmath48 . we will also parameterize the motion of the particle in terms of the coordinate time @xmath49 , instead of proper time @xmath26 . that is , the particle world - line , @xmath27 , will be given by @xmath50 . taking this into account , we can introduce the expansion ( [ scalarfieldharmonics ] ) into the scalar field equation ( [ geo ] ) and find that the equations for the different harmonic coefficients @xmath51 decouple and have the form of a 1 + 1 wave - type equation : @xmath52 where @xmath53 is the regge - wheeler potential for scalar fields on the schwarzschild geometry , given by @xmath54\,,\end{aligned}\ ] ] and @xmath55 is the coefficient of the singular source term generated by the particle : @xmath56 where an overbar denotes complex conjugation . on a hypersurface @xmath57 we can prescribe initial data for @xmath58 , @xmath59 , and then find the corresponding solution ( given that @xmath23 satisfies a wave equation , the problem is well - posed ) . there are a couple of comments in order : first , the solution found in this way corresponds to the @xmath45 contribution to the full retarded solution . second , this solution will be finite and continuous at the particle location , but it will not be differentiable at the particle location in the sense that the radial derivative from the left and from the right of the particle yields different values . moreover , the sum of the multipole coefficients over @xmath60 will diverge at the particle location . this can be fixed by extracting , multipole by multipole , the singular part of the scalar field . since we are interested on regularizing the self - force , which is defined at the particle location , let us introduce first a multipolar decomposition of the gradient of the scalar field @xmath61 so that the gradient of the retarded field is given by @xmath62 obviously , @xmath63 also diverges at the particle worldline although the @xmath64 are finite . following the discussion of the previous section , the gradient of @xmath23 , and hence the self - force , can be regularized by splitting the full retarded field into a singular part and a regular part ( see , for instance , @xcite ) , such that they satisfy equations ( [ eqsingular])-([regparticlemotion ] ) . that is , the regular part of the gradient of the scalar field will be given , on the particle location , by @xmath65 the singular field @xmath66 is known in a neighborhood of the particle worldline . in particular , the multipoles of the singular part of the gradient of the scalar field at the particle worldline are given by @xcite : @xmath67\ , , \label{l_sing}\end{aligned}\ ] ] where @xmath68 , @xmath69 , @xmath70 , @xmath71 , are called the regularization parameters @xcite . they are independent of @xmath60 , but depend on the particle dynamics . the singular part corresponds to the first three terms which lead to quadratic , linear , and logarithmical divergences when we sum over @xmath60 . the remaining terms form a convergent series that does not contribute to the self - force ( each of them ) . in our approximate calculations , we maintain the @xmath71 term as it accelerates the convergence of the series as we increase the number of multipoles included @xcite . for the case of interest of this work , circular geodesics , the non - vanishing regularization parameters are @xmath72 , @xmath73 , and @xmath74 @xcite , which can be written as follows : @xmath75 @xmath76\ , , \label{br}\end{aligned}\ ] ] @xmath77 where @xmath78 is a sign that takes the value @xmath79 when the limit in equation ( [ l_sing ] ) is performed from the _ right _ ( @xmath80 ) and @xmath81 when the limit is performed from the _ left _ ( @xmath82 ) . the coefficients @xmath83 can be computed in terms of hypergeometric functions as follows ( see @xcite , equation ( 15.1.1 ) ) : @xmath84 since the only non - vanishing component of the regularization coefficients is the radial one , this is the only component of the self - force that is actually singular , and hence the only one to be regularized . in this way the regularized self - force is : @xmath85 in this section we introduce all the ingredients of a new time - domain method to evolve the equation(s ) for a scalar field generated by a charged particle orbiting a nonrotating black hole , and also to compute the self - force from the evolved solution . we are going to introduce some mathematical developments needed in order to compute , using numerical algorithms based on the psc method , the self - force on a charged particle on circular geodesics . we discuss first the particular formulation that we implement numerically . since our numerical framework deals with multiple domains it is useful to adopt a first - order formulation of the scalar field equation . the main reason is that first - order formulations of pdes can be adapted to the hyperbolic character of the underlying equation , and hence they are suitable for a correct communication between domains . we perform a first - order reduction of ( [ master ] ) by introducing the following new variables @xmath86 the evolution equations for @xmath87 , that is for a given harmonic @xmath45 , written in matrix form , are : @xmath88 where @xmath89 and @xmath90 it can be seen that ( [ evosystem ] ) is a first - order symmetric hyperbolic system of pdes , that is , it has the maximum degree of hyperbolicity , as expected of a system that comes from the reduction of a wave - type equation . the characteristic structure is described by the matrix @xmath91 . given that discontinuities in the solution are only allowed across characteristics , and given that due to the singular character of the source some of our variables will have jumps across the radial particle location , it is important to study these jumps using the formulation just introduced . to approach this question it is convenient to divide the spatial domain ( the radial direction as parametrized by the coordinate @xmath19 ) into two disjoint regions , one to the left of the particle ( @xmath92 ) , and one to the right of the particle ( @xmath93 ) . then , we can write the solution of the equations ( [ evosystem ] ) in the form ( see also @xcite ) : @xmath94 where @xmath95 denotes the heaviside step function . introducing ( [ globalsolution ] ) into ( [ evosystem ] ) we can derive the jump across the particle location , defined as @xmath96^{}_{p } } = \mathop{\lim } \limits_{{r^{\ast}}\to { r^{\ast}}_{p}}\lambda^{}_{+}(t , { r^{\ast } } ) -\mathop{\lim } \limits_{{r^{\ast}}\to { r^{\ast}}_{p}}\lambda^{}_{-}(t , { r^{\ast}})\,,\end{aligned}\ ] ] for the different variables of our problem . we find the following result : @xmath97^{}_{p } } = 0\,,~~ { \left[\,\phi^{m}_{\ell}\,\right]^{}_{p } } = 0\,,~~ { \left[\,\varphi^{m}_{\ell}\,\right]^{}_{p } } = \frac{s^{m}_{\ell}}{f(r_p)}\ , . \label{jump}\end{aligned}\ ] ] these are the conditions we have to impose in our numerical evolution algorithm . in practice , these conditions have to be imposed on the _ characteristic _ fields , the fields associated to the characteristics . in our problem there are two different types of characteristics : ( i ) the @xmath98 surfaces , and ( ii ) the null surfaces @xmath99 . the associated characteristic fields are @xmath100 , and @xmath101 , respectively . finally , our system of equations has to be complemented with initial conditions and boundary conditions . the initial conditions , since we are dealing with a first - order system of pdes , consist in prescribing @xmath102 . we need boundary conditions at the ends of the spatial domain . the physical spatial domain is @xmath103 , with @xmath104 corresponds to the horizon location , while @xmath105 corresponds to spatial infinity . in our numerical computations we will take a _ truncated _ domain : @xmath106 $ ] , and therefore we need to prescribe outgoing boundary conditions at the ends of that domain ( also called _ absorbing _ boundary conditions ) . as an approximation we take the conditions that are exact for the case in which there is no potential , that is @xmath107 & & \phi^{m}_{\ell}(t,{r^{\ast}}_{{{\mbox{\tiny i } } } } ) + \varphi^{m}_{\ell}(t,{r^{\ast}}_{{{\mbox{\tiny i } } } } ) = 0\ , . \label{outgoingbcs}\end{aligned}\ ] ] this is the leading approximation for the outgoing boundary conditions . they can be improved by analyzing the solution near @xmath108 . however , we can always take values of @xmath109 such that the boundaries are not in causal contact with the particle location , avoiding contamination of the solution due to propagation of unphysical modes from the boundaries . we now describe in some detail the numerical techniques we use to solve for the full retarded scalar field and to compute from it the regularized self - force . in order to solve for the pdes that describe the scalar field , equations ( [ evosystem ] ) , we use the psc to discretize in space . once this is done we obtain a system of ordinary differential equations ( odes ) that can be solve by using the method of lines ( see , e.g. @xcite ) applying a convenient ode solver . in general , spectral methods can approximate solutions of pdes by expanding the variables in a given basis of functions , and then by using an appropriate criterium that forces this expansion to approach the exact solution as we increase the number of functions included in the expansion . in the case of the psc method , the criterium consist in imposing the solution exactly at a set of _ collocation _ points ( see , e.g. @xcite , for details on the psc method ) . in our work , we use the chebyshev polynomials , @xmath110 ( @xmath111 $ ] ) , as the basis functions ( see appendix [ chebyshev ] for definitions ) , and for the collocation points we take a _ lobatto - chebyshev _ grid . if we want to use a grid of @xmath112 collocation points , the lobatto - chebyshev grid is made out of the zeros of the following polynomial ( see appendix [ chebyshev ] ) : @xmath113 where the prime indicates differentiation with respect to @xmath114 . the zeros can be written as follows : @xmath115 now , we need to map the domain in which the chebyshev polynomials are defined ( we will call it the _ spectral domain _ ) to the physical domain . before doing that , we have to mention that for computational reasons that we discuss later , we split the computational domain , @xmath116 $ ] , into a number of subdomains ( @xmath117 ) : @xmath118\,,\ ] ] where @xmath119 and @xmath120 are the left and right boundaries of the subdomain @xmath121 . they are disjoint subdomains , that is @xmath122 ( see figure [ multidomain ] and the caption there ) . ( this coordinate will have a finite value and is meant to approach the bh horizon at @xmath123 ) and @xmath124 ( this coordinate will have too a finite value and is meant to approach spatial infinity at @xmath125 ) . the particle is at the boundaries of two different boundaries ( which are identified ) , which have coordinates @xmath120 and @xmath126 that satisfy the relation ( [ particleatinterface]).,scaledwidth=50.0% ] we apply the psc method to each subdomain , in the sense that our variables have different expansions in chebyshev polynomials in each subdomain . these different expansions are related by using appropriate boundary conditions that we discuss in section [ evolutionalgorithm ] . to apply the psc method to each subdomain , we map the physical subdomain @xmath127 ( @xmath128 ) to the spectral domain , @xmath129 $ ] , using a linear mapping @xmath130 \longrightarrow \left[-1,1\right]\,,$ ] with @xmath131 the inverse correspondence is another linear mapping , @xmath132 \longrightarrow \left [ { r^{\ast}}_{a , l } , { r^{\ast}}_{a , r}\right]\,,$ ] with @xmath133 the variables of our problem are arranged in a vector , @xmath134 . then , at a given domain @xmath127 ( @xmath128 ) , we have the following spectral expansion of our variables : @xmath135 where the @xmath136 are ( time - dependent ) vectors that contain the spectral coefficients of the expansion of our variables . in the psc method , we have also a _ physical _ expansion , which looks as follows : @xmath137 where @xmath138 are the _ cardinal _ functions @xcite associated with our choice of basis functions ( chebyshev polynomials ) and set of collocation points ( lobatto - chebyshev grid ) [ see appendix [ chebyshev ] for details ] . the cardinal functions have the following remarkable property : @xmath139 in this way , the time - dependent ( vector ) coefficients , @xmath140 , of the expansion ( [ physicalrepresentation ] ) are the values of our variables at the collocation points @xmath141 these are the variables that one looks for in the psc method . the spectral ( [ spectralrepresentation ] ) and physical ( [ physicalrepresentation ] ) representations are related via a matrix transformation ( see appendix [ chebyshev ] ) . the computations ( float - point operations ) required to change representation scale , with the number of collocation points , as @xmath142 , as it can be deduced from the fact that it is a matrix transformation . however , using the change of spectral coordinate given in equation ( [ changespectralcoord ] ) , we can perform the change of representation by means of a discrete fourier transform using a fast - fourier transform ( fft ) algorithm . in our numerical codes , we use the routines of the fftw library @xcite . then , the number of computations required for a change of representations scales as @xmath143 with the number of collocation points . changing between representations is useful in order to perform some operations . for instance , differentiation is easier in the spectral representation , so we can transform from the physical to the spectral representation , compute derivatives there , and finally transform back to the physical representation . in the case of a chebyshev psc method , the differentiation process can be described by the following scheme @xmath144 where @xmath145 are the spectral coefficients associated with the spatial derivative , and their relation to the coefficients of the variables , @xmath146 , is given by ( see , e.g. @xcite ) @xmath147 and @xmath148 1 & \text{otherwise\ , . } \end{array}\right . \label{cns}\end{aligned}\ ] ] in the psc method , we find a discretization of our system of equations ( [ evosystem ] ) by imposing them at every collocation point . in practice , this is done by introducing the expansion ( [ physicalrepresentation ] ) into the equations ( [ evosystem ] ) , and then we evaluate the result at every collocation point of our chebyshev - lobatto grid ( [ chebyshevlobattogrid ] ) . we obtain a system of odes for the variables @xmath149 @xmath150 where the dot denotes differentiation with respect to the time coordinate @xmath49 , and @xmath151 has to be interpreted according to the scheme in equation ( [ pscdifferentiation ] ) . in this section we discuss the details of the time evolution of the discrete equations we derived in the previous section . in particular , we describe how to introduce the particle in the multi - domain psc method that we propose . the main argument is the following : if we put the particle inside one of the subdomains @xmath121 we are introducing in the equations ( [ evosystem ] ) a singular term that , in the case of a scalar charged particle , will produce solutions that are not differentiable ( in the sense that the derivative with respect to @xmath19 is not single - valued at the particle location ) . that is , our solution will not be smooth and hence we can not expect the psc method to converge exponentially to the true solution . that would spoil one of the main motivations for using this numerical technique , that is , the accuracy that the psc method provides . to avoid this ( and this is part of the reason for using a multi - domain scheme ) we put the particle at the interface between two subdomains . if the subdomains are @xmath121 and @xmath152 , we have : @xmath153 since we have restricted ourselves to the case of circular orbits , once the grid has been set we do not need to change it during the evolution . in the case of generic orbits , in order to maintain the particle at the interface between two subdomains we have either to make a coordinate change or implement a moving grid scheme ( a similar situation was confronted in @xcite ) . we discuss this further in section [ discussion ] . for each subdomain @xmath121 ( @xmath128 ) , we evolve the set of odes ( [ odes ] ) independently . since we are locating the particle at the interface between two subdomains , the last term in ( [ odes ] ) , @xmath154 , does not appear . this term is the source term that accounts for the particle s energy density [ see equation ( [ source ] ) ] . the main implication of this setup is that we have to solve the homogeneous field equations , which are equations with smooth solutions and hence , the advantages of the psc method are preserved . then , in our framework , the contributions of the particle to the solution appear as boundary conditions between the subdomains . in summary , we evolve the ( homogeneous ) field equations for each subdomain independently and connect their solutions by boundary conditions at the interfaces . the equations are evolved using a runge - kutta ( rk ) solver ( see @xcite ) , typically a rk4 algorithm . the key point in the evolution is the imposition of boundary conditions at the interfaces between subdomains . there are two possible situations for a given interface : ( i ) the particle is not there . in this case we only have to impose the continuity of the solution [ that would correspond to imposing the junction conditions given in equation ( [ jump ] ) with zero right - hand sides ] . ( ii ) the particle is there . in this case we have to impose the junction conditions in equation ( [ jump ] ) where @xmath55 is given in equation ( [ source ] ) . the question is then how to impose these boundary conditions numerically within our framework . there are different ways of doing this , and in this work we have chosen to impose this boundary conditions at the interfaces between subdomains in a dynamical way , via _ penalty _ terms . the penalty method is a well - known technique and has been applied to several numerical schemes to solve pdes ( for the psc method , see @xcite and references therein ) . it is relatively simple to implement it for elliptic - type problems , that is , for problems that do not require evolution _ in time_. for time - dependent problems , it can also be implemented but the result is not always numerically stable in our framework , the implementation of the penalty method goes as follows : first , let us consider the two domains around the particle s location , say @xmath121 and @xmath152 , so that equation ( [ particleatinterface ] ) holds . then , let us consider the solutions at these two subdomains , @xmath155 and @xmath156 . the equations for the inner points are just equations ( [ odes ] ) ( with @xmath157 ) , and the equations for the nodes at the interface between these subdomains , @xmath120 and @xmath126 ( which are identified ) , are modified in the following way : for the subdomain @xmath121 ( we have simplified the notation by dropping the harmonic indices @xmath60 and @xmath33 ) @xmath158\ , , \label{penaltyr1}\\ \partial^{}_t \phi^{}_{a , r } & = & \partial^{}_{{r^{\ast } } } \varphi^{}_{a , r } - v^{}_{p}\,\psi^{}_{a , r } -\frac{\tau^{a , r}_{\phi}}{2}\left\ { \phi^{}_{a , r}+\varphi^{}_{a , r}\right . \nonumber \\ & - & \left . ( \phi^{}_{a+1,l } + \varphi^{}_{a+1,l } ) - { \left[\,\varphi\,\right]^{}_{p } } \right\}\ , , \label{penaltyr2 } \\ \partial^{}_t\varphi^{}_{a , r } & = & \partial^{}_{{r^{\ast}}}\phi^{}_{a , r } -\frac{\tau^{a , r}_{\varphi}}{2}\left\{\phi^{}_{a , r } + \varphi^{}_{a , r } \right.\nonumber \\ & - & \left . ( \phi^{}_{a+1,l } + \varphi^{}_{a+1,l } ) - { \left[\,\varphi\,\right]^{}_{p } } \right\}\ , , \label{penaltyr3}\end{aligned}\ ] ] and for the subdomain @xmath152 @xmath159\ , , \label{penaltyl1 } \\ \partial^{}_t\phi^{}_{a+1,l } & = & \partial^{}_{{r^{\ast}}}\varphi^{}_{a+1,l } - v^{}_{p}\ , \psi^{}_{a+1,l}-\frac{\tau^{a+1,l}_{\phi}}{2}\left\ { \phi^{}_{a+1,l}\right . \nonumber \\ & - & \left . \varphi^{}_{a+1,l } - ( \phi^{}_{a , r } - \varphi^{}_{a , r } ) - { \left[\,\varphi\,\right]^{}_{p } } \right\}\ , , \label{penaltyl2 } \\ \partial^{}_t\varphi^{}_{a+1,l } & = & \partial^{}_{{r^{\ast}}}\phi^{}_{a+1,l } -\frac{\tau^{a+1,l}_{\varphi}}{2}\left\{\phi^{}_{a+1,l } -\varphi^{}_{a+1,l}\right . \nonumber \\ & - & \left . ( \phi^{}_{a , r } - \varphi^{}_{a , r } ) - { \left[\,\varphi\,\right]^{}_{p } } \right\}\ , , \label{penaltyl3}\end{aligned}\ ] ] where @xmath160 , the quantity @xmath161^{}_{p}}$ ] is the analytic jump imposed by the dynamics and given by equation ( [ jump ] ) , and @xmath162 and where @xmath163 and @xmath164 are ( constant ) penalty parameters . the structure of the penalty terms in equations ( [ penaltyr1])-([penaltyl3 ] ) obeys the following rationale : the main idea behind of the penalty terms is to drive the dynamical system to satisfy a set of conditions that are not part of the original evolution equations ( like constraints on the variables that have to be satisfied for all times or boundary conditions ) , and the strength of the driving ( penalty ) terms is controlled by the penalty parameters @xmath163 and @xmath164 . in our case , in section [ mathematicalstuff ] , we mentioned the fact that the junction conditions ( [ jump ] ) have to be imposed on the characteristic field of our system of pdes , and these characteristic fields and their associated characteristic surfaces where given there . therefore , we have constructed the penalty terms so that the junction conditions that are satisfied are those corresponding to the characteristic fields , that is [ restoring the @xmath45 indices ] : @xmath97^{}_{p } } = 0\,,~~ { \left[\,\phi^{m}_{\ell}\pm \varphi^{m}_{\ell}\,\right]^{}_{p } } = \pm\ , \frac{s^{m}_{\ell}}{f(r_p)}\ , . \label{characteristicjump}\end{aligned}\ ] ] on the other hand , @xmath165 can be seen as a _ subsidiary _ variable , in the sense that we can first evolve the equations for @xmath166 and @xmath167 and then , use the result to evolve @xmath165 . then , we can use a different way of imposing the continuity of @xmath165 , instead of the penalty method we can just replace the right - hand sides of equations ( [ penaltyr1 ] ) and ( [ penaltyl1 ] ) by @xmath168 which ensures the continuity of @xmath165 by construction and we have seen in our numerical experiments that it is numerically stable ( see sec . [ results ] ) . regarding the global boundary conditions ( _ near _ the horizon , @xmath169 , and _ near _ spatial infinity , @xmath170 ) , we impose them directly at the corresponding nodes , without using the penalty method ( which is another option ) . since we have made a first - order reduction of our original pde ( [ master ] ) , we have to adapt the outgoing / absorbing boundary conditions to the set of variables @xmath171 . then , the boundary conditions at @xmath169 are @xmath172 and the boundary conditions at @xmath170 are @xmath173 and we remark that the first subdomain is number @xmath174 and the last one is number @xmath117 , the total number of subdomains . in this section we describe the results from a series of numerical experiments that show the potential of the methods just proposed . to that end , we have developed a numerical code that implements the techniques previously described . this code is based on the c language and the use of the gnu scientific library @xcite , mainly for calculations with special functions , and the fftw library @xcite for performing ffts . in the implementation of the physical domain in the numerical code we have adopted a comoving tortoise coordinate : @xmath175 . in this way , the particle location ( and we emphasize again that we only deal with circular orbits ) is always @xmath176 . the first test we have performed is to study the evolution of a simple wave equation using the formulation and methods described in the previous sections . this case corresponds to : @xmath177 , @xmath178 ( which implies @xmath179 ) , and @xmath180 ( that is , no source , @xmath181 ) . the test consists in following the propagation of an initial gaussian packet in a multidomain grid and to study the convergence of the numerical scheme as the number of collocation points per subdomain , @xmath112 , increases . in figure [ vacuumconvergence ] , we show a convergence plot for a simulation that uses two subdomains , @xmath182\cup [ 0,550\,m]$ ] , connected by the penalty method as described in subsection [ evolutionalgorithm ] . the truncation error has been computed in the subdomain where the gaussian packet is present at a chosen time . as one can see , the truncation error , estimated as the absolute value of the last spectral coefficient , @xmath183 , decreases exponentially with the number of collocation points , as expected in the psc method for smooth solutions . , associated with the field @xmath165 ) with respect to the number of collocation points . this error corresponds to an snapshot of the evolution of the classical wave equation , for an initial condition given by a moving gaussian packet . the data indicates the exponential convergence of the numerical method.,scaledwidth=50.0% ] the same convergence follows for an initial gaussian wave packet propagating on a schwarzschild background , that is , for the case in which only @xmath181 . this is expected as the mathematical structure of the equation is essentially the same . the additional ingredient is the excitation of quasinormal modes of the black hole by the initial wave packet . when we introduce the sco , i.e. the particle , the situation is conceptually different . if we think in global terms , the presence of the particle implies that the global solution ( the solution in the whole computational domain ) will not be smooth . hence , we can not expect exponential convergence for the solution of our problem . however , as we have argued above , our multidomain framework avoids the presence of dirac delta distributions in the equations by locating the particle in the interface between subdomains . then , the presence of the particle enters through boundary conditions , actually matching conditions between subdomains . therefore , at each subdomain we will have a smooth solution , and hence we expect our numerical solution to converge exponentially towards it . in our numerical experiments with a particle ( remember we restrict ourselves to the case of circular orbits ) , we take _ zero initial data _ , that is @xmath184 this initial data is obviously not consistent with einstein s equations , and as a consequence the evolution produces an initial unphysical burst . we have to wait until this unphysical feature propagates away in order to analyze the solution and obtain physically relevant results . in figure [ detailsevolution ] we show snapshots of the evolution of a scalar charged particle in circular motion at the last stable orbit ( lso ) , in principle the most demanding case ( in schwarzschild the lso is located at @xmath185 ) . the figure includes details of the different variables , @xmath186 for @xmath187 , near the particle location . these snapshots illustrate the ability of our method to capture the structure of the solution near the particle , in particular the ability of resolving the jump in the radial derivative of the field ( snapshot on the right in figure [ detailsevolution ] ) . in order to further validate our numerical code , we have performed simulations with the particle at the lso changing the number of collocation points , while leaving the number of subdomains fixed . in this way we have checked that our method can achieve the exponential convergence in each individual subdomain . in figure [ convergencewithparticle ] we show a convergence plot obtained from simulations that use four subdomains : @xmath188\cup[-20\,m,0]\cup [ 0,20\,m]\cup [ 20\,m,550\,m]$ ] . the number of collocation points , @xmath112 , is the same at each subdomain , and in this way we have more resolution near the particle , where it is most needed . the figure shows that indeed our numerical scheme has exponential convergence . ) with respect to the number of collocation points for simulations of circular motion at the lso ( @xmath185 ) corresponding to the @xmath189 mode , in particular for the field @xmath190 . the data , taken from the subdomain @xmath191 $ ] indicates the exponential convergence of the numerical method . [ convergencewithparticle],scaledwidth=50.0% ] the multidomain feature of our method is useful for another important reason , namely computational cost . it is not the same having @xmath112 points in @xmath117 domains that having @xmath192 points in one single domain . this is an important fact to be considered in a situation where the need for resolution comes only from some isolated regions of the computational domain . this is what happens in our problem and the type of multidomain structure can have dramatic consequences in the calculations . first of all , for computations in a given time step , the first option involves less calculations . second , from the evolution point of view , in the psc method , the courant - friedrichs - lax ( cfl ) condition on the allowed size of the time step , @xmath193 , is more stringent than in finite differences ( fd ) schemes . for the psc method the maximum allowed time step is @xmath194 ( this is set by the minimum distance between two collocation points , which occurs at the boundaries of the subdomains @xcite ) , that is , it goes like @xmath195 with respect to the number of collocation points , whereas in fd schemes it goes like @xmath196 . then , dividing the domain in subdomains can help in having a bigger @xmath197 . in addition , the multidomain scheme can be seen as a way of adaptivity , in the sense that we can construct small subdomains for the regions that need to be well resolved , essentially near the particle , and large subdomains for the regions that do not need to have high resolution , essentially far away from the particle . in order to illustrate how the multidomain feature of our numerical framework works , we have performed simulations with a fixed number of collocation points but changing the number of subdomains . the aim is to show how the solution improves by adding new subdomains . in our simulations , the most demanding region for resolution is clearly near the particle . apart from this we have the waves leaving the particle location and going both towards the horizon and towards spatial infinity . these waves are in principle easy to resolve , but for modes with high @xmath60 and @xmath33 we have that the source term oscillates like @xmath198 ( where @xmath199 is the coordinate angular velocity of the particle ) , and hence the wavelength gets reduced so that we have many more waves that for low @xmath33 . these waves are moving away and need to be resolved . in figure [ multidomainshow ] we show snapshots of evolutions with @xmath200 collocation points per subdomain , but with different number of subdomains , namely from @xmath201 to @xmath202 subdomains ( half of them to the left of particle and the other half to the right ) . the figure shows the mode @xmath203 and @xmath204 ( so that the period of the waves is @xmath205 ) for the three fields @xmath206 . we can see how for few subdomains the waves are unresolved , and that as we increase the number of subdomains the solution converges . therefore , the multidomain structure is a very useful tool to achieve accurate results with a reasonable computational cost . the key point is to realize what is the optimal number of subdomains , their size , and their distribution over the whole computational domain . given that in our case the period of the waves changes with the harmonic number @xmath33 , the optimal strategy for setting the subdomain will depend on it . however , since the main aim of this work is to show the principal ingredients of the method and to illustrate its performance , we have not explored the possibility of changing the subdomain structure as @xmath33 changes . however , this is something that should be done for optimizing the computational cost in the case of systematic calculations of the self - force . in this sense , it would be convenient to have a deep understanding of the computational parameter space in order to automatically adapt these parameters to the physical parameters and optimize in this way the calculations . this would require an extensive investigation that will be done and presented elsewhere . the next step in the validation of the code is to compute the components of the self - force acting on the charged scalar particle moving in circular geodesics around a nonrotating black hole . this is the goal of this numerical scheme , that is , to provide accurate computations of the self - force with reasonable computational cost . the calculation of the self - force require to compute the evolution of the real and imaginary parts [ note that the source term in the evolution equation ( [ master ] ) is complex and hence both the real and imaginary parts are needed ] of the @xmath45 harmonic components of @xmath207 . since @xmath23 is a real scalar we have that for each @xmath45 the relation @xmath208 holds and hence we do not need to compute the modes with @xmath209 . given that we need to truncate the sum over @xmath45 at a given @xmath210 , the total number of evolutions that we need to run for a single self - force calculation is @xmath211 $ ] , where here the brackets denote the the nearest integer to the argument . for the particular case @xmath212 , a typical case in our calculations , we need to perform @xmath213 evolutions . in table [ self_force ] we present the results of the computations of the self - force vector ( actually of the gradient of the regular scalar field ) acting on a scalar particle in circular ( geodesic ) motion around a black hole at the following radii : @xmath214 @xmath215 , @xmath216 , @xmath217 , @xmath218 , @xmath219 , and @xmath220 . for these simulations we have used our multidomain framework with subdomains that contain @xmath200 collocation points . we have used the same number of subdomains to the right and to the left of the particle . the size of these subdomains is typically @xmath221 , specially those near the particle . as we get far away enough from the particle towards the boundaries ( typically at @xmath222 ) , the size of the subdomains is increased since we need less resolution in those regions . the total number of subdomains that we have used ranges from @xmath223 . we have observed that at some point increasing the number of subdomains ( maintaining the size of @xmath224 ) does not change significantly the results . another important point to mention is the fact that we are computing the self - force right at the particle location . but since the particle is at the interface between two subdomains , we obtain two values of the self - force components , one from the subdomain to the left [ let us call it @xmath127 as in equation ( [ particleatinterface ] ) ] , @xmath225 , and the other one from the subdomain to the right ( @xmath226 ) , @xmath227 . this also provides us with a test of the numerical calculations as both values have to agree to a good gegree of precision . in table [ self_force ] we show our results and compare them with two types of calculations in the literature : ( i ) calculations based on a time - domain method that uses a characteristic formulation of the scalar field equations @xcite ; ( ii ) calculations based on a frequency - domain method @xcite . as we can see , we can get a good numerical approximation to the self - force components using a modest amount of computational resources . it is important to mention that these computations have a relatively low computational cost ( relative to the computational cost of time - domain simulations of this sort ) . the average time for a full self - force calculation of the type just described in a computer with two quad - core intel xeon processors at @xmath228ghz is always in the range @xmath229 minutes . [ cols="^,^,^,^,^,^ " , ] in this paper we have introduced a new time - domain technique for the simulations of emris . the main ingredient of the method is to use a multi - domain framework in which the sco , described as a point - like object , is located at the interface between two subdomains . in this way we have shown that this technique enjoys the exponential convergence property of the psc method and its accuracy and , at the same time , it is also an efficient method to make time - domain computations of the self - force . we have shown that we can achieve a good accuracy in this computations by using a relatively low number of collocation points . in this sense , the multidomain framework allows us to locate more collocation points in the region where they are most needed , i.e. around the particle , by choosing appropriately the number of subdomains , and their size and location . another positive property of our numerical scheme is that it can be easily parallelized to be used in supercomputers , as we can assign the work of each subdomain to different cpus . the only information that we need to communicate is the one necessary to satisfy the matching conditions between subdomains . the calculations performed for this paper show the potential of this technique for the description of emris . these results still leave room for improvement as we have not explore yet the full parameter space of the numerical method ( number of collocation points per subdomain , number and distribution of subdomains , parameters of the spectral filter , penalty parameters , etc ) , which is wide enough . in this regard , it would be convenient to be able to adjust the computational parameters to the physical parameters of each mode in an automatic way . in addition to this , there are several additions / modifications to the present numerical framework that may improve the present accuracy and efficiency of the calculations , some of which we will investigated in future calculations by the present authors . we list here some of them : ( i ) _ compactification of the computational domain_. we can change the mapping between the spectral and physical domains incorporating the two infinities ( the horizon location at @xmath230 and spatial infinity at @xmath231 ) into the calculation . in practice , this mapping only would need to be performed in the first and last subdomains . by doing this , the outgoing boundary conditions ( [ outgoingbcs ] ) would be exact instead of approximate as they are know . an alternative to this could be to improve the boundary conditions by analyzing the solution near the two infinities , like for instance in @xcite in a frequency domain framework . ( ii ) _ reduction of the time step_. as we have mentioned before , the psc method has a more strict cfl condition as a fd scheme , namely @xmath232 versus @xmath233 . this is because of high density of collocation points near the boundaries and the fact that the minimum time step come from the minimum separation between points . a way of changing this is again to change the mapping between the spectral and physical domains . one known way of doing this is the modification introduced by kosloff and tal - ezer @xcite , which was shown to lead to a time step restriction like in the fd case , that is @xmath233 . richardson extrapolation_. in our calculations we have computed harmonic components of the scalar field , @xmath207 , up to a certain @xmath234 , in this work @xmath235 . in other words , we have truncated the expansion in spherical harmonics . we can try to improve the final numerical results for the scalar field and derived quantities , like the self - force , by profiting from our analytical knowledge of the expansions in inverse powers of the harmonic number @xmath60 . this can be done using numerical techniques based on the richardson extrapolation , like in @xcite , which can provide an estimation of the different quantities of interest as @xmath236 . beyond the case studied in this paper , circular orbits of a charged scalar particle , we will study in the future how to extend our time - domain computational framework in order to include generic ( eccentric ) orbits . the main difficulty is obviously to deal with a particle moving the radial direction as our present framework assumes that it is located at a fixed value ( circular motion ) . there are several ways in which we can attack this problem . one is to try to implement some type of moving grid technique ( as it was done in a similar situation in @xcite ) , but the reconstruction of the grid at every time step could increase significantly the computational cost of the evolution . another possibility is to try to make a change of coordinates to coordinates comoving with the particle , so that the techniques presented in this paper can be implemented in a straightforward way . on the other hand , we expect that mbhs sitting at galactic centers will have considerable spins ( @xmath237 or bigger , where @xmath238 is the kerr spin parameter ) and hence the mbh should be described by the kerr metric instead of the schwarzschild metric . this means to have a less symmetric background , instead of the spherical symmetry of schwarzschild just the axisymmetry of kerr . in that case one can separate the dependence on the azimuthal angle and is left with @xmath239 wave - type equations with singular terms . the main difficulty in this case is that each @xmath33-mode ( coming from the separation of the azimuthal angle ) diverges logarithmically at the particle location . then , before trying to transfer the techniques presented in this paper to the kerr case , one has to apply before a regularization procedure as it has been done in @xcite . apart from this , we expect that most of the methods presented in this paper will be helpful in achieving efficient simulations of emris in the case of a spinning mbh . we would like to thank leor barack , jos luis jaramillo and eric poisson for helpful discussions and encouragement . cfs acknowledges support from the ramn y cajal programme of the ministry of education and science of spain and by a marie curie international reintegration grant ( mirg - ct-2007 - 205005/phy ) within the 7th european community framework programme . pcm is supported by a predoctoral fpu fellowship of the spanish ministry of science and innovation ( micinn ) . financial support from the spanish ministry of science and education contract esp2007 - 61712 is gratefully acknowledged . this research has been partly performed using the resources of the centre de supercomputaci de catalunya ( cesca ) . here , we summarize the main conventions used for the special functions involved in calculations of this paper . the expression we use for the scalar spherical harmonics is : @xmath240 where @xmath241 are the associated legendre polynomials [ we use the same expressions as in @xcite , equations ( 8.6.6 ) and ( 8.6.18 ) ] @xmath242 where @xmath60 is a non - negative integer and @xmath33 is an integer restricted to the following range : @xmath243 . the chebyshev polynomials can be expresses as follows : @xmath244 and are defined in the interval @xmath245 $ ] with @xmath246 , where @xmath247 is the degree of the polynomial . chebyshev polynomials are orthogonal in the _ continuum _ in the following sense : @xmath248 where the coefficients @xmath249 are given in equation ( [ cns ] ) . the set of collocation points for the spatial discretization of our pdes by means of the psc method are the extrema of the chebyshev polynomials , together with the end points @xmath250 [ that is , the zeros of the polynomial given in equation [ chebyshevlobattogrid ] ) ] . these points form , in the interval @xmath129 $ ] , a chebyshev - lobatto collocation grid . the cardinal functions associated with them are @xmath251 : @xmath252 once this set of collocation points is adopted , and taking into account the properties of the gauss - lobatto - chebyshev quadratures ( see , e.g. @xcite ) , the chebyshev polynomials have another orthogonality relation , this time in the _ discrete _ , in the following sense ( @xmath253 ) : @xmath254= \sum_{i=0}^{n } w^{}_i\ , t^{}_n(x^{}_i)\ , t^{}_m(x^{}_{i } ) = \nu_n^2\,\delta^{}_{nm}\ , , \label{scalarp}\end{aligned}\ ] ] where @xmath255 are the weights associated with the chebyshev - lobatto grid , @xmath256 , and where the @xmath257 s are normalization coefficients given by @xmath258 1 & \text{otherwise\ , . } \end{array}\right.\end{aligned}\ ] ] finally , the constants @xmath259 in ( [ scalarp ] ) are given by @xmath260 . on the other hand , introducing a new variable , @xmath261 , the chebyshev polynomials look like @xmath262 then , an expansion in chebyshev polynomials can be mapped to a cosine expansion . in order to reduce the spurious high - frequency components of our numerical solutions , we apply a spectral filter of the exponential type to the solution after every time step , that is , after every full rk step . the scheme for the action of the spectral filter is @xmath263 where @xmath264 are the values of the solutions at the collocation points after the rk step ; @xmath265 are their corresponding spectral components ; @xmath266 are the filtered spectral components ; and @xmath267 are the filtered values of the solution at the collocation points . the exponential filter is defined by its action on the spectral coefficients @xmath265 to yield the spectral coefficients @xmath266 . this action is given by @xmath268 where @xmath269 is the exponential filter @xmath270 & \mbox{for~ } n^{}_c < n\leq n\ , , \end{array } \right . \end{aligned}\ ] ] where @xmath271 is the cut - off mode number , @xmath27 is the order of the filter ( typically chosen to be of the order of the number of collocation points , @xmath112 ) , and @xmath272 is the machine accuracy parameter , which is related to the machine accuracy , @xmath273 , by @xmath274 .

the turbulent magnetic fields of a large set of fast solar wind streams measured onboard _ ace _ and _ stereo a _ and _ b _ are analyzed in an effort to identify the effects of the turbulence - level broad variations on the orientations of the local , time - averaged magnetic fields . the power level of turbulence , roughly defined as the power in the transverse field fluctuations normalized to the medium - scale average background field , tightly orders the location of the peaks in the probability distribution functions ( pdfs ) of the angles between local fields and parker spiral . as a result , the broad variations in the power level of turbulence cause a steep dependence of the average power level of turbulence on the angle of the local field to the parker spiral , with the highest turbulence levels found near the normal to the parker spiral and the lowest levels near the parker spiral direction . generalized quasilinear estimates of the mean cross - field displacements adapted to intermittent time - varying turbulence lead to accurate fits of the observed angle pdfs at all stable levels of turbulence , supporting the idea that isotropic turbulence could account for the observed angle pdfs . modeling of the angles between local fields and radial direction , from the pdfs of the angles between local and background fields under an assumption of axisymmetry of the turbulent fields around a background field in or near the direction of the parker spiral , also produces fairly good fits of the observed pdfs of , thereby validating the assumption . finally , local field reversals are found to be quite common even within very broad streams of `` unipolar '' fast solar wind .

1312.4880

local spectral analysis of the solar wind ( sw ) magnetic fields reveals strong time variations in the power level of the turbulent field fluctuations . this variability of the power level of the field fluctuations is responsible for the strong non - gaussianity of the observed pdfs of field variations associated with intermittency ( ragot 2013 ) . the strong non - gaussianity of the observed pdfs of field variations is unlikely to be the only observable consequence of the power - level variability . here , we investigate the magnetic field orientations as functions of the highly varying power level of turbulence in the fast sw , and vice versa , where the power level of turbulence is defined with the field fluctuations normalized to the medium - scale ( @xmath2min ) average background field . sw magnetic fields @xmath3 are measured along sw flow lines and the time series of these in situ field measurements can be averaged on a timescale @xmath4 to obtain mean `` magnetic field vectors '' @xmath5 . if @xmath6 is the time resolution of the measurements and @xmath7 , ( t , t ) = 1 n+1 _ n =- n/2^n/2 ( t+n t ) . [ bav ] what is the meaning of these mean `` magnetic field vectors '' @xmath8 ? do they somehow give the direction of a real magnetic field within the sw ? how do they relate to the direction of `` real '' magnetic field lines ? the answer depends on how much magnetic field lines diverge from each other , locally , and on how much transport they undergo collectively . if the directions of neighboring field lines strongly deviate from each other rather than being dominated by one common direction of transport , then it is dubious that the mean `` magnetic field vectors '' @xmath9 can be ascribed any other meaning than that of time averages of in situ magnetic fields along sw flow lines . however , if the directions of neighboring field lines mostly run `` parallel '' to each other , being dominated by one common direction of transport , then the mean `` magnetic field vectors '' @xmath9 give the mean local common direction of transport , or mean local direction of a field line that passes through one of the measurement points . in a magnetic turbulence with background field @xmath10 , such as the sw , both the field - line dispersal ( relative field - line divergence ) and global field - line transport ( common transport ) can be assessed theoretically and modeled ( e.g. , ragot 1999 , 2006a , 2006b , 2009 , 2010a , 2010b , 2011 ) . field - line dispersal and global field - line transport are measured by the variations in field - line separations and by the field - line displacements across the background - field direction , or cross - field displacements , @xmath11 , respectively . it appears that in slow and fast sw streams , the association between time - averaged field vector direction and mean direction of an actual field line is justified on most timescales because the cross - field displacements @xmath11 far exceed the variations in field - line separations on the background field - aligned length scales @xmath12 cm ( e.g. , ragot 2010b , knowing that @xmath13 on these @xmath14 scales).@xmath15 cm apart may grow tenfold on a @xmath14 scale of @xmath16 cm , or even hundredfold on a @xmath14 scale of @xmath17 cm ( see figures 3 and 5 of ragot 2011 ) , and statistically , by factors of @xmath18 and 20 ( from figures 1 , 2 and 4 of ragot 2011 ) , but the variation of that separation remains only a small fraction of the cross - field displacements undergone by both field lines on these @xmath14 scales , here about 0.9 and 3.6 percent at @xmath19 and @xmath17 cm , respectively , in the flux - tube simulation of ragot ( 2011 ) , and only 0.4 and 0.7 percent , statistically ( from figure 1 of ragot 2006b , @xmath20 and @xmath21 cm at @xmath19 and @xmath17 cm , respectively ) . ] for a sw speed @xmath22 , the length scales @xmath14 relate to the timescales @xmath4 through @xmath23 , the angle @xmath24 between parker spiral and radial direction being defined below in equation ( [ spiral ] ) . the time - averaged field directions obtained by integrating in situ fast sw magnetic fields over time intervals of duration @xmath25s ( depending on @xmath22 ) should therefore match the local field - line directions on the background field - aligned length scale @xmath26 cm . and the angles @xmath1 [ see equation ( [ tan_a ] ) in section 4 ] between these time - averaged in situ fields and the background field @xmath10 should match the angles between the mean directions of local individual field - line segments of @xmath10-projection @xmath14 and @xmath10 , whose tangents are given by the ratios @xmath27 . again , this would not be the case if the field lines were too strongly diverging from each other . these premises are at the basis of the modeling of section 4 for the pdfs of the angles @xmath1 at the scale @xmath28 cm . a first - order approximation for the large - scale geometry of the sw magnetic fields is given by the parker - spiral model for interplanetary magnetic fields ( parker 1958 , 1963 ) . the angle @xmath24 of the parker spiral to the radial direction at heliocentric distance @xmath29 is related to the solar rotation rate @xmath30 and sw speed @xmath22 through ( e.g. , burlaga 1995 ) _ r . [ spiral ] the local magnetic fields actually deviate from the parker spiral by large angles , but the parker spiral direction generally represents a good axis of symmetry for long statistics of the fields . pdfs of the azimuthal angles of the magnetic fields in - ecliptic projections were determined from in situ measurements early on ( ness & wilcox 1966 ; hirshberg 1969 ; see also , burlaga & ness 1997 ) . more recently , pdfs of the angles between sw magnetic fields and parker spiral ( or background field ) were also determined both in three- and two - dimensional space from in situ measurements , and using the measured fourier spectra of magnetic fluctuations , also theoretically and numerically ( ragot 2006b ) . these determinations , however , were made from only one single sw stream in each of four sw conditions ( slow and fast sw at @xmath31 and @xmath32au ) , and the time variability of the power level of turbulence was not considered . here we focus on the fast sw near @xmath33au and extend our analysis to a large set of streams ( over one hundred ) observed onboard _ ace _ and _ stereo a _ and _ b_. most importantly , we analyze the directions of the local magnetic fields as functions of the local level of turbulence , rather than a characteristic average value of that level . an earlier study by podesta ( 2009 ) showed that the turbulent energy observed in fast sw streams is on the average higher when the local magnetic field is normal to the sw flow . the conclusion of that earlier study was that the higher levels of turbulent energy seen when the local magnetic field is normal to the sw flow are indicative of anisotropic turbulence with a wavevector distribution dominated by the wavevectors normal to the field . we are somewhat skeptical of this conclusion that variations of turbulent energy with magnetic field angles necessarily reflect an anisotropy of the wavevector distribution . our skepticism is driven by the facts that ( 1 ) the variations in the turbulent energy and turbulence level are actually so much greater than the variations seen in the above anisotropy argument ( orders of magnitude , to be compared to a factor of the order of a few ) , and ( 2 ) the orientation of the local magnetic field is in large measure determined by the turbulent energy and turbulence level ( see abstract , next paragraph and section 3 for a definition of the power level of turbulence or turbulence level ) , as we confirm here through our analysis and modeling of the field orientations for a broad range of turbulence levels . through our analysis and modeling , we will test whether a turbulence model of isotropic wavevector distribution could account for the observed pdfs of local field angles at any of the measured turbulence levels . we will also test whether the pdfs of angles relative to the radial can be recovered from the pdfs of angles relative to the background under an assumption of axisymmetry and parker - spiral background field . here , we distinguish between turbulent energy and turbulence level . the turbulent energy is basically the power level of the magnetic field fluctuations , computed at either one frequency or integrated over a broader frequency range . by contrast , to anticipate our discussion of section 3.5 , the turbulence level is the power level of the magnetic field fluctuations _ normalized to the medium - scale average background field . _ while the turbulent energy does say little about the stochasticity of the fields and the amount of field - line wandering actually taking place in the sw , especially if the background field undergoes strong fluctuations , the turbulence level tightly parametrizes field stochasticity and field - line wandering . in fast sw streams , it appears that the background field does undergo strong fluctuations . therefore , the turbulence level as defined above and in greater detail in section 3.5 , is a far better ordering parameter for the orientations of the fast sw magnetic fields . whether the turbulence level is defined at one frequency , integrated over a range of frequencies , or involves a more complex combination of frequencies through the formula for the mean cross - field displacement ( see sections 3.4 and 3.5 ) does not affect that basic fact . our data selection is presented in section 2 and the variations in the power level of magnetic fluctuations are shown in section 3 , where the power level of intermittent turbulence is also defined . in section 4 , through modeling of the pdfs of field - line directions , we model the pdfs of the angles @xmath1 between local and background ( or parker spiral ) fields at series of stable levels of turbulence . section 5 deals with the transition from the @xmath1 angles to the angles @xmath0 between local fields and radial direction . section 6 presents some consequences of the power variability on the field orientations and the apparent `` anisotropy '' of the turbulence . the conclusion follows in section 7 . the study presented in this paper requires high statistics of fast sw turbulent magnetic fields . to reach sufficient statistics , we use data from _ ace _ , from year 2000 to 2011 , and from the two _ stereo _ spacecraft , _ stereo ahead _ ( or _ a _ ) and _ stereo behind _ ( or _ b _ ) , from year 2007 to 2012 . we selected a total of 121 intervals with sw speeds in excess of @xmath34 km s@xmath35 ( but mostly @xmath36 km s@xmath35 ) and durations ranging from 1 to 8 days , for a total duration of 307.5 days . stream leading and trailing edges were avoided as far as possible . some of the selected intervals obviously belong to the same sw stream , but for our purpose , they can be considered as independent because of the broad separation of the spacecraft and/or the time lapsed between recurrences . @xmath33s - averaged interplanetary magnetic field data from the mag magnetometer on _ ace _ were downloaded from the ace science center at www.srl.caltech.edu/ace/asc . @xmath37ms - resolution magnetic field data from the impact / mag magnetometers on _ stereo a _ and _ b _ were downloaded from www.ssl.berkeley.edu and averaged on 8 consecutive points to match the averaging timescale of the _ ace _ data . in the following , all magnetic field data have a time resolution @xmath6 of @xmath33second . one of the longer intervals selected for our analysis is the fast sw stream observed onboard _ ace _ between day 70 and 76 of year 2008 . we use this interval as an example to illustrate in figures 1@xmath383 how the fourier power level of the transverse magnetic fields fluctuates with time . for our analysis , we define the directions @xmath39 , @xmath40 and @xmath41 as the directions along the background magnetic field @xmath10 ( or parker spiral ) and normal to @xmath10 , in the ecliptic plane ( @xmath40 ) and normal to it ( @xmath41 ) . the angle @xmath24 of the parker spiral to the radial direction is determined from equation ( [ spiral ] ) using the @xmath42s resolution proton speeds from the swepam experiment . ( for sw streams observed onboard _ stereo a _ or _ b _ , we use the @xmath33min resolution proton bulk speeds from the plastic experiment.)min , with no noticeable variation in the results . so the exact averaging timescale for vsw does not seem to affect the results of this paper , which probably reflects the low variations of @xmath22 over each of the fast sw intervals selected for the analysis . ] we project the magnetic field vectors ( in rtn coordinates ) on these three directions and compute the spectra of the @xmath43 and @xmath44 components , formally introduced in section 3.5 , the @xmath43 and @xmath44 fluctuations dominate the @xmath45 fluctuations and a modeling based on only the @xmath43 and @xmath44 fluctuations is appropriate . at the lowest values of @xmath46 , where @xmath45 fluctuations can become dominant , we will include the effects of the``compressive '' field - aligned fluctuations in our modeling of the field angles @xmath47 ( see section 4 and end of 3.4 ) . ] by fast fourier transform , with a hanning window to minimize truncation effects . we make this analysis time - dependent by using a window of width @xmath48 and letting the window slide by @xmath6 increments . taking the module of the results , squaring and multiplying by the window width @xmath49 , we obtain the power levels of @xmath43 and @xmath44 before `` smoothing , '' a_x , y ( , t , t_w ) & & t_w |1 n_w _ n=0^n_w-1 1 2 . b_x , y(t+n t ) e^i 2 n t / n_w |^2 , [ axy ] and compute their sum , @xmath50 . we then reduce the noise in the result @xmath51 by smoothing . for that , we make use of the very long - time reference power level , @xmath52 , which we compute by including the entire interval of data ( in this particular case , @xmath53days ) , to determine the spectral slopes and right the power level @xmath51 before averaging locally in frequencies . the reference power level itself can easily be smoothed out at the higher frequencies by averaging over 200 successive frequencies . in that particular case , the separation @xmath54 between harmonics is so small that we do not have to worry about the spectral slopes when averaging over the neighboring frequencies . because we find that the spectral slopes of @xmath51 vary little with time , we smooth the power level of transverse magnetic fluctuations by computing : a(_0 , t , t_w ) _ n_s =- n_s^(1- ) n_s a(_0+n_s , t , t_w ) ( _ 0+n_s ) , [ asmooth ] where @xmath55 and @xmath56 is the separation between fourier harmonics of @xmath51 , not @xmath52 . in figures 1@xmath383 , @xmath57 and @xmath58 . the smoothing operation is basically a local averaging over the frequencies , after correction for the spectral slope using the very long - time reference power level , @xmath52 . with this time - frequency analysis , we are keeping control of both the frequency @xmath59 and the window width or time resolution of the fourier analysis , @xmath49 . one may argue that a wavelet analysis ( e.g. , daubechies 1992 ) is superior , but both fourier and wavelet analysis methods produce similar results ( podesta 2009 used a wavelet analysis and our results are consistent , see section 6 ) . by using a fourier spectral analysis with sliding window , we are just being consistent with our earlier work on the wandering of magnetic field lines , which is most relevant to the study of the field orientations and whose results we are using in section 4 to model the pdfs of the angles @xmath1 [ see equation ( 27 ) of ragot 2006a for the definition of the fourier transform used in that earlier work ; see also section 3.4 ] . in most theoretical calculations , it is far easier to fourier transform back and forth than to wavelet analyze . the fourier analysis has for itself its simplicity and `` ease of interpretation . '' with this time - frequency analysis , we are not only determining the power level of the magnetic field fluctuations as a function of time ( see sections 3.2 and 3.3 ) , but we are also checking the shapes of the frequency spectra . to our analysis of the field directions and turbulence it is important that we keep track of the time variations or lack thereof of the spectral slopes ( see figure 2 of ragot 2009 or figure 4 of ragot 2006d ) . indeed , the spectral indexes are still needed to evaluate the mean field - line directions analytically ( see section 3.4 ) . figure 1 shows the time variations of the power level @xmath60 computed at frequency @xmath61hz with a shifting window of width @xmath49 decreasing from @xmath62 the full duration of the sw stream ( violet line ) down to @xmath63 of it ( light - blue line ) by three successive factors of about @xmath64 . the black line gives the very long - time reference power level , @xmath65 , to which the running power level @xmath66 is normalized . we see that the shorter the shifting window is , that is , the higher the time resolution of the spectral analysis is , the stronger the fluctuations of the power level are . the higher - resolution fluctuations far exceed the lower - resolution fluctuations , but roughly follow them in their time average , fractal - like . figure 2 presents a zoom out of figure 1 , for the first day of data . the electronic version of figure 1 can also be zoomed in to follow the four fluctuating lines and check how each is fluctuating around the line of next lower time resolution . the power - level fluctuations of figure 1 are smoothed on 200 successive frequencies . the effect of that smoothing is to reduce the amplitude of the fluctuations ( see figures 3 and 4 of ragot 2013 ) , especially at the lower values of power level , which can be most affected by noise . we checked that the effect of this reduction in fluctuation amplitude is to eliminate the tiny excess `` bump '' at the top of the pdfs of field variations presented in figures 5@xmath388 of ragot ( 2013 ) . when computing the power level at a given frequency @xmath59 , the value of @xmath59 constrains the width @xmath49 of the window that can be used for the fourier analysis . typically , @xmath49 has to exceed a few times @xmath67 , and in order to capture the time variations of the power , @xmath49 must also be chosen as short as possible . because the average slope of the turbulence spectrum appears to vary little with time ( again , see figure 2 of ragot 2009 or figure 4 of ragot 2006d ) , and because our primary concern here is to capture the precise time variations of the power over a broad range of scales , we now compute the power at a higher frequency @xmath68 with a window of width @xmath69 , and extrapolate the result to the lower frequency @xmath59 by using the spectral slope obtained with a much longer window , covering the entire sw stream . figure 3 shows the resulting time variations of the coefficients @xmath70 , @xmath71 , @xmath72 , @xmath73 and @xmath74 in cm nt@xmath75 [ @xmath76 , see also section 3.4 ] for the spectral amplitudes of transverse magnetic field , starting with the lowest wavenumber @xmath77 cm ( black line ) , where the turbulence spectrum becomes steeper than @xmath78 , and ending with @xmath79 cm ( red line ) . for a wave vector @xmath80 making an angle @xmath81 with the sw flow velocity , the wavenumber @xmath82 is related to the frequency @xmath59 through @xmath83 . here we are considering a series of wave vectors @xmath84 along the direction of the background magnetic field , that is , along the parker spiral , and corresponding to a series of frequencies @xmath85 . the flow velocity being practically radial in the sw , the wavenumbers @xmath86 are therefore related to the frequencies @xmath85 through @xmath87 , with @xmath24 the angle of the parker spiral to the radial direction at heliocentric distance @xmath29 , as defined by equation ( [ spiral ] ) . the wavenumbers @xmath86 and wave vectors @xmath84 are parallel wavenumbers and wave vectors along the parker spiral , as needed for the modeling ( see sections 3.4 and 4 ) , which calls for a one - dimensional projected spectrum along the background field [ see also equation ( [ cl ] ) and following discussion ] . as expected from figures 1 and 2 , in figure 3 the time variations at the higher frequencies ( or wavenumbers ) roughly follow in their time average those of the lower frequencies , but locally display much more intense fluctuations . with the time - varying coefficients @xmath88 , we can estimate in section 3.4 the magnetic field - line mean cross - field displacement , @xmath89 , in the intermittent turbulence of the sw . the cross - field displacements @xmath11 on a background field - aligned scale @xmath14 are central to our study of magnetic field directions because in the turbulence of fast and slow sw streams on the scales @xmath90 cm , the ratios @xmath91 are expected to closely match ( see introduction ) the tangents of the angles @xmath1 between the measured magnetic field , averaged on the corresponding timescale @xmath92 , and the background field @xmath10 . that expectation of a close match between time - averaged field direction , obtained from in situ data , and direction of an actual field line is justified by our earlier study of field - line dispersal , which shows that on the scales @xmath12 cm in the sw , the cross - field displacements far exceed the variations in field - line separations ( see , e.g. , ragot 2010b , and introduction ) . again , this would not be the case if the field lines were too strongly diverging from each other . the cross - field displacement @xmath11 is not defined from the data . however , it is a useful quantity whose root mean square , @xmath89 , the mean cross - field displacement , can be evaluated from the spectral amplitudes @xmath88 at wavenumber @xmath86 using predictions of the generalized quasilinear ( gql ) theory ( see below ) . @xmath89 is central to our modeling . with @xmath89 , we can estimate in section 4 the pdfs of the field - line directions , and compare them to the pdfs of the angles @xmath47 between the time - averaged in situ fields and the background field @xmath10 . the displacements @xmath11 of turbulent magnetic field lines across the direction of a background field @xmath10 have been extensively studied for a general background field - aligned scale @xmath14 ( ragot 1999 , 2006a , 2010a ) . the square root of their variance or mean cross - field displacement , @xmath89 , can be accurately estimated , both theoretically and numerically , in self - similar turbulence , that is , in a turbulence of single `` constant - at - all - scales '' spectrum of magnetic fluctuations . both fully nonlinear and gql estimates can be made for @xmath89 . on background field - aligned scales @xmath90 cm and for turbulence spectral shapes typical of slow or fast sw streams , nonlinear corrections to the gql result in a background field @xmath10 ( r)^2 ^_b_0 & = & 4 b_0 ^ 2 _ + & & \{1 - |k_l z|^1- ( 1- ) ( -1 + ) ( 2 ) . + & & . - f_p , q } + & & - c_l+1 + & & \ { 1 - |1 z|^1- ( 1- ) ( -1 + ) ( 2 ) . - f_p , q } ) + [ dr ] of ragot ( 1999 , 2006a ) [ see also equation ( a13 ) of ragot 2010a ] + @xmath93 when @xmath14 approaches @xmath94 , @xmath94 and @xmath95 being the + parallel and perpendicular correlation lengths of the turbulence . ] are found to be negligible , even for much enhanced power levels ( up to at least a factor of 40 at @xmath16 cm , which would easily cover the range of enhancement values observed in figures 1@xmath383 , and much more at shorter @xmath14 scales , see ragot 2010a ) . on scales @xmath90 cm , accurate estimates of the mean cross - field displacement should therefore be obtained in the intermittent turbulence of the sw with the same gql expression as given by equation ( [ dr ] ) , but with coefficients @xmath88 now varying with time and broadly departing from their long - time average value , as shown in figure 3 , rather than constant as in a self - similar turbulence . the expression of equation ( [ dr ] ) should only remain valid , however , as long as strong variations of the amplitudes @xmath88 giving the dominant contribution do not occur on the timescales shorter than @xmath4 , since constant amplitudes have been assumed on the corresponding lengthscale @xmath14 in the integration leading to equation ( [ dr ] ) . we return to that point in section 4 . in equation ( [ dr ] ) , @xmath88 and @xmath96 are the amplitudes @xmath97 and spectral indexes of the one - dimensional or projected fourier spectrum ( ) d _ _ ^2 ( , _ ) [ cl ] on the background field direction ( parallel direction ) at parallel wavenumbers @xmath98 , the assumption being that of a piecewise power - law spectrum with a well - defined single spectral index on each interval @xmath99 $ ] . in equation ( [ cl ] ) , @xmath100 denotes the theoretical three - dimensional fourier transform ( with sliding window ) of the transverse component of magnetic field . equation ( [ dr ] ) applies whatever the three - dimensional distribution of wavevectors , as long as it produces the projected spectrum of equation ( [ cl ] ) . the sw measurements , however , give the spectrum projected on the sw flow direction rather than the background field . so the measured amplitudes @xmath88 of figure 3 may not strictly coincide with the required @xmath101 , and the spectral indices @xmath96 used in our further estimates of @xmath102 are not necessarily the spectral indexes of the projected fourier spectrum of equation ( [ cl ] ) . only if the three - dimensional distribution of wavevectors is nearly isotropic ( as found in slow sw by narita et al . 2010 at the scale of @xmath103 cm ) should both sets of amplitudes and spectral indexes a priori coincide . the anisotropy found in the cluster data analysis of narita et al . ( 2010 ) becomes significant below the scale of @xmath103 cm , with a @xmath104 percent variation in the energy distribution from the parallel to the perpendicular directions at @xmath105 cm , but these shorter turbulent scales do not contribute ( or only very little ) to the field - line displacements at the scales @xmath26 cm considered in this paper , nor to the mean cross - field displacements at any other scale . we therefore are hopeful that the discrepancy in the measurement direction of the one - dimensional spectra will not become an issue as we further make use of equation ( [ dr ] ) in our analysis of section 4 . and spectral indices @xmath96 obtained from measurements made along a direction that differs from that of the parker spiral . but we do obtain quite reasonable fits ( see figures 8 , 10 and 11 in section 4 ) . ] in equation ( [ dr ] ) , @xmath106 denotes the hypergeometric function , computed as a generalized hypergeometric series f_p , q[\ { } , \ { , } ; ] & & 1 + + ( + 1 ) ^2 ( + 1 ) ( + 1 ) 2 ! + & & + ( + 1 ) ( + 2 ) ^3 ( + 1 ) ( + 2 ) ( + 1 ) ( + 2 ) 3 ! + [ fpq ] ( e.g. , gradshteyn & ryzhik 1980 ) and @xmath107 is the gamma function ( euler s integral of the second kind ) defined for @xmath108 by the integral ( ) _ 0^+ dx e^-x x^-1 . [ gamma ] the gql result of equation ( [ dr ] ) assumes a constant background field of magnitude @xmath109 . however , the starting equation for the field - line tangent ( see equation ( 1 ) of ragot 1999 or 2006a ) calls for the actual @xmath45 field projection on the background @xmath10 , and provided that @xmath45 does not fluctuate too fast , or is averaged on a timescale long enough , it can obviously substitute for @xmath110 in equation ( [ dr ] ) . when there are significant long - scale fluctuations in the value of @xmath45 , that substitution should greatly improve the estimate of the mean cross - field displacement . therefore we write the mean cross - field displacement as ( r)^2 ^1/2 = ( r)^2 ^1/2 _ ( r)^2 ^1/2_b_0 , [ dr_bz ] where @xmath111 is given by equation ( [ dr ] ) . throughout the rest of this paper , unless otherwise specified , @xmath89 denotes @xmath112 , the subscript @xmath113 of the medium - scale averaged @xmath45-field being omitted to lighten the notations . the background field @xmath110 , or very long - time stream - average field , is computed for each of our fast sw streams by averaging the @xmath45-field over the entire stream . shorter - scale fluctuations @xmath114 of @xmath45 can be taken into account perturbatively in the evaluation of the mean cross - field displacement by expanding @xmath115^{-1}$ ] , which we later do to estimate the correction due to `` compressibility effects . '' an appropriate medium scale for averaging the background field and computing @xmath113 remains to be chosen . we return to the averaging scale of the local background field @xmath113 in section 3.6 . from the multiple time - resolution and multiple - frequency fourier spectral analysis of sections 3.2 and 3.3 , we realize that the `` power level '' of the intermittent sw turbulence is an extremely complex function of time and frequency , with wild fluctuations of large amplitude . in our study of the magnetic field orientations with varying turbulence level , we are therefore faced with a first serious challenge that consists in defining a turbulence level . our choice of a proper physical parameter that would both be a good representation of the level of turbulence at a given time and provide a good parametrization of the field orientations is guided here by our earlier theoretical and numerical studies of magnetic field - line wandering ( ragot 1999 , 2006a , 2010 ) . we know from these earlier studies that the displacements @xmath11 across the direction of the background field @xmath10 on a given field - aligned scale @xmath14 , which should give good estimates of the directions of the in situ magnetic fields averaged on the corresponding timescale @xmath92 , is not dependent on the fourier power level at just one frequency , but at a broad range of frequencies [ see section 3.4 and equation ( [ dr ] ) ] . because the spectral power level is so wildly varying with time and frequency ( see figures 13 ) , studying the field directions as functions of the spectral power level at any one single frequency may therefore not reveal the most meaningful trends of the data . by instead studying the field directions as functions of a linear combination of the fourier power levels @xmath88 at a broad range of frequencies ( or wavenumbers ) that happen to produce the variance @xmath102 of the field - line displacements [ see equations ( [ dr ] ) and ( [ dr_bz ] ) ] , we believe that we are optimizing our chances of uncovering the underlying organization of the field directions , and of modeling these field directions successfully . the use of the mean cross - field displacement , @xmath89 , which is a linear combination of the spectral amplitudes @xmath88 , rather than that of just one of the spectral amplitudes @xmath88 , presents the advantage of incorporating the variability effects at all frequencies . also , unlike the spectral amplitudes @xmath88 , which may vary by up to a couple of orders of magnitude with time but actually tell us little about the turbulent behavior of the fields , the mean cross - field displacement has an obvious simple physical meaning that is directly related to the turbulent behavior of the fields . it is the quantity that we will be using here to quantify the turbulence level . more precisely , we introduce the parameter @xmath116 defined by c^1/2 , [ c ] with the mean cross - field displacement @xmath112 given by equations ( [ dr_bz ] ) and ( [ dr ] ) . the quantity @xmath117 is the mean cross - field displacement estimated with a background field magnitude equal to the average @xmath118 and divided by the field - aligned scale @xmath14 . the parameter @xmath116 is a linear combination of the spectral amplitudes @xmath88 of magnetic field fluctuations [ see equation ( [ dr ] ) ] . the product @xmath119 is not a diffusion coefficient because the fields do not diffuse on the scales @xmath90 cm . also , due to intermittency , @xmath116 strongly varies with time . a related parameter , which we will be abundantly using throughout this paper , is the angle ( c^1/2 ) . [ xi ] note that if the turbulence were self - similar with only one `` constant - at - all - scales '' spectrum of magnetic fluctuations , then @xmath102 and @xmath116 would simply be proportional to any one of the amplitudes @xmath88 , that is , to what is ordinarily understood as the energy level of turbulence in self - similar , non - intermittent turbulence . it is only because the sw turbulence is strongly intermittent that involving the mean cross - field displacement @xmath89 in the definition of the turbulence level presents an advantage . an appropriate scale for computing the local average background field @xmath113 remains to be chosen . the pdfs of the turbulence - level parameter @xmath46 , computed for a @xmath45 field averaged over each complete sw stream ( violet ) , and over scales of @xmath2min ( red ) and @xmath103 cm ( blue ) , are presented in figure 4 for our entire data set ( 121 intervals , 307.5 days).-component of the field , not its magnitude . when averaged on a complete sw stream to obtain @xmath110 , the @xmath45-component does not `` cancel out '' because we are considering fast sw streams of mostly unipolar fields . ] [ the values of @xmath46 are computed from the @xmath88 amplitudes of figure 3 using equations ( [ dr ] ) and ( [ dr_bz ] ) . ] a complete sw stream is much too long ( one to 8 days ) , while @xmath103 cm , the scale over which we are computing the mean cross - field displacement , is too short . we find that an averaging scale of @xmath2min is a good compromise between the two . the pdfs of @xmath120 and @xmath121 differ relatively little from each other , meaning that a @xmath2min average already captures most background field fluctuations . also , a scale of @xmath2min in fast sw of speed exceeding @xmath122 km s@xmath35 is a safe 40 to 50 times longer than the scale @xmath28 cm . in the rest of this paper , unless otherwise specified , @xmath113 denotes the @xmath2min average of @xmath45 . figure 5 now compares the pdfs of @xmath123 , where @xmath113 is averaged on scales of @xmath103 cm ( blue ) and @xmath2min ( red ) , in four contiguous intervals of the parameter @xmath46 covering a total of 80 . the parameter @xmath46 is estimated with an averaging scale of @xmath2min for @xmath113 for both the blue and red pdfs . the polarization sign @xmath124 of the background field is computed from each complete sw stream . a negative value for @xmath123 indicates a reversal of the field relative to the unperturbed parker spiral field . while reversals are rare events at @xmath125 , they become relatively frequent at the higher values of @xmath46 . at the highest levels of turbulence ( @xmath126 , bottom - right panel of figure 5 ) , short - scale reversals are so frequent that on the shorter scale of @xmath103 cm , reversed fields are practically as common as normal - polarity fields . this of course does not mean that field reversals are ubiquitous throughout the fast sw . the occurrence rate of @xmath127 is relatively low ( see red curve in figure 4 ) . still , it is far from negligible , and we find that if the unperturbed field is indeed along the parker spiral direction , then field reversals are not rare occurrences in the fast sw . on the whole , again if the unperturbed field is along the parker spiral direction , the field appears to be reversed between 4 and 10 percent of the time in fast sw streams . note also that unlike for the @xmath128cm - average , the pdfs for the @xmath2min - average fall to zero at @xmath129 . from the mean cross - field displacement of equations ( [ dr ] ) and ( [ dr_bz ] ) , we can model the pdf of the angles @xmath1 between local and background fields . obtained from the model and from the measurements , since they could be very different quantities , were it not for the fact that in sw streams and on the scales @xmath90 cm , the cross - field displacements far exceed the variations in field - line separations ( see intoduction and beginning of section 3.4 ) . the angles of the model are the angles @xmath130 between segments of magnetic field lines with projection @xmath14 on the background field , and the background field , while the measured angles are the angles @xmath1 between the time - averaged in situ fields and the background field @xmath10 or parker spiral . these facts being clear , we will be omitting the subscript @xmath131 of the angles @xmath130 . ] as long as the gql result holds , we expect gaussian distributions for the displacements @xmath132 and @xmath133 in each of the directions perpendicular to the background field and a distribution , f_r(r ) = 2 r ( r)^2 e^-(r)^2/(r)^2 , [ fr ] for the cross - field displacements @xmath134^{1/2}$ ] . from this distribution , we derive in section 4.2 of ragot ( 2006b ) the pdf , f_a ( ) = z ^2 f_r , [ fth ] for the angles @xmath1 between local and background fields . the gql expression for the mean cross - field displacement in equation ( [ dr ] ) assumes a turbulence spectrum that does not vary on the scale @xmath14 over which @xmath102 is estimated . for a fixed value of the parameter @xmath102 , one should therefore expect our theoretical prediction to fit the observations only when the power level is stable on the scale @xmath14 . from our large data set , we extract all the time intervals over which such conditions are satisfied . in figure 6 , we show the pdfs of the `` power level '' @xmath46 for both the entire data set ( red ) and the subset of time intervals over which the `` power level '' is stable , that is , the subset of time intervals over which @xmath46 remains within one 2-wide bin for a time exceeding @xmath135s ( which is slightly larger than @xmath136 ) . the contribution from the highest turbulence levels are much reduced in figure 6 in the `` stable '' subset because the peaks of higher power are usually of short duration or involve very steep time variations . figure 7 shows similar pdfs for a stricter definition of stability whereby the edges of the stability intervals are removed ( some @xmath137s on each end ) . this eliminates time intervals over which the computed turbulence power level may still be influenced by time intervals with different dominant power level . the contribution of each subinterval is now more peaked around its center , but the pdf shape is not otherwise significantly modified . the purpose of these subsets of stable turbulence power levels is to check whether the angles between local and background fields can be accurately modeled and understood when the basic model requirement of stable turbulence levels is satisfied . for more generally varying turbulence power levels , the modeling would require a more complex convolution to include the contributions from all power levels . this is beyond the scope of the current paper . here , we just want to find out whether the model has a chance to work in the first place and check the basics . we thus deconvolve the problem by decomposing the data into distinct turbulence power levels , instead of convolving the model . on the @xmath14 scales less than @xmath138 cm , the cross - field displacements far exceed the variations in field - line separations ( see , e.g. , ragot 2010b ) . the cross - field displacements obtained by integrating the @xmath43 and @xmath44 measured in situ field components over time intervals of duration @xmath4 should therefore match the cross - field displacements of real magnetic field lines relatively well , and give good estimates of the actual angles @xmath1 between local and background magnetic fields in the sw . again , this would not be the case if the field lines were too strongly diverging from each other ( see introduction , including footnotes ) . in figure 8 , we compare the model to the measured pdfs of the angle @xmath1 between local and background magnetic fields ( see footnote 8) . the measured angles @xmath1 are obtained from = \{^2 + ^2}^1/2 s _ n =- n/2^n/2 b_z(t+nt ) , [ tan_a ] with the background field polarity sign , @xmath139 , for each sampling time @xmath140 , excluding the first and last @xmath141 measurements of each sw stream ( see also footnote 7 ) . the value of @xmath142 slightly varies with time . it is given by the ratio of @xmath28 cm over the local product @xmath143 at time @xmath140 . [ if @xmath142 is an odd number , the sums are taken from @xmath144 to @xmath145 . ] the pdfs computed from the entire data set are shown in red , while those computed for the subset of intervals with stable turbulence levels are shown in black . in each of the four panels , a different range of turbulence levels is included , from @xmath146 in the upper - left panel to all @xmath46 values in the lower - right panel . the reason for this distinction is that at the lowest values of @xmath46 , typically below 28 , `` compressive '' field - aligned fluctuations become dominant . to the `` non - compressive '' transverse model based on equation ( [ dr ] ) , we then have to add a correction . this is done by expanding @xmath115^{-1}$ ] , and roughly results in the multiplication of the transverse result for @xmath102 by a factor @xmath147 . numerically , it amounts to a factor of the order of @xmath148 in front of @xmath113 . the results for the pdfs of @xmath1 are the blue lines shown in the bottom panels of figure 8 . the model ( black lines in upper panels and dark - blue lines in bottom panels ) fits the measured pdfs well for the subset of stable turbulence levels , confirming the validity of the model . as expected , however , the model does not fit the measured pdfs so accurately ( see red and light - blue pdfs ) when intervals of fast varying turbulence level are included in the data . we now decompose the @xmath1 pdfs into four ranges of turbulence levels ( @xmath149 in violet , @xmath150 in green , @xmath151 in blue and @xmath152 in red ) . in figure 9 , we compare the pdfs of each subrange for the subset of stable turbulence levels and for the entire data set , both for a @xmath45-field averaged on @xmath103 cm ( top panel ) and on @xmath2min [ bottom panel , where the sum in the denominator of equation ( [ tan_a ] ) is made from -450 to 450 instead of -n/2 to n/2 ] in the definition of @xmath153 . the measured @xmath1 pdfs differ little at the lower turbulence levels , but more significantly at the higher levels . with the same decomposition into four ranges of turbulence levels , we compare in figure 10 the measured normalized @xmath1 pdfs of the stable subset with the model pdfs . the statistics in the highest @xmath46-range are very limited and the error bars on the measured pdf very large . we therefore consider the model a reasonable fit in all four ranges of turbulence levels . while in figure 10 , the angles @xmath1 are computed with a @xmath45-field averaged on @xmath103 cm , in figure 11 the @xmath45-field is averaged on a scale of @xmath2min in the denominator of equation ( [ tan_a ] ) ( sum made from -450 to 450 instead of -n/2 to n/2 ) . the comparison data / modeling remains satisfactory . figures 8 , 10 and 11 reveal no problem with our modeling of the magnetic fields and of their orientations relative to the background field , as long as the turbulence level varies sufficiently slowly . on all subsets of stable turbulence level , we obtain fits of the measured pdfs that are actually quite good , even at the highest turbulence levels where the theory was not originally intended . our gql modeling and results do not assume anything about the turbulence isotropy or lack thereof . the only requirement for our turbulence model is that the three - dimensional distribution of wavevectors produces the right projected spectrum of turbulence [ see equation ( [ cl ] ) ] . so the positive result of our modeling actually supports an isotropic model of turbulence by not excluding it . because the direction of the radial is known while that of the background field is in some measure assumed , the orientations of the sw magnetic fields are more often studied relative to the radial than to the background . here , we go from the angles @xmath1 relative to the background to the angles @xmath0 relative to the radial , compare the pdfs of the two at a number of turbulence levels , and investigate how one goes from one to the other . returning to the entire data set , we show in figure 12 the pdfs of field angles at a series of turbulence levels @xmath154 , @xmath155 , @xmath156 , @xmath157 and @xmath158 . in figure 12 , only the central pdf at @xmath156 is normalized . the other pdfs are given relative to that pdf and have much lower integrals at the lower and higher ends of the turbulence power levels . the pdfs of measured @xmath1 angles between local and background fields are shown in dashed lines while the pdfs of measured @xmath0 angles between local field and radial direction are shown in solid lines . the measured angles @xmath0 are obtained from _ r = \{^2 + ^2}^1/2 ( b_0_r / |b_0_r| ) _ n = n/2^n/2 b_r(t+nt ) , [ tan_ar ] where @xmath159 , @xmath160 and @xmath161 are the measured magnetic field components in the rtn coordinate system ( @xmath162 : along the radial , pointing away from the sun ; @xmath163 : orthoradial in the ecliptic , pointing west ; @xmath142 : normal to the ecliptic , pointing north ) . @xmath164 is the r - component of the very long - time , stream - average field , the ratio @xmath165 giving again the field polarity in each sw stream . to test our understanding of the measured fields and a couple of simple hypotheses , we now use the pdfs of measured @xmath1 to simulate the angles @xmath0 and their pdfs . through monte - carlo simulation , we generate sets of @xmath1 angles with the measured pdfs . assuming ( 1 ) a background field along the parker spiral direction and ( 2 ) axisymmetry of the turbulent fields around that background field , we then transform the angles @xmath1 into the angles @xmath0 between local field and radial direction . finally , we compute the pdfs of these new angles and show them in dotted lines in figure 12 . the modeled pdfs of @xmath0 appear to fit the measured pdfs reasonably well as soon as @xmath166 . at the highest turbulence levels , above @xmath167 , we note that the @xmath0 pdfs , both measured and modeled , are nearly flat on a broad range of angles @xmath0 between @xmath168 and @xmath169 . this we now believe can be attributed to the broad and nearly axisymmetric distribution of the turbulent fields about the background - field direction . in the cases @xmath170 and @xmath171 , where the modeled pdfs are less than satisfactory , we do an additional transformation with a slightly different background field direction , shifted by some 7 and 8 relative to the parker spiral . the results are also shown in dotted lines , but with a slightly different color . they appear to improve the fit to the measured pdfs of @xmath0 , but the fit at the lowest turbulence level remains less than satisfactory . the issue there again might be connected to the measured `` compressibility '' of the turbulence . we also note that the statistics in the central power bin ( @xmath172 ) are relatively high , and that a large number of sw streams contribute to these statistics , with a spread in the parker spiral direction of more than 4 degrees around the mean value of @xmath173 . it may be affecting our result , but reduced statistics with more constrained values of @xmath22 and therefore @xmath24 did not seem to change that result . this issue will need further investigation . from the simple test of this section , we conclude that axisymmetry of the turbulent fields around a background field in or near the direction of the parker spiral is a reasonably good assumption that produces fairly accurate fits of the measured pdfs of @xmath0 angles , away from the lowest turbulence levels where @xmath45 turbulence becomes dominant , that is , above @xmath174 . this `` axisymmetry '' result appears to us difficult to reconcile with an @xmath0 anisotropy of turbulence power driven by a wavevector anisotropy . figure 13 displays the pdfs of `` turbulence levels '' @xmath46 for series of angles @xmath1 ( left panels ) and @xmath0 ( right panels ) measured between 0 and 90 ( top panels ) and between 0 and 180 ( bottom panels ) . the left panels show pdfs peaking at the highest levels of turbulence near the normal to the background field ( green pdf ) , and at the lowest levels near the direction of the background field ( black , yellow ) , with a clear shifting of the peak with the angles @xmath1 and @xmath175 . magnetic fields near the normal to the background dominate at the highest levels of turbulence , while magnetic fields close to the background field direction dominate at the lowest turbulence levels @xmath46 , which can also be seen in figure 12 ( see also dashed - line histograms for entire data set of figure 9 ) . as a consequence , because the angle between background and radial direction is relatively small in fast sw ( @xmath176 ) , it is also true that magnetic fields at large angles to the radial are preponderant at higher @xmath46 , while magnetic fields closer to the radial are more frequent at lower @xmath46 . but the effect is less striking in the right panels of figure 13 than that for the angles @xmath1 relative to the background in the left panels . the strong `` anisotropy '' shown in figures 12 and 13 , and further in figure 17 , of course is related to our choice of the parameter @xmath46 to characterize the turbulence level [ see equations ( [ c][xi ] ) ] . this parameter closely orders the peaks of the @xmath1 and @xmath175 pdfs because it is related to the mean cross - field displacement of the turbulent field lines . but we argue that the power level has to be divided by the squared magnitude of the background magnetic field , as it is done in the definition of @xmath116 and therefore @xmath46 , because in itself , the power in @xmath177 tells us nothing about turbulence ; it is the power in @xmath178 ( of the fluctuations relative to the background ) that does . no matter how high the power in @xmath177 , if @xmath113 is high enough , it will `` quench '' the turbulence . figures 14 and 15 are the `` equivalent '' of figures 12 and 13 with a turbulence level defined using the field fluctuations @xmath179 at only one frequency ( close to @xmath180 ) , rather than using the mean cross - field displacement @xmath181 as in section 3.5 . similar conclusions can be drawn from these figures . so our conclusion that the highest turbulence levels are found near the normal to the parker spiral and the lowest levels near the parker spiral direction is _ not _ due to the use of @xmath182 in place of @xmath179 in the definition of the turbulence level . figure 16 also shows the pdfs of the angles @xmath1 for a series of intervals of the parameter @xmath183 . the pdfs of figure 16 are to be compared to the dashed - line histograms of the top panel of figure 9 for the entire data set . again , the results are similar whether the definition of the turbulence level involves @xmath182 or only @xmath179 . we note , however , that the parameter @xmath184 @xmath185 used throughout this paper is a more meaningful ordering parameter because the turbulent magnetic field lines are affected by the turbulent field fluctuations at a broad range of frequencies , down to where the spectrum becomes flatter than @xmath78 , and not just by the fluctuations at the frequency inverse of the averaging timescale for the computation of the mean local field direction . the parameter @xmath46 , in all instances where the transverse fluctuations dominate , that is , above about @xmath186 , gives the approximate location of the peak in the pdfs of the angles @xmath1 between mean local and background parker fields . and @xmath116 would simply be proportional to any one of the amplitudes @xmath88 , that is , to what is ordinarily understood as the energy level of turbulence in self - similar , non - intermittent turbulence . it is only because the sw turbulence is strongly intermittent that involving the mean cross - field displacement @xmath89 in the definition of the turbulence level presents an advantage . ] knowing the bivariate pdfs @xmath187 for @xmath188 , @xmath175 , @xmath0 and @xmath189 , we can now estimate the average turbulence level at each angle @xmath190 by computing the integral ( ) _ 0^ _ d p ( , ) ^2 , [ cav ] with a cutoff at @xmath191 . the results relative to @xmath192 in the first bin of the @xmath190 histograms , are presented in figure 17 for @xmath193 . the peaks heights are very sensitive to the value of the cutoff @xmath194 . for instance , a value @xmath195 would produce a peak in @xmath196 over 4 times higher than found here in figure 17 , and if all turbulence levels present in our data set were included , that peak would exceed 500 . the value @xmath193 was chosen because it is already sufficiently large to produce high peaks , yet still low enough to keep the rising phase of the curves at low @xmath190 clear . most importantly , we note that the ratio @xmath197 is much greater for @xmath1 and @xmath175 than it is for @xmath0 and @xmath189 . the dependence on @xmath1 of the turbulence level is the main effect , that of the turbulence level on @xmath0 a mere consequence of it . the steeper dependence on @xmath1 , or stronger `` anisotropy '' in @xmath1 , is due to the fact that the @xmath1 pdfs remain peaked at all turbulence levels @xmath46 , whereas the @xmath0 pdfs become flat at the higher @xmath46 ( see figure 12 ) as a result of the broad and presumably nearly axisymmetric distribution of field directions about the background field direction . also , the use of the amplitudes @xmath198 rather than @xmath116 as defined in equation ( [ c ] ) would lead to much steeper curves and higher peaks , depending on the frequency ( see figure 18 ) . the higher the frequency , the higher the peaks , consistent with the findings of podesta ( 2009 ) . finally , when the running @xmath2min average @xmath113 of the background field magnitude is replaced by the long - time average @xmath110 in the definition of @xmath46 , the sharp dependence of the average turbulence level @xmath199 on the angles @xmath1 and @xmath0 is damped out . it is reduced in figure 19 to a mere factor @xmath200 increase in @xmath0 , and is almost entirely suppressed in @xmath1 . the result of figure 19 for the dependence on @xmath0 is very much consistent with the results shown in the top panel of figure 7 in podesta ( 2009 ) . the strong `` anisotropy '' presented in this section is in no way related to any anisotropy in the wavevector distribution of the turbulence . the fact that the turbulence levels are observed to be much higher in the directions close to the normal to the parker spiral than in the directions close to the parker spiral itself is strictly due to the facts that ( 1 ) the field directions deviate more strongly from the parker spiral direction when the turbulence is higher , and ( 2 ) the turbulence level varies strongly with time , enabling the observation of a wide range of field directions and turbulence levels . so the strong apparent `` anisotropy '' presented in this section is strictly a consequence of the turbulence - level time variability . in an effort to identify the effects of the broad variations in turbulence levels on the orientations of the local mean fields , we have analyzed the turbulent magnetic fields of a large set of fast sw streams measured onboard _ ace _ and _ stereo a _ and _ b_. our multiscale fourier analysis of the transverse turbulent fields reveals time variations of the spectral power at all frequencies and on many timescales . the higher the time resolution of the spectral analysis is , the stronger the fluctuations of the power level are . the higher - resolution fluctuations far exceed the lower - resolution fluctuations , but roughly follow them in their time average ( see figures 1@xmath383 ) . from this multiscale fourier analysis , we estimate as a linear combination of the fourier amplitudes @xmath88 ( see figure 3 ) the gql mean cross - field displacement of the magnetic field lines in the time - varying intermittent turbulence . this estimate of the mean cross - field displacement in intermittent turbulence is given by the square root of equation ( [ dr ] ) with a local medium - scale ( @xmath201min ) average background field @xmath113 substituting for the stream average @xmath110 . it includes the effects of all turbulent scales and fluctuates on many timescales . this estimate of the mean cross - field displacement , once divided by the field - aligned scale @xmath14 , defines the square root @xmath117 of the power level of the intermittent turbulence in equation ( [ c ] ) , and the related angle , @xmath202 , used throughout the paper to parametrize the power level of the intermittent turbulence . provided `` compressibility effects '' are included at the lowest power levels of turbulence ( @xmath203 ) , modeling the pdfs of the angles @xmath1 between local and background fields produces satisfactory fits of the observed pdfs at all stable power levels of turbulence ( see figures 8 , 10 and 11 ) , that is , on all time intervals with variations in the power level of turbulence that are sufficiently slow . because the highest power levels of turbulence happen in short bursts , the statistics for stable power levels are biased toward the lower turbulence levels and smaller angles ( see figures 6@xmath389 ) . but the fits remain quite accurate at all levels @xmath46 , even for the very low statistics of the higher @xmath46 . because our gql modeling does not assume anything about the three - dimensional distribution of wavevectors in the turbulence , and therefore does not preclude an isotropic distribution of wavevectors , it follows from these fits that an isotropic turbulence could account for the measured pdfs of the angles @xmath1 between local and background fields at all stable levels of turbulence . both the direct measure of the field projection on the background field ( parker spiral direction ) and the pdfs of the measured angles @xmath1 between local and background fields reveal local field reversals that are quite common even within very broad streams of `` unipolar '' fast sw ( see figures 5 and 12 ) . on the whole , they happen some 4 to 10 percent of the time , but at the highest levels of turbulence , short - scale reversals are so frequent that on the scale of @xmath103 cm , reversed fields are practically as common as normal - polarity fields . modeling through monte - carlo simulation the pdfs of the angles @xmath0 between local fields and radial direction from the measured pdfs of the angles @xmath1 , we find that axisymmetry of the turbulent fields around a background field in or near the direction of the parker spiral is a reasonably good assumption . the modeling made under this simple assumption does indeed produce fairly accurate fits of the measured pdfs of @xmath0 angles , away from the lowest turbulence levels where @xmath45 fluctuations become dominant , that is , above @xmath174 . at the highest turbulence levels , above @xmath204 , both the model and observed @xmath0 pdfs are nearly flat on a broad range of angles @xmath0 between @xmath168 and @xmath169 ( see figure 12 ) . unsurprisingly , magnetic fields near the normal to the background field or parker spiral dominate at the highest turbulence levels , while magnetic fields close to the parker spiral direction dominate at the lowest turbulence levels @xmath46 , with a peak of the @xmath1 ( @xmath46 ) angle pdfs smoothly shifting from the parallel to the normal direction ( from the low to the high turbulence levels ) as @xmath46 ( @xmath1 ) increases [ figure 12 ( 13 ) ] . this results in a very steep dependence of the average power level of turbulence @xmath205 of equation ( [ cav ] ) on the angle to the parker spiral , and in a more moderate dependence on the angle to the radial ( figure 17 ) . the steeper dependence on @xmath1 is due to the fact that the @xmath1 pdfs remain peaked at all @xmath46 whereas the @xmath0 pdfs , due to the broad and nearly axisymmetric pdfs of the field directions relative to the background , become flat at the higher @xmath46 . the dependence is found to be even steeper for the higher - frequency amplitudes @xmath206 ( figure 18 ) , but strongly reduced for @xmath199 ( figure 19 ) . our conclusion that magnetic fields near the normal to the background parker spiral dominate at the highest turbulence levels , while magnetic fields close to the parker spiral direction dominate at the lowest turbulence levels is not modified by the use of a single - frequency turbulence - level definition @xmath207 in the place of @xmath208 ( figures 1416 ) . we do not presume to know whether the @xmath0 `` anisotropy '' of @xmath199 presented in figure 19 is a residual effect of the strong variability in the power level of turbulence . but we can certainly conclude that the extreme `` anisotropy '' seen in figures 17 and 18 is unambiguously caused by the time variability in that power level of turbulence , and not by any anisotropy in the wavevector distribution of the turbulence . clearly , a purely isotropic model of turbulence , that is , a model of turbulence with an isotropic distribution of wavevectors @xmath80 , _ can _ reproduce the pdfs of the angles @xmath1 and @xmath0 between local and background fields and between local field and radial direction presented in this paper for series of the power level of turbulence . by this we mean that an isotropic model of turbulence can not be excluded on the basis of the observed dependences of the angle pdfs on the power level of turbulence , or of the average power levels of turbulence on the angles . these dependences are entirely imputable to the time variability in the power level of turbulence observed in the fast sw streams . these results are consistent with the near isotropy found by narita et al . ( 2010 ) from four - point magnetic field measurements of the cluster spacecraft at the turbulence scales close to @xmath103 cm in the sw . the parameter @xmath46 defined by equations ( [ c ] ) and ( [ xi ] ) does order the directions of the local fields best because it physically represents the local _ statistical _ average directions . this tight ordering of the local field directions with the parameter @xmath46 brings the observed `` anisotropies '' of the average power level of turbulence to extremes much higher than otherwise found with the power level _ undivided _ by the local background field . but again , we argue that the power level has to be divided by the squared magnitude of the background magnetic field because in itself , the power in @xmath177 tells us nothing about turbulence , it is the power in @xmath178 that does . no matter how high the power in @xmath177 , if @xmath113 is high enough , it will `` quench '' the turbulence . the variations of @xmath113 itself are likely caused , in large part , by the non - uniformity of the emerging fields within each of the coronal holes at the source of the fast sw streams . we suggest that some of the variations of @xmath113 , in particular the local field reversals ( figure 5 ; section 3.6 ) , may also be signs of intermixed magnetic fields of opposite polarity , originating from a different sw stream , perhaps even signs of ongoing reconnection with these opposite - polarity fields , or remnants of reconnections that took place earlier . though it is of course also possible and even likely that some of these local field reversals are due to the reconnection , at the basis of the corona , within coronal holes , of open field lines with closed magnetic loops . this research was funded by nsf under the grant 0940976 of the solar terrestrial research program . the _ ace _ mag and _ stereo _ impact / mag data were downloaded from the _ ace _ science center at www.srl.caltech.edu/ace/asc and from sprg.ssl.berkeley.edu/impact .

we extend the generalised langevin equation ( gle ) method [ phys . rev . b * 89 * , 134303 ( 2014 ) ] to model a central classical region connected to two realistic thermal baths at two different temperatures . in such nonequilibrium conditions a heat flow is established , via the central system , in between the two baths . the gle-2b ( gle two baths ) scheme permits us to have a realistic description of both the dissipative central system and its surrounding baths . following the original gle approach , the extended langevin dynamics scheme is modified to take into account two sets of auxiliary degrees of freedom corresponding to the mapping of the vibrational properties of each bath . these auxiliary variables are then used to solve the non - markovian dissipative dynamics of the central region . the resulting algorithm is used to study a model of a short al nanowire connected to two baths . the results of the simulations using the gle-2b approach are compared to the results of other simulations that were carried out using standard thermostatting approaches ( based on markovian langevin and nose - hoover thermostats ) . we concentrate on the steady state regime and study the establishment of a local temperature profile within the system . the conditions for obtaining a flat profile or a temperature gradient are examined in detail , in agreement with earlier studies . the results show that the gle-2b approach is able to treat , within a single scheme , two widely different thermal transport regimes , i.e. ballistic systems , with no temperature gradient , and diffusive systems with a temperature gradient .

1512.03705

nanoscale devices and materials are becoming increasingly important in the development of novel technologies . in most applications of these new nanotechnologies , the central system is part of a more complex set up where driving forces are present to establish heat and/or particle flows . the understanding of the corresponding non - equilibrium properties is of utmost importance . this is especially true when one considers potential applications based on the thermal conductivity of materials @xcite and the heat transport within nanodevices @xcite . being able to describe the dynamics and dissipation of such nanoscale atomic systems is central for modern nanoscience . for that , one has to consider the central region of interest as an open system surrounded by a heat bath ( an environment ) which is in contact with the system and is kept at a given temperature . for studying the transport properties , one has to consider the proper non - equilibrium conditions , i.e. the central region is connected to two ( or more ) independent heat baths ( kept at their own temperatures ) . hence a heat flow ( transient or stationary depending on the experimental conditions ) is established between the two baths via the central region . an appropriate general approach for treating this kind of systems is based on the so - called generalised langevin equation ( gle)@xcite . the gle is an equation of motion containing non - markovian stochastic processes where the particle ( point particle with mass ) has a memory effect to its velocity . the gle has been derived for a realistic system of @xmath0 particles coupled to a single realistic ( harmonic ) bath , i.e. a bath described at the atomic level @xcite . the non - markovian dynamics is obtained for the central system with gaussian distributed random forces and a memory kernel that is exactly proportional to the random force autocorrelation function @xcite . solving the gle for complex heterogeneous and extended systems is still a challenge . a major step towards the solution of this problem for a realistic application has been recently given @xcite . in particular , an efficient and transferable algorithm has been developed in ref . [ ] to solve the gle numerically while taking into account the two fundamental features of the gle @xmath1 a time - dependent memory kernel and the presence of a coloured noise , which are absolutely essential for the description of the bath at the atomic level . however , the previous tools have been developed for a single bath only . in order to treat properly the presence of a ( transient or steady ) flow of heat current through the central system , one has to take into account the proper non - equilibrium conditions . that is , one has to consider the presence of at least two baths ( at their own temperatures ) in contact with the central region . the aim of the present paper is to extend the previous gle approaches developed in refs . [ ] to the systems consisting of a central region connected to two spatially separated thermal baths . with such an approach , we can study the non - equilibrium processes in nanoscale systems by using molecular dynamics ( md ) simulations . the dissipative processes are properly described since the system can exchange energy ( heat ) with the environment . the environment consists of the two baths whose dynamical properties are described more thoroughly than when standard thermostats ( i.e. langevin , nose - hoover , velocity renormalisation , etc ) @xcite are used in conventional md simulations . note that , for a system at equilibrium , the langevin dynamics applied only to a part of the system can be derived from the gle assuming short - range atomic interactions and the markovian approximation @xcite . it was also shown in [ ] that thermostatting only some of the degrees of freedom by using the langevin dynamics brings the system , in equilibrium , to the corresponding canonical distribution . it can also be shown that the application of the nose method to only a part of the system ( i.e. to a subset of atomic degrees of freedom ) also performs correct thermostatting of the entire system to the target temperature . the paper is organised as follows . in sec . [ sec : generalisation ] , we generalise the methodology of refs . [ ] to the cases of two independent baths @xmath2 each having its temperature @xmath3 . the generalisation , called gle-2b , includes the use of two sets of auxiliary degrees - of - freedom ( dof ) corresponding to each bath and their stochastic dynamics . this is performed by the use of a multi - variate markovian stochastic process for the position and momentum of the dof of the central region and the corresponding `` position '' and `` momentum '' of the two sets of auxiliary dof @xcite . the resulting algorithm is explicitly described in appendix [ app : algo ] and has been implemented in the code lammps @xcite following our previous work on the gle with a single bath @xcite . in sec . [ sec : gle_ex ] , we consider some applications of the gle-2b for a specific realistic system . it consists of a short al nanowire connected to two al baths . each bath is represented by a set of auxiliary dof generated from a model solid , i.e. one half - sphere of an al fcc lattice . in this first application of our gle-2b approach , we concentrate on the steady - state properties of the system . first , we consider the equilibrium condition and an artificially thermally decoupled system to perform a first validation of our methodology . then we treat the proper non - equilibrium conditions when @xmath4 . we also compare our results with other possible thermostatting procedures . in sec . [ sec : ne ] we interpret our results for the temperature profiles through the system in terms of the properties of integrable versus non - integrable cases . in the latter case , a full temperature gradient is established in the system , while in the former case there is no temperature gradient built up in the system . in terms of transport properties , a perfect thermal conductor ( a ballistic thermal system ) has an infinite conductance ( in the thermodynamic limit ) and , therefore , there is no temperature gradient within the central part of the system . whenever the system presents some form of thermal resistance ( finite conductance ) , a temperature gradient exists in the system . finally , in sec . [ sec : ccl ] , we conclude and discuss further developments of the present study . regions at temperature @xmath3 . because the forces @xmath5 and the quantities @xmath6 are of finite range ( not necessarily short ranged ) , one can perform the mapping of @xmath7 on a finite region of space : the bath reduced region ( in pink ) , one for each bath . , width=264 ] we consider a central system ( of finite size ) interacting with two independent baths @xmath2 [ ] , see fig . [ fig : sys ] for a schematic representation of the system . the corresponding classical lagrangian is given by @xmath8 , where @xmath9 @xmath10 @xmath11 the positions of the atoms , labelled @xmath12 with mass @xmath13 , of the central system are given by vectors @xmath14 with components @xmath15 ( @xmath16 indicating the appropriate cartesian coordinate ) . @xmath17 is the lagrangian of the system with potential energy @xmath18 . the lagrangian @xmath19 describes the harmonic bath ( bath @xmath20 ) . the bath s atoms are labelled @xmath21 and have masses @xmath22 . we introduce a shorthand notation for the labels of the bath degrees of freedom ( dof ) @xmath23 , where @xmath24 indicates the cartesian coordinate . the corresponding potential energy @xmath25 is harmonic with respect to the displacements @xmath26 of the bath atoms from their equilibrium positions . its expression is @xmath27 where @xmath28 are the elements of the dynamic matrix of the bath @xmath20 . the lagrangian @xmath29 for the bath @xmath30 is similar to @xmath19 and obtained from @xmath19 by swapping the bath index @xmath31 . finally the interaction between the central system and the baths is a linear superposition of the interaction between the system and each bath ( we recall that the baths are independent ) . the individual contribution @xmath32 is taken to be linear in the bath displacements @xmath33 with the following expression : @xmath34 note that the dependence of such an interaction on the system dof , via @xmath35 , remains arbitrary . we can now derive the equations of motion for the central system and baths dof from the lagrangian , eqs . ( [ eq : lagrangian-1]-[eq : lagrangian-3bis ] ) , following refs . we find , for the central system dof , that @xmath36 where @xmath37 . for the bath dof , we find two sets of equations which can be solved analytically , since the lagrangian @xmath38 is harmonic in the bath dof @xmath33 . these sets of equations are given by @xmath39 for @xmath2 . equation ( [ eq : bath_eom ] ) can be solved by introducing the kernel of the differential equation defined from the eigenstates @xmath40 and eigenvalues @xmath41 of the dynamical matrix @xmath42 . the solution of eq . ( [ eq : bath_eom ] ) is then substituted into eq . ( [ eq : sys_eoms ] ) to obtain a closed equation in terms of the system dof only . we consider the initial positions and velocities of the bath atoms , appearing in the solution of eq . ( [ eq : bath_eom ] ) , being stochastic . it permits us to derive a generalised langevin - like equation of motion ( eom ) for the system dof @xcite : @xmath43 the dynamics of the system dof is governed by an effective potential @xmath44 , two memory kernels @xmath45 and stochastic forces @xmath46 corresponding to each independent bath @xmath47 . the potential energy @xmath48 is given by the nominal potential energy @xmath49 inside the central system plus the potential energy between the central system and the two frozen baths . there is also a `` polaronic '' correction energy due to the coupling between the system atoms and the harmonic displacements of the baths atoms around their equilibrium positions : @xmath50 the memory kernel for the bath @xmath47 is given by @xmath51 the polarisation matrix @xmath52 entering the above definitions is obtained from the eigenstates and eigenvalues of the dynamical matrix of the corresponding bath @xmath47 as follows : @xmath53 finally , the stochastic ( and hence non - conservative ) forces @xmath46 are functions of the initial positions and velocities of the dof of the bath @xmath47 . following refs . [ ] , we can now assume that each bath @xmath47 , described by the combined lagrangian @xmath54 , is in thermodynamic equilibrium at temperature @xmath3 . therefore , the stochastic forces @xmath55 can be treated as random variables . from these assumptions , the dissipative forces are well described by a multi - dimensional gaussian stochastic process with correlation functions @xcite : @xmath56 following ref [ ] , we transform the gle given by eq . ( [ eq : gle ] ) into a more convenient set of markovian langevin dynamics ( with white noise ) by introducing a set of auxiliary dof for each bath @xcite . first we proceed with a mapping of the memory kernel of each bath by transforming the @xmath57 matrices as follows @xcite @xmath58 we then introduce two sets of auxiliary dof ( adof ) @xmath59 corresponding to each independent bath @xmath2 . they are associated with the corresponding mapping coefficients @xmath60 with @xmath61 and @xmath62 with @xmath63 . note that the frequencies @xmath64 used in the mapping eq . ( [ eq : gle_mapping_vdof ] ) of the matrix @xmath57 are not directly related to the eigenvalues @xmath41 of the dynamical matrix @xmath65 , as explained in detail in ref . [ ] . there would be a one - to - one correspondence only when the number of adof , @xmath66 , is exactly equal to the number of eigenvalues . for a memory kernel of the type given in eq . ( [ eq : gle_mapping_vdof ] ) , solving the gle eq . ( [ eq : gle ] ) is equivalent to solving the following extended variable dynamics @xcite : @xmath67 the corresponding total vector state @xmath68 $ ] follows a multivariate markovian process , where @xmath69 are independent wiener stochastic processes with ( white noise ) correlation functions @xmath70 we recall that even if the total vector state @xmath71 corresponds to markovian processes , a subset of its components , for example the vector @xmath72 $ ] , is not necessarily following markovian processes @xcite . this was clearly shown in the previous section where the random noise of the corresponding gle is actually a coloured noise given by the memory kernel of the gle . for such classes of non - markovian processes ( that are components , or functions of one or more components , of multivariate markovian processes ) self - consistency is guaranteed as the chapman - kolmogorov equation is satisfied @xcite . following the scheme given in ref . [ ] , we now develop a fokker - planck ( fp ) approach to derive a set of equations which are equivalent to the equations of the extended langevin dynamics given by eq . ( [ eq : extended_gle ] ) . we consider the probability distribution function ( pdf ) @xmath73 of the total vector state @xmath71 . such a pdf follows a fp dynamical equation which can be written as follows @xmath74 where @xmath75 is the corresponding fp liouvillian . we split the liouvillian @xmath75 in two parts @xmath1 a conservative part @xmath76 and a dissipative part @xmath77 . the dynamics generated by the conservative part @xmath76 corresponds to the eom of the dof and adof given in eq . ( [ eq : extended_gle ] ) if one omits all the terms containing the @xmath78 parameters @xcite . the remaining dissipative part @xmath77 generates the eom of the adof , given by the following generic form : @xmath79 for such a stochastic dynamics there exists an exact integration algorithm @xcite . in order to obtain an integration algorithm ( see details in appendix [ app : algo ] ) , we consider the time evolution of the pdf over an elementary time - step @xmath80 @xmath81 we use the splitting of @xmath75 and a second - order trotter expansion to decompose the time evolution operator as follows @xcite : @xmath82 the first and last time evolution operators @xmath83 with half a time step @xmath84 generate steps ( a ) and ( h ) in the algorithm given in appendix [ app : algo ] . furthermore , we use a second trotter expansion of the term @xmath85 by splitting @xmath76 in two parts @xmath1 @xmath86 and @xmath87 . the part in @xmath87 generates the time evolution of the system dof @xmath88 and of the adof @xmath89 over half a timestep @xmath84 , see steps ( c ) and step ( g ) in appendix [ app : algo ] . the part in @xmath86 generates the time evolution of the system dof @xmath15 and of the adof @xmath90 over @xmath91 , see steps ( d ) and ( f ) in appendix [ app : algo ] . in order to perform the mapping given by eq . ( [ eq : gle_mapping_vdof ] ) , we first fourier transform the equation into : @xmath93 . \end{split } \label{eq : mapping_pi_matrix}\ ] ] for each bath , the set of parameters @xmath94 is obtained from a fitting procedure ( described in detail in ref . [ ] ) based on the vibrational properties of the bath . as shown in ref . [ ] , the polarisation matrices @xmath95 are related to the imaginary part of the phonon bath propagator @xmath96 as follows : @xmath97 the bath propagator is defined from the dynamical matrix @xmath98 of the bath @xmath47 as @xmath99^{-1}_{b_\nu , b_\nu^\prime } , \nonumber\ ] ] where @xmath100 . we use a real - space method , based on the lanczos algorithm , to calculate the inverse of the matrix defining @xmath101 and the fitting procedure , described in ref . [ ] , to get the values of the parameters @xmath94 . once the system is defined , the calculations of the dynamical matrices and the mapping procedures are performed individually for each bath @xmath47 . it should be noted that the generalisation to the case of the central system connected to @xmath102 independent baths is straightforward . one can expand the gle-2b by introducing @xmath102 auxiliary sets of dof @xmath90 and @xmath89 . the eom of these adof are given in the 3rd and 4th lines of eq . ( [ eq : extended_gle ] ) . the eom for the momenta of the central system will include the contribution of the @xmath102 baths via the set of adof . it is defined by the sum over all the bath indices instead of just a sum over @xmath2 , i.e. @xmath103 the mapping and fitting procedures of the polarisations matrices @xmath92 will be performed for each individual bath . ) and right ( @xmath104 ) baths . for the gle calculations , the @xmath105 and @xmath104 bath reduced region contains 68 atoms each . for the calculations of the dynamical matrix and the mapping of the polarisation matrix , larger baths ( each containing 203 atoms ) were considered @xcite . the embedded atom method ( eam ) is used for the inter - atomic effective potential . for convenience , we now call the baths @xmath106 instead of @xmath2 . , width=226 ] we now apply the gle-2b approach to a model system , which consists of a short al nanowire connected to two independent baths ( represented by two half spheres ) as shown in figure [ fig : system ] . for obvious reasons , we now call the left and right baths @xmath106 ( instead of @xmath2 ) . the electronic transport properties of similar al nanowires have been studied some decades ago @xcite , it is now interesting to study their thermal transport properties under the proper non - equilibrium conditions . figure [ fig : system_gle ] provides more information about the central system ( labelling of different layers of the system ) and the bath reduced regions . the system is built using a fcc lattice and the distance between layer a and layer c ( fig . [ fig : system_gle ] ) corresponds to the lattice parameter of 4.05 . the al nanowire ( layers a to g ) has the length of 12.15 . the central system ( layers l3-l1 , a - g , r1-r3 ) for which the gle-2b simulations are performed has a length of 20.25 . we have taken the embedded atom method @xcite to model the metallic al system . the tabulated interatomic potential is provided by the nist interatomic potential repository project @xcite . in order to compare to the results obtained with our gle-2b approach , we also consider two different thermostatting approaches for the baths . these are more widely used and consist of stochastic dynamics for the atoms in the bath regions , while the central region follows the common deterministic newtonian dynamics ( see fig . [ fig : system_gle ] and appendix [ app : algo ] ) . we consider two stochastic dynamics , i.e. a simple langevin ( langevin - gauss lg ) dynamics @xcite and the dynamics generated by a nose - hoover ( nh ) thermostat which are already implemented in the md code lammps . for the simple lg approach , the stochastic dynamics for the atoms in the bath regions are given by @xmath107 with the momentum vector @xmath108 of the atoms in the bath @xmath106 and the white noise vector @xmath109 ( following a gaussian distribution with a width related to the temperature @xmath3 ) . the single parameter @xmath24 characterises the friction ( damping @xmath110 ) for all atoms in the bath regions @xcite . for the nh thermostating approach , each is also characterized by a damping parameter @xmath111 ( in unit of time ) in the lammps implementation . as a first application of our gle-2b , we concentrate on studying the steady - state properties of the system . for all stochastic dynamics , we consider some initial conditions ( values of the baths temperatures @xmath112 and @xmath113 and of the velocities of the atoms in the central system ) and let the system evolve in time until the total kinetic energy reaches a plateau , i.e. a constant value up to the corresponding thermal fluctuations . for the different gle-2b calculations , the system takes roughly 80 ps to reach a steady state regime ( i.e. @xmath114 40000 time steps for a @xmath80 = 2 fs ) . for the lg and nh thermostat calculations , the steady state can be reached in less time - steps since the characteristic damping time is adjustable by the user . and @xmath113 respectively . for the other thermostatting approaches , the bath reduced regions are described by their own stochastic dynamics ( langevin - gauss ( lg ) or nose - hoover ( nh ) thermostatting ) at their own temperatures @xmath112 and @xmath113 ) . the central system ( l1-l3 , a - g , r1-r3 ) evolves according to a newtonian ( nve ) dynamics . the remaining outer atoms are kept at fixed positions to ensure the structural stability of the system in the lg and nh thermostatting calculations . , width=264 ] in order to validate our gle-2b approach and the corresponding algorithm , we first need to check that when the bath temperatures are equal , @xmath115 , we obtain the correct results for the energy and/or the velocity distributions as expected from equilibrium statistical mechanics . we have performed different calculations for different temperatures @xmath116 where @xmath117 = 100 , 150 , 200 , 300 , 400 k. figure [ fig : equi_veloc_gle ] shows the velocity distribution functions for different atomic layers of the central system @xcite . the results for the velocity distributions show that the system indeed reaches the expected thermal equilibrium , where the velocity distributions follow the maxwell equilibrium distribution . . ( top left panel ) layers l3 , l2 , l1 of the central system . ( top right panel ) layers r1 , r2 , r3 of the central system . ( bottom panel ) layers a to g. the gle calculations are performed for @xmath115 = 200 k. all distributions ( histograms ) fit perfectly the equilibrium maxwell distribution ( solid red line ) as expected . , width=283 ] it should be noted that for temperatures above @xmath114 400 k , our model system shows some structural instabilities @xcite . therefore calculations are performed only with temperatures lower than 400 k. to further complement the validity of our approach , we have also performed calculations for pseudo double equilibrium conditions . this is done by considering the two baths at two different temperatures , and keeping fixed the atomic coordinates of the central layer d of the central system . this creates a thermal decoupling between the two sides , i.e. the frozen layer d acts as a perfect reflective barrier for the thermal transport between the two baths . figure [ fig : veloc_frozenmid_gle ] shows the results obtained for the velocity distribution when the bath temperatures are @xmath112 = 200 k and @xmath113 = 125 k [ ] . one clearly sees that one side of the system ( layers l3 , l2 , l1 , a , b , c ) has velocity distributions that are well represented by the equilibrium maxwell distributions obtained for @xmath112 , while the other side ( layers e , f , g , r1 , r2 , r3 ) has equilibrium velocity distributions obtained from @xmath113 . similar results are also obtained when using the lg or nh thermostatting procedures for the bath stochastic dynamics . . the gle calculations are performed for two different bath temperatures @xmath112 = 200 k and @xmath113 = 125 k [ ] . the central layer d is frozen , i.e. atomic positions fixed , and acts as a perfect thermal barrier . the distributions ( histograms ) for the atomic layers on both sides of the frozen layer d , namely layers ( l3 , l2 , l1 , a , b , c ) and layers ( e , f , g , r1 , r2 , r3 ) , fit perfectly the equilibrium maxwell distributions obtained for the two temperatures @xmath112 ( lines in the top left and bottom panels ) and @xmath113 ( lines in the top right and bottom panels ) . , width=264 ] with this preliminary set of calculations , we feel confident that our approach and its numerical implementation in the md code lammps are correct and , therefore , we move onto discussing our out of equilibrium calculations . in this section , we consider the proper non - equilibrium conditions when the temperatures of the two baths are different @xmath118 . more specifically , we consider the steady state regime when , after some relaxation time , the total kinetic energy of the system reaches a `` constant '' value ( up to the thermal fluctuations ) . in a nanowire connecting two thermal baths when @xmath118 , it has been shown that a temperature gradient may or may not build up across the system . temperature profile measurements are usually performed , by using thermal probe afm , on mesoscopic scale systems ( a few microns in length ) or on multiwall carbon nanotubes ( 14 nm diameter and 4 microns length ) @xcite . such measurements have not yet been performed on nanoscale objects . however , a lot of theoretical work on model systems can be found in the literature . in the following paragraphs , we will review briefly the causes of the presence or absence of a temperature gradient as reported in previous theoretical studies . since the seminal work of rieder _ et al . _ [ ] , it has been known that in an one - dimensional homogeneous harmonic system ( also referred to as an integrable system ) , the thermal conductivity diverges in the thermodynamic limit . no temperature gradient is formed in the bulk of the system , since the dominating energy `` carriers '' are not scattered and propagate ballistically . a large variety of harmonic ( integrable ) systems have been studied in the classical @xcite and quantum @xcite limits , using analytical and/or numerical approaches . all these studies show that there is no temperature gradient inside the system ( except small regions in the vicinity of the contacts between the central system and the baths ) . one usually obtains a constant - temperature profile @xcite in the central system around the averaged temperature @xmath119 . on the other hand , in classical or quantum non - integrable systems , a temperature gradient is formed inside the system . the temperature gradient is uniform , and the heat conductivity is finite ( it means that these systems obey fourier s law ) . the general trend is that one has to break the integrability of the system in order to build up a temperature gradient there . this can be done by introducing : ( a ) any form of anharmonicity @xcite , ( b ) extra local stochastic processes @xcite or extra collision processes @xcite , ( c ) dephasing for quantum systems @xcite , or ( d ) mode coupling for classical systems @xcite . the introduction of topological / configurational defects @xcite or disorder @xcite in harmonic systems can also lead to the build up of a temperature gradient . this result can be understood from the fact that defect / disorder introduces some form of localisation of the vibration modes @xcite . such modes do not favour ballistic transport , as phonons get scattered by impurities or boundaries @xcite . hence the system has a finite thermal conductivity and presents a temperature gradient across itself . . the gle calculations are performed for proper non - equilibrium conditions with the bath temperatures being @xmath112 = 150 k and @xmath113 = 50 k. the lineshape of the different distributions ( histograms ) is similar to that of the equilibrium maxwell - boltzmann distribution . all distributions fit well onto the maxwell distribution obtained from the average temperature @xmath120 100 k. , width=264 ] we now test our gle-2b method to investigate the thermal transport properties of the al nanowire in the context of ballistic versus diffusive transport regimes . we consider different temperatures @xmath112 and @xmath113 for the left and right baths respectively . the temperatures are chosen between 50 and 300 k. figure [ fig : veloc_ne_gle_tl150tr050 ] shows the velocity distribution for a non - equilibrium calculation performed with @xmath112 = 150 k and @xmath113 = 50 k. this is a typical result , and the following discussion can be applied to other combinations of bath temperatures @xmath112 and @xmath113 ( not shown in the paper ) . the corresponding time evolution of the total kinetic energy of the central system and its statistical distribution is shown in appendix [ app : ekin_andco ] . first of all , we can see that the lineshape of the different velocity histograms for different atomic layers is similar to the lineshape of the equilibrium maxwell distribution . such a behaviour permits us to define a local temperature for each different layer . second , all velocity histograms appear to follow the same equilibrium maxwell distribution corresponding to the average temperature @xmath120 100 k ( even for the layers l3 and r3 embedded in the @xmath105 and @xmath104 bath respectively ) . the overall results clearly indicate that , within such a non - equilibrium regime , we are dealing with an integrable system , essentially a complex harmonic system which perfectly transmits the heat flux between the two baths . to simplify the analysis of our calculations , we now consider only the temperature profiles along the central system . such profiles are calculated for each layer by fitting the local velocity histograms onto the maxwell distribution . we then extract the corresponding local temperature for each layer . , there is no temperature gradient in the system , i.e. it behaves like a perfectly integrable system . there is an inherent uncertainty in the evaluation of the local temperature from a fit to the maxwell distribution . all temperatures have an error bar of @xmath121 k. as a visual guide , this corresponds to the size of the symbols . note that we also have performed calculations swapping @xmath122 and obtained the same flat temperature profiles . , width=283 ] figure [ fig : tprofile_gle_harm ] shows the temperature profiles in the central system for different sets of temperatures @xmath123 . from the procedure of fitting the velocity histograms to the maxwell distribution , we estimate an uncertainty of approximately @xmath121 k on the temperature values . even in the presence of such an error , one can see that there is no temperature gradient in the central system for the different temperatures considered in fig . [ fig : tprofile_gle_harm ] . this implies that , for such temperatures , the system behaves like a perfect harmonic ( integrable ) thermal conductor . we can also compare our gle-2b approach with the other lg and nh thermostatting approaches ( fig . [ fig : gle_and_otherthermostats ] ) . in such approaches , the atoms in the bath reduced regions are allowed to move and follow a dissipative dynamics ruled by a simple lg dynamics or by a nh thermostat . the atoms in the central region ( while interacting with themselves and with the moving atoms of the two baths ) follow a newtonian nve dynamics . , comparison between the gle-2b approach and the other lg and nh thermostatting approaches . while the gle-2b calculations provide a uniform temperature profile inside the system , the lg and nh thermostats show either the building up of a temperature gradient in the central system or a flatter temperature profile depending on the value used for the damping parameter . the lg calculations performed with the damping parameter @xmath125 ps show a temperature gradient , while almost no gradient is obtained for @xmath126 ps . the nh calculations show a strong temperature gradient for @xmath127 ps , and a flatter temperature gradient for @xmath128 ps . there is still a gradient for nh thermostats ( with @xmath127 ps ) even for smaller @xmath129 with @xmath130 and @xmath131 ( see orange curve ) . , width=283 ] from the results shown in figure [ fig : tprofile_gle_and_otherthermo ] , we can see that while the gle-2b calculations provide a uniform temperature profile inside the system , the lg and nh thermostatting approaches display two different kind of behaviour depending on the chosen value of damping parameter . note that in all lg and nh calculations , we have performed the calculations with as many time - steps as required to reach a stationary state . the larger the value of the damping parameter , the longer the dynamics is needed to reach the stationary state ( see also appendix [ app : ekin_andco ] for the time evolution of the kinetic energy ) . for small values of the damping parameter @xmath132 ps ( @xmath127 ps ) , we obtain a temperature gradient in the central system . the nh thermostats appear to provide an almost perfect linear gradient in the whole system , while the simple langevin thermostatting corresponds to a small temperature gradient . however for larger values of the damping parameter @xmath126 ps ( @xmath128 ps ) , both lg and nh thermostats provide a flatter temperature profile . the lg calculations result in a temperature profile almost similar to that of the gle-2b calculations . it is interesting to see that changing the value of the damping can lead to completely different physical results , i.e. the presence of a temperature gradient or its absence in the central system . such a dilemma does not exist in our gle-2b approach since it does not contain any adjustable parameter . one should note that the small values of the damping parameter has been chosen in order to reproduce a relaxation of the kinetic energy ( for the lg and nh thermostatting ) similar to the relaxation obtained from the gle-2b approach ( see also appendix [ app : ekin_andco ] ) . it seems fair to say that fitting the values of the damping parameter ( to reproduce the evolution of the kinetic energy ) is not enough for obtaining the same temperature profiles with the lg , nh thermostats and with the gle-2b approach . as mentioned above , the presence of a temperature gradient is a signature of the breaking down of the integrability of the system , and corresponds to a system with a finite thermal conductance . it also implies the introduction of anharmonic effects , configurational disorder or the introduction of other ( uncontrolled ) random processes . in order to understand if one can obtain a temperature gradient with the gle-2b approach , we have performed calculations in different situations where the harmonicity of the system is broken . this can be achieved be considering higher temperatures . in these cases , the atoms of the central system move in an effective potential which goes beyond the harmonic potential well and consequently one can obtain a finite temperature gradient , as shown in figure [ fig : tprofile_gle_anharm_desorder ] . one can see that larger deviations of the temperatures from the average @xmath119 are obtained on the side of the hotter bath . one can further increase such effects by introducing artificially some configurational disorder in the system . for that , we have considered the same system and at random we have given one atom in each layer l1 , a - g and r1 a mass 20@xmath133 larger ( smaller ) than the mass of the other atoms in the system . the corresponding temperature profile is shown in figure [ fig : tprofile_gle_anharm_desorder ] and presents , as expected for such a disordered and/or anharmonic system , a finite temperature gradient . in appendix [ app:1dmodel ] we present calculations for a model of a one - dimensional al chain . the gle-2b calculations obtained for such a simple model confirm , as expected , the analysis we have performed in this section . and different atomic masses . it is possible to break the integrability of the system by considering larger temperatures ( red curves ) and/or by introducing configurational disorder ( blue curve ) . the disorder is introduced by changing the mass ( @xmath134 ) of some atoms picked up randomly in the central system . for higher temperatures , parts of the system are `` driven '' beyond the harmonic limit . in such cases , the system is not entirely ballistic and presents a finite thermal conductivity , leading to the build up of a temperature gradient . , width=283 ] furthermore , it is also crucial to understand the importance of anharmonic effects . for that we introduce into the gle-2b method some form of anharmonic effects in the baths by taking a finite life - time for the phonon modes of the bath . it means that vibrational excitations do not have an infinite life - time ( as should be the case for an integrable harmonic system ) but rather that they decay ( dissipate ) in time . the simplest way to treat such effects is to introduce a `` self - energy '' in the phonon bath propagator @xmath135^{-1}_{b_\nu , b_\nu^\prime } \ . \nonumber\ ] ] furthermore , we consider that @xmath136 is purely imaginary and simply modifies the linewidth of the spectral features in @xmath137 . in practice , such effects can be implemented in a rather straightforward way : once the mapping of @xmath138 is established , we take the values of the fitted parameters @xmath78 and make them smaller . the features of the corresponding vibration modes in the phonon bath propagator are then broadened @xcite . . such anharmonic effects lead to a small temperature gradient on the side of the hotter bath ( @xmath105 side ) . the building up of the temperature gradient is however not as strong as for the lg / nh thermostatting calculations . , width=283 ] the results of such calculations , shown in figure [ fig : tprofile_gle_anharm_baths ] , suggest that anharmonic effects in the baths tend to favour the build up of the temperature gradient across the system . finally we would like to add that short nanowires can be ballistic ( harmonic regime ) for a range of temperatures @xmath123 . however , such a behaviour might not be true for much larger ( longer ) and strongly heterogeneous systems @xcite . indeed , in such systems , one may expect to observe the presence of disorder , of more localized vibrational modes or , more importantly , of vibrational mode `` mixing '' effects ( interaction between phonons ) which lead to the building up of a temperature gradient in the system . for example , the process of mode coupling has been studied in model three - dimensional systems @xcite and has been shown to lead to the presence of a temperature gradient in the system . we have developed a generalised langevin equation ( gle ) approach to treat non - equilibrium conditions when a central classical region is connected to two realistic thermal baths at two different temperatures . the method is called gle-2b for generalised langevin equation with two baths . following the original gle approach @xcite , the extended langevin dynamics scheme is modified to take into account two sets of auxiliary degrees of freedom , each of which characterises the vibrational properties of the baths . these auxiliary degrees of freedom are then used to solve the non - markovian dissipative dynamics of the central region . we have developed the corresponding algorithm for md simulations and implemented it within the md code lammps . as a first application , we have studied the heat transport properties of a short al nanowire , that connects the left and the right al baths , in the steady - state regime . we have mostly considered the establishment of a local temperature profile within the system when the two bath temperatures are different . our results are interpreted in terms of the properties of harmonic versus non - harmonic systems , and the presence or the absence of defects . in agreement with earlier studies , we found that in a purely harmonic ( ballistic ) thermal conductor ( with spatially extended normal modes ) , there is no temperature gradient across the central part of the system . whenever the system presents some form of thermal resistance ( finite conductance ) due to anharmonic effects , disorder , or extra random processes , a temperature gradient is present in the system . furthermore a concrete example of such effects in a model of a one - dimension al chain is provided in appendix [ app:1dmodel ] . we have also compared the results of the simulations using the gle-2b approach to the results of other simulations that were carried out using standard thermostatting approaches ( based on markovian langevin and nose - hoover thermostats , see fig . [ fig : gle_and_otherthermostats ] ) . in the latter cases , either a flat temperature profile or a temperature gradient across the central system can be obtained depending on the value used for the damping parameter . upon the choice of this parameter , two different physical results can be obtained . such a dilemma does not exist in the gle-2b approach as it does not contain any adjustable parameters . furthermore , we have shown that the gle-2b is able to treat , within the same scheme , two widely different transport regimes , i.e. systems which have ballistic ( with no temperature gradient ) or diffusive ( with temperature gradient ) thermal transport properties . this is a crucial point since the crossover between ballistic and diffusive transport regimes has been observed experimentally @xcite in organic molecules of different lengths connecting two electrodes , after having been predicted theoretically @xcite . penultimately we would like to add that the gle-2b has also another advantage over the more commonly used thermostatting approaches . this method has been derived explicitly in order to be able to treat inherently non - equilibrium properties which can not be simulated ( in principle ) by the nh thermostats . furthermore , we have already shown in appendix d of ref . [ ] that we can derive the gle dynamics with a coloured noise which is not simply proportional to the memory kernel ( as is the case for the classical limit of the equilibrium fluctuation - dissipation theorem ) . this means that quantum effects of the baths can be incorporated in the gle dynamics . the importance of such effects has been considered in refs . [ , ] . finally , it should be noticed that our gle-2b approach is also perfectly appropriate to study time - dependent phenomena . such interesting phenomena , which involve proper dynamical behaviour of systems , will be the subject of future studies . ) is performed only on the atoms of the central region . the @xmath105 and @xmath104 baths ( displaced further to the left and right for clarity ) enter into the gle calculations via the sets of fitting parameters @xmath94 and the @xmath139 quantities . in such calculations , the positions of the bath atoms are fixed at their equilibrium positions shown in fig . [ fig : system_gle ] . ( _ lower panel _ ) the atoms in the central region follow newton s eom and the atoms in the @xmath105 and @xmath104 baths follow a dissipative lg or nh dynamics . the atoms with a black cross are kept fixed to ensure the overall stability of the system . the two approaches , gle-2b and lg / nh , represent two different kinds of stochastic processes . , title="fig:",width=340 ] + ) is performed only on the atoms of the central region . the @xmath105 and @xmath104 baths ( displaced further to the left and right for clarity ) enter into the gle calculations via the sets of fitting parameters @xmath94 and the @xmath139 quantities . in such calculations , the positions of the bath atoms are fixed at their equilibrium positions shown in fig . [ fig : system_gle ] . ( _ lower panel _ ) the atoms in the central region follow newton s eom and the atoms in the @xmath105 and @xmath104 baths follow a dissipative lg or nh dynamics . the atoms with a black cross are kept fixed to ensure the overall stability of the system . the two approaches , gle-2b and lg / nh , represent two different kinds of stochastic processes . , title="fig:",width=226 ] we acknowledge financial support from the uk epsrc , under grant no . ep / j019259/1 . hn , cdl and lk acknowledge the stimulating research environment provided by the epsrc centre for doctoral training in cross - disciplinary approaches to non - equilibrium systems ( canes , ep / l015854/1 ) . finally , ag would like to acknowledge the department of physics at king s college london for funding the summer internship which resulted in her contribution to this project . finally the authors thank one of the referees for a careful and critical analysis of our results and for suggestions that strengthened the value of the present work . following the markovian equations derived in section [ sec : glevdof ] and prescriptions given in ref . [ ] , we use the following algorithm for a single time - step @xmath80 . the algorithm is derived , in a manner similar to the verlet algorithm , from a different splitting and a trotter - like decomposition of the total liouvillian for the extended langevin dynamics of the system dof , @xmath15 , and the auxiliary dof @xmath140 associated with the two independent baths @xmath2 . such a decomposition has been shown to provide a more appropriate description of the velocity correlation functions @xcite . algorithm : @xmath141 where the different forces , @xmath142 are explained below . the force @xmath143 is the force acting on the system dof @xmath144 due to the interaction between the atoms in the system and in the bath region(s ) ; the `` polaronic '' force @xmath145 @xmath146 ( with @xmath147 for dof the @xmath47-th bath ) is the force acting on the system dof @xmath144 due to the interaction between the system and bath regions which induces a displacement of the positions of the harmonic oscillators characterising the baths . in eq . ( [ eq : fpol ] ) , we used the fact that @xmath148 is the inverse fourier transform ( evaluated at @xmath149 ) of @xmath7 given by eq . ( [ eq : mapping_pi_matrix ] ) . the force @xmath150 acts on the system dof @xmath144 and arises from the generalised langevin equations : @xmath151 the force @xmath152 acts on the adof @xmath153 and also arises from the generalised langevin equations @xmath154 the integration of the dissipative part of the dynamics of the adof ( see steps ( a ) and ( f ) in the algorithm ) includes the coefficients @xmath155 and @xmath156^{1/2}$ ] and the uncorrelated random variable @xmath157 corresponding to the white noise . [ app : ekin_andco ] in this appendix we show the time evolution of the kinetic energy and potential energy for a system under nonequilibrium conditions . we also calculate the statistical distribution of the kinetic energy and briefly discuss the convergence of the numerical calculations for the gle-2b and lg approaches . figure [ fig : time_evol_ekintot ] shows the time evolution of the total kinetic energy @xmath158 and of the total potential energy @xmath159 for the non - equilibrium conditions @xmath160 k and @xmath161 k. this is a typical result , and the following discussion can be applied ( qualitatively ) to other combinations of bath temperatures [ ] . for all calculations , @xmath158 reaches an asymptotic ( nonequilibrium ) stationary value . for the gle-2b calculations , the stationary regime is obtained after 80 ps ( around 40000 timesteps ) . while for the lg calculations , the number of timesteps needed to reach the stationary regime is strongly dependent on the value used for the damping parameter @xmath162 . the asymptotic value of @xmath158 is around 0.8 ev , which is completely different than the corresponding values expected from equilibrium and the equipartition principle for the two bath temperatures , i.e. @xmath163 ev and @xmath164 ev . this is indeed not surprising as the central system is not in an equilibrium state . and of the total potential energy @xmath159 of the system shown in fig . [ fig : system ] versus time ( @xmath165 fs ) [ ] . the calculations are performed with the bath temperatures @xmath166 , @xmath124 . the horizontal dashed blue lines correspond to the averages of the energy over the time range 100 to 300 ps . the convergence is fairly well obtained after @xmath114 80 ps ( 40000 timesteps ) . ( b ) : evolution of the total kinetic energy @xmath158 of the central system only ( 59 atoms from layers l3 to r3 ) . for the lg calculations , the convergence towards the stationary value occurs after roughly 80 ps for the damping time @xmath132 ps . the stationary state is reached after more timesteps , as expected , for the larger damping time @xmath126 ps ( see inset where it is apparent that the stationary value is reached after 500 ps or 250000 timesteps ) . , title="fig:",width=283 ] and of the total potential energy @xmath159 of the system shown in fig . [ fig : system ] versus time ( @xmath165 fs ) [ ] . the calculations are performed with the bath temperatures @xmath166 , @xmath124 . the horizontal dashed blue lines correspond to the averages of the energy over the time range 100 to 300 ps . the convergence is fairly well obtained after @xmath114 80 ps ( 40000 timesteps ) . ( b ) : evolution of the total kinetic energy @xmath158 of the central system only ( 59 atoms from layers l3 to r3 ) . for the lg calculations , the convergence towards the stationary value occurs after roughly 80 ps for the damping time @xmath132 ps . the stationary state is reached after more timesteps , as expected , for the larger damping time @xmath126 ps ( see inset where it is apparent that the stationary value is reached after 500 ps or 250000 timesteps ) . , title="fig:",width=283 ] , @xmath124 . the averaged kinetic energy @xmath167 ev is the energy reference . , width=283 ] interestingly , for the nonequilibrium conditions , the choice of the value of the damping parameter @xmath162 for the lg calculations is crucial to obtain the proper physics . one can not simply use the best @xmath162 to reproduce the time evolution of the kinetic energy , as done for the equilibrium case @xcite . the influence of the value of @xmath162 is reflected in the distribution of the kinetic energy shown in figure [ fig : histo_ekintot ] . only the lg calculations performed with a large value of @xmath162 reproduce the distribution of @xmath158 obtained from the gle-2b approach . the lg calculations performed with @xmath132 ps produce a much broader distribution . furthermore , in order to obtain the correct temperature profile given by the gle-2b approach , one needs to use a large @xmath162 in the lg approach . the lg calculations made with @xmath132 ps , which result in a gradient in the temperature profile ( i.e. a completely different physics behaviour , see fig . [ fig : tprofile_gle_and_otherthermo ] ) , show a time evolution of @xmath158 that is much more similar to that obtained from the gle-2b method . these results confirm that the choice of the adjustable parameter for the lg method is crucial for being able to simulate the proper physical behaviour . this is also true for the nh thermostatting approach ( results not shown ) . in this section , we consider another way of coupling the al nanowire to the thermal baths . we treat a system similar to the system shown in figure [ fig : system ] and figure [ fig : system_gle ] by including into the central system only the 7 layers labelled a to g ( system containing 31 atoms ) . the atoms of the layers l1 , l2 and l3 ( r1 , r2 and r3 ) have now been incorporated in the l ( r ) bath regions themselves . for this sytem , we obtain the same physical results for the temperature profile as those presented in the main text [ ] . for low temperatures and small @xmath168 , the central system is harmonic and no temperature gradient is obtained across the system , see figure [ fig : tprofile_alwire2 ] . once more , the lg thermostats provide two different temperature profiles depending upon the value of the damping parameter @xmath162 . for small values of the damping parameter ( @xmath132 ps ) , we obtain a small temperature gradient in the central system while for larger values of the damping parameter @xmath126 ps , the lg thermostats provide a flat temperature profile . in terms of convergence versus the baths size , one should first note that , in the gle-2b approach , we do not choose which group of atoms are connected to a thermal baths , as this is usually performed when using more conventional lg or nh thermostats . the coupling of the central system to the thermal bath @xmath47 is obtained by the coupling of the dof @xmath169 to the adof @xmath90 via the matrix elements @xmath6 , see eq . ( [ eq : gle_pia_nbath ] ) and eq . ( [ eq : extended_gle ] ) . such a coupling exists only when the matrix element @xmath6 is non - zero . we recall that @xmath37 is the derivative , with respect to the coordinate of the dof @xmath169 , of the force @xmath5 felt by the atom @xmath170 of the bath @xmath47 with dof @xmath171 . and [ fig : gle_and_otherthermostats ] . the different bath temperatures are @xmath172 , @xmath173 . while the gle-2b calculations provide a uniform temperature profile inside the system , the lg thermostats show either the building up of a temperature gradient in the central system ( @xmath132 ps ) or a flat temperature profile ( @xmath126 ps ) . , width=283 ] the range of the matrix elements @xmath6 is determined by the cut - off of the inter - atomic potential used in the calculations . the range of the quantities @xmath6 is actually smaller than the cut - off of the inter - atomic potential , as the former is the second derivative ( versus spatial coordinates ) of the latter . hence there is no need to increase the size of the reduced bath region , as long as dofs of the central system are properly coupled to the existing atoms in the reduced bath region . indeed , adding atoms in the reduced bath region ( which corresponds to @xmath174 elements ) will not change the dissipative dynamics of the atoms of the central region . however , one should not forget that the infinite spatial extension of the baths has already been taken into account . in particular , the continuous vibrational spectra of the infinite baths has been obtained through the calculation of their dynamical matrix and subsequently in the mapping procedure described in sections [ sec : glevdof ] and [ sec : calc_pi_mat ] ( for more detail , see also ref . [ ] ) . the coupling to the thermal baths comes directly from the construction of the geometry of the system itself and from the cut - off of the inter - atomic potential . it is not controlled by the user as usually done with lg or nh thermostats . this is , once more , one of the main differences between the consistent ( and more elaborate ) gle-2b approach and the main - stream lg or nh thermostatting approach . as far as the convergence versus the size of the central system is concerned , we have already mentioned in the main text that nonballistic transport properties can be obtained for much longer and heterogeneous systems @xcite . the presence of long wavelength accoustic modes and their indirect coupling ( via the baths ) with other delocalised vibrational modes can also lead to the establishment of a more diffusive transport property associated with the presence of a temperature profile in the central system . the study of the transport properties versus the size of the central system is important , but out of the scope of the present paper , and will be treated elsewhere . in this section , we consider a toy model for a one - dimensional system : a chain made of 11 al atoms connected to two baths as shown in fig . [ fig:1dsystem ] . this is a simpler system than the three - dimensional wire considered in the main body of the paper . we use the gle-2b approach for the dynamics of the central chain and study the local temperature profile of the perfect chain and of the chains containing one or two defects . to model the defect , we simply change the mass of the corresponding atom while conserving the same eam inter - atomic potential for all atoms in the gle-2b calculations . to make the system simpler , we constrain the dynamics of the atoms in the central chain to a purely one - dimensional problem , i.e. the atoms can move only along the main axis ( the @xmath175-axis ) of the chain . therefore each atom in the central region is associated with only one degree of freedom . for a system at equilibrium , @xmath115 , the gle-2b calculations provide a velocity distribution for each atom of the central region which follows , as expected , the statistical gaussian distribution of a single degree of freedom given by @xmath176 ) and right ( @xmath104 ) baths . for the gle calculations , the @xmath105 and @xmath104 bath reduced region contains 30 atoms each . for the calculations of the dynamical matrix and the mapping of the polarisation matrix , larger baths ( containing each 95 atoms ) were considered . the mapping of the polarisation matrix elements is done using 78 adof with a set of 78 parameters @xmath94 for each bath @xmath106 . , width=264 ] for the non - equilibrium conditions , @xmath118 , the velocity distributions , calculated from the gle-2b , for the atoms of the central chain also follow the lineshape of a gaussian distribution . however the associated temperature varies across the chain . . the labels lbath ( rbath ) represent the temperatures at the left ( right ) bath . the other labels a to k indicate different atoms in the central chain . for temperatures @xmath123 , the system is harmonic and one gets a flat temperature profile as expected for integrable systems . for higher temperatures , the motion of some of the atoms sample the anharmonic part of the interatomic potential , and therefore the system is non - integrable and a temperature gradient is built up . , width=283 ] figure [ fig : tprofile_1dchain ] shows the temperature profile across the chain when the masses of all atoms in the central chain are equal . for low temperatures @xmath123 , the homogeneous system behaves as a harmonic system and one obtains a flat temperature profile as expected for integrable systems . for higher temperatures , the motion of some of the atoms start to sample the anharmonic part of the inter - atomic potential . therefore some parts of the system are not harmonic any more and a temperature gradient starts to build up . note that because the dynamics of the system is strongly constrained , no structural instability is possible and higher local effective temperatures ( in comparison with the 3d short wire considered in the main part of the paper ) can be investigated in order to achieve the anharmonic regime . is increased by 10 , 20 , 30 or 40% . the case of two defects ( atoms c and h with their mass increased by 40% ) is also shown . for the temperatures considered , the chain is in the harmonic regime . however the presence of defects , and their associated more localised vibration modes , leads to the build up of a temperature gradient across the chain . , width=283 ] figure [ fig : tprofile_1dchain_chmass ] shows the temperature profile across the chain when one introduces a localised defect in the chain . calculations are performed when the mass of the atom labelled h in the chain is increased by 10 to 40% and when the mass of both atoms c and h is increased by 40% . one can see that the introduction of a defect in the chain ( in the harmonic regime ) leads to the build - up of a temperature gradient . such an effect is clearly obtained for a mass increase larger than 10 % and in the cases of more than one defect present in the chain . we have checked that , in the presence of defects , the vibration modes of the chain are slightly more localised ( around the defects ) than for the perfect chain . furthermore the amplitudes of the vibration modes at the ends of the chain are also different in the presence of defects in the chain . such effects are thought to hinder the heat transport from one bath to the other and hence lead to the absence of a flat temperature profile across the chain . , , and , * * , ( ) , , , and , * * , ( ) and , * * , ( ) and , * * , ( ) , , , and , * * , ( ) , , and , * * , ( ) . , , and , * * , ( ) . and , * * , ( ) . a. majumdar , annu . * 29 * , 505 ( 1999 ) . there exists different possible choice for the trotter factorization of the liouvillian . from our understanding of ref . [ ] , the trotter decomposition we use to obtain our gle - md algorithm has been shown to provide a more appropriate description of the velocity correlation functions . an accurate statistical description of the velocities is crucial in the present paper , as we extract the local temperature from the velocities statistical distributions . it should noted that there can not be any form of coupling between the two baths . otherwise they would not be independent and therefore it would not be possible to consider them as being kept at their own thermodynamic equilibrium characterised by the temperature @xmath3 . the velocity distributions are calculated from the averages of several realisations of each velocity distribution obtained at a given timestep , after the system is thermalised . for @xmath117 = 200 k , the thermalisation is reached after 78 ps of gle dynamics , i.e. after 40000 time - steps with @xmath80 = 2 fs . the distributions are averaged over all the time - steps in the time range @xmath177 $ ] ps . calculation using the lg thermostats and the gle-2b approach were performed at equilibrium when the system is uniformly thermostatted at @xmath178 . from the results of these calculations , we observed that the al nanowire melts down at t = 500 k for the eam inter - atomic potential we used . during the md run , we observe the formation of the smaller constriction in the middle of the nanowire and eventually all the atoms of the layers a to f end up on the surfaces of the l and r bath regions . at temperatures t = 400 - 450 , we also obtain the formation of a smaller constriction in the middle of the nanowire ( with atoms moving towards the left and right sides of the constriction ) . the constriction consists of 3 layers of only 2 atoms in each layer . this is what we call `` structural instabilities '' , meaning that the central system does not have the same atomic configuration ( layers made of 4 or 5 atoms ) . such strong modifications of the configuration of the system make it difficult to perform a consistent analysis of the ( layer - by - layer ) temperature profiles . one should notice that the melting of al surfaces , depending on the miller indices @xmath179 , may occur even at temperatures which are 150 to 200 k below the bulk melting temperature [ see for example p. stoltze , j. k. nrskov , and u. landman , phys . lett . * 61 * , 440 ( 1988 ) ] , i.e. roughly around 730 - 780 k. the coordination of an al atom at the surface is roughly half of that of an atom in the bulk . therefore we believe that it is not surprising that the melting of the nanowire happens at even lower temperature as the coordination of the atoms inside the nanowire is even smaller that the coordination of surface atoms . the atoms of the bath reduced regions are present in the gle-2b method ( implemented in the lammps code ) , as one needs to calculate the matrix elements @xmath180 for the instantaneous atomic positions of the central system . however the atoms in the bath reduced regions are kept fixed during the gle-2b runs , hence they do not contribute to the total kinetic energy of the system shown in fig . [ fig : system ] . futhermore the contribution , to the total potential energy , from these atoms is just a constant throughout the gle-2b runs . the gle-2b runs , as well as the lg runs with @xmath132 ps , are performed for 120 ps ( 120000 timesteps with @xmath181 fs ) . the steady state is reached around 80 ps . the velocity histograms ( from which the local temperature is extracted ) are obtained from statistical averaging over the time range 100 to 120 ps . for the lg runs with @xmath126 ps , the steady state is reached after a longer time , @xmath114 300 ps . the velocity histograms are obtained from statistical averaging over the time range 450 to 600 ps .

we consider the low scale implications in the extended mssm ( umssm ) . we restrict the parameter space such that the lightest supersymmetric particle ( lsp ) is always the lightest neutralino . in addition , we impose quasi yukawa unification ( qyu ) at the grand unification scale ( ) . qyu strictly requires the ratios among the yukawa couplings as , , and . we find that the need of fine - tuning over the fundamental parameter space of qyu is in the acceptable range ( ) , even if the universal boundary conditions are imposed at , in contrast to cmssm and nuhm . umssm with the universal boundary conditions yields heavy stops ( tev ) , gluinos ( tev ) , and squarks from the first two families ( tev ) . similarly the stau mass is bounded from below at about tev . despite this heavy spectrum , we find , which is much lower than that needed for the minimal supersymmetric models . in addition , umssm yield relatively smallterm , and the lsp neutralio is mostly form by the higgsinos of mass gev . we obtain also bino - like dark matter ( dm ) of mass about 400 gev . wino is usually found to be heavier than higgsinos and binos , but there is a small region where tev . we also identify chargino - neutralino coannihilation channel andresonance solutions which reduce the relic abundance of lsp neutralino down to the ranges compatible with the current wmap measurements .

1604.06430

even if the minimal supersymmetric extension of the standard model ( mssm ) is compatible with the current experimental measurements for the higgs boson , recent studies show that realizing a higgs boson of mass around 125 gev in minimal models such as constrained mssm ( cmssm ) and models with non - universal higgs masses ( nuhm ) requires a heavy supersymmetric particle spectrum . the higgs boson of mass about 125 gev leads to the stop quark mass in multi - tev range @xcite , or necessitates a large soft supersymmetry breaking ( ssb ) trilinear term @xmath15 @xcite . in addition to the higgs boson results , also absence of a direct signal in the experiments conducted in the large hadron collider ( lhc ) has lifted up the mass bounds on the supersymmetric particles , especially in the color sector . for instance , the current results exclude the gluino of mass lighter than @xmath16 tev when @xmath17 @xcite , which becomes severer when @xmath18 , where @xmath19 denotes the squarks from the first two families . even though these bounds are mostly for r - parity conserved cmssm , they are applicable for a large class of supersymmetric models . while there are numerous motivations behind the supersymmetry ( susy ) searches , such heavy spectrum has brought naturalness under scrutiny . it is clear that the recent experimental constraints can not be satisfied in the natural region identified with @xmath20 gev @xcite . even though it is possible to find @xmath21 gev @xcite , @xmath22 needs to be very heavy because of the necessity of large mixing . apart from the natural region , one might measure how much fine - tuning is required by considering the @xmath23boson mass ( @xmath24 gev ) @xmath25 where @xmath26 is the bilinear mixing of the mssm higgs doublets , @xmath27 , ratio of vacuum expectation values ( vevs ) , @xmath28 are the radiative effects from the higgs potential and @xmath29 are the soft supersymmetry breaking ( ssb ) mass terms for the higgs doublets @xmath30 . a recent work @xcite has defined the following parameter to quantify the fine - tuning measure @xmath31 where @xmath32 the fine - tuning can be interpreted as the presence of some missing mechanisms and its amount measures the effects of such missing mechanisms . their effects can be reflected within susy models by considering non - universality or adding extra sectors to the theory @xcite . in this respect , it is interesting to probe the models beyond the mssm models in light of the current experimental results . note that in contrast to the natural region characterized with the stop and sbottom masses , the fine - tuning does not depend on these masses directly . from moderate to large @xmath33 values , @xmath34 is needed in order to obtain correct z boson mass @xmath35 ; hence , the fine - tuning is mostly determined by @xmath36 , unless @xmath26 is so small that the large radiative corrections to @xmath37 are needed in eq.([zmass ] ) . thus , large stop or sbottom masses can still yield an acceptable amount of fine - tuning . the conclusion that the fundamental parameter spaces of cmssm and nuhm need to be highly fine - tuned is raised due to the strict universality in the boundary conditions of these models . in this work we consider the mssm extended by an additional @xmath38 group ( umssm ) in the simplest form . a general extension of mssm by a u(1 ) group can be realized from an underlying gut theory involving a gauge group larger than su@xmath39 . for instance the following symmetry braking chain @xmath40 where @xmath41 is the mssm gauge group , and u(1)@xmath42 can be expressed as a general mixing of @xmath43 and @xmath44 as @xmath45 emergence of so@xmath46 and/or su@xmath39 allows one to imply a set of boundary conditions , which can be suited in these groups . for instance , the supersymmetric particle masses can be set universal at the grand unified scale ( @xmath1 ) in @xmath47 , while two different mass scales can be imposed to the fields in @xmath48 and @xmath49 representations of @xmath50 . in exploring this extension , we briefly aim to analyze the effects only from having another gauge sector , which is not included in the minimal susy models , by imposing universal boundary conditions at @xmath1 . in addition to the boundary conditions imposed on the fundamental parameters , we also restict the yukawa sector such that the yukawa couplings , especially for the third family matter fields , are determined by the minimal @xmath51 ( or @xmath47 ) unification scheme . if a model based on e@xmath52 gauge group @xcite is constructed in a minimal fashion in a way that all the matter fields of a family are resided in a @xmath53 dimensional representation and the higgs fields in another @xmath53 , such a model also proposes unification of the yukawa couplings ( yu ) as well as the gauge couplings . this elegant scheme of unification can be maintained if e@xmath52 is broken down to the mssm gauge group via so@xmath46 , since models based on the so@xmath46 gauge group reserves yu . even though it is imposed at @xmath1 , yu is also strongly effective at the low scale , since it requires the threshold corrections at the low scale @xcite . relaxing yu to @xmath54 yu does not weaken its strength on the low scale implications , since @xmath55 still requires large and negative susy threshold corrections @xcite . despite its testable predictions at lhc @xcite , yu rather leads to contradictory mass relations such that @xmath56 and @xmath57 , @xmath58 , and @xmath59 . one way to avoid this contradiction and obtain realistic fermion masses and mixing is to propose vector - like matter multiplets at the gut scale @xcite , which are allowed to mix with fermions in 16-plet representation of @xmath47 . this approach is also equivalent to introduce non - renormalizable couplings along with non - zero vacuum expectation value ( vev ) of a non - singlet so@xmath46 field @xcite . another way is to extend the higgs sector with an assumption that the mssm higgs doublets are superpositions of fields from different so@xmath46 representations @xcite . even though yu for the third family can be consistently maintained under assumptions that the extra fields negligibly interact with the third family and the mssm higgs doublets solely reside in 10 dimensional representation of so@xmath46 , these two approaches , in general , break yu in so@xmath46 . on the other hand , if one can formulate the asymptotic relation among the yukawa couplings , then the contributions can be restricted such that the quasi - yu ( qyu ) can be maintained . for instance , it was shown in ref . @xcite that in the presence of higgs fields from @xmath60 in addition to those from @xmath61 of the pati - salam model @xcite yukawa couplings at @xmath1 can be expressed as @xmath62 the gauge group of the pati - salam model , @xmath63 is the maximal subgroup of @xmath47 , and hence these extra higgs fields can be employed in @xmath47 gut models . the parameter @xmath64 denotes the contributions to yukawa couplings from the extra higgs fields , and restricting these contributions as @xmath65 , eq.([qyu ] ) refers to the qyu condition . note that @xmath64 can be either positive or negative , but it is possible to restrict it to positive values without lose of generality by adjusting the phase of the representations @xmath66 and @xmath67 . qyu yield significantly different low scale phenomenology @xcite than the exact yu . in addition , qyu can provide an interesting scenario in respect of the fine - tuning , since a better fine - tuning prefers that the ratios of yukawa couplings are different from unity @xcite , when the universal boundary conditions are imposed at @xmath1 . in this work we analyze the fine - tuning requirements in umssm with qyu condition imposed at the gut scale . the outline of the paper is the following . we will briefly describe umssm in section [ sec : model ] . after summarizing our scanning procedure and the experimental constraints we employ in our analysis in section [ sec : scan ] , we present our results in the fundamental parameter space of qyu in section [ sec : qyufun ] . the mass spectrum of the supersymmetric particles and dark matter ( dm ) implications are considered in section [ sec : spec ] . finally , we summarize and conclude our results in section [ sec : conc ] . in this section , we briefly summarize the @xmath51 based supersymmetric @xmath38 models whose symmetry breaking patterns and resultant gauge group is given in eq.([e6chain ] ) ( for a detailed consideration see @xcite ) . the superpotential in such models can be given as @xmath68 where @xmath69 and @xmath70 denote the left - handed chiral superfields for the quarks and leptons , while @xmath71 , @xmath72 and @xmath73 stand for the right - handed chiral superfields of u - type quarks , d - type quarks and leptons , respectively . @xmath74 and @xmath75 mssm higgs doublets and @xmath76 are their yukawa couplings to the matter fields . finally @xmath77 denote a chiral superfield , which does not exist in mssm . this field is singlet under the mssm group and its vev is responsible for the braking of @xmath38 symmetry . the invariance under @xmath38 requires an appropriate charge assignment for the mssm fields . table [ charges ] displays the charge configurations for @xmath43 and @xmath44 models . note that eq . ( [ umix ] ) allows infinite number of different charge configurations depending on @xmath78 . .charge assignments for the fields in several models . [ cols="^,^,^,^,^,^,^,^,^",options="header " , ] eq.([suppot1 ] ) is almost the same as the superpotential in mssm except the last term . as is well known , a bilinear mixing of the mssm higgs doublets are introduced with the term @xmath79 in mssm , and @xmath26-term plays an essential role in the electroweak symmetry breaking ( ewsb ) . however , in mssm , @xmath80term preserves the susy , and hence ; it can be at any scale , despite its connection with the ewsb . this is so - called @xmath80problem in mssm . on the other hand , if an extra @xmath38 group , under which the mssm fields have non - trivial charges , is introduced , the invariance principle forbids to introduce such terms like @xmath79 , since @xmath74 and @xmath75 are charged under @xmath38 , and their charges do not have to cancel each other . rather , another term can be introduced such as @xmath81 , where @xmath82 is a dynamical field , and its non - zero vev breaks the extra @xmath38 symmetry , while it also induces a bilinear mixing between @xmath74 and @xmath75 with @xmath83 . in this picture , the @xmath11term can be related to @xmath38 breaking scale and it can be generated dynamically . before proceeding , one of the important task about @xmath38 models is to deal with the anomalies and make sure that the model under concern is anomaly free . there are several attempts @xcite by either adding exotics or imposing non - universal charges to the families . the charge assignments given in table [ charges ] corresponds to the universal charge configurations for the families . in this case , one should carefully consider the exotics , since their existence may bring back the @xmath84 or break the gauge coupling unification . the gauge coupling unification can be maintained if another ( @xmath85 ) is added and assumed to yield only mssm higgs - like doublets can be light @xcite . even if the exotics are heavy and they decouple at a high energy in compared to @xmath1 , they can still contribute to the proton decay . in this case , one can consider umssm along with @xmath47 which forbids baryon and lepton number violating processes @xcite . finally we should note that the existence of right - handed neutrinos . we neglect the contributions from the right - handed neutrinos , since these contributions are suppressed due to smallness of the established neutrino masses @xcite , unless the inverse seesaw mechanism is imposed @xcite . in addition to the mssm particle content , umssm yields two more particles at the low scale , one of which is the gaugino associated with the gauge fields of @xmath38 , and the other is the supersymmetric partner of the mssm singlet @xmath82 . since these two particles are of no electric charge , they mix with the mssm neutralinos after ewsb , which enriches the dark matter implications in umssm @xcite . ewsb also yields a mixing @xmath86 , where @xmath87 is the gauge boson associated with @xmath38 . hence , one can expect some effects from interference of @xmath87 , but since the mass bound on @xmath87 is strict , these effects are highly suppressed by its heavy mass . finally , the content of the charged sector of mssm remain the same , but @xmath88 and @xmath89 are effective in this sector , since they generate the @xmath11term effectively , which also determined the mass of higgsinos . a minimal @xmath51 model , in which the matter fields are resided in @xmath90plet , and the mssm higgs fields in ( @xmath85 ) , also proposes yu via @xmath91 in the superpotential . the discussion on the contradictory mass relations in the fermion sector can be handled by extending the higgs sector with @xmath92 and @xmath93-plets within the @xmath51 framework @xcite . in our work , we assume the minimal lay out for the @xmath51 model . however , the higgs fields emerging from ( @xmath85 ) can also break @xmath47 to the pati - salam model @xcite which is based on the gauge group @xmath94 . in such a framework , the yukawa sector may include also interactions between the matter fields and the higgs fields from @xmath60 representation of @xmath95 . if one assumes that @xmath95 breaks into the mssm gauge group at about the gut scale , the known yukawa couplings can be stated as given in eq.([qyu ] ) at @xmath1 . note that emergence of @xmath95 in the breaking chain allows non - universal gaugino masses at the gut scale such that @xcite @xmath96 in this case , @xmath97 can be varied over the parameter space as a free parameter , and hence , the tension from the heavy gluino mass bound can be significantly relaxed , which yield drastic improvement in regard of the fine - tuning . however , as stated above , we restrict ourselves with the universal boundary conditions , and we will impose only one ssb mass term for all the three gauginos . a solution can be analyzed if it is consistent with qyu or not by considering a parameter defined as @xmath98 where @xmath99 where @xmath100 are yukawa couplings at @xmath1 , and @xmath101 denote the contributions to these couplings . the consistency with qyu requires @xmath102 , i.e @xmath103 . however , yukawa couplings can receive some contributions from the interference of @xmath82 , @xmath87 @xcite and even exotics at @xmath1 as well as unknown threshold corrections from the symmetry breaking . even though these contributions can be neglected , we allow utmost @xmath104 uncertainty in @xmath105 to count for such contributions . hence , a solution compatible with qyu satisfies @xmath106 as well as @xmath107 . we have employed spheno 3.3.3 package @xcite obtained with sarah 4.5.8 @xcite . in this package , the weak scale values of the gauge and yukawa couplings presence in umssm are evolved to the unification scale @xmath1 via the renormalization group equations ( rges ) . @xmath1 is determined by the requirement of the gauge coupling unification through their rge evolutions . note that we do not strictly enforce the unification condition @xmath108 at @xmath1 since a few percent deviation from the unification can be assigned to unknown gut - scale threshold corrections @xcite . with the boundary conditions given at @xmath1 , all the ssb parameters along with the gauge and yukawa couplings are evolved back to the weak scale . note that the gauge coupling associated with the @xmath109 symmetry is determined by the unification condition at the gut scale by imposing @xmath110 , where @xmath111 is the gauge coupling associated with the @xmath112 gauge group . we have performed random scans over the following parameter space @xmath113 where @xmath114 is the universal ssb mass term for all the scalar fields including @xmath74 , @xmath75 , @xmath82 fields , and similarly @xmath115 is the universal ssb mass term for the gaugino fields including one associated with @xmath0 gauge group . @xmath116 is the ratio of vevs of the mssm higgs doublets , @xmath117 is the ssb trilinear scalar interaction term . similarly , @xmath118 is the ssb interaction between the @xmath82 and @xmath30 fields , which is varied free from @xmath117 in our scans . finally , @xmath119 denotes the vev of @xmath82 fields which indicates the @xmath0 breaking scale . recall that the @xmath11term of mssm is dynamically generated such that @xmath120 . its sign is assigned as a free parameter in mssm , since rewsb condition can determine its value but not sign . on the other hand , in umssm , it is forced to be positive by @xmath88 and @xmath119 . finally , we set the top quark mass to its central value ( @xmath121 gev ) @xcite . note that the sparticle spectrum is not too sensitive in one or two sigma variation in the top quark mass @xcite , but it can shift the higgs boson mass by @xmath122 gev @xcite . the requirement of radiative electroweak symmetry breaking ( rewsb ) @xcite puts an important theoretical constraint on the parameter space . another important constraint comes from the relic abundance of the stable charged particles @xcite , which excludes the regions where charged susy particles such as stau and stop become the lightest supersymmetric particle ( lsp ) . in our scans , we allow only the solutions for which one of the neutralinos is the lsp and rewsb condition is satisfied . in scanning the parameter space , we use our interface , which employs metropolis - hasting algorithm described in @xcite . after collecting the data , we impose the mass bounds on all the sparticles @xcite , and the constraint from the rare b - decays such as @xmath123 @xcite , @xmath124 @xcite , and @xmath125 @xcite . in addition , the wmap bound @xcite on the relic abundance of neutralino lsp within @xmath126 uncertainty . these experimental constraints can be summarized as follows : @xmath127 we have emphasized the bounds on the higgs boson@xcite and the gluino @xcite , since they have drastically changed since the lep era . even though the mass bound on @xmath87 can be lowered through detailed analyses @xcite , we require our solutions to yield heavy @xmath87 , since it is not directly related to our considerations . one of the stringent bounds listed above comes from the rare b - meson decay into a muon pair , since the supersymmetric contribution to this process is proportional to @xmath128 . we have considered the high @xmath33 region in the fundamental parameter space as given in eq.([paramsp ] ) , and @xmath129 needs to be large to suppress the supersymmetric contribution to @xmath130 . besides , the wmap bound is also highly effective to shape the parameter space , since the relic abundance of neutralino lsp is usually high over the fundamental parameter space . one needs to identify some coannihilation channels in order to have solutions compatible with the wmap bound . the dm observables in our scan are calculated by micromegas @xcite obtained by sarah @xcite . finally , we impose the fine - tuning condition as @xmath5 . we present our results for the fundamental parameter space in light of the experimental constraints mentioned in the previous section . figure [ fig1 ] illustrates the qyu parameter space in a correlation with the usual cmssm fundamental parameters in the @xmath131 , @xmath132 , @xmath133 , and @xmath134 planes . all points are consistent with rewsb and neutralino lsp . green points satisfy the mass bounds and the constraints from the rare b - decays . orange points form a subset of green and they are compatible with the wmap bound on the relic abundance of neutralino lsp within @xmath126 . finally , the blue points are a subset of orange , which are consistent with the qyu and fine - tuning condition . the dashed lines indicates @xmath135 . as seen from @xmath131 , qyu requires the universal scalar mass parameter larger than 2 tev as in the case of mssm with non - universal gauginos imposed at @xmath1 , with the current experimental bounds @xmath114 is expected to be much larger in the cmssm framework . similarly , @xmath132 plane indicates that @xmath115 can only be as light as @xmath136 gev . this bound is not strictly imposed by the qyu condition , rather the heavy gluino mass bound requires heavy @xmath115 , when the universal gaugino masses are imposed . qyu condition mostly restrict the @xmath33 parameter to the values larger than about 54 as seen from the @xmath134 plane , as happens in the mssm . finally , @xmath117 values are mostly find in the negative region , while it is possible to realize qyu with small positive @xmath137 values . in addition to the fundamental parameters of cmssm , figure [ fig2 ] displays the results in umssm parameters with plots in the @xmath138 and @xmath139 . the color coding is the same as figure [ fig1 ] . the @xmath138 plane shows that the qyu solutions accumulate mostly in the region with @xmath140 , while it can be enlarged to about 0.3 with a good statistics . on the other hand , the region with @xmath141 is excluded by the current experimental constraints ( green ) . the plane @xmath139 shows that the lowest scale for the @xmath0 breaking is about 5 tev . this breaking scale is restricted to @xmath142 tev by qyu and the fine - tuning condition ( blue ) . since , the breaking scale along with @xmath88 generates the @xmath80term , it is worth to consider how large a @xmath80term can be realized in umssm . figure [ fig3 ] represents our results with plots the @xmath143 , @xmath144 , @xmath145 , and @xmath146 . the color coding is the same as figure [ fig1 ] . the @xmath143 plane shows that the alignment between @xmath119 and @xmath88 allows the range @xmath147 gev , which yields low fine - tuning ( @xmath148 ) compatible with the qyu condition as seen from the @xmath144 plane . such a low fine - tuning can be achieved even when @xmath149 tev and @xmath150 tev as shown in the bottom panels of figure [ fig3 ] . figure [ fig4 ] displays the ratios of the yukawa couplings with plots in @xmath151 , @xmath152 , @xmath153 , and @xmath154 planes . the color coding is the same as figure [ fig1 ] . qyu requires certain ratios among the yukawa couplings . even though @xmath155 can lie from @xmath156 to about @xmath157 , qyu rather restricts this ratio as @xmath2 . similarly , it restricts @xmath3 and @xmath4 as seen from the @xmath152 and @xmath153 planes . these ratios hold for any value of the fine - tuning parameter . finally the @xmath154 indicates that @xmath33 can be as high as 58 without disturbing the yukawa coupling ratios and raising amount of the fine - tuning . this section will present the sparticle spectrum compatible with qyu . we start with the color sector as well as staus and @xmath129 as shown in figure [ fig5 ] with plots in the @xmath158 , @xmath159 , @xmath160 , and @xmath161 planes . the color coding is the same as figure [ fig1 ] . as seen from the @xmath158 plane , the gluino can be as light as about 2 tev , while the region with @xmath162 tev is not compatible with the qyu condition . even though it is possible to realize the higgs boson of mass about 125 gev with light stops in the umssm framework , such light stop solutions are mostly excluded by the heavy gluino mass spectrum . similarly , the squarks from the first two families are required to be heavier than about 3 tev as seen from the @xmath159 plane . the @xmath160 plane shows that even though one can realize the stau mass almost degenerate with the lsp neutralino consistently with the wmap bound ( orange ) , the qyu together with the fine - tuning condition requires @xmath163 tev . the last panel of figure [ fig5 ] shows @xmath129 in a correlation with the @xmath33 parameter . the results in the @xmath161 plane shows that @xmath14boson can be as light as 400 gev compatible with qyu , despite the high @xmath33 values . even if one can also impose a mass bound as @xmath164 for high @xmath33 values , there are still a significant number of solutions compatible with qyu and escape from this bound . figure [ fig6 ] displays the sparticle spectrum in the @xmath165 and @xmath166 planes . the color coding is the same as figure [ fig1 ] . diagonal line indicates the region where the plotted sparticles are degenerate in mass . the @xmath165 shows that the chargino and lsp neutralino are mostly degenerate in mass in the region where @xmath167 gev . this region may indicate the higssino dm , and the degeneracy can arise from the degeneracy of two higgsinos . these solutions favors the chargino - neutralino coannihilation processes which reduce the relic abundance of lsp neutralino such that the solutions can be consistent with the wmap bound . this region also yields @xmath14resonance solutions as seen from the @xmath166 plane . it is also possible to realize lighter lsp neutralino solutions ( @xmath168 gev ) . there is no mass degeneracy between the lsp neutralino and chargino in this region . hence , one can conclude the light lsp neutralino region that the lsp neutralino is bino - like , and the wmap bound on the relic abundance of lsp neutralino is satisfied through @xmath14resonance solutions , in which two neutralinos annihilate into an @xmath14boson . the lsp neutralino composition can be seen better from the @xmath169 and @xmath170 planes shown in figure [ fig7 ] . the color coding is the same as figure [ fig1 ] . the diagonal line indicates the region where @xmath171 ( @xmath172 ) in the left ( right ) plane . the @xmath169 plane shows that the @xmath11parameter is smaller than @xmath173 over most of the parameter space . the lsp neutralino is formed by the higgsinos in this region . such solutions also yield high scattering cross - section at the nuclei used in the direct detection experiments , since the interactions between quarks in nuclei and the lsp neutralino happen via yukawa interactions . the higgsinos and bino are almost degenerate in mass in the region around the diagonal line , and this region indicate bino - higgsino mixing in formation of the lsp nutralino . it is also possible to realize bino - like dm as represented with the solutions below the diagonal line where @xmath174 . one can also check if it is possible to have wino mixture in formation of the lsp neutralino . the @xmath170 plane shows that wino is usually heavier than @xmath26 and hence @xmath173 . however , there could be some solutions at about @xmath175 tev , for which the higgsinos and wino are nearly degenerate in mass . comparing with solutions shown in the @xmath169 plane , @xmath173 is seen to be at about 1 tev for this solutions ; i.e. @xmath176 , and wino mixture in the formation of the lsp neutralino becomes as significant as the bino and higgsinos . we explore the low scale implications in the @xmath0 extended mssm ( umssm ) . we restrict the parameter space such that the lightest supersymmetric particle ( lsp ) is always the lightest neutralino . in addition , we impose quasi yukawa unification ( qyu ) at the grand unification scale ( @xmath1 ) . the fundamental parameters of umssm are found to be in a large range such as @xmath149 tev , @xmath177 gev . the @xmath178 parameter is mostly restricted to the region where @xmath179 by the qyu condition . also , qyu strictly requires the ratios among the yukawa couplings as @xmath2 , @xmath3 , and @xmath4 . in addition , the breaking of @xmath0 group takes a place at the energy scales from about 5 tev to 10 tev . we find that the need of fine - tuning over the fundamental parameter space of qyu is in the acceptable range , even if the universal boundary conditions are imposed at @xmath1 , in contrast to cmssm and nuhm . such a set up yields heavy stops ( @xmath6 tev ) , gluinos ( @xmath7 tev ) , and squarks from the first two families ( @xmath8 tev ) . similarly the stau mass is bounded from below at about @xmath9 tev . despite this heavy spectrum , we find @xmath10 , which is much lower than that needed for the minimal supersymmetric models . in addition , umssm yield relatively small @xmath11term , and the lsp neutralio is mostly formed by the higgsinos of mass @xmath12 gev . we obtain also bino - like dark matter ( dm ) of mass about 400 gev . wino is usually found to be heavier than higgsinos and binos , but there is a small region where @xmath13 tev . we also identify chargino - neutralino coannihilation channel and @xmath14resonance solutions which reduce the relic abundance of lsp neutralino down to the ranges compatible with the current wmap measurements . csu would like to thank qaisar shafi for useful discussions about the gut models based on the @xmath51 gauge group . this work is supported in part by the scientific and technological research council of turkey ( tubitak ) grant no . mfag-114f461 ( csu ) . m. a. ajaib , i. gogoladze , f. nasir and q. shafi , phys . b * 713 * , 462 ( 2012 ) doi:10.1016/j.physletb.2012.06.036 [ arxiv:1204.2856 [ hep - ph ] ] . m. carena , s. gori , n. r. shah and c. e. m. wagner , jhep * 1203 * , 014 ( 2012 ) doi:10.1007/jhep03(2012)014 [ arxiv:1112.3336 [ hep - ph ] ] ; a. djouadi , phys . * 459 * , 1 ( 2008 ) doi:10.1016/j.physrep.2007.10.005 [ hep - ph/0503173 ] , and references therein . the atlas collaboration , atlas - conf-2015 - 067 . r. kitano and y. nomura , phys . b * 631 * , 58 ( 2005 ) [ hep - ph/0509039 ] ; r. kitano and y. nomura , phys . rev . d * 73 * , 095004 ( 2006 ) [ hep - ph/0602096 ] ; c. brust , a. katz , s. lawrence and r. sundrum , jhep * 1203 * , 103 ( 2012 ) [ arxiv:1110.6670 [ hep - ph ] ] ; m. papucci , j. t. ruderman and a. weiler , jhep * 1209 * , 035 ( 2012 ) [ arxiv:1110.6926 [ hep - ph ] ] . d. a. demir and c. s. n , phys . d * 90 * , 095015 ( 2014 ) doi:10.1103/physrevd.90.095015 [ arxiv:1407.1481 [ hep - ph ] ] ; h. baer , v. barger , p. huang , d. mickelson , a. mustafayev and x. tata , phys . rev . d * 87 * , no . 3 , 035017 ( 2013 ) doi:10.1103/physrevd.87.035017 [ arxiv:1210.3019 [ hep - ph ] ] . for an incomplete list , see + l. j. hall , d. pinner and j. t. ruderman , jhep * 1204 * ( 2012 ) 131 [ arxiv:1112.2703 [ hep - ph ] ] ; f. brummer and w. buchmuller , jhep * 1205 * ( 2012 ) 006 ; s. caron , j. laamanen , i. niessen and a. strubig , jhep * 1206 * , 008 ( 2012 ) ; z. kang , j. li and t. li , arxiv:1201.5305 [ hep - ph ] ; t. cheng , j. li , t. li , d. v. nanopoulos and c. tong , arxiv:1202.6088 [ hep - ph ] ; m. asano and t. higaki , arxiv:1204.0508 [ hep - ph ] ; h. m. lee , v. sanz and m. trott , jhep * 1205 * ( 2012 ) 139 [ arxiv:1204.0802 [ hep - ph ] ] ; f. brummer , s. kraml and s. kulkarni , jhep * 1208 * , 089 ( 2012 ) [ arxiv:1204.5977 [ hep - ph ] ] ; j. l. feng and d. sanford , arxiv:1205.2372 [ hep - ph ] ; k. blum , r. t. dagnolo and j. fan , arxiv:1206.5303 [ hep - ph ] ; l. randall and m. reece , arxiv:1206.6540 [ hep - ph ] ; b. kyae and j. -c . park , arxiv:1207.3126 [ hep - ph ] ; h. baer , v. barger , p. huang , a. mustafayev and x. tata , arxiv:1207.3343 [ hep - ph ] ; j. e. younkin and s. p. martin , phys . d * 85 * , 055028 ( 2012 ) ; s. antusch , l. calibbi , v. maurer , m. monaco and m. spinrath , arxiv:1207.7236 [ hep - ph ] ; h. abe , t. kobayashi and y. omura , phys . d * 76 * ( 2007 ) 015002 ; d. horton , g. g. ross , nucl . * b830 * ( 2010 ) 221 ; i. gogoladze , m. u. rehman and q. shafi , phys . d * 80 * , 105002 ( 2009 ) ; i. gogoladze , f. .nasir and q. .shafi , int . j. mod . a * 28 * , 1350046 ( 2013 ) [ arxiv:1212.2593 [ hep - ph ] ] ; i. gogoladze , f. nasir and q. shafi , jhep * 1311 * , 173 ( 2013 ) [ arxiv:1306.5699 [ hep - ph ] ] ; e. hardy , jhep * 1310 * , 133 ( 2013 ) [ arxiv:1306.1534 [ hep - ph ] ] ; e. hardy , jhep * 1403 * , 069 ( 2014 ) [ arxiv:1311.2944 [ hep - ph ] ] ; d. m. ghilencea and g. g. ross , nucl . b * 868 * , 65 ( 2013 ) [ arxiv:1208.0837 [ hep - ph ] ] ; d. m. ghilencea , h. m. lee and m. park , jhep * 1207 * , 046 ( 2012 ) [ arxiv:1203.0569 [ hep - ph ] ] ; s. cassel , d. m. ghilencea and g. g. ross , nucl . phys . b * 825 * , 203 ( 2010 ) [ arxiv:0903.1115 [ hep - ph ] ] ; d. m. ghilencea , nucl . b * 876 * , 16 ( 2013 ) [ arxiv:1302.5262 [ hep - ph ] ] ; l. calibbi , t. li , a. mustafayev and s. raza , arxiv:1603.06720 [ hep - ph ] ; t. li , s. raza and k. wang , phys . d * 93 * , no . 5 , 055040 ( 2016 ) doi:10.1103/physrevd.93.055040 [ arxiv:1601.00178 [ hep - ph ] ] ; t. li , s. raza and x. c. wang , arxiv:1510.06851 [ hep - ph ] ; g. du , t. li , d. v. nanopoulos and s. raza , phys . d * 92 * , no . 2 , 025038 ( 2015 ) doi:10.1103/physrevd.92.025038 [ arxiv:1502.06893 [ hep - ph ] ] . for detailed analyses , see + g. lazarides , c. panagiotakopoulos and q. shafi , z. phys . c * 34 * , 553 ( 1987 ) ; doi:10.1007/bf01679872 q. shafi , phys . b * 79 * , 301 ( 1978 ) ; doi:10.1016/0370 - 2693(78)90248 - 4 f. gursey , p. ramond and p. sikivie , phys . lett . b * 60 * , 177 ( 1976 ) ; doi:10.1016/0370 - 2693(76)90417 - 2 b. bajc and v. susi , jhep * 1402 * , 058 ( 2014 ) doi:10.1007/jhep02(2014)058 [ arxiv:1311.0775 [ hep - ph ] ] ; o. k. kalashnikov and s. e. konshtein , nucl . phys . b * 166 * , 507 ( 1980 ) ; doi:10.1016/0550 - 3213(80)90210 - 2 f. buccella and g. miele , phys . b * 189 * , 115 ( 1987 ) ; doi:10.1016/0370 - 2693(87)91280 - 9 y. achiman and b. stech , phys . lett . b * 77 * , 389 ( 1978 ) ; doi:10.1016/0370 - 2693(78)90584 - 1 h. ruegg and t. schucker , nucl . b * 161 * , 388 ( 1979 ) ; doi:10.1016/0550 - 3213(79)90219 - 0 r. barbieri and d. v. nanopoulos , phys . b * 91 * , 369 ( 1980 ) ; and references there in . doi:10.1016/0370 - 2693(80)90998 - 3 s. raza , q. shafi and c. s. n , phys . rev . d * 92 * , no . 5 , 055010 ( 2015 ) doi:10.1103/physrevd.92.055010 [ arxiv:1412.7672 [ hep - ph ] ] . i. gogoladze , r. khalid and q. shafi , phys . d * 79 * , 115004 ( 2009 ) doi:10.1103/physrevd.79.115004 [ arxiv:0903.5204 [ hep - ph ] ] . i. gogoladze , r. khalid , s. raza and q. shafi , jhep * 1012 * , 055 ( 2010 ) . i. gogoladze , q. shafi and c. s. un , phys . b * 704 * , 201 ( 2011 ) ; i. gogoladze , q. shafi and c. s. un , jhep * 1208 * , 028 ( 2012 ) doi:10.1007/jhep08(2012)028 [ arxiv:1112.2206 [ hep - ph ] ] ; i. gogoladze , q. shafi and c. s. un , jhep * 1207 * , 055 ( 2012 ) doi:10.1007/jhep07(2012)055 [ arxiv:1203.6082 [ hep - ph ] ] ; m. adeel ajaib , i. gogoladze , q. shafi and c. s. un , jhep * 1307 * , 139 ( 2013 ) doi:10.1007/jhep07(2013)139 [ arxiv:1303.6964 [ hep - ph ] ] ; m. a. ajaib , i. gogoladze , q. shafi and c. s. n , jhep * 1405 * , 079 ( 2014 ) doi:10.1007/jhep05(2014)079 [ arxiv:1402.4918 [ hep - ph ] ] ; i. gogoladze , a. mustafayev , q. shafi and c. s. un , phys . d * 91 * , no . 9 , 096005 ( 2015 ) doi:10.1103/physrevd.91.096005 [ arxiv:1501.07290 [ hep - ph ] ] . see , incomplete list of references , l. j. hall , r. rattazzi and u. sarid , phys . rev . d * 50 * , 7048 ( 1994 ) ; b. ananthanarayan , q. shafi and x. wang , phys . rev . d * 50 * , 5980 ( 1994 ) ; r. rattazzi and u. sarid , phys . d * 53 * , 1553 ( 1996 ) ; t. blazek , m. carena , s. raby and c. wagner , phys . d * 56 * , 6919 ( 1997 ) ; j. l. chkareuli and i. g. gogoladze , phys . d * 58 * , 055011 ( 1998 ) ; t. blazek , s. raby and k. tobe , phys . d * 62 * , 055001 ( 2000 ) ; h. baer , m. brhlik , m. diaz , j. ferrandis , p. mercadante , p. quintana and x. tata , phys . rev . d * 63 * , 015007(2001 ) ; c. balazs and r. dermisek , jhep * 0306 * , 024 ( 2003 ) ; u. chattopadhyay , a. corsetti and p. nath , phys . d * 66 * 035003 , ( 2002 ) ; t. blazek , r. dermisek and s. raby , phys . * 88 * , 111804 ( 2002 ) ; m. gomez , t. ibrahim , p. nath and s. skadhauge , phys . d * 72 * , 095008 ( 2005 ) ; k. tobe and j. d. wells , nucl . b * 663 * , 123 ( 2003 ) ; i. gogoladze , y. mimura , s. nandi , phys . * b562 * , 307 ( 2003 ) ; w. altmannshofer , d. guadagnoli , s. raby and d. m. straub , phys . b * 668 * , 385 ( 2008 ) ; s. antusch and m. spinrath , phys . d * 78 * , 075020 ( 2008 ) ; h. baer , s. kraml and s. sekmen , jhep * 0909 * , 005 ( 2009 ) ; s. antusch and m. spinrath , phys . rev . d * 79 * , 095004 ( 2009 ) ; k. choi , d. guadagnoli , s. h. i m and c. b. park , jhep * 1010 * , 025 ( 2010 ) ; m. badziak , m. olechowski and s. pokorski , jhep * 1108 * , 147 ( 2011 ) ; j. s. gainer , r. huo and c. e. m. wagner , jhep * 1203 * , 097 ( 2012 ) ; h. baer , s. raza and q. shafi , phys . lett . b * 712 * , 250 ( 2012 ) ; i. gogoladze , q. shafi , c. s. un and , jhep * 1207 * , 055 ( 2012 ) ; m. badziak , mod . phys . lett . a * 27 * , 1230020 ( 2012 ) ; g. elor , l. j. hall , d. pinner and j. t. ruderman , jhep * 1210 * , 111 ( 2012 ) ; h. baer , s. kraml , s. sekmen and h. summy , jhep * 0803 * , 056 ( 2008 ) doi:10.1088/1126 - 6708/2008/03/056 [ arxiv:0801.1831 [ hep - ph ] ] ; h. baer , m. haider , s. kraml , s. sekmen and h. summy , jcap * 0902 * , 002 ( 2009 ) doi:10.1088/1475 - 7516/2009/02/002 [ arxiv:0812.2693 [ hep - ph ] ] . see , for instance , e. witten , phys . lett . * b91 * , 81 ( 1980 ) ; s. m. barr , phys . d * 21 * , 1424 ( 1980 ) ; y. nomura and t. yanagida , phys . rev . d * 59 * , 017303 ( 1999 ) ; m. frigerio , p. hosteins , s. lavignac and a. romanino , nucl . b * 806 * , 84 ( 2009 ) ; s. m. barr , phys . d * 76 * , 105024 ( 2007 ) ; m. malinsky , phys . d * 77 * , 055016 ( 2008 ) ; m. heinze and m. malinsky , phys . d * 83 * , 035018 ( 2011 ) ; k. s. babu , b. bajc and z. tavartkiladze , phys . rev . d * 86 * , 075005 ( 2012 ) and references therein . g. anderson , s. raby , s. dimopoulos , l. j. hall and g. d. starkman , phys . rev . d * 49 * , 3660 ( 1994 ) . k. s. babu and r. n. mohapatra , phys . lett . * 70 * , 2845 ( 1993 ) doi:10.1103/physrevlett.70.2845 [ hep - ph/9209215 ] . m. e. gomez , g. lazarides and c. pallis , nucl . phys . b * 638 * , 165 ( 2002 ) doi:10.1016/s0550 - 3213(02)00483 - 2 [ hep - ph/0203131 ] . j. c. pati and a. salam , phys . d * 10 * , 275 ( 1974 ) erratum : [ phys . rev . d * 11 * , 703 ( 1975 ) ] . doi:10.1103/physrevd.10.275 , 10.1103/physrevd.11.703.2 n. karagiannakis , g. lazarides and c. pallis , phys . d * 92 * , no . 8 , 085018 ( 2015 ) doi:10.1103/physrevd.92.085018 [ arxiv:1503.06186 [ hep - ph ] ] ; q. shafi , . h. tanyldz and c. s. un , nucl . b * 900 * , 400 ( 2015 ) doi:10.1016/j.nuclphysb.2015.09.019 [ arxiv:1503.04196 [ hep - ph ] ] ; r. armillis , g. lazarides and c. pallis , phys . d * 89 * , no . 6 , 065032 ( 2014 ) doi:10.1103/physrevd.89.065032 [ arxiv:1309.6986 [ hep - ph ] ] ; n. karagiannakis , g. lazarides and c. pallis , int . j. mod . a * 28 * , 1330048 ( 2013 ) doi:10.1142/s0217751x13300482 [ arxiv:1305.2574 [ hep - ph ] ] ; n. karagiannakis , g. lazarides and c. pallis , phys . d * 87 * , 055001 ( 2013 ) doi:10.1103/physrevd.87.055001 [ arxiv:1212.0517 [ hep - ph ] ] ; n. karagiannakis , g. lazarides and c. pallis , j. phys . ser . * 384 * , 012012 ( 2012 ) doi:10.1088/1742 - 6596/384/1/012012 [ arxiv:1201.2111 [ hep - ph ] ] ; n. karagiannakis , g. lazarides and c. pallis , phys . b * 704 * , 43 ( 2011 ) doi:10.1016/j.physletb.2011.08.058 [ arxiv:1107.0667 [ hep - ph ] ] ; s. dar , i. gogoladze , q. shafi and c. s. un , phys . d * 84 * , 085015 ( 2011 ) doi:10.1103/physrevd.84.085015 [ arxiv:1105.5122 [ hep - ph ] ] ; m. e. gomez , g. lazarides and c. pallis , phys . d * 67 * , 097701 ( 2003 ) doi:10.1103/physrevd.67.097701 [ hep - ph/0301064 ] . s. antusch , l. calibbi , v. maurer , m. monaco and m. spinrath , phys . rev . d * 85 * , 035025 ( 2012 ) doi:10.1103/physrevd.85.035025 [ arxiv:1111.6547 [ hep - ph ] ] . s. m. barr , phys . lett . * 55 * , 2778 ( 1985 ) ; doi:10.1103/physrevlett.55.2778 j. l. hewett and t. g. rizzo , phys . * 183 * ( 1989 ) 193 ; doi:10.1016/0370 - 1573(89)90071 - 9 m. cvetic and p. langacker , phys . d * 54 * , 3570 ( 1996 ) doi:10.1103/physrevd.54.3570 [ hep - ph/9511378 ] ; g. cleaver , m. cvetic , j. r. espinosa , l. l. everett and p. langacker , phys . d * 57 * , 2701 ( 1998 ) doi:10.1103/physrevd.57.2701 [ hep - ph/9705391 ] ; g. cleaver , m. cvetic , j. r. espinosa , l. l. everett and p. langacker , nucl . b * 525 * , 3 ( 1998 ) doi:10.1016/s0550 - 3213(98)00277 - 6 [ hep - th/9711178 ] ; d. m. ghilencea , l. e. ibanez , n. irges and f. quevedo , jhep * 0208 * , 016 ( 2002 ) doi:10.1088/1126 - 6708/2002/08/016 [ hep - ph/0205083 ] ; s. f. king , s. moretti and r. nevzorov , phys . rev . d * 73 * , 035009 ( 2006 ) doi:10.1103/physrevd.73.035009 [ hep - ph/0510419 ] ; r. diener , s. godfrey and t. a. w. martin , arxiv:0910.1334 [ hep - ph ] ; p. langacker , rev . phys . * 81 * , 1199 ( 2009 ) doi:10.1103/revmodphys.81.1199 [ arxiv:0801.1345 [ hep - ph ] ] ; p. langacker and j. wang , phys . d * 58 * , 115010 ( 1998 ) doi:10.1103/physrevd.58.115010 [ hep - ph/9804428 ] . h. c. cheng , b. a. dobrescu and k. t. matchev , phys . lett . b * 439 * , 301 ( 1998 ) doi:10.1016/s0370 - 2693(98)01052 - 1 [ hep - ph/9807246 ] ; h. c. cheng , b. a. dobrescu and k. t. matchev , nucl . b * 543 * , 47 ( 1999 ) doi:10.1016/s0550 - 3213(99)00012 - 7 [ hep - ph/9811316 ] ; j. erler , nucl . b * 586 * , 73 ( 2000 ) doi:10.1016/s0550 - 3213(00)00427 - 2 [ hep - ph/0006051 ] ; p. langacker and m. plumacher , phys . d * 62 * , 013006 ( 2000 ) doi:10.1103/physrevd.62.013006 [ hep - ph/0001204 ] ; v. barger , c. w. chiang , p. langacker and h. s. lee , phys . b * 580 * , 186 ( 2004 ) doi:10.1016/j.physletb.2003.11.057 [ hep - ph/0310073 ] ; d. a. demir , g. l. kane and t. t. wang , phys . d * 72 * , 015012 ( 2005 ) doi:10.1103/physrevd.72.015012 [ hep - ph/0503290 ] . r. wendell _ et al . _ [ super - kamiokande collaboration ] , phys . rev . d * 81 * , 092004 ( 2010 ) doi:10.1103/physrevd.81.092004 [ arxiv:1002.3471 [ hep - ex ] ] . r. n. mohapatra and j. w. f. valle , phys . d * 34 * , 1642 ( 1986 ) ; doi:10.1103/physrevd.34.1642 m. c. gonzalez - garcia and j. w. f. valle , phys . b * 216 * , 360 ( 1989 ) ; doi:10.1016/0370 - 2693(89)91131 - 3 s. khalil , phys . rev . d * 82 * , 077702 ( 2010 ) doi:10.1103/physrevd.82.077702 [ arxiv:1004.0013 [ hep - ph ] ] . m. frank and s. mondal , phys . d * 90 * , no . 7 , 075013 ( 2014 ) doi:10.1103/physrevd.90.075013 [ arxiv:1408.2223 [ hep - ph ] ] , and references therein k. s. babu , b. bajc and v. susi , jhep * 1505 * , 108 ( 2015 ) doi:10.1007/jhep05(2015)108 [ arxiv:1504.00904 [ hep - ph ] ] . r. howl and s. f. king , jhep * 0801 * , 030 ( 2008 ) doi:10.1088/1126 - 6708/2008/01/030 [ arxiv:0708.1451 [ hep - ph ] ] . c. s. un and o. ozdal , phys . d * 93 * , 055024 ( 2016 ) doi:10.1103/physrevd.93.055024 [ arxiv:1601.02494 [ hep - ph ] ] . w. porod , comput . commun . * 153 * , 275 ( 2003 ) [ hep - ph/0301101 ] ; w. porod and f. staub , comput . commun . * 183 * , 2458 ( 2012 ) doi:10.1016/j.cpc.2012.05.021 [ arxiv:1104.1573 [ hep - ph ] ] . f. staub , arxiv:0806.0538 [ hep - ph ] ; f. staub , comput . commun . * 182 * , 808 ( 2011 ) [ arxiv:1002.0840 [ hep - ph ] ] . j. hisano , h. murayama and t. yanagida , nucl . b * 402 * , 46 ( 1993 ) [ hep - ph/9207279 ] ; y. yamada , z. phys . c * 60 * , 83 ( 1993 ) ; j. l. chkareuli and i. g. gogoladze , phys . d * 58 * , 055011 ( 1998 ) [ hep - ph/9803335 ] . l. e. ibanez and g. g. ross , phys . b110 * ( 1982 ) 215 ; k. inoue , a. kakuto , h. komatsu and s. takeshita , prog . . phys . * 68 * , 927 ( 1982 ) [ erratum - ibid . * 70 * , 330 ( 1983 ) ] ; l. e. ibanez , phys . * b118 * ( 1982 ) 73 ; j. r. ellis , d. v. nanopoulos , and k. tamvakis , phys . lett . * b121 * ( 1983 ) 123 ; l. alvarez - gaume , j. polchinski , and m. b. wise , nucl . phys . * b221 * ( 1983 ) 495 . k. nakamura _ et al . _ [ particle data group collaboration ] , j. phys . g * 37 * , 075021 ( 2010 ) . doi:10.1088/0954 - 3899/37/7a/075021 t. e. w. group [ cdf and d0 collaborations ] , arxiv:0903.2503 [ hep - ex ] . i. gogoladze , r. khalid , s. raza and q. shafi , jhep * 1106 * , 117 ( 2011 ) doi:10.1007/jhep06(2011)117 [ arxiv:1102.0013 [ hep - ph ] ] . i. gogoladze , q. shafi and c. s. un , jhep * 1208 * , 028 ( 2012 ) doi:10.1007/jhep08(2012)028 [ arxiv:1112.2206 [ hep - ph ] ] ; m. adeel ajaib , i. gogoladze , q. shafi and c. s. un , jhep * 1307 * , 139 ( 2013 ) doi:10.1007/jhep07(2013)139 [ arxiv:1303.6964 [ hep - ph ] ] . g. belanger , f. boudjema , a. pukhov and r. k. singh , jhep * 0911 * , 026 ( 2009 ) ; h. baer , s. kraml , s. sekmen and h. summy , jhep * 0803 * , 056 ( 2008 ) . k. a. olive _ et al . _ [ particle data group collaboration ] , chin . phys . c * 38 * , 090001 ( 2014 ) . doi:10.1088/1674 - 1137/38/9/090001 r. aaij _ et al . _ [ lhcb collaboration ] , phys . lett . * 110 * , no . 2 , 021801 ( 2013 ) [ arxiv:1211.2674 [ hep - ex ] ] . y. amhis _ et al . _ [ heavy flavor averaging group collaboration ] , arxiv:1207.1158 [ hep - ex ] . d. asner _ [ heavy flavor averaging group collaboration ] , arxiv:1010.1589 [ hep - ex ] . g. hinshaw _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 208 * , 19 ( 2013 ) doi:10.1088/0067 - 0049/208/2/19 [ arxiv:1212.5226 [ astro-ph.co ] ] . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 716 * , 1 ( 2012 ) ; s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 716 * , 30 ( 2012 ) . g. belanger , f. boudjema , a. pukhov and a. semenov , comput . commun . * 176 * , 367 ( 2007 ) doi:10.1016/j.cpc.2006.11.008 [ hep - ph/0607059 ] ; g. belanger , f. boudjema , a. pukhov and a. semenov , comput . commun . * 185 * , 960 ( 2014 ) doi:10.1016/j.cpc.2013.10.016 [ arxiv:1305.0237 [ hep - ph ] ] ; g. blanger , f. boudjema , a. pukhov and a. semenov , comput . phys . commun . * 192 * , 322 ( 2015 ) doi:10.1016/j.cpc.2015.03.003 [ arxiv:1407.6129 [ hep - ph ] ] . e. accomando , d. becciolini , a. belyaev , s. moretti and c. shepherd - themistocleous , jhep * 1310 * , 153 ( 2013 ) doi:10.1007/jhep10(2013)153 [ arxiv:1304.6700 [ hep - ph ] ] ; w. abdallah , j. fiaschi , s. khalil and s. moretti , jhep * 1602 * , 157 ( 2016 ) doi:10.1007/jhep02(2016)157 [ arxiv:1510.06475 [ hep - ph ] ] ; e. accomando , a. belyaev , j. fiaschi , k. mimasu , s. moretti and c. shepherd - themistocleous , jhep * 1601 * , 127 ( 2016 ) doi:10.1007/jhep01(2016)127 [ arxiv:1503.02672 [ hep - ph ] ] .

we present a multiwavelength study of the star - forming region using infrared ( ir ) observations in ukirt _ jhk _ bands , _ spitzer _ irac bands , and _ herschel _ pacs bands ; 2.12 m h narrow - band imaging ; and radio continuum observations from gmrt ( 610 and 1280 mhz ) , in a field of view ( fov ) of 34 40 . archival _ spitzer _ observations in conjunction with near - ir ( nir ) observations are used to identify 1162 classii / iii and 40 class i sources in the fov . the nearest - neighbour stellar surface density analysis shows that majority of these young stellar objects ( ysos ) constitute the embedded cluster centered on the high - mass source irs1a south . some ysos , predominantly younger population , are distributed along and trace the filamentary structures at lower stellar surface density . the cluster radius is obtained as 0.44pc - matching well with the extent of radio emission - with a peak density of 650pc . the _ jhk _ data is used to map the extinction in the region which is subsequently used to compute the cloud mass . it has resulted in 126m and 71m for the central cluster and the northern irs5 region , respectively . narrow - band imaging displays significant emission , which prominently resembles fluorescent emission arising at the borders of dense regions . radio continuum analysis shows this region as having blister morphology , with the radio peak coinciding with a protostellar source . free - free emission spectral energy distribution ( sed ) analysis is used to obtain physical parameters of the overall photoionized region and the irs5 sub - region . this multiwavelength scenario is suggestive of star formation having resulted from merging of multiple filaments to form a hub . star formation seems to have taken place in two successive epochs , with the first epoch traced by the central cluster and the high - mass star(s ) - followed by a second epoch which is spreading into the filaments as uncovered by the class i sources and even younger protostellar sources along the filaments . the irs5 h ii region displays indications of swept - up material which has possibly led to the formation of protostars .

1309.7127

@xcite ( also known as ) is an optically visible h ii region @xcite in the serpens - aquila rift region . located at @xmath6 , @xmath7 , it is visible as a bipolar nebula at mid and far ir wavelengths ; harbouring complex features such as filamentary structures and an embedded arc - shaped nebula ( figure [ fig_colourcompositepacs160100irac8 ] ) . early co observations of this region showed the w40 h ii region to be located at the edge of the molecular cloud @xcite , resulting in the well - known blister morphology . even so , the full extent of the molecular gas in this region is not mapped so far . only the central few arcminutes of this cloud are mapped , using isotopologues of co , finding the mass of central cloud core to be @xmath2 100@xmath8 @xcite . w40 also has an associated dense molecular clump of @xmath2 20 diameter (; @xcite ) . a large scale , weak molecular outflow is also found , through co observations , to originate in the molecular cloud @xcite . located at a galactic latitude of @xmath2 3.5@xmath9 , this region is above the main galactic plane , and the distance estimates to it vary from 300 to 900 pc ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) . in the present work , we have adopted a distance of 500 pc from @xcite . the central region of the w40 molecular cloud and h ii region is known to host an embedded cluster of young stars , with a significant population of early - type sources @xcite , with the earliest spectral type of about o9.5 @xcite . this cluster of early - type sources also partially reveals itself as a cluster of compact radio sources , coexisting with other compact radio sources which are classified as candidate ultra - compact h ii regions and radio variable sources @xcite . together , these observations show that the w40 molecular cloud / h ii region is one of the few nearby regions with active star formation away from the galactic plane , hosting an embedded cluster including high - mass stars , and thus representing a template laboratory to investigate the process of star formation . recent observations of the larger serpens - aquila rift in the ( sub)millimeter wavelengths include this region , and have produced a systematic , unbiased sample of starless as well as protostellar dense cores within this cloud @xcite . also , high resolution x - ray observation using _ chandra _ has revealed the near - complete census of young stellar population within the embedded cluster @xcite . low frequency radio observations show newer compact sources in this region @xcite . despite the multitude of individual studies of the w40 molecular cloud , embedded cluster and h ii region , at wavelengths ranging from x - rays to radio , an analysis of the overall star formation scenario in w40 is pending . in this paper , the aim is to fill this lacuna by using the vast archival dataset together with new nir and radio observations . in section [ section_observations - and - analysis ] , we present the observations ( archival or otherwise ) and data reduction procedures . we discuss the results pertaining to stellar sources ( classification and their analysis ) , the morphology , and the physical parameters of the region from various datasets in sections [ section_ysos ] , [ section_morphology ] , and [ section_physicalparameters ] , respectively ; followed by a discussion on the possible star formation scenario in section [ section_star - formation - scenario ] . the main conclusions are presented in section [ section_conclusions ] . + + + + + nir photometric imaging observations were carried out using the wide field camera ( wfcam ; @xcite ) mounted on the 3.8 m united kingdom infrared telescope ( ukirt ) . wfcam contains four rockwell hawaii - ii ( hgcdte 2048 @xmath3 2048 pixel ) arrays spaced by 94% in the focal plane . with a pixel scale of 0.4 , the fov of each array is @xmath2 13.7 . _ jhk _ band data were obtained in the service mode during the period 2009 december 15 - 17 . observations at four positions are required to observe a contiguous ( @xmath2 0.75 deg@xmath10 ) square field - called a wfcam tile . observation at each of these positions was taken using a nine point jitter pattern , and a 2@xmath32 microstepping pattern , with an exposure time of 5s each . the seeing was in the range 1.3 - 1.8 . we use the starlink software to do the initial processing , followed by mosaic making using the routine makemos in a manner similar to @xcite ( see this paper for details of wfcam processing ) . the sextractor software @xcite was used to detect all the stellar sources in the region . source detection and deblending is highly effective with this software . a detection threshold of 3 , and a deblend contrast factor of 32 was used for our purposes . subsequent to first pass of detection , the detection catalogue was filtered using the full - width - at - half - maxima ( fwhm ) , stellarness parameter , and ellipticity to remove spurious detections . finally , some leftover spurious detections were removed manually to create a reliable detection catalogue . aperture photometry for the detected sources was performed using the phot routine in the iraf apphot package . the aperture radius was set to 3 pixels (= 1.2 ) , and the sky annulus was located 9pixels away - with a width of 7pixels . sources without any centering error ( cier ) , sky fitting error ( sier ) and photometric error ( pier ) were retained for further work . instrumental magnitudes are in the wfcam native system ( the mauna kea consortium filter system or mko system ) . absolute photometric calibration of the _ j , h , _ and _ k _ photometry was done using the two micron all sky survey ( 2mass ) stars in the fov as explained below . we obtained 2mass data with the constraints of `` phqual = aaa '' and `` ccflag = 000 '' for all the stars in the fov . this ensures detections which have the highest photometric quality without neighbouring source confusion in all the three bands . we first constructed a _ j - h / h - k _ colour - colour ( cc ) diagram using this data to identify only the main sequence ( ms ) stars in the region . the ms stars obtained using the cc diagram were then used to carry out the photometric calibration of the wfcam data . the 2mass data was converted to wfcam system using the transformation equations given by @xcite . wfcam data saturates for magnitudes brighter than 11.5mag in all the three bands , and these saturated points were replaced by 2mass magnitudes . completeness limit tests were carried out for all the three bands by choosing subimages of size 5@xmath3 5 located in the cloud region and field region . the 90% completeness limits for the _ j , h _ , and _ k _ band data are 19 , 18 , and 17 mag - respectively - for the field region ; and 18 , 17 , and 16 mag - respectively - for the cloud region ; giving the mean as 18.5 , 17.5 , and 16.5 mag , respectively . these early observations of wfcam have small differences in background matching between multiple chips and tiles , primarily resulting from the presence of large scale nebula contaminating the jittered frames . therefore , the _ j , h _ , and _ k _ catalogs were clipped at 18 , 17 and 16 mag limits for nir cc , nir colour - magnitude ( cm ) , and extinction map analyses to obtain conservative and robust classifications of both young stellar objects ( ysos ) and extinction measurements . since all nir magnitudes had low errors associated with them , they were not filtered by any error criteria . the magnitudes of bright sources which were saturated in the wfcam image , but had good quality ( phqual = aaa ) detections in 2mass , were taken from the 2mass catalogue after converting them to the wfcam system using @xcite . the _ spitzer _ archival data for the gould belt survey of ( * ? ? ? * pid : 30574 ) in all the four irac bands - 3.6 ( ch1 ) , 4.5 ( ch2 ) , 5.8 ( ch3 ) , and 8.0 ( ch4 ) @xmath0 m was obtained using the _ spitzer _ heritage archive . final image mosaics were created with the mopex data reduction software , by using the corrected basic calibrated data ( cbcd ) ( s18.5 processed version ) , imask ( bimsk ) , and uncertainty ( cbunc ) files . mosaic images in all the four irac bands were created with a final pixel scale of 0.6 . aperture photometry was carried out for the four irac bands . we first used sextractor for source detection as described above ( see section [ section_nirphotometry ] ) and filtered to build a complete detection catalogue . the detection catalogue was used as input to the phot task in iraf to obtain the aperture photometry of irac images . the aperture size , radius of the inner annulus , and radius of the outer annulus were taken as 2.4@xmath11 , 2.4@xmath11 , and 7.2@xmath11 , respectively . the zero point magnitudes ( with aperture corrections applied ) were taken as 18.593 , 18.090 , 17.484 , and 16.700 for the four irac bands , respectively ; calculated using the irac instrument handbook ( version 2.0.2 , march 2012 ) . average completeness limits were determined using the peak of the luminosity function for the four bands . as is elucidated in the final delivery document for irac and mips data ( see its section 3.4.2 ) , for c2d legacy project @xcite , the peak of the observed luminosity function can be assumed to estimate the 90% completeness limit . we constructed the luminosity functions with 0.5 mag bin size for each band , and the peak ( and thus the approximate 90% completeness limit ) was found to be at @xmath2 14 , 14 , 11.5 , and 11.5 for ch1 , ch2 , ch3 , and ch4 , respectively . the south - east region of the wide - field image where the extinction is low was the only region where both wfcam and _ spitzer _ had a high density of sources causing source crowding problems . however , as mentioned in above sections , double - care was exercised in filtering first at the level of sextractor and then at photometry allowing us to be quite certain that the detected sources were all real and uncontaminated . finally , the cross - matching makes it even more robust . the number of source detections in individual photometric catalogues of each band are different due to respective band extinctions , point spread function ( psf ) , sensitivity limits and area of coverage . different cross - matched catalogues were produced for different analyses . the final analysis is done on an area of size @xmath2 34.33@xmath12 40.00@xmath13 , centered at @xmath14 18@[email protected]@xmath17 , @xmath18 -02@[email protected]@xmath11 . we obtained the following cross - matched and merged catalogues for various analyses : * _ ch1-ch2-ch3-ch4 catalogue : _ the irac ch1 , ch2 , ch3 and ch4 band catalogues were cross - matched ( within 0.6 ) with the constraint that all individual band errors are @xmath19 0.15 mag to produce an all - irac catalogue of 1874 sources . this was used in the classification of ysos . * _ h - k - ch1-ch2 catalogue : _ not all sources are detected in irac ch3 and ch4 bands , but have good quality detections at _ h _ and _ k _ wavelengths . for classification of such sources into different evolutionary stages , we obtain a cross - matched catalogue of _ h _ , _ k _ , ch1 , and ch2 magnitudes ( nir _ h _ and _ k _ catalogues were matched with 0.4 matching radius , ch1 and ch2 catalogues were matched with 0.6 matching radius ; followed by matching the resultant nir and _ spitzer _ catalogues with 0.6 radius ) with good quality detections in irac ch1 and ch2 bands , i.e. sources with irac ch1 and ch2 magnitude errors @xmath19 0.15 mag . the resulting catalogue contains 8585 cross - matched sources . * _ j - h - k catalogue : _ next , we obtained a cross - matched ( within 0.4@xmath11 ) _ j - h - k _ catalogue which was further clipped by conservative _ j , h _ , and _ k _ completeness limits 18 , 17 , and 16 mag as elucidated in section [ section_nirphotometry ] . this filtered nir catalogue containing 10990 sources was used for the identification of ysos and to produce the extinction map ( see section [ section_extinction ] ) of the region . * _ h - k catalogue : _ many embedded nir sources are detected only at _ h _ and _ k _ wavelengths . for identification of additional ysos from amongst these , cross - matched nir sources ( within 0.4 ) with detections in only _ h _ and _ k _ bands ( clipped at _ h _ and _ k _ conservative completeness limits of 17 and 16 , repectively ) were obtained . the above described catalogues are not mutually exclusive . respective ysos were identified using these in section [ section_identification ] . narrow - band imaging in the 2.12 @xmath0 m h@xmath1 ( v=10 ) s(1 ) filter and broad band imaging in the _ _ k__@xmath20 for continuum subtraction was obtained for a 19@xmath314 region centered on the w40 embedded cluster . these data were obtained with the calar - alto 3.5 m telescope during the night of 2001 june 13 using the omega prime camera . the camera used an hgcdte detector with 1024@xmath31024 pixels with a plate scale of 0.39 pixel@xmath21 resulting in a [email protected] fov . observations were carried out using a 9 point jitter pattern . the exposure time was 30s and 3s - per jitter pointing - in the nb2122 and _ _ k__@xmath20 filters respectively , resulting in total integration times of 180s and 18s per pixel , respectively . standard image reduction of dark subtraction , flat - fielding and sky - subtraction was carried out for each pointing . k__@xmath20 images were psf matched and scaled to the h@xmath1 narrow - band image to perform continuum subtraction . _ herschel _ space observatory , a 3.5 m telescope , was launched to carry out observations in the wavelength range 51 - 670 @xmath0 m using the instruments photodetector array camera and spectrometer ( pacs ) , spire , and hifi @xcite . we obtained the publicly available level2_5 processed archival data for pacs 100 @xmath0 m ( 3.2pixel@xmath21 ) and 160 @xmath0 m ( 6.4pixel@xmath21 ) bands - obtained as part of the proposal id : kgbt_pandre_1 . the madmap@xcite processed products are used , as opposed to photproject processing , since our aim is to analyse extended features in the region . a detailed overview of procedures involved , data reduction pipeline , and data products is available on the _ herschel _ website we use the images to examine the filamentary structures vis - a - vis the distribution of ysos and radio continuum emission . c c c date of obs . & 2011 nov . 15 & 2011 nov . 18 + phase center & @xmath22=18@[email protected]@xmath25 & @xmath22=18@[email protected]@xmath25 + & @xmath26=-02@[email protected]@xmath11 & @xmath26=-02@[email protected]@xmath11 + flux calibrator & 3c286 , 3c48 & 3c286 , 3c48 + phase calibrator & 1822 - 096 & 1822 - 096 + cont . bandwidth & 32 mhz & 32 mhz + primary beam & 24@xmath13 & 43@xmath13 + resolution of maps & & + used for fitting & [email protected]@xmath11 & [email protected]@xmath11 + peak flux density & 0.45 jy / beam & 0.35 jy / beam + rms noise & 6.34 mjy / beam & 5.52 mjy / beam + total integrated & & + flux density & & + within 4@xmath29 area & 7.55 jy & 4.07 jy + integrated flux & & + density within & & + 50 radius of & & + irs5 radio peak & 109 mjy & 121 mjy + [ table_gmrtobservation ] new radio continuum observations are obtained using the giant metre - wave radio telescope ( gmrt ) in two bands centered at 1280 mhz and 610 mhz . data was obtained on 2011 november 15 , and 2011 november 18 ( project i d 21_015 ) , respectively , the details of which are given in table [ table_gmrtobservation ] . the gmrt array consists of 30 antennae located in an approximately y shaped configuration . each antenna has a diameter of 45 m . a central region of about 1 km @xmath3 1 km consists of 12 ( randomly distributed ) antennae , while the rest are distributed along three radial arms ( 6 along each arm ) upto a distance of @xmath2 14 km . it results in a baseline of @xmath2 100m25 km , providing sensitivity to features @xmath30 8@xmath13 and 17@xmath13 in the 1280 mhz and 610 mhz bands , respectively . details about the gmrt array can be found in @xcite . data reduction was carried out with the aips software . bad data were flagged using aips tasks `` tvflg '' and `` uvflg '' . multiple rounds of flagging and calibration were done to improve the calibration successively . calibrated object data ( w40 ) was subsequently split from the raw file ( which contains phase and flux calibrator data also ) , following which low resolution ( @xmath2 45@xmath31 45@xmath11 ) images were finally made using facet imaging with the help of the task `` imagr '' . a few rounds of ( phase ) self - calibration were carried out to remove the ionospheric phase distortion effects . the ysos were identified using sources detected in irac bands first , followed by using the catalogue of sources with detections in nir and irac bands , and lastly using the nir catalogue of sources . following is an elucidation of the set of steps followed : * identification of ysos from the irac data was accomplished by computing the irac spectral index for the sources with detections in all four irac bands ( i.e sources from _ ch1-ch2-ch3-ch4 _ catalogue ; see section [ section_crossmatching ] ) . the irac spectral index ( @xmath32 ; * ? ? ? * ) , for each source , was calculated using linear regression . thereafter , using the spectral index limits as given in @xcite ( class i : 0 @xmath33 , and class ii : -2 @xmath34 0 ) , sources were classified into class i and class ii categories ( see figure [ fig_spectralindex ] ) . * _ h - k - ch1-ch2 _ cross - matched catalogue ( see section [ section_crossmatching ] ) was used for the identification of additional class i and class ii sources using the classification scheme of ( * ? ? ? * ; * ? ? ? * see their appendix a - phase 2 ) . for this , the dereddened ( k-[3.6 ] ) and ( [ 3.6]-[4.5 ] ) colours , calculated using the colour excess ratios from @xcite , were used . as is stated in @xcite , to filter out possible dim extragalactic contaminants , we also applied the limits on dereddened ch1 band magnitude to be @xmath19 15 and @xmath19 14.5 for class i and class ii categories , respectively . we used the average extinction value estimated in section [ section_extinction ] ) ( converted to a@xmath35 using * ? ? ? * ) to calculate the extinction for ch1 band in this region , and thus intrinsic ch1 magnitudes , to impose this condition . the reddening law from @xcite for serpens ( since it is closest to our region ) was used for this purpose . figure [ fig_hkch1ch2_ccd ] shows the location of these class i and class ii sources - which have been marked as filled black circles - on the ( _ k_-[3.6])@xmath36 vs. ( [ 3.6]-[4.5])@xmath36 cc diagram . + c c c c c c c c c c + 277.635620 & -2.305891 & 14.637 @xmath37 0.038 & 11.474 @xmath37 0.025 & 9.499 @xmath37 0.025 & 8.417 @xmath37 0.002 & 8.249 @xmath37 0.002 & 7.810 @xmath37 0.004 & 7.860 @xmath37 0.004 & class2 + 277.638031 & -2.118266 & & & & 11.351 @xmath37 0.006 & 10.719 @xmath37 0.006 & 10.156 @xmath37 0.016 & 9.422 @xmath37 0.016 & class2 + 277.638184 & -2.349572 & 14.582 @xmath37 0.048 & 11.551 @xmath37 0.025 & 9.273 @xmath37 0.020 & 7.709 @xmath37 0.001 & 6.973 @xmath37 0.001 & 6.402 @xmath37 0.002 & 6.179 @xmath37 0.001 & class2 + 277.638580 & -2.190570 & & & & 10.309 @xmath37 0.004 & 9.623 @xmath37 0.003 & 8.998 @xmath37 0.009 & 8.210 @xmath37 0.004 & class2 + 277.641449 & -1.932326 & & 14.189 @xmath37 0.002 & 12.462 @xmath37 0.001 & & & & & class2 + * in the next step , the _ j - h - k _ photometric catalogue was used as follows . for the sources in this catalogue , a _ j - h / h - k _ cc diagram was constructed . figure [ fig_jhk_ccd ] shows the nir _ j - h / h - k _ cc diagram . the classical t - tauri star ( ctts ) locus ( blue line ) was taken from @xcite ( which is derived from * ? ? ? * ) , while the locus for dwarfs ( red curve ) in mko system was taken from @xcite . similar to @xcite , the reddening laws of @xcite ( a@xmath38/a@xmath39 = 0.282 , a@xmath40/a@xmath39 = 0.175 , and a@xmath35/a@xmath39 = 0.112 ) were used . the dashed green lines show the reddening vectors . the sources which were located in the t and p regions - similar to the method of @xcite , @xcite - were taken as nir class ii / iii sources . there may be an overlap between the herbig ae / be stars and class ii sources in this t region @xcite . to decrease any contamination in the p region , we took only those sources which were above @xmath41 0.89 threshold . * additionally , those nir sources detected only in the _ h _ and _ k _ bands ( i.e. using the _ h - k _ catalogue from section [ section_crossmatching ] ) were also used to identify ysos using the _ h / h - k _ cm diagram ( figure [ fig_hk_cmd ] ) . from the _ h / h - k _ cm diagram , only the sources with an ir excess upwards of _ h - k _ = 1.5 were selected . this cut - off was chosen as - in the cm diagram - it marks a clear low density gap between the field sources branch and the ysos branch . cm diagrams of nearby non - nebular regions also show the sources in those regions ( i.e. the field sources ) to be confined below this value . ysos selected using the nir photometry mostly fall into the class ii / iii category . the ysos selected from each of the above steps are not mutually exclusive and there are overlapping ysos between various methods . the selected ysos were hence merged ( with the class of a yso taken as what it was identified as first in the above order of steps ) to obtain a final catalog of 1202 ysos , of which 40 are in the class i category and 1162 in the class ii / iii category . the central high - mass irs sources ( 1a south , 2b , 3a , and 5 ) were detected as classii in the above set of steps , but were expunged from the yso catalogue as they have already been identified as main - sequence sources by @xcite . table [ table_yso ] gives the ysos detected in the fov . a sample of table [ table_yso ] is given here , whereas the complete table is available in electronic form as part of the online material . . [ fig_extmap ] the nir photometry from the new wfcam data is deeper by two magnitudes compared to the 2mass data . the nir _ j - h / h - k _ cc diagram of the region was used to select the reddened ms sources . sources in the f region ( see figure [ fig_jhk_ccd ] ) , which were not classified as ysos in section [ section_identification ] , were selected for this analysis ( a total of 10011 sources used here ) . we estimated the average visual extinction in the region by using a method similar to the near - ir - colour - excess ( nice ) method of @xcite . first , the _ ( h - k ) _ colour excess for each of the selected sources was calculated by dereddening them , along the reddening vectors , to the dwarf locus which was approximated by a straight line asymptote . a@xmath39 , for each source , was subsequently calculated using the reddening laws of @xcite . the dereddenned extinction values were used to produce an extinction map of the region as shown in figure [ fig_extmap ] . in producing this extinction map , at each position on the regular grid , a median extinction from the 20 valid nearest neighbours was computed . the choice of median instead of average value is because median works like an outlier rejection . figure [ fig_extmap ] displays two distinct features : a ) a low extinction cavity around the ionizing stars , and b ) higher extinction towards west and north . the filamentary structures ( whose locations have been marked with arrows ) are partially traceable ( with the one along north - east much more distinct than the one towards south ) . although the filaments are found to be very dense from pacs 160 @xmath0 m emission ( see section [ section_filamentarystructures ] ) , the extinction map fails to reproduce the same result because the nir data is not deep enough to effectively detect many stars behind the filament . the depth of the nir data is nevertheless good enough to trace extinctions of the order of a@xmath42 20mag , which is the maximum value found along the north - eastern filamentary structure and northern regions . the average extinction of the diffuse low density remnant cloud covering most of the w40 h ii region is found to be a@xmath42 15mag . note that the low extinction cavity around the irs sources in figure [ fig_extmap ] displays a@xmath43 12mag which agree well with the previous extinction estimates for the dominant irs sources ( a@xmath44 8 - 10 mag ) in the region @xcite . we carried out the yso surface density analysis of the region using the nearest neighbour ( nn ) method @xcite . figure [ fig_surfacedensity ] shows contours for 50 and 20 nn density overlaid on pacs 160 @xmath0 m image and irac 3.6 @xmath0 m image , respectively . a large value of nn is used to examine large scale structures , while a lower value of nn is more sensitive to smaller - scale density variations and is used to trace sub - structures . this can be thought of some kind of smoothing process . here , the 50 nn density contours are fairly circularly symmetric , and were thus used to calculate the cluster radius ( as opposed to using 20 nn map ) . the dense central region for 50 nn map was fitted by a gaussian profile whose fwhm comes out to be @xmath2 3 ( @xmath2 0.44 pc at a distance of 500 pc ) . there are 170 ysos ( from section [ section_identification ] ) within this cluster radius . the 20 nn density contours ( being more sensitive to local density fluctuations ) reveal the sub - structures of the cluster . the contours mirror the filamentary structures to a considerable degree showing that the filaments also contain ysos , albeit at low densities . the centers of both 50 and 20 nn density contour maps are coincident with irs1a . figure [ fig_densityhistograms ] shows the cumulative frequency distribution and the histogram for the 20 nn and 50 nn distances in solid black line / curve and dark - grey dotted line / curve , respectively . as is expected , each feature ( kink , peak , dip , etc ) which is observed in 50 nn histogram / frequency polygon at , say , x nn distance , is observed at @xmath2 x/2 nn distance in 20 nn histogram / frequency polygon . the cluster radius has been marked in vertical dashed line . the cluster radius seems to be near an inflection point in the rising portion of the 50 nn frequency polygon . @xcite have carried out an analysis of clusters in nearby star - forming regions - , , and molecular clouds - using _ spitzer _ the cluster peak for the w40 region ( @xmath2 650 pc@xmath4 for 20 nn ) seems to be much higher than the clusters in these molecular clouds , except for that in serpens core(a ) ( which is 1045 pc@xmath4 ) , and comparable to in ophiucus ( which is 509 pc@xmath4 ) ; while the cluster radius is much smaller ( at least a factor of @xmath2 0.5 or more lesser ) than all of them , except for that of ophiucus centre cluster ( which has a radius of 0.52 pc ) . from the compilation of @xcite for different clusters ( though it must be kept in mind that this is pre-_spitzer _ and mostly just using nir ) , we can see that among the few clusters at similar distances and with similar cluster radii ( e.g. , , , ) , the star count is mostly much lower than for w40 region ( which is 170 ) . even among regions at larger distances and similar radii ( e.g. , , , , , , ) , the star counts are much lower . the ysos are not in particular found to be correlated with the lobes of the bipolar nebula , suggesting that the bipolar nebula is relatively young , and not enough material has accumulated along the edge of the lobes to lead to star formation . there appears to be a prominent compact nebula ( marked with a black arrow on 3.6 @xmath0 m image in figure [ fig_surfacedensity ] ) located at the edge of the southern lobe . the source associated with this compact nebula was not classified as any yso ( though there are class ii / iii ysos in its vicinity ) . it is likely to be a reflection nebula illuminated by the embedded star , and seems to have been overrun by the expanding southern wall of the bipolar nebula . figure [ fig_centralarea ] shows the zoomed - in view of the central region with overlaid ysos . class i and class ii / iii sources have been marked in blue squares and green circles , respectively . while classii / iii sources are distributed throughout the image , the classi sources are located mostly along the filamentary structures extending upto beyond the cluster radius ( details in section [ section_filamentarystructures ] ) . additionally , protostars and starless cores from @xcite have been marked in blue diamond symbols and crosses , respectively . we would like to note here that none of the class 0/i sources or starless cores from @xcite had any yso counterpart from our analysis within 1 matching radius , probably because these @xcite sources are younger ( and/or are oriented edge - on ) as opposed to the ysos identified in section [ section_identification ] which are class i or class ii / iii sources . it is possible that smoothing by @xcite to 13 beam - while we are only matching upto 1 here - might be the cause of mismatch . moreover , the classification by @xcite is using envelope mass vs. bolometric luminosity diagram , inducing further uncertainty . the high - mass irs sources have been marked with arrows and labelled . irs1a south - the main ionizing source - was found , using cc / cm diagrams , to exhibit an ir excess emission . this could be because it is located in the central region surrounded by high density material . irs 2b , 3a , and 5 ( other high - mass sources , but of comparatively much lower mass than 1a south ) were found to have a negative sed slope , with the respective spectral indices being @xmath2 -1 for irs2b , -0.4 for irs3a , and -1.9 for irs5 . another noticeable feature is the arc - shaped nebula around the irs5 high mass source . @xcite have estimated the spectral type of this source as b1 type . the protostars and starless cores from @xcite seem to be distributed along the circumference of this arc - shaped nebula . this probably indicates that material has been pushed by the expansion of h ii region of irs5 and collection of dense material due to it has led to the formation of these protostellar sources and cores . the zoomed - in central region shown in figure [ fig_centralarea ] with overlaid sources shows the presence of manifold bright - rimmed cloud ( brc ) like structures with all their heads pointing in almost the same direction - towards the center - suggestive of the fact that they could have been carved out by the radiation of central high - mass star(s ) ; though alternatively , these illuminated bright - rims might just be the borders of normal density filaments carved out by the ionizing radiation and hence pointing towards the central region . there are very few ysos associated with the heads of these brcs , unlike what has been observed in many regions @xcite . figure [ fig_withhalpha ] shows a colour composite image made using pacs 160 @xmath0 m ( red ) , irac 3.6 @xmath0 m ( green ) , and h@xmath45 image ( blue ) from supercosmos h@xmath45 survey @xcite . the dashed white box in the image marks the irs5 arc - shaped nebular region . the image shows the current star formation activity ( traced by h@xmath45 emission ) vis - a - vis the dispersed molecular material by the high - mass sources ( traced by pah in 3.6 @xmath0 m band ) . as can clearly be seen , there is no correlation between h@xmath45 emission and the 3.6 @xmath0 m emission ; though this could partially be the effect of variable extinction . the h@xmath45 emission regions are also not correlated with the location of our detected ysos . h@xmath45 is mainly present in the irs5 nebular region , and in selected portions to the south of the midriff . some of the h@xmath45 emission is correlated with brcs and the so - called elephant trunks ( to the east ) . it is likely that the stellar objects in these h@xmath45 emission regions are in extremely young stages . we determined the mass of the cloud within the central cluster radius and the irs5 region using the extinction map , assuming the empirically determined gas - to - dust ratio of @xmath46 molecules cm@xmath4 mag@xmath21 @xcite . this expression has been derived assuming that the total - to - selective extinction ratio ( r@xmath39 ) is 3.1 , and that most of the gas is in form of molecular hydrogen @xcite . for the central cluster region ( which has a radius of 3 centered on irs1a ) , first the column density at each pixel ( of size 20 ) was determined using the a@xmath39 value in that pixel . thereafter , column density values in all the pixels within the cluster area were integrated to obtain the total mass in that area . same method was used to calculate the mass of the irs5 region - with the area taken as that extending from the edge of the central cluster circumference till approximately the edge of the arc - shaped nebula . the areas used for the calculation of masses in the central cluster and the irs5 region are mutually exclusive . the resulting masses obtained for the central cluster region and irs5 region were @xmath2 126 m@xmath5 and @xmath2 71 m@xmath5 , respectively . these estimates are most likely lower limits as a not unsignificant fraction of the molecular gas might have been expelled during the course of evolution of the cluster . also , some of the gas will be present in form of the ionised gas . another caveat to be kept in mind is that r@xmath39 can , in general , have wide variations @xcite . figure [ fig_filamentarystructure ] shows _ herschel _ pacs 160 @xmath0 m image of the larger region ( @xmath2 45.5@xmath3 63.0 ) with overlaid ysos , and radio sources ( in green plus symbols ) from @xcite . the location of the molecular clump tgu279-p7 from @xcite has been marked in a green cross . apart from the circular lobes which join at the midriff , filamentary structures are also seen emanating towards north - eastern and southern directions from the central region . in figure [ fig_filamentarystructure ] , three prominent parsec - scale filaments , marked on the image , can clearly be made out . due to high extinction in the region and by virtue of their intrinsically dense nature , the filamentary structures start becoming visible in emission only at _ herschel _ pacs 100 and 160 @xmath0 m bands and at submillimeter spire wavelengths ( see * ? ? ? * for the image of the larger region encompassing w40 ) . they can also be seen in absorption against the dominant pah ( plus the continuum ) emission at 11.3 @xmath0 m and 12.7 @xmath0 m in wise 12 @xmath0 m band image ( not shown here ) . [ fig_h2narrowband ] filament 1 is about @xmath48 2.4 pc long , and seems to be splitting into two parts near the midriff where it ploughs into the cloud . possible signatures are the two bow shocks ( the southern portion of filament 1 , marked with an arrow ) visible here ( also see figure [ fig_surfacedensity]_(left ) _ for clarity ) . filament 2 , which is comparatively diffuse , extends upto @xmath48 1.2 pc to the north of the central region , where it bifurcates into a filament to the north ( @xmath48 1.6 pc in length ) and the other to the west ( @xmath48 1 pc in length ) . filament 3 is the longest visible structure , extending for @xmath48 3 pc from the western part of the central region , and seemingly cutting across the southern lobe of the bipolar nebula . according to the column density map of @xcite , the value throughout - for all the three filamentary structures - is of the order of @xmath49 10@xmath50 cm@xmath4 . the filamentary structures are also seen in dust temperature as well as extinction maps ( * ? ? ? * see their figures 1 and 3 ) . figure [ fig_filamentarystructure ] also shows overlaid ysos , with class i sources , protostars , and starless cores marked in green squares , cyan diamond symbols , and cyan crosses , respectively . as can be seen on the image , most of the sources seem to be distributed along two radially outward filaments 1 and 3 from the central region which has a concentration of these ysos . filament 2 hardly contains any of these youngest sources , probably because of its diffuse nature , suggesting that it is most likely still not jeans critical . @xcite had proposed a hub - filament model according to which a central hub with a column density of @xmath2 10@xmath50 cm@xmath4 radiates filaments which can be seen upto a column density of @xmath2 10@xmath51 cm@xmath4 . it was suggested that the hub should have a minimum surface density of 25 ysos pc@xmath4 . the filamentary structures of the w40 region resemble this hub - filament structure . a possible hub region is marked in figure [ fig_centralarea ] by a white rectangle . this hub region has a stellar density of @xmath2 358 ysos pc@xmath4 ( total number of ysos divided by total area ) , while the mean column density over this hub region is _ _ n__@xmath52 @xmath2 1.8 @xmath3 10@xmath50 cm@xmath4 @xcite . among the sources from our analyses and literature , a total of 12 class i sources , 6 protostars , and 126 class ii / iii sources are located in this hub . this translates to the fraction of youngest sources ( class i / protostars ) being @xmath2 0.125 , and thus , according to the criteria used by @xcite ( that this fraction should be @xmath53 0.1 ) , this region can be deemed young . there are filaments to the west of this bipolar nebula which are a part of serpens south cluster ( not shown here ) . though they appear to be spatially correlated with region , this could be a projection effect ; as the distance to serpens south is about 260 pc @xcite , while the distance to - though with varying values based on different techniques - has been estimated to be about 500 pc @xcite . in figure 9 we display the continuum - subtracted h@xmath1 narrow - band image using inverted grey scale . the brightest emission from the _ herschel _ pacs 160 @xmath0 m image is overlayed using dashed contours . several important features are marked on this figure which are explained below . comparing h@xmath1 emission with 160 @xmath0 m contours indicates that most of the observed h@xmath1 emission traces the boundaries of filaments and dense gas traced by the 160 @xmath0 m contours . the h@xmath1 emission also reveals a faint bipolar shaped cavity centered on the w40 embedded cluster . the northern lobe of this cavity is clearly brighter than the southern lobe , however , the 160 @xmath0 m emission is also stronger in this northern region . brcs can be seen along the southern boundary of this bipolar shaped cavity in irac images ( see figure [ fig_centralarea ] ) . the morphology of the observed h@xmath1 emission features are indicative of fluorescent excited emission in most of the observed field . we note that the arc - shaped bipolar cavity in the northern region , partially surrounding the irs5 source , is also the region where several starless cores were mapped by @xcite ( see section [ section_spatialdistribution ] ) . it may therefore represent a dense shell swept by the radiation from the central embedded cluster . while the lack of collimated bipolar features indicate absence of shocked emission in general , confirmation of the fluorescent or shocked nature for the diffuse h@xmath1 emission is not possible without obtaining spectroscopic observations . figure [ fig_irac1withradio ] shows the _ spitzer _ 3.6 @xmath0 m image of the central portion of the w40 region . overlaid are the 610 mhz and 1280 mhz radio contours from gmrt . the early - type stars in the region - irs 1a south , 2b , 3a , and 5 have been marked with white squares and labelled . the @xmath54co(j=1 - 0 ) peaks from @xcite have been marked with plus symbols . it can clearly be seen that one of the @xmath54co(j=1 - 0 ) peaks lies at the western edge of the h ii region . the @xmath55co(j=2 - 1 ) and @xmath55co(j=3 - 2 ) maps from @xcite also reveal the presence of molecular cloud to the west of ionizing region . this molecular cloud to the west of the h ii region has also been observed in various molecular lines and in continuum at millimeter wavelengths @xcite . the region thus appears to be density bounded to the west as is evidently borne out by the co isotopologues peaks . but , there is no such constraint towards the east , leading to the proliferation of ionizing photons towards this direction . this is indicative of blister morphology , also referred to as champagne flow @xcite , in the region . the radio contours depicted on the image also support this blister morphology for the region , with the head of the ionized flow towards the west ( density - bounded ) , and an extended emission towards the east . the peak of 1280 mhz h ii emission is about 1.24@xmath13 to the south - west of the peak at 408 mhz from @xcite , but this could be due to their lower beamsize ( @xmath2 3@xmath13 ) . as can be seen in figure [ fig_irac1withradio ] , there are no _ spitzer _ class i ysos near this peak , only class 0/i sources or starless cores from @xcite are seen . also , one of the class 0/i sources from @xcite coincides with the 1280 mhz radio peak . this is a probable indication of some high - mass star formation going on in this region , which is also supported by the outflow seen by @xcite . the timescale of the outflow from @xcite ( @xmath2 6 @xmath3 10@xmath56 yr ) is of the order of the age estimate from @xcite ( @xmath2 4 - 9 @xmath3 10@xmath56 yr ) of class 0/i sources . however , it is plausible that the outflow observed by @xcite might simply be the material swept - up and pushed into the bipolar shaped nebula . the h ii emission does not extend all the way up to the edge of the bipolar lobes ( which marks the edge of the photo - dissociation region ) or trace them at gmrt sensitivity levels . the extent of the central h ii region is almost same as the cluster radius at half - peak surface density ( see section [ section_spatialdistribution ] and figure [ fig_surfacedensity ] ) , which is expected . the central ionizing region is limited to @xmath2 2@xmath13 ( @xmath2 0.29 pc ) on either side of the midriff of the bipolar lobes . though isolated , non - thermal radio emission sources have been discovered by @xcite ( see figure [ fig_filamentarystructure ] ) , it seems unlikely that they are related to the w40 region due to their non - thermal nature . another feature of note is the sub - region around irs 5 which shows distinct radio emission with a roughly circularly symmetric ionisation region around this source . the presence of these various morphological structures support the schematic for this region by ( * ? ? ? * see their figure 2 ) . the high - mass stars formed are probably located at the edge of the parental giant molecular cloud towards the observer , and as these sources have ionized the natal medium , the whole h ii region has broken out of the parental cloud . the filaments seem to be emanating from the midriff of the w40 region . this suggests that the density was high along the equatorial axis , as compared to the polar directions , leading to a high density contrast . thus the lyman continuum radiation has escaped along the polar directions , flinging the ionised gas ( which forms the bipolar nebula ) partially towards the observer . this flung - off material seems to have encountered more resistance towards the north - as seems likely by the smaller diameter of the northern lobe , more extincted regions towards the north , and brighter 160 @xmath0 m emission from the northern lobe ( see figures [ fig_filamentarystructure ] ) . this probably indicates that the main body of the cloud lies to the north of the w40 midriff region . the initial orientation and location of the high - mass stars seem such that this presence of molecular cloud to the north has proven harder to disrupt , while the southern thin shell of molecular material ( between the high - mass stars and the ism ) has been blown away relatively easily . we note that the southern lobe could be appearing larger than the northern lobe due to the projection effect from the viewing angle , i.e. the region might not be being viewed exactly face - on , but from a non - zero inclination angle . as has also been evidenced in @xcite for various other regions , these northern and southern lobes could be more like rings than spheres , opening up the possibility that the molecular cloud could be oblate or sheet - like . the morphology of the filamentary structure shows its complete extent till the filaments join the central region . since the filaments can be seen right till they join the midriff , they are most likely in front of the progenitor molecular cloud . radio continuum observation data was used to derive the characteristic physical parameters like emission measure , electron density , dynamical age , and strmgren radius for the region . since this region contains multiple radio sources @xcite , we carried out a global analysis using the low resolution images ( @xmath2 45@xmath31 45@xmath11 ) to get an estimate of the parameters . integrated flux density was calculated for both the frequencies within 4@xmath29 contours using the aips task `` tvstat '' ; giving us a value of 4.07 jy and 7.55 jy at 610 mhz and 1280 mhz , respectively . thereafter , using the model for free - free emission of @xcite , the flux density in a region is given by ( adapted from * ? ? ? * ) : @xmath57 @xmath58 where , @xmath59 is the integrated flux density in jy , @xmath60 is the electron temperature of the ionized core in k , @xmath61 is the frequency in mhz , @xmath62 is the electron density in @xmath63 , @xmath64 is the extent of the ionized region in pc , @xmath65 is the optical depth , @xmath66 is the correction factor , and @xmath67 is the solid angle subtended by the beam in steradian here . @xmath68 , called the emission measure , measures the optical depth in the medium ( in @xmath69pc ) . for our calculation , we took @xmath60 to be 8500 k @xcite , and thus the correction factor @xmath66 as 0.99 ( * ? ? ? * table 6 ) . using the two data points for 610 mhz and 1280 mhz , we fit the above equations using a non - linear regression with emission measure ( @xmath68 ) as a free parameter ( figure [ fig_em - fit - all ] ) . this yielded the value of emission measure as 1.63 @xmath37 0.09 @xmath3 10@xmath70 @xmath69pc . the extent ( @xmath64 ) of the central h ii region is approximately 7.0@xmath13 , equivalent to about 1.02 ( @xmath37 0.41 ) pc at a distance of 500 ( @xmath37 200 ) pc ; and thus the electron density turns out to be @xmath2 1265 @xmath37 218 @xmath63 for this central region . the lyman continuum luminosity ( photons s@xmath21 ) required for ionizing the gas was calculated by using the formulation of ( * ? ? ? * see their equation 5 ) for the optically thin regime . as is evident from figure [ fig_em - fit - all ] , 610 mhz and 1280 mhz emission lie in the optically thick and thin regimes , respectively . therefore , using the flux density of 7.55 jy estimated at 1280 mhz , the lyman continuum luminosity was calculated to be @xmath2 1.67 @xmath3 10@xmath71 photons s@xmath21 . @xcite have estimated that the radio emission is dominated by an o9.5 type star ( irs 1a south , see figure [ fig_irac1withradio ] ) with a luminosity of 6.3 @xmath3 10@xmath71 photons s@xmath21 . they place an upper limit on the lyman continuum luminosity to be 1.6 @xmath3 10@xmath72 photons s@xmath21 . @xcite , using ir and low resolution ( @xmath2 11@xmath13 ) radio observation of @xcite inferred a luminosity of @xmath2 1.5 @xmath3 10@xmath72 photons s@xmath21 . our lyman continuum luminosity estimate differs by a factor of the order of unity from the calculated flux of an o9.5 type star in @xcite and the tabulated value from @xcite . however , it is about 10% of the value from @xcite and of the upper limit in @xcite . this could be because of absorption of photons by the dust in the region , implying the presence of dense gas in the central region . as has also been noticed by @xcite , presence of dust absorption can lead to underestimation of radio luminosity . @xcite have estimated that dust absorption can range from 60% upto 96% of the ionizing photons . similar underestimation of ionizing flux by radio data has been observed for other regions in @xcite . the luminosity calculated here , therefore , is most likely just the lower limit . in the first stage during the ionization of a region by massive star(s ) , the ionization front propagates outwards leading to an expansion of the h ii region , till an equilibrium is reached between the number of ionizations and recombinations . the radius of the h ii region at this point , assuming that the ambient medium has spatially uniform density , is given by @xcite : @xmath73 where @xmath74 is called the strmgren radius ( in cm ) . @xmath75 is the initial ambient density ( in @xmath63 ) , and @xmath76 is the total recombination coefficient to the first excited state of hydrogen . the value of @xmath76 , for a @xmath60 of 8500 k , was taken to be 2.94@xmath310@xmath77 @xmath78 s@xmath21 @xcite . in the second stage of expansion , after the strmgren radius has been reached , the shock front overtakes the ionization front due to the vast pressure difference between material inside and outside of the ionization front . in this stage , the radius of the h ii region is given by @xcite : @xmath79 where @xmath80 denotes the extent of the h ii region at time @xmath81 , and @xmath82 is the speed of sound , taken as 11@xmath310@xmath83 cm s@xmath21 @xcite . since we have no way of gauging the initial ambient density ( @xmath84 ) of the molecular cloud , we have plotted the parameters for ambient density ranging from 1000 to 15000 @xmath63 to get an estimate of the ranges . figure [ fig_radiusage ] shows strmgren radius ( @xmath74 ) and dynamical age ( @xmath81 ) plotted as a function of initial ambient density of the medium . strmgren radius varies from @xmath2 0.17 to 0.03 pc , while the dynamical age varies from @xmath2 0.19 to 0.78 myr . if we were to take the higher estimates of @xmath85 ( the values discussed above ) , then , for a fixed density , the @xmath86 will increase and the @xmath81 will decrease . it should be noted that there is more than one massive star in the region , and each massive star will have its own strmgren radius and age . the multiple 3.6 cm vla sources found in the region @xcite will also contribute to the total flux . also , the radio morphology is not spherical here . therefore the above calculations should be taken as a representative value for the strmgren radius and age estimate , and @xmath62 should be taken as the averaged out electron density for the ionized region . since radio emission around irs 5 source is distinct from the rest of the morphology ( see figure [ fig_irac1withradio ] ) , we can carry out a separate calculation to determine the physical parameters around this source . for this , we first determined the 610 mhz and 1280 mhz integrated flux density in this region in a circle of 50 radius centered at the peaks of respective frequencies . this yielded a flux density of 121 and 109 mjy for 610 mhz and 1280 mhz , respectively . @xcite have detected irs 5 source using vla observations at 3.6 cm . however , they calculated a flux of 0.92 mjy for sub - arcsecond area , while here we are concerned with the extended emission . another vla source from ( * ? ? ? * source 2 in their table 2 ) lies close to the irs 5 source . however , it has a flux of 0.90 mjy ( at 3.6 cm ) , and is hence insignificant for our calculation here . following a similar set of steps as above ( section [ section_overallregion ] ) , we determined the emission measure by fitting the free - free emission sed ( see figure [ fig_em - fit - irs5 ] ) to be @xmath2 2.01 @xmath37 0.04 @xmath3 10@xmath56 @xmath69 pc . the extent of radio contours for this region is @xmath2 100 ( @xmath2 0.24 @xmath37 0.10 pc at a distance of 500 @xmath37 200 pc ) which gives the electron density as @xmath2 288 @xmath37 55 @xmath63 . we estimated the lyman continuum luminosity using the formulation of @xcite and 1280 mhz data point as in section [ section_overallregion ] ( in this case , since 610 mhz data point is in the optically thin regime too , it yields similar value ) , which results in a value of @xmath2 2.40 @xmath3 10@xmath87 photons s@xmath21 . comparing log(lyman continuum luminosity ) ( 45.38 ) with the tabulated values from ( * ? ? ? * table ii ) , we get the most probable spectral type of irs 5 as b1v , in agreement with the values from @xcite . the slightly lower value of the calculated lyman continuum luminosity , as compared to the value from @xcite is probably due to the absorption of photons by dust in the region . though the radio contours are seen within @xmath2 50 radius of the respective radio peaks ( figure [ fig_irac1withradio ] ) , it can be seen that the ionised region extends beyond that . this weak extended emission region is traced by pah emission at 7.7 @xmath0 m and 8.7 @xmath0 m in irac ch4 and at 3.3 @xmath0 m in irac ch1 ( which , it must be kept in mind , also contain continuum emission ) ( also see figure [ fig_centralarea ] ) . most of class 0/i sources and starless cores from @xcite are approximately along the circumference of a circle of radius @xmath2 80 centered at respective radio peaks . therefore , taking the extent to be 160 and the ambient density ( @xmath75 ) to be a typical value of @xmath2 1000 @xmath63 @xcite , we use equation [ equation_stromgrenradius ] in conjunction with equation [ equation_dynamicalage ] to obtain the dynamical age of the h ii region around the irs5 arc - shaped nebula to be @xmath2 0.11 myr . we note that though the ionised region appears more extended than what we have assumed in the above calculation , and the ambient density of the medium could be larger , a higher value of both these parameters will only increase the value of the dynamical age . therefore , the 0.11 myr should be considered a lower limit for the dynamical age of this sub - region . the nature of the ysos with their respectively distinct spatial distributions suggest a scenario in which ( spontaneous ) star formation has taken place at different epochs for ysos at different evolutionary stages . it is very likely that class ii / iii and the central early - type sources formed in an earlier epoch as the central region became jeans critical , followed by the formation of younger sources - class 0/i sources and starless cores from @xcite and class i ysos from our analysis - along filaments in a later epoch . the lifetimes of the dominant early type star in this region ( o9.5 ; * ? ? ? * ) is of the order of @xmath2 10 myr @xcite , while the lifetimes of class ii / iii sources is also of the order of a few myr . the fact that the two lifetimes are comparable and that this region is not very old , makes it likely that the majority of the class ii / iii and central high - mass sources formed at the same epoch . @xcite , based on the velocity of molecular hydrogen , have opined that star formation might have been initiated several myrs ago due to an external shock . it is probable that this could have played a role too and affected the earlier epoch of star formation - one which led to the formation of class ii / iii sources and central high - mass stars . this seems to have been followed by star formation leading to younger ysos . alternatively , it is possible that the formation of high - mass stars in the center could have driven a compressional wave into the individual filaments , leading the high - density regions in the filaments to become jeans critical , thus subsequently initiating the formation of class i sources , protostars , and starless cores . the stellar distribution seen has also been observed by @xcite , who , using the rosette molecular cloud as a template , have done a density analysis to conclude that almost all ir clusters lie at the junction of filaments . simulations by @xcite also show that the star formation during the course of evolution of molecular clouds to be mostly confined to the filaments and their junctions . @xcite , assuming the distance to w40 to be 260 pc , have estimated the mass of the molecular cloud ( g28.74 + 3.52 ) associated with the w40 h ii region to be about 1.1 @xmath3 10@xmath88 m@xmath5 using 2mass extinction map . the mass estimated using extinction map depends on the distance as a square of it . @xcite have recently ruled out distances below 340 pc for this region . also , mass estimation in high optical depth regions can often get underestimated using this method . so , the mass estimated by @xcite is most likely a lower limit and the actual mass could be slightly higher , probably between a few 10@xmath88 m@xmath8910@xmath90 m@xmath5 . the irs5 region ( see figure [ fig_centralarea ] ) also seems to be hosting fresh star formation at the edge of the cavity excavated by the high - mass source . the starless cores and class 0/i sources from @xcite are found to be distributed along the edge of the h ii region of this arc - shaped nebula . the arc of ( seemingly ) collected molecular material , along the edge of this arc - shaped nebula , has also been observed in various molecular lines @xcite , as well as at submillimeter wavelengths @xcite and h@xmath1 narrow - band image ( see section [ section_h2narrowbandresults ] ) . the farthest yso from the irs5 source in figure [ fig_centralarea ] is located at a distance of @xmath2 1.8 ( @xmath2 0.26 pc ) . therefore , using the isothermal sound speed of 11kms@xmath21 for the ionised region ( see section [ section_overallregion ] ) , it can be seen that the high - mass star could have lead to the sweeping - up of material which subsequently might have provided impetus to the formation of protostars and starless cores in @xmath2 0.02 myr . from the radio analysis ( see section [ section_irs5_radiocalculations ] ) , the lower limit for the dynamical age of this sub - region was calculated to be @xmath2 0.11 myr . the ages of class0 and classi ysos are generally of the order of a few 0.01 myr and 0.1 myr , respectively , and as such , their formation ( and those of starless cores ) could have been influenced by the expanding h ii region . we add a caveat that though the numbers are consistent with a scenario where the high - mass star could have influenced the star formation at the periphery , this might not be the correct scenario necessarily owing to possible complex and hitherto unexamined morphological structures in this sub - region . based on the above discussions , the following appears to be a plausible chronology of star formation in this region . as the filaments formed and joined , they led to the formation of dense core at their junction , referred to as midriff in this paper . this seems to have initiated the formation of the central high - mass star(s ) and other class ii / iii sources . at this point , an external shock - as mentioned by @xcite - might have given a boost to the star formation going on . the high - mass source irs 1a south was formed at the junction , as well as the entire cluster which is centered upon this source . due to lower density along the polar direction as opposed to the equatorial direction , when the high - mass star(s ) formed , the lyman continuum radiation escaped along the polar directions leading to the formation of the bipolar nebula . fresh ( spontaneous ) star formation seems to be going on along the filamentary structures and their junction ( albeit at lower stellar density ) as evidenced by the younger sources in class 0/i stages and starless cores . a few of brcs and elephant - trunks exhibiting h@xmath45 emission to the south of the midriff might have undergone evolution under the influence of the ionizing region which has chiselled them out . the high - mass source irs5 , which is distinct from the rest of the radio emission region , seems to have led to the formation of a separate arc - shaped nebula , and swept - up material to its periphery . future observations and analysis of the whole region in various molecular lines , radial velocity measurements , analysis of cluster properties using minimum - spanning tree method , understanding the mass and luminosity functions of the region , observations of indivdual brcs , and spectral observations of other nir bright sources will help in putting the star formation scenario on a firm footing . in this paper , we carried out a multiwavelength study of the galactic star - forming w40 h ii region . our main conclusions are as follows : 1 . using the mir data from _ spitzer _ in conjunction with nir data from ukirt , 1202 ysos were identified in the region , out of which 40 are class i sources and 1162 are class ii / iii sources . analysis of the yso distribution and nearest - neighbour surface density yields the cluster radius as @xmath2 0.44 pc and peak surface density as 650 pc@xmath4 . mass calculation using extinction map yields a value of @xmath2 126 m@xmath5 within this radius . 2 . the filamentary structures were examined to reveal 3 parsec scale filaments emanating from the midriff . two of them ( filaments 1 and 3 in our labelling ) contain most of the youngest ysos aligned along their lengths . filament 2 was found to be relatively diffuse with hardly any of the youngest ysos . 3 . sed fitting using the radio continuum emission at 610 mhz and 1280 mhz for the total emission region yielded the value of electron density to be @xmath2 1265 @xmath37 218 @xmath63 and the total lyman continuum luminosity to be @xmath2 1.67 @xmath3 10@xmath71 photons s@xmath21 . the dynamical age , for the ambient density ranging from 1000 to 15000 @xmath63 , ranges from @xmath2 0.19 to 0.78 myr . the irs5 arc - shaped nebular region was found to be distinct from the rest of the emission region , and was thus examined in radio separately . the electron density was obtained to be 288 @xmath37 55 @xmath63 and a lower limit on the dynamical age to be @xmath2 0.11 myr . a comparison of radio continuum photon luminosity with the tabulated values from @xcite shows irs5 to be of b1v spectral type , reaffirming previous estimate from @xcite . extinction map gives a value of @xmath2 71 m@xmath5 as the mass for this arc - shaped nebula . 5 . the star formation seems to have taken place in two successive epochs , with the formation of relatively - older class ii / iii and the central high - mass sources - resulting in a cluster which is centered around the high - mass star irs 1a south - followed by that of the youngest sources ( class 0/i , and starless cores ) . a distinct case is of the irs5 nebular region , where material seems to have been swept - up to the edge of this arc - shaped nebula by the expanding h ii region of the irs5 source . we thank the anonymous referee for a thorough and critical reading of the manuscript , and for the suggestions which helped in improving this paper . k.k.m . , m.s.n.k . , d.k.o . , and m.r.s . acknowledge support from marie curie irses grant ( 230843 ) under the auspices of which this work was carried out . k.k.m . would like to thank hendrik linz , mpia ( heidelberg ) for his invaluable help in obtaining the _ herschel _ data . this research made use of data products from the _ spitzer _ space telescope archive . these data products are provided by the services of the infrared science archive operated by the infrared processing and analysis centre / california institute of technology , funded by the national aeronautics and space administration and the national science foundation .

we examine the feasibility of creating and measuring large relative number squeezing in multicomponent trapped bose - einstein condensates . in the absence of multimode effects , this squeezing can be arbitrarily large for arbitrarily large condensates , but a range of processes limit the measurable squeezing in realistic trap configurations . we examine these processes , and suggest methods to mitigate them . we conclude that high levels of squeezing with large numbers of atoms is feasible , but can realistically only be achieved in particular trap geometries . we also introduce a method of maximising the measurable squeezing by using a-pulse during the process to improve spatial mode - matching .

1307.6631

quantum degenerate ultracold atomic sources have been used to study fundamental processes such as quantum phase transitions @xcite , quantum non - equilibrium thermodynamics @xcite and entanglement of massive particles @xcite . their ability to be highly isolated from their environment and controlled precisely by optical , magnetic and rf fields makes them an ideal platform for engineering and manipulating non - trivial quantum states . in quantum optics , the most common way to produce entangled states has been through squeezing , and it has been recognised that the intrinsic nonlinearities of atoms naturally lead to quadrature squeezing @xcite . number squeezing can be achieved with spin squeezing techniques @xcite , molecular dissociation of bose - einstein condensates ( becs ) @xcite , four - wave mixing @xcite , double - well potentials @xcite , or mixing quadrature - squeezed states using ramsey interferometry @xcite . the creation of these non - classical atomic states has relevance for fundamental tests of quantum mechanics such as the epr paradox @xcite , as well as atom interferometry @xcite . while these techniques have resulted in significant experimentally detectable number squeezing , they have been limited to small numbers of atoms , typically a few hundred @xcite . ultracold atomic interferometers are also reaching state of the art sensitivities in precision measurements @xcite . while all current atom interferometry is limited by technical noise sources , it appears that bec - based coherent sources offer the best chance of overcoming some of these issues , for the same reasons that optical lasers are typically preferred for optical interferometry . in particular , it appears that bose - condensed sources allow for higher fidelity mirrors and beamsplitters @xcite , are less sensitive to the effects of distortions in the optical wavefronts of the beamsplitters and mirrors and the coriolis effect @xcite , and are robust to the adverse effects of strong outcoupling when feedback - cooled @xcite . if these technical noise sources can be reduced and/or controlled , atom interferometers will be limited by atomic shot noise , which leads to a sensitivity that scales as the square root of the atomic flux . from that point , sensitivity will only be improved by increasing the atomic flux and brightness @xcite , and/or using non - classical quantum states such as squeezed states to go below the shot noise limit @xcite . producing squeezed states for metrological purposes is therefore only relevant when it can be achieved in a context of large numbers of atoms . this paper examines the generation of atomic squeezing using the inherent kerr - like nonlinearities of ultracold atoms , and demonstrates a trap geometry in which metrologically relevant levels of squeezing are present in trap geometries that are compatible with large atomic number . the creation , detection and application of squeezing requires precise mode - matching , and is therefore very sensitive to uncontrolled coupling between spatial modes of the atomic field . atomic squeezing experiments have typically attempted to operate in the single - mode regime , where the trap is tight enough such that all higher modes are frozen out @xcite . however , tight traps limit the total number of atoms , as the 3-body recombination rate places an effective maximum atomic density in the gas , and well before that limit , the interatomic interactions break the single - mode operation . both of these effects limit the number of atoms that can be squeezed effectively , so we will examine a broader range of trap geometries , and calculate the dynamics in the presence of multiple spatial modes . many similar previous simulations have been performed in one or two spatial dimensions , by integrating out some of the remaining dimensions @xcite . for simulations of semiclassical fields this can be a reasonable approximation , but it can ignore spontaneous scattering processes in full quantum field theory calculations , so we take care to include full 3d simulations . in section [ sectwomodeanalytic ] we examine the ideal behaviour of a nonlinear bosonic two - state system when that system possesses only a single spatial mode , extracting full analytic solutions for the number squeezing in each of the two states , as well as the number difference squeezing between the two states . these solutions of the ideal case provide a check on the multimode cases we consider later in the paper , as well as providing a bound on the best possible squeezing for the system . in section [ sec : mmdescription ] we show how multimode behaviour becomes important in physical systems , and introduce our model for simulating higher dimensional trap geometries . section [ sec:1dsim ] analyses one - dimensional models , highlighting some of the issues regarding optimising squeezing parameters and mode - matching in squeezing experiments . section [ sec : pipulse ] shows that the application of a @xmath0-pulse during the process can significantly improve mode - matching issues . section [ sec:2d3dsqueezing ] investigates measurable squeezing in 2d and 3d , and uses a bogoliubov analysis to explain the scaling with system size and dimensionality . section [ sectionconclusion ] describes the conclusion that large volume , flat - bottomed , three - dimensional traps , which are obviously compatible with large atom number , can produce extremely high degrees of squeezing . our goal is to describe the number squeezing possibilities of a bec comprising two relevant internal states denoted by @xmath1 and @xmath2 . the hamiltonian for a pair of coupled atomic internal modes is the spatial integral of the hamiltonian density : @xmath3 where @xmath4 is the annihilation field operator for a particle at position @xmath5 and in internal state @xmath6 , @xmath7 is the mass of the atoms , @xmath8 is the trapping potential for atoms in internal state @xmath6 , @xmath9 describes the coupling between the two states , and the @xmath10 describe the various inter- and intra - state nonlinearities and are given by @xmath11 where @xmath12 are the inter- and intra - species s - wave scattering lengths . we begin by considering a simplified version of the problem , assuming that only one of these spatial modes is relevant for each of the condensate s components . this is done by writing the multimode atomic field operators in terms of a set of basis functions @xmath13 and @xmath14 , where @xmath15 and @xmath16 annihilate a particle with normalised spatial mode @xmath17 and internal state @xmath1 or @xmath2 respectively . with this expansion , the nonlinear term in the multimode hamiltonian for the self - interaction of the @xmath1 internal state becomes @xmath18 when the system remains predominantly in spatial mode @xmath19 , we can approximate this sum as the single term @xmath20 where @xmath21 we proceed similarly for the other terms . in the parameter regimes we are interested in , the detuning between the two spatial modes will be irrelevant , so setting our zero of energy appropriately , we are left with only a standard kerr - type nonlinearity arising from atom - atom interactions , as well as a linear coupling between the two modes . the hamiltonian governing this reduced system is given by @xmath22 where the @xmath23 describe the various inter- and intra - mode nonlinearities , and @xmath24 couples the two modes and allows for population transfer between them . when considering squeezing , the hamiltonian ( [ eqtwomodehamiltonian ] ) has typically been handled within the framework of spin squeezing , which has a long history @xcite . this approach involves exploiting the equivalence between the algebra of two harmonic oscillators and that of angular momentum , and writing @xmath25 using these variables eq . ( [ eqtwomodehamiltonian ] ) becomes @xmath26 where @xmath27 is the total number operator , @xmath28 and @xmath29 . for initial states which are total number fock states , the last two terms are physically meaningless phase shifts and the first term can be removed by an appropriate choice of detuning . in this case eq . reduces to the josephson hamiltonian with zero detuning @xcite @xmath30 the system described by @xmath31 is then solved using angular momentum algebra . the reduction of eq . to the josephson hamiltonian has led to the common misconception that no squeezing can be generated in this system if @xmath32 as the effects of the @xmath33 term are usually dismissed as an irrelevant detuning . this is not true in general for initial states other than a total number fock state as the @xmath34 term leads to a number - dependent detuning . we demonstrate in this paper that for the case of an initial coherent state this term leads to squeezing . a spin - echo pulse applied half - way through the experiment which exchanges the population of modes @xmath35 and @xmath36 ( which is the protocol used in most spin squeezing experiments ) also cancels the effects of this term , and in this case squeezing is only generated if @xmath37 . in this work we directly solve eq . demonstrating that squeezing is generated ( in the absence of a spin - echo pulse ) even if @xmath38 . the system is prepared with all the population initially in @xmath1 , and vacuum in @xmath2 . we choose the initial state @xmath1 to be a coherent state with average number @xmath39 . as the system is insensitive to the initial phase of the coherent state , this is equivalent to a mixture of coherent states of uncertain phase , or a mixture of poisson - distributed number states . such a state is consistent with bec coherence experiments @xcite . at time @xmath40 , the coupling @xmath24 is applied until time @xmath41 , resulting in a portion of the population being transferred into mode @xmath2 . we assume this transfer process is fast relative to the nonlinear energy time scale , ensuring that , as the initial states were coherent states , after the population transfer both @xmath1 and @xmath2 are still described by coherent states with mean number @xmath42 and @xmath43 respectively . the coupling @xmath24 is then switched off until time @xmath44 , while the atoms interact solely through the nonlinear terms . after this hold time @xmath45 , the coupling @xmath24 is switched back on until time @xmath46 , with a phase shift @xmath47 compared to its first application . during this last stage , the two modes exchange population and the quadrature variances are converted into number variance , in the same way that homodyne measurements are used in quantum optics to convert quadrature squeezing into number squeezing , which can be directly measured . in this interpretation , @xmath47 is the relative phase of the strong local oscillator , which allows specific phase angles of the quadrature squeezing to be examined . this experiment can also be interpreted as a ramsey interferometer with a final beam splitter phase of @xmath47 . the entire sequence of pulses and the resulting populations are shown schematically in figure [ figpulsescheme ] . , which is where the majority of the squeezing occurs.,width=302 ] we solve this system analytically , and derive expressions for absolute and relative number squeezing . we present the solutions here ; the full derivation can be found in [ appendixtwomodederivation ] . if we define @xmath48 , \\ a_2 & = & n_a \exp [ n_a ( e^{-2i(\lambda_{aa}-\lambda_{ab } ) } - 1 ) ] , \\ b & = & -i e^{i \phi}\sqrt{n_b } \exp [ n_b ( e^{-i(\lambda_{bb}-\lambda_{ab } ) } - 1 ) ] , \\ b_2 & = & -n_b e^{2i\phi } \exp [ n_b ( e^{-2i(\lambda_{bb}-\lambda_{ab } ) } - 1 ) ] , \\ d & = & ab^ * - a^*b , \label{eqddef}\end{aligned}\ ] ] then the expectation values for number , number variance , and number difference variance after applying the final coupling field are given by @xmath49(t_3 ) & = & n_a c^2 + n_b s^2 + 2 n_a n_b c^2 s^2 \nonumber\\ & & + c^2 s^2 ( d^2 - e^{i(\lambda_{aa } - \lambda_{bb } ) } b_2 a_2^ * - e^{-i(\lambda_{aa } - \lambda_{bb } ) } b_2^ * a_2 ) \nonumber \\ & & + i c s d ( 1 - 2 n_a c^2 -2 n_b s^2 ) \nonumber\\ & & + 2 i c^3 s n_a ( e^{-i(\lambda_{aa } - \lambda_{ab } ) } a b^ * - e^{i(\lambda_{aa } - \lambda_{ab } ) } a^ * b ) \nonumber \\ & & + 2 i c s^3 n_b ( e^{i(\lambda_{bb } - \lambda_{ab } ) } a b^ * - e^{-i(\lambda_{bb } - \lambda_{ab } ) } a^ * b ) , \\ { \mathrm{var } } [ n_b](t_3 ) & = & n_a s^2 + n_b c^2 + 2 n_a n_b c^2 s^2 \nonumber \\ & & + c^2 s^2 ( d^2 - e^{i(\lambda_{aa } - \lambda_{bb } ) } b_2 a_2^ * - e^{-i(\lambda_{aa } - \lambda_{bb } ) } b_2^ * a_2 ) \nonumber \\ & & + i c s d ( -1 + 2 n_a s^2 + 2 n_b c^2 ) \nonumber \\ & & + 2 i c^3 s n_b ( e^{-i(\lambda_{bb } - \lambda_{ab } ) } a^ * b - e^{i(\lambda_{bb } - \lambda_{ab } ) } a b^ * ) \nonumber \\ & & + 2 i c s^3 n_a ( e^{i(\lambda_{aa } - \lambda_{ab } ) } a^ * b - e^{-i(\lambda_{aa } - \lambda_{ab } ) } a b^ * ) , \\ { \mathrm{var } } [ n_a - n_b](t_3 ) & = & n_a + n_b + 4 i c s d ( n_a - n_b)(s^2 - c^2 ) \nonumber \\ & & + 4 c^2 s^2 ( 2 n_a n_b + d^2 - e^{i(\lambda_{aa } - \lambda_{bb } ) } a_2^ * b_2 - e^{-i(\lambda_{aa } - \lambda_{bb } ) } a_2 b_2^ * ) \nonumber \\ & & + 4 i c s ( c^2 - s^2 ) \left ( n_a e^{-i(\lambda_{aa } - \lambda_{ab } ) } a b^ * - n_a e^{i(\lambda_{aa } - \lambda_{ab } ) } a^ * b \right . \nonumber \\ & & \,\,\,\,\ , \left . + n_b e^{-i(\lambda_{bb } - \lambda_{ab } ) } a^ * b - n_b e^{i(\lambda_{bb } - \lambda_{ab } ) } a b^ * \right ) . \label{eqnumdiffvariance}\end{aligned}\ ] ] in the case where all the nonlinearities are equal , such that @xmath50 , the number variances all reduce to those of a coherent state , i.e. @xmath51=n_a$ ] , and no number squeezing is possible . it should also be noted that , although there are three independent nonlinearities @xmath23 in the hamiltonian ( [ eqtwomodehamiltonian ] ) , the expressions for the number variances only depend on differences . this means that number squeezing is parameterised by only two independent quantities , for example @xmath52 and @xmath53 . in figure [ figtwomodeanalyticexamples ] we plot a solution for the normalised number variance in mode @xmath1 as a function of @xmath47 , choosing a typical set of nonlinearities . this example illustrates the basic features of the system : during the final coupling period , the normalised number variance undergoes cyclic variation with a period @xmath54 , and has a series of minima . it is necessary to choose both the correct phase @xmath47 as well as the correct length of time for the coupling to be applied in order to obtain optimum squeezing . as mentioned previously , squeezing is possible in this system even when the nonlinearity in the reduced josephson hamiltonian @xmath55 . figure [ figzeroeffectivechi ] demonstrates that significant squeezing is possible in this system if a spin - echo pulse is not used . this two - mode analytic model allows arbitrarily good number squeezing , provided there are no limits to the number of particles in the system or to the hold time . the two - mode approximation can be valid in tight traps with very low particle number , and this result correctly predicts the squeezing demonstrated by gross _ unfortunately , for larger numbers of atoms in realistic traps there are more than two modes present , and these multimode effects limit the level of squeezing that can be achieved . /n_a$ ] in mode @xmath56 as a function of final recombination time plotted on a logarithmic scale , using four different mixing angles @xmath47 . parameters : @xmath57 , @xmath58s . solid line , @xmath59 ; dash - dotted line , @xmath60 ; dashed line , @xmath61 ; dotted line , @xmath62 . these particular parameters yield 21db of number squeezing for the @xmath63 and @xmath60 cases.,width=377 ] ( solid line ) and number difference squeezing between states @xmath1 and @xmath2 ( dashed line ) as a function of final recombination time in the case where @xmath64 . parameters : @xmath65 , @xmath66s . solid line , @xmath67 ; dashed line , @xmath68.,width=377 ] the zero - dimensional , two - mode model described in the previous section shows that arbitrarily good squeezing can be achieved using the intrinsic nonlinearities in a bec , but it ignores the existence of multiple spatial modes . there are three problems caused by the existence of multiple modes , and these will limit the achievable squeezing . the first problem is present even if the spatial modes are uncoupled . the two - mode model shows that the parameters required to obtain best squeezing are dependent on the strength of the nonlinearity . in the multimode case , the effective nonlinearity of a mode is a function of the mode shape , as is shown by eq . ( [ eqchiuequivalence ] ) . the optimum hold time @xmath69 depends on the product of the effective mode nonlinearity and the number of particles in that mode , as does the recombination phase @xmath47 . different modes in a multimode environment will typically have varying effective nonlinearity - particle number products and therefore squeeze at different rates , with each mode having a different optimum @xmath69 and optimum recombination phase @xmath47 . this means that the best squeezing will be achieved by choosing values of @xmath69 and @xmath47 that , averaged across all the spatial modes present , result in the overall lowest total number variance . clearly , however , this averaged number variance will be still be higher than the specific mode that exhibits the best squeezing . the second problem is due to the number - dependent dynamics of the spatial modes , which is observable even in a semiclassical simulation . due to the nonlinear term in the hamiltonian , the local phase evolution of the atomic field is density dependent . consequently the mode shape of the atomic field changes depending on the nonlinearity - density product , which means that unless @xmath70 for all @xmath71 , the mode shapes will not overlap perfectly during the final coupling pulse . this degraded mode - matching will result in lower efficiency when converting the quadrature squeezing to number squeezing . we might speculate that the effects of these first two problems would be lessened by trap geometries that lead to near - constant density profiles for the bec . this should enable higher squeezing . we can not make strong conclusions without also considering the effects of coupling between spatial modes , however , which leads to the third potential problem : coupling between modes is inevitable in the presence of a nonlinearity , meaning that the modes can not be analysed independently . coupling between modes will disturb mode - matching , as well as mixing fields with different phase evolution , which can rapidly destroy squeezing . an analytic solution to the multimode , higher - dimensional problem is intractable , so we must turn to numerical solutions . we use stochastic methods based on phase space representations to simulate the dynamics of the multimode quantum fields . these methods alleviate the problem of directly simulating elements of the hilbert space , the size of which increases exponentially with the number of spatial modes @xcite . stochastic methods achieve this by finding a sufficiently well - behaved quasi - probability representation for the density matrix , which can then be simulated as the average behaviour of a number of low - dimensional samples . these sample objects are the same size as the equivalent classical field . in our case , the only tractable phase space representation is the functional wigner representation @xcite , which is well behaved when the quantum field is approximately gaussian , as it is for coherent and squeezed states . stochastic methods based on the functional wigner representation have been used to model the behaviour of becs and atom lasers @xcite , as the wigner representation allows the mapping of the system to a fokker - planck equation with positive semi - definite diffusion matrix ( unlike the @xmath72-representation @xcite ) , and does not suffer exponential path - weighting problems that would constrain it to very short simulation intervals ( as would the positive-@xmath72 representation @xcite ) . the equation of motion for the density matrix is defined by the hamiltonian in eq . ( [ eqfieldhamiltonian ] ) . this defines the evolution of the functional wigner distribution of the system , which has a one - to - one correspondence with the density matrix . the equation of motion for the functional wigner distribution contains derivatives of third order which we assume to be negligible . this uncontrolled approximation is called the truncated wigner approximation ( twa ) , and while it has been used widely on ultracold gases @xcite , care must be taken that the simulation remains valid . for our simulations we checked the results of our simulations against the analytic two - mode solution , as well as checking for typical indications of twa breakdown such as the appearance of negative densities in lightly populated modes . we were also able to use the bogoliubov analysis described in section [ sec:2d3dsqueezing ] as an independent estimate of the validity of the twa . we found that the twa remained valid except in some 3d simulations with high nonlinearities , where there were a large number of modes , and the number of particles per mode could drop to ten or less in the densest regions where the squeezing was generated . over very long time scales , this low mode occupation began to result in twa breakdown . fortunately , the best squeezing was typically found in regions where the twa was valid . under the twa , the equation of motion for the functional wigner distribution is a fokker - planck equation , and can therefore be sampled by a set of stochastic equations , as described in @xcite . these stochastic fields @xmath73 are related to field operator expectation values by @xmath74 , \label{eqexpectationrelations}\ ] ] where @xmath75 denotes symmetric ordering of the operators , and @xmath76 denotes a stochastic average over the variables @xmath73 . while we see below that the evolution of these fields is deterministic , the initial state still requires a random element , so multiple realisations are required . when approximating these equations on a discrete grid , the magnitude of initial noise , the relationship between stochastic averages and the expectation values , and the equations of motion all become grid - dependent . this is not a signature of a fundamental problem , as the hamiltonian in eq . ( [ eqfieldhamiltonian ] ) assumes a contact potential between the atoms , which is not correct below a length scale that can probe the details of the true potentials . furthermore , it is not a problem in practice , as the calculation gives grid - independent predictions for physical observables well before that regime is reached . these stochastic equations take the form @xmath77 where @xmath78 , @xmath79 , @xmath80 is the volume element of the grid on which the simulation is carried out , and the terms proportional to @xmath81 correspond to vacuum corrections . it is a peculiarity of the wigner representation that , up to the vacuum correction terms , eqs . ( [ eqstochasticequations ] ) look identical to those of the coupled semi - classical nonlinear schrdinger equation . the quantum statistical nature of the wigner equations enters due to the fact that eqs . ( [ eqstochasticequations ] ) are run many times and stochastically averaged , with each run using different set of random initial conditions describing ( in this case ) the noise on a coherent state . we integrate eqs . ( [ eqstochasticequations ] ) numerically in one , two and three spatial dimensions using the numerical package xmds2 @xcite . for the one- and two - dimensional simulations , the dimensional reduction is achieved by estimating the mode shape in three dimensions , and using dimensionally reduced values for @xmath10 that match the zero - dimensional reduction given by eq . ( [ eqchiuequivalence ] ) . to make relevant comparisons between equivalent situations with different numbers of spatial dimensions , this mode shape is typically chosen to match the chemical potentials of the initial states . a one - dimensional simulation is sufficient to examine the hypothesis that the squeezing will be higher for a bec where the spatial mode has a more constant density profile . we compare squeezing for fields starting in the ground states of a harmonic trap with negligible nonlinearity ( a gaussian ) , a harmonic trap with strong nonlinearity ( thomas - fermi ) , and a constant potential ( constant density ) . using eq . ( [ eqchiuequivalence ] ) we find the following equivalences @xmath82 where @xmath83 , @xmath84 , @xmath85 are the angular frequencies of the harmonic trap , and @xmath86 is the number of particles in the mode of interest . choosing parameters such that the squeezing occurs fast enough that the mode shape is reasonably stable over the simulation gives the best squeezing , and allows us to make a fair comparison between the squeezing achievable by different mode shapes without being concerned that different modes will change shapes at different rates , ruining mode - matching . the results of this simulation are shown in figure [ figmodalsqueezingeffects1d ] , which clearly confirms the hypothesis that maximum squeezing is obtained when the mode shape is closest to constant density . as a function of the time the final coupling pulse is applied . we compare the analytic two - mode solution ( thick red ) , the constant density mode ( solid blue ) , the thomas - fermi mode ( dashed ) and the gaussian mode ( dash - dotted ) . the nonlinearities of the various 1d simulations have been adjusted for their mode shape so that they are equivalent to that of the 0d , two - mode mode nonlinearity . parameters used were @xmath87 , @xmath88s , @xmath89 . maximum squeezing is achieved when the atomic density profile is the most uniform.,width=302 ] one way to avoid any requirement to engineer and maintain specific atomic modes is to post - select a spatial mode . rather than considering the variance in the total number of atoms in each internal state , it may be possible to consider only the statistics in a spatially filtered area of the atomic cloud . for example , suppose the atoms were in a gaussian atomic distribution , but we consider a mode that only includes the atoms within @xmath90 standard deviations of the centre . for low @xmath90 , this mode shape is considerably closer to the constant density case than a standard gaussian . the equivalence between the single mode case and this mode is given by @xmath91 such spatial filtering can easily produce higher squeezing , but there is the obvious disadvantage that it significantly reduces the effective number of atoms . this is a serious issue for most applications of squeezed sources , such as interferometry , where the signal to noise scales with the square root of the flux . we will focus on methods for producing squeezing with large atomic number . modal mismatch is generated by the different atomic potentials seen by the two internal states . a straightforward method of minimising this effect is to apply a @xmath0-pulse to swap the populations of the internal states after half of the evolution . as can be seen by eqs . and , such a pulse also leaves the quantum statistics of each state unchanged . the idea of this scheme is to make the effective potential seen by the atoms in state @xmath35 during the second half of the hold time very similar to that seen by the atoms that were in state @xmath35 during the first half of the hold time , and will only work if the modes @xmath35 and @xmath36 have an equal occupation at the start of the hold time , or if @xmath92 . in the latter case , this reduces to a spin - echo pulse , and as the squeezing effects of the @xmath33 term in eq . are cancelled in this case , there will be no squeezing . however , in the limit of short hold times , for equal occupations of the two modes , and when @xmath93 , this technique reduces the difference in the mode shapes after the hold time . in practice this will never work perfectly , as @xmath94 and @xmath95 will typically have evolved by the time the @xmath0-pulse is applied , and so the spatial dynamics in the second half of the hold time will not be quite the same as those during the first half . nonetheless , in many regimes , such a @xmath0-pulse can produce some improvement . more importantly , as this scheme accepts that the two modes will change shape , and attempts to make them change in a similar fashion , it allows the possibility of much longer hold times . that is , if we are no longer constrained to hold times short enough such that the mode shapes of the atoms in states @xmath56 and @xmath96 do not change significantly , much more squeezing can be extracted from the system . these effects are illustrated in fig . [ pipulsefig ] , which shows the best squeezing obtainable from a specific system with and without the @xmath0-pulse , as well as the extra squeezing that can be obtained by allowing a longer hold time as well as the @xmath0-pulse . -pulse halfway through the hold time . common parameters used were @xmath97 , @xmath98 , @xmath99 . solid red line : analytic two - mode solution , @xmath100s , @xmath101 ( 10.9db squeezing ) ; solid blue line : no @xmath0-pulse , @xmath100s , @xmath101 ( 4.9db squeezing ) ; dashed blue line : with @xmath0-pulse , @xmath100s , @xmath101 ( 5.8db squeezing ) ; dash - dotted blue line : with @xmath0-pulse and the longer hold time @xmath102s allowed by the application of the pulse , @xmath103 ( 7.2db squeezing).,width=302 ] .5 , @xmath104 , @xmath105 , @xmath100s , @xmath106 . ( b ) @xmath97 , @xmath98 , @xmath107 , @xmath100s , @xmath101.,title="fig:",width=226 ] [ figdimensionalsqueezingeffects : sub1 ] .5 , @xmath104 , @xmath105 , @xmath100s , @xmath106 . ( b ) @xmath97 , @xmath98 , @xmath107 , @xmath100s , @xmath101.,title="fig:",width=226 ] [ figdimensionalsqueezingeffects : sub2 ] the simulations in the previous two sections were one dimensional , and show how having spatially constant density and using a @xmath0-pulse will improve the resultant squeezing . while these conclusions remain valid for simulations in two or three dimensions , adding each additional dimension increases the number of spatial modes in a given energy range . for becs of different dimension at the same energy per particle , this leads to a degradation of squeezing , even when the potential ( and the resulting atomic density ) is constant across the bulk of the trap . this is demonstrated in fig . [ figdimensionalsqueezingeffects ] , where we show the squeezing generated in 1d , 2d and 3d becs for two different parameter regimes . in both cases we take the bec to be trapped in a box potential , which in the thomas - fermi limit leads to a constant density profile across the bec . in figure [ figdimensionalsqueezingeffects]a we consider the case where @xmath108 , @xmath109 , while in figure [ figdimensionalsqueezingeffects]b we consider the case @xmath110 , @xmath111 . in both cases it is clear that number squeezing is best in lower dimensions , and becomes degraded or non - existent as we move to a three dimensional system ; the effect is worse if more than one nonlinearity is non - zero . to find the trap geometries that produce the best squeezing , it is important to understand the root cause of this loss of squeezing in higher dimensional traps . the dependence on dimensionality is not explained by the processes described in section [ sec : mmdescription ] , which focus on the effects of inhomogeneity within the trap . fundamentally , this degradation of the squeezing is due to nonlinearity - induced coupling between different momentum modes . we can understand the origin and scaling of this degradation of squeezing by considering the response of the condensate to small fluctuations about the mean field . we do this using bogoliubov theory @xcite , which considers the first order quantum - mechanical fluctuations about the mean field . to simplify the analysis we consider the case in which @xmath112 , @xmath113 and the condensate has a spatially - constant density in a box of side lengths @xmath114 , @xmath115 and @xmath116 . in this limit the occupation of the non - zero momentum modes ( the non - condensed fraction ) in the @xmath35 internal state at the end of the pulse sequence is ( a derivation is given in [ appendixbogderivation ] ) @xmath117 ^ 2 , \label{eqbogoccupation}\end{aligned}\ ] ] where the bogoliubov mode frequency is @xmath118 , and the free particle frequency is @xmath119 . the non - zero momentum modes can be expected to have an impact on the squeezing of the system when their occupation is a non - negligible fraction of the total number of atoms . the fractional occupation of these modes is @xmath120 where the sum over @xmath121 is taken over the available non - zero momentum modes @xmath122 with @xmath123 , @xmath124 and @xmath125 arbitrary integers that are not all zero . as a function of the hold time @xmath126 . the blue dashed line is the bogoliubov integral approximation given in eq . ( [ eqbogfraction1d ] ) , the purple dot - dashed line is the bogoliubov summation expression given in eq . ( [ eqcondensatedepletion ] ) , and the solid black line is the result of a truncated wigner simulation . parameters : @xmath127 , @xmath128 , @xmath129 , @xmath130 , @xmath99 , @xmath131.,width=377 ] figure [ figbogsqueezingvalidation ] demonstrates the agreement between truncated wigner simulations and the bogoliubov theory for the non - condensed fraction for parameters where the occupation of these modes approach 15% of the total occupation of the system . the disagreement for larger non - condensed fractions occurs because the bogoliubov theory used in deriving is not number - conserving as it neglects the effect of the non - zero momentum modes on the condensate . to minimise the damaging effects of the non - condensed fraction it is better to operate in a regime where the total occupation of the non - zero momentum modes @xmath132 is small , which is also fortuitously the regime in which the bogoliubov theory is a good approximation , so we can investigate in more detail . in 1d eq . is @xmath133 we approximate this sum by the integral @xmath134 where @xmath135 and @xmath136 is the length of the 1d box condensate . this approximation will be valid in the limit that @xmath137 , where @xmath138 as for small @xmath139 the allowed values @xmath140 will closely sample the region @xmath141 where @xmath142 is significant . changing integration variables to @xmath143 and defining @xmath144 , the non - condensed fraction can be written as @xmath145 where @xmath146 in 2d and 3d the equivalent expressions are @xmath147 where @xmath148 and @xmath149 is the area of the two - dimensional system and @xmath150 is the volume of the three - dimensional system . figure [ figbogsqueezingvalidation ] demonstrates the agreement in 1d between the integral approximation to @xmath151 in , the summation expression in and a truncated wigner simulation . we wish to examine the scaling of the non - condensed fraction @xmath152 with the physical size of the condensate . other system parameters such as the total number @xmath86 and the occupation of the @xmath35 and @xmath36 states at @xmath153 ( @xmath154 and @xmath155 , respectively ) are kept constant and chosen to maximise the squeezing of the equivalent two - mode model presented in section [ sectwomodeanalytic ] . in particular , this fixes the values of @xmath156 , @xmath47 , and @xmath157 . as the size of the condensate is scaled , @xmath158 will change , but we will vary @xmath126 to keep @xmath159 constant . therefore @xmath160 will also be constant . with this in mind , we can write @xmath126 in terms of the system size and the nonlinearity @xmath161 . the non - condensed fraction can then be shown to scale with the physical size of the condensate as : @xmath162 the non - condensed fraction @xmath151 thus scales differently with trap size in each of 1d , 2d and 3d . for one - dimensional traps , the non - condensed fraction increases with system size , whereas for 2d traps it is constant . for three - dimensional traps , the non - condensed fraction _ decreases _ with volume . this suggests that the best system to use will be a three - dimensional box potential , and the system size can be increased until the condensate depletion becomes negligibly small . figures [ fig1d_squeezing_and_bog_scaling][fig3d_bog_mode_occupation_scaling ] show the non - condensed fraction predicted by the bogoliubov analysis compared to that predicted by a full numerical simulation in one , two and three dimensions , as a function of the trap size . also shown in the 1d and 2d cases is the maximum amount of number squeezing obtainable , as predicted by the numerical simulation . the full squeezing scaling curve is not shown in 3d , as for the smaller trap sizes the twa method breaks down , and results are unreliable . for the larger trap sizes with small non - condensate fractions , however , the twa simulation remained valid , and we obtained relative number squeezing of 12.7 db . the traps are assumed to have a constant potential , which minimises the effects of the multimode issues described in section [ sec : mmdescription ] . we see that the squeezing does indeed degrade at large trap sizes in 1d , but remains constant in 2d . for small trap sizes , the squeezing in 1d asymptotically approaches that of the ideal 0d system . we note that the non - smooth nature of the squeezing prediction in 2d is due to the stochastic nature of the simulations , and the necessity of averaging over fewer quantum trajectories due to the very high number grid points in this regime . the non - smooth nature of the non - condensate mode fraction in 3d , on the other hand , is due to the number of modes being both discrete and low in this regime , and counting only the non - condensed fraction in modes with an energy less than than @xmath163 , the chemical potential . the simulation shown in fig . [ fig3d_bog_mode_occupation_scaling ] demonstrates that provided particle number is kept fixed , the number of accessible states for three - dimensional traps decreases with volume in 3d . this is because the energy of the ground state of the bec ( the chemical potential , @xmath163 ) is reducing faster than the density of states at low energies is increasing , which in turn results in fewer states that can become dynamically occupied . this scaling of the estimated bogoliubov fraction in 3d suggests that strong squeezing will be found in very large , weakly trapped condensates , provided the trapping potentials are box - like . such box - like traps have already been demonstrated and shown to be capable of producing becs @xcite . although systematic analysis of squeezing across a wide range of large 3d traps and interaction strengths is impossible due to the breakdown of the twa method , fortunately the method is accurate when the highest squeezing is produced . while inter - state scattering lengths of most atomic species and states are not well characterised , extremely high squeezing can be found in systems with large , three - dimensional box potentials , and asymmetry in the three scattering lengths . as a demonstration of the plausibility of such a scheme , we carried out a full 3d simulation using the geometry of the optical box potential described in ref . @xcite , using the same potentials ( a cylindrical `` box '' with length 63@xmath163 m and diameter 30@xmath163 m ) and condensate number ( @xmath164 @xmath165rb atoms ) . we set one s - wave scattering length to @xmath166 nm and the others to zero : @xmath167 nm , @xmath168 . the associated nonlinearities are given by eq . ( [ equij ] ) . this system produced number squeezing of 15db , and an extremely low non - condensed fraction of @xmath169 . both of these numbers are in agreement with the bogoliubov calculations . .48 in a 1d trap . ( left ) the proportion of the non - condensed fraction in state @xmath1 relative to total particle number , with the bogoliubov theory model in red and the full numerical solution in blue . ( right ) relative number squeezing in mode @xmath1 as a function of trap size . parameters : @xmath170 , @xmath171jm , @xmath111 , @xmath172s.,title="fig : " ] .48 in a 1d trap . ( left ) the proportion of the non - condensed fraction in state @xmath1 relative to total particle number , with the bogoliubov theory model in red and the full numerical solution in blue . ( right ) relative number squeezing in mode @xmath1 as a function of trap size . parameters : @xmath170 , @xmath171jm , @xmath111 , @xmath172s.,title="fig : " ] .48 in a 2d trap . ( left ) the proportion the non - condensed fraction in state @xmath1 relative to total particle number , with the bogoliubov theory model in red and the full numerical solution in blue . ( right ) relative number squeezing in mode @xmath1 as a function of trap size . parameters : @xmath170 , @xmath173jm@xmath174 , @xmath111 , @xmath175s.,title="fig : " ] .48 in a 2d trap . ( left ) the proportion the non - condensed fraction in state @xmath1 relative to total particle number , with the bogoliubov theory model in red and the full numerical solution in blue . ( right ) relative number squeezing in mode @xmath1 as a function of trap size . parameters : @xmath170 , @xmath173jm@xmath174 , @xmath111 , @xmath175s.,title="fig : " ] in a 3d trap . ( left ) the proportion the non - condensed fraction in state @xmath1 relative to total particle number , with the bogoliubov theory model in red and the full numerical solution in blue . ( right ) relative number squeezing in mode @xmath1 as a function of trap size . parameters : @xmath176 , @xmath177jm@xmath178 , @xmath111 , @xmath179s.,width=264 ] the large kerr - type nonlinearities present between neutral atoms in bec present the possibility of significant squeezing , but they also cause potentially complicated multimode behaviour when large becs are used . furthermore , the nonlinearities between multiple internal atomic states are not yet known for many atomic species , and often have sensitive dependence on external magnetic fields due to feshbach resonances @xcite . this paper analyses the requirements for producing useful squeezing via these nonlinearities for systems with large numbers of atoms . we have presented an analytic solution to the two - mode model of atomic fields with a strong kerr nonlinearity , allowing for a rapid estimate of the atomic number difference squeezing that can be obtained from these systems as scattering length data is obtained . in general , it predicts arbitrarily good squeezing in the two - mode limit . we then examined trap geometries in which this limit can be approached . using a constant potential in the trap ( a ` box ' trap ) allows the bec density to be spatially uniform throughout the process . this is shown to alleviate the difficulties with choosing correct hold times and phases for the coupling pulses , and also alleviates the majority of the mode - matching issues for the final recombination of the two modes . if such a box potential is not available , the application of a @xmath0-pulse part way through the hold period can also improve mode matching by imparting the same modal dynamics to both states . even in the best case , where a box potential is used , the existence of multiple possible momentum modes allows the possibility of cross - mode scattering via nonlinear processes , which can degrade or totally remove any number squeezing . this degradation is visible only in full 3d simulations , and not lower dimensional approximations , due to its dependence on the number of available modes . it is worth noting that despite their difficulty , in multimode systems with large atom number , full 3d simulations need to be carried out to accurately determine the presence of number or spin squeezing . excluding these modes by making them energetically unfavourable can reduce the system to a single mode problem , making our analytic solution valid , but requires extremely tight trapping potentials . such a regime also strongly limits the number of atoms that can be trapped , partly due to the rate at which the required trapping frequencies grow with atom number , and partly due to inelastic scattering rates . using a bogoliubov analysis , however , it was shown that the coupling between different modes could be minimised in a second regime , namely the _ weak _ trapping limit for 3d traps . this latter regime is in fact ideally compatible with large atomic numbers , and shows the prospect of extremely high squeezing . a 3d twa simulation using an experimentally demonstrated trapping potential yielded @xmath180db of squeezing with @xmath181 particles . in practice , the only limit to the number of atoms in such traps would be the stability of the trap , as the bec would become increasingly sensitive to fluctuations as the energy per particle is reduced as the volume increases . jh acknowledges the support of an arc future fellowship . in this appendix we derive the equations for the number variances . we use the model described in section [ sectwomodeanalytic ] , where we consider a two mode system with @xmath86 particles , initially prepared as a coherent state with all the population in mode @xmath1 at time @xmath40 . a coupling pulse is then applied , assumed to be on resonance , so that the system evolves under the hamiltonian @xmath182 until time @xmath41 . from time @xmath153 until time @xmath183 the coupling is turned off , and the system evolves under the hamiltonian @xmath184 finally , from time @xmath44 until time @xmath185 , the coupling is once again applied with a phase offset of @xmath47 relative to the first time , and the system evolves via @xmath186 we will work in the heisenberg picture , and derive expressions for @xmath187 and @xmath188 in terms of @xmath189 and @xmath190 , which in turn can be expressed in terms of @xmath191 and @xmath192 . we will show that the state of the system at time @xmath153 consists of separate coherent states in modes @xmath1 and @xmath2 , enabling us to calculate expectation values at time @xmath46 in terms of the known state at @xmath153 . in order to have more compact expressions we will use the notation @xmath193 beginning with the hamiltonian @xmath194 , during the period @xmath195 mode @xmath196 will evolve as @xmath197 utilizing the hadamard lemma @xmath198 + \frac{1}{2!}[\hat{x},[\hat{x},\hat{y } ] ] + \frac{1}{3!}[\hat{x},[\hat{x},[\hat{x},\hat{y } ] ] ] + \ldots \label{eqhadamard}\ ] ] we obtain @xmath199 similarly , we find @xmath200 from the form of ( [ eqa1 ] ) and ( [ eqb1 ] ) it is clear that as both modes @xmath196 and @xmath201 began in coherent states , they will remain in coherent states , with only their amplitudes changing . specifically , at time @xmath153 the system is in a product of coherent states , which we denote @xmath202 with @xmath203 and @xmath204 with @xmath205 . next we consider evolution under @xmath206 . taking mode @xmath196 first , we note that as it commutes with the @xmath207 term , and since @xmath208 commutes with @xmath209 we have @xmath210 where we have defined @xmath211 . using ( [ eqhadamard ] ) to move @xmath212 through the exponential , with some algebra one can show that @xmath213 \hat{a}_1 \nonumber \nonumber \\ & = & \exp[-i \lambda_{aa } \hat{a}_1^{\dagger } \hat{a}_1 ] \hat{a}_1.\end{aligned}\ ] ] to handle the cross - nonlinearity term in ( [ eqa2evolution ] ) we use the identity @xcite @xmath214 to obtain @xmath215 similarly , from permutation symmetry , we have @xmath216 finally , to obtain @xmath217 and @xmath218 we use ( [ eqa1 ] ) and ( [ eqb1 ] ) with the phase factor attached to the @xmath201 operator to obtain @xmath219 we are now in a position to evaluate the number variance of the system throughout the period of the final coupling . the number variances of the fields are defined as @xmath220 & = & \langle \hat{a}^{\dagger}_3 \hat{a}_3 \hat{a}^{\dagger}_3 \hat{a}_3 \rangle - \langle \hat{a}^{\dagger}_3 \hat{a}_3 \rangle ^2 \label{eqnavariance } \\ { \mathrm{var}}[n_b ] & = & \langle \hat{b}^{\dagger}_3 \hat{b}_3 \hat{b}^{\dagger}_3 \hat{b}_3 \rangle - \langle \hat{b}^{\dagger}_3 \hat{b}_3 \rangle ^2 . \label{eqnbvariance}\end{aligned}\ ] ] making use of ( [ eqa3 ] ) and using the shorthand notation @xmath221 \\ c & = & \cos[\omega ( t_3 - t_2)]\end{aligned}\ ] ] we have @xmath222 \nonumber \\ & & + i c^3 s [ 2 e^{-i \phi } \hat{b}^{\dagger}_2 \hat{a}^{\dagger}_2 \hat{a}_2 \hat{a}_2 - 2 e^{i \phi } \hat{a}^{\dagger}_2 \hat{a}^{\dagger}_2 \hat{a}_2 \hat{b}_2 + e^{-i \phi } \hat{b}^{\dagger}_2 \hat{a}_2 - e^{i \phi } \hat{a}^{\dagger}_2 \hat{b}_2 ] \nonumber \\ & & + i c s^3 [ 2 e^{-i \phi } \hat{b}^{\dagger}_2 \hat{b}^{\dagger}_2 \hat{a}_2 \hat{b}_2 - 2 e^{i \phi } \hat{a}^{\dagger}_2 \hat{b}^{\dagger}_2 \hat{b}_2 \hat{b}_2 + e^{-i \phi } \hat{b}^{\dagger}_2 \hat{a}_2 - e^{i \phi } \hat{a}^{\dagger}_2 \hat{b}_2 ] \end{aligned}\ ] ] @xmath223 \nonumber \\ & & + i c^3 s [ 2 e^{i \phi } \hat{a}^{\dagger}_2 \hat{b}^{\dagger}_2 \hat{b}_2 \hat{b}_2 - 2 e^{-i \phi } \hat{b}^{\dagger}_2 \hat{b}^{\dagger}_2 \hat{b}_2 \hat{a}_2 + e^{i \phi } \hat{a}^{\dagger}_2 \hat{b}_2 - e^{-i \phi } \hat{b}^{\dagger}_2 \hat{a}_2 ] \nonumber \\ & & + i c s^3 [ 2 e^{i \phi } \hat{a}^{\dagger}_2 \hat{a}^{\dagger}_2 \hat{b}_2 \hat{a}_2 - 2 e^{-i \phi } \hat{b}^{\dagger}_2 \hat{a}^{\dagger}_2 \hat{a}_2 \hat{a}_2 + e^{i \phi } \hat{a}^{\dagger}_2 \hat{b}_2 - e^{-i \phi } \hat{b}^{\dagger}_2 \hat{a}_2 ] \end{aligned}\ ] ] clearly , to calculate ( [ eqnavariance ] ) and ( [ eqnbvariance ] ) we will need the expectation values of terms quartic and quadratic in @xmath224 and @xmath225 . specifically , we require @xmath226 , @xmath227 , @xmath228 , @xmath229 , @xmath230 , @xmath231 , @xmath232 , @xmath233 , and @xmath234 , as well as their hermitian conjugates . as @xmath235 and @xmath236 commute with the hamiltonian @xmath206 , they are constants of motion so we have @xmath237 @xmath238 @xmath239 furthermore , @xmath240 and @xmath241 also commute with @xmath242 , giving @xmath243 @xmath244 the remaining terms are not constants of motion , so we proceed by making use of ( [ eqa2 ] ) and ( [ eqb2 ] ) . for @xmath228 we have @xmath245 to calculate the expectation value of this expression , the creation and annihilation operators must be normally ordered . making use of the identity @xcite @xmath246 allows us to compute expectation values of coherent states of the form @xmath247 . \label{eqexpectationvalueofexponential}\end{aligned}\ ] ] this gives @xmath248 \exp[|\beta|^2 ( e^{-i ( \lambda_{bb } - \lambda_{ab } ) } -1)].\end{aligned}\ ] ] we now consider @xmath232 . using ( [ eqa2 ] ) and ( [ eqb2 ] ) and the fact that @xmath196 and @xmath201 commute , we have @xmath249 where we have made multiple uses of the identity @xmath250 which follows from ( [ eqefeidentity ] ) . finally , applying ( [ eqexpectationvalueofexponential ] ) we obtain @xmath251 \exp[|\beta|^2(e^{-2i ( \lambda_{bb } - \lambda_{bb } ) } -1)].\ ] ] using the same techniques one can show that @xmath252 \exp[|\beta|^2(e^{i ( \lambda_{bb } - \lambda_{ab } ) } -1 ) ] \\ \langle \hat{a}^{\dagger}_2 \hat{b}^{\dagger}_2 \hat{b}_2 \hat{b}_2 \rangle & = & |\beta|^2 \alpha^ { * } \beta e^{-i(\lambda_{bb}-\lambda_{ab } ) } \exp[|\alpha|^2(e^{i ( \lambda_{aa } - \lambda_{ab } ) } -1 ) ] \exp[|\beta|^2(e^{-i ( \lambda_{bb } - \lambda_{ab } ) } -1)].\end{aligned}\ ] ] employing the definitions ( [ eqsdef])([eqddef ] ) we obtain @xmath253 \nonumber \\ & & + i c^3 s [ 2 n_a e^{-i(\lambda_{aa}-\lambda_{ab } ) } ] a b^ * - 2 n_a e^{i(\lambda_{aa}-\lambda_{ab } ) } a^ * b + a b^ * - a^ * b ] \nonumber \\ & & + i c s^3 [ 2 n_b e^{i(\lambda_{bb}-\lambda_{ab } ) } ] a b^ * - 2 n_b e^{-i(\lambda_{bb}-\lambda_{ab } ) } a^ * b + a b^ * - a^ * b ] \\ \langle \hat{b}^{\dagger}(t_3 ) \hat{b}(t_3 ) \hat{b}^{\dagger}(t_3 ) \hat{b}(t_3 ) \rangle & = & ( n_b^2 + n_b ) c^4 + ( n_a^2 + n_a ) s^4 \nonumber \\ & & + c^2 s^2 [ n_a + n_b + 4 n_a n_b - e^{-i(\lambda_{aa}-\lambda_{bb } ) } b_2^ * a_2 - e^{i(\lambda_{aa}-\lambda_{bb } ) } b_2 a_2^ * ] \nonumber \\ & & + i c^3 s [ 2 n_b e^{-i(\lambda_{bb}-\lambda_{ab } ) } ] a^ * b - 2 n_b e^{i(\lambda_{bb}-\lambda_{ab } ) } a b^ * - a^ * b + a b^ * ] \nonumber \\ & & + i c s^3 [ 2 n_a e^{i(\lambda_{aa}-\lambda_{ab } ) } ] a^ * b - 2 n_a e^{-i(\lambda_{aa}-\lambda_{ab } ) } a b^ * - a^ * b + a b^ * ] . \end{aligned}\ ] ] when substituted into ( [ eqnavariance ] ) and ( [ eqnbvariance ] ) these expectation values give the absolute number variances stated in section [ sectwomodeanalytic ] . to calculate the number difference variance , we note @xmath254 = { \mathrm{var}}[n_a ] + { \mathrm{var}}[n_b ] - 2\langle \hat{a}^{\dagger}_3 \hat{a}_3 \hat{b}^{\dagger}_3 \hat{b}_3\rangle + 2\langle \hat{a}^{\dagger}_3 \hat{a}_3 \rangle \langle \hat{b}^{\dagger}_3 \hat{b}_3\rangle \label{eqnumdiffvarianceappendix}.\ ] ] we write ( [ eqnumdiffvarianceappendix ] ) in terms of @xmath224 and @xmath255 by using using ( [ eqa3 ] ) and ( [ eqb3 ] ) , and then use the expectation values calculated earlier in this appendix . this procedure gives the expression ( [ eqnumdiffvariance ] ) given in section [ sectwomodeanalytic ] . we consider the bogoliubov model presented in section [ sec:2d3dsqueezing ] taken in the limit of a homogeneous condensate with @xmath256 , @xmath99 . in this limit , the hamiltonian for the internal state @xmath35 during the hold time is @xmath257 \,dv . \label{eqbogoriginalhamiltonian}\end{aligned}\ ] ] to find the response of the condensate to small fluctuations we define fluctuation operators for the non - zero momentum modes @xmath258 where @xmath259 is the condensate mean field ( @xmath260 for @xmath121 as the condensate is homogenous ) . the hamiltonian is then expanded in terms of @xmath261 retaining only the lowest order contributions . the lowest order contribution is second order : @xmath262.\end{aligned}\ ] ] this hamiltonian is diagonalised with the standard bogoliubov transformation @xcite @xmath263 where @xmath264 is the frequency of a free particle , and @xmath118 are the frequencies of the bogoliubov modes : @xmath265 the expectation value of the number of atoms in momentum mode @xmath266 after the hold time is then found to be @xmath267 { \ensuremath{\left<\delta \hat{a}^\dagger_\mathbf{k}(t_1 ) \delta\hat{a}_\mathbf{k}(t_1)\right > } } \\ & \phantom{\mathrel{=}}- 4 u_\mathbf{k } v_\mathbf{k } \sin(\omega_\mathbf{k}\tau_\text{hold } ) { \operatorname{im}}\left\{\left[1-i2 u_\mathbf{k}^2 \sin(\omega_k \tau_\text{hold})\right ] { \ensuremath{\left<\delta\hat{a}_{-\mathbf{k}}(t_1 ) \delta\hat{a}_\mathbf{k}(t_1)\right>}}\right\}\\ & \phantom{\mathrel{= } } + 4 u_\mathbf{k}^2 v_\mathbf{k}^2 \sin^2(\omega_\mathbf{k}\tau_\text{hold } ) . \end{split}\end{aligned}\ ] ] if the non - zero momentum modes are in a vacuum state at @xmath41 then this reduces to @xmath268 ^ 2.\end{aligned}\ ] ] after the final rabi coupling pulse the occupation of the non - zero momentum mode @xmath266 is @xmath269 ^ 2\end{aligned}\ ] ] which is the expression given earlier in eq . . 10 dring d , mcdonald g , debs j e , figl c , altin p a , bachor h - a , robins n p , and close j d 2010 quantum - projection - noise - limited interferometry with coherent atoms in a ramsey - type setup _ phys . rev . _ a * 81 * 043633 jaskula j c , bonneau m , partridge g b , krachmalnicoff v , deuar p , kheruntsyan k v , aspect a , boiron d , and westbrook c i 2010 sub - poissonian number differences in four - wave mixing of matter waves _ phys . lett . _ * 105 * 190402 meyer v , rowe m a , kielpinski d , sackett c a , itano w m , monroe c , and wineland d j 2001 experimental demonstration of entanglement - enhanced rotation angle estimation using trapped ions _ phys . lett . _ * 86 * 5870 leibfried d , barrett m d , schaetz t , britton j , chiaverini j , itano w m , jost j d , langer d , wineland d j 2004 toward heisenberg - limited spectroscopy with multiparticle entangled states _ science _ * 304 * 1476

in this paper we prove that the poincar map associated to a lorenz like flow has exponential decay of correlations with respect to lipschitz observables . this implies that the hitting time associated to the flow satisfies a logarithm law . the hitting time is the time needed for the orbit of a point to enter for the first time in a ball centered at , with small radius . as the radius of the ball decreases to its asymptotic behavior is a power law whose exponent is related to the local dimension of the srb measure at : for each such that the local dimension exists , holds for almost each . in a similar way it is possible to consider a quantitative recurrence indicator quantifying the speed of coming back of an orbit to its starting point . similar results holds for this recurrence indicator .

0901.0574

it is well known that in a chaotic dynamics the pointwise , future behavior of an initial condition is unpredictable and even impossible to be described by using a finite quantity of information . on the other hand many of its statistical properties are rather regular and often described by suitable versions of classical theorems from probability theory : law of large numbers , central limit theorem , large deviations estimations , correlation decay , hitting times , various kind of quantitative recurrence and so on . in this article we consider a class of flows which contain the celebrated geometric lorenz flow and we will study some of its statistical features by a sharp estimation for the decay of correlations of its first return map on a suitable poincar section . this will give a quantitative recurrence estimation and an estimation for the scaling behavior of the time which is needed to hit small targets ( logarithm law ) . let @xmath9 be a @xmath10 flow in @xmath11 . quantitative recurrence estimations and logarithm laws can be seen in the following framework : we are interested in a quantitative estimation of the speed of approaching of a certain orbit @xmath12 ( starting from the point @xmath1 ) of the system to a given target point @xmath13 . let @xmath14 be a ball with radius @xmath4 centered at @xmath13 . we consider the time @xmath15 needed for the orbit of @xmath1 to enter in @xmath14 for the first time and the asymptotic behavior of @xmath16 as @xmath4 decreases to @xmath5 . often this is a power law of the type @xmath17 and then it is interesting to extract the exponent @xmath18 by looking at the behavior of @xmath19 in this way , we have a hitting time indicator for orbits of the system . as @xmath20 ( for the equivalence see @xcite ) . ] if the orbit @xmath21 starts at @xmath13 itself and we consider the second entrance time in the ball @xmath22(because the orbit trivially starts inside the ball ) with the same construction as before , we have a quantitative recurrence indicator . if the dynamics is chaotic enough , often the above indicators converge to a quantity which is related to the local dimension of the invariant measure of the system and in the hitting time case this relation is called _ logarithm law_. hitting time results of this kind ( sometime replacing balls with other suitable target sets ) have been proved in many continuous time dynamical systems of geometrical interest : geodesic flows , unipotent flows , homogeneous spaces , etc . ( see e.g. @xcite ) . for discrete time systems this kind of results hold in general if the system has fast enough decay of correlation ( @xcite ) . mixing is however not sufficient , since this relation does not hold in some slowly mixing system having particular arithmetical properties ( @xcite ) . some further connections with arithmetical properties are shown in interesting examples as rotations and interval exchange maps ( see e.g. @xcite ) . this kind of problem is also connected with the so called dynamical borel cantelli results ( see @xcite and e.g. @xcite ) . moreover , in the symbolic setting , similar results about the hitting time are used in information theory ( see e.g. @xcite ) . about quantitative recurrence , our approach follows a set of results connecting a quantitative recurrence estimation with local dimension ( see e.g. @xcite ) . we remark that the speed of correlation decay for lorenz like flows is not yet known ( although some are proved to be mixing , see @xcite ) hence quantitative recurrence and hitting time results can not be proved directly using this tool , instead of this we will consider a poincar section , estimating its correlation decay and work with return times . let @xmath23 $ ] be a unit interval , we consider a flow @xmath24 on @xmath25 having a poincar section on a square @xmath26 satisfying the following properties : 1 ) : : the flow induces a first return map @xmath27 of the form @xmath28 ( preserves the natural vertical foliation of the square ) and : 1.a ) : : there is @xmath29 and @xmath30 such that , if @xmath31 are such that @xmath32 $ ] then @xmath33 1.b ) : : @xmath34 is @xmath35-lipschitz with @xmath36 ( hence is uniformly contracting ) on each vertical leaf @xmath37 : @xmath38 1.c ) : : @xmath39 is onto and piecewise monotonic , with two @xmath40 increasing branches on the intervals @xmath41,@xmath42 $ ] and @xmath43 where it is defined can be relaxed to @xmath44^{-1}<1 $ ] provided that the map @xmath45 is eventually expanding in the sense of @xcite , chapter 3 . ] . moreover @xmath46 @xmath47 . 1.d ) : : @xmath48 has bounded variation . by the statistical properties of the map @xmath45 , which is piecewise expanding , under the above assumptions , it turns out that @xmath49 has a unique srb measure @xmath50 we then ask the following property for the flow : 2 ) : : the flow @xmath24 is transversal to the section @xmath51 and its return time to @xmath52 is integrable with respect to @xmath53 . in section [ sec : constr - geometr - model ] we will describe the geometric lorenz system and we show that it satisfies these properties . the main results of the paper concern some statistical properties of @xmath24 and @xmath49 , more precisely : * theorem a ( decay of correlation for the poincar map ) * _ the unique srb measure _ @xmath54 _ of _ @xmath49 _ _ has exponential decay of correlation with respect to lipschitz observables . _ this result is proved in section [ 2p1 ] ( theorem [ resuno ] ) where the reader can also find a precise definition of correlation decay . the proof also uses a regularity estimation for the invariant measure @xmath54 which can be found in the appendix i ( lemma 8.1 ) and is proved by sort of lasota - yorke inequality . we remark that a stretched - exponential bound for the decay of correlation for a two dimensional lorenz like map was given in @xcite and @xcite . we say that a point @xmath55 is _ regular _ if there are @xmath56 and @xmath57 such that @xmath58 induces a diffeomorphism between a neighborhood of @xmath59 and a neighborhood of @xmath3 . in section [ sec : sbrfluxo ] we recall how to construct an srb ergodic invariant measure for the flow @xmath24 which will be denoted by @xmath60 . it turns out that this measure has the following property * theorem b ( logarithm law for the flow ) * _ for each regular _ @xmath13 _ such that the local dimension _ @xmath61 _ _ is defined it holds _ _ @xmath62__for a.e . starting point _ _ @xmath1__. _ _ this is proved in section [ sec : loglaw ] ( theorem [ main1 ] ) and uses the above decay of correlation estimation for the first return map @xmath49 , a result from @xcite giving the hitting time estimation for systems having faster than polynomial decay of correlations and finally the integrability of return time is used to get the result for the flow . using the main result of @xcite , by a similar construction , if the flow also satisfies the following property 3 ) : : the map @xmath45 has derivative bounded by a power law near @xmath63 : there is a @xmath64 s.t . @xmath65 is bounded in a neighborhood of @xmath63 we prove ( in section [ sec : recurrpoli ] , corollary 7.4 ) the following estimation for the return time * theorem c ( quantitative recurrence ) * _ if the flow satisfies conditions 1),2 ) , 3 ) above , then for a.e . _ @xmath1 _ it holds _ @xmath66 in the appendix ii we give an auxiliary result , using a theorem by steinberger @xcite showing that the local dimension is defined a.e . for the geometric lorenz system . in this section we will introduce and motivate the so - called geometric lorenz system . this is the main example where our results will be applied . indeed we will see that assumption 1.a), ... ,1.d ) and 2 ) of the introduction are verified for this model . the results in this section are however not strictly necessary for the proofs of our main theorems . the reader familiar with the construction of such models can skip it and start at section [ sec : sbrfluxo ] . in 1963 the meteorologist edward lorenz published in the journal of atmospheric sciences ( @xcite ) an example of a parametrized @xmath67-degree polynomial system of differential equations @xmath68 as a very simplified model for thermal fluid convection , motivated by an attempt to understand the foundations of weather forecast . numerical simulations performed by lorenz for an open neighborhood of the chosen parameters suggested that almost all points in phase space tend to a _ chaotic attractor_. an _ attractor _ is a bounded region in phase - space , invariant under time evolution , such that the forward trajectories of most ( positive probability ) or , even , all nearby points converge to . and what makes an attractor _ chaotic _ is the fact that trajectories converging to the attractor are _ sensitive with respect to initial data _ : trajectories of two any nearby points get apart under time evolution . lorenz chaotic attractor , width=340 ] lorenz s equations proved to be very resistant to rigorous mathematical analysis , and also presented serious difficulties to rigorous numerical study . as an example , the above stated existence of a chaotic attractor for the original lorenz system where not proved until the year 2000 , when warwick tucker did it with a computer aided proof ( see @xcite ) . in order to construct a class of flows having properties which are very similar to the lorenz system and are easier to be studied , afraimovich , bykov and shilnikov @xcite , and guckenheimer , williams @xcite , independently constructed the so - called _ geometric lorenz models _ for the behavior observed by lorenz . these models are flows in @xmath69-dimensions for which one can rigorously prove the existence of a chaotic attractor that contains an equilibrium point of the flow , which is an accumulation point of typical regular solutions . recall that @xmath70 is a regular solution for the flow @xmath71 if @xmath72 for all @xmath73 and @xmath74 . the accumulation of regular orbits near an equilibrium prevents such sets from being hyperbolic @xcite . furthermore , this attractor is robust ( in the @xmath10 topology ) : it can not be destroyed by any small perturbation of the original flow . we point out that the robustness of this example provides an open set of flows which are not morse - smale , nor hyperbolic , and also non - structurally stable ( see @xcite ) . [ sec : constr - geometr - model ] the results of the paper will be given for a class of three dimensional flows which will be defined axiomatically . to show that these axioms are verified in the geometric lorenz models we give a detailed introduction to this model . we first analyze the dynamics in a neighborhood of the singularity at the origin , and then we complete the flow , imitating the butterfly shape of the original lorenz flow ( see figure [ fig - lorenz3d ] and compare with figure [ l3d ] ) . in the original lorenz system the origin @xmath75 is an equilibrium of saddle type for the vector field defined by equations ( [ e - lorenz - system ] ) with real eigenvalues @xmath76 , @xmath77 satisfying @xmath78 ( in the classical lorenz system @xmath79 , @xmath80 , @xmath81 ) . if certain nonresonance conditions are satisfied ( see @xcite ) this vector field is smoothly linearizable in a neighborhood of the origin . to construct a model which is similar to the original lorenz one we start with a linear system @xmath82 , with @xmath76 , @xmath83 satisfying relation ( [ eigenvalues ] ) this vector field will be considered in the cube @xmath84 ^ 3 $ ] containing the origin . for this linear flow , the trajectories are given by @xmath85 where @xmath86 is an arbitrary initial point near @xmath87 . consider @xmath88 and @xmath89 @xmath52 is a transverse section to the linear flow and every trajectory crosses @xmath52 in the direction of the negative @xmath90 axis . consider also @xmath91 with @xmath92 . for each @xmath93 the time @xmath94 such that @xmath95 is given by @xmath96 which depends on @xmath97 only and is such that @xmath98 when @xmath99 . hence , using ( [ tempo ] ) , we get ( where @xmath100 for @xmath101 ) @xmath102 , we have @xmath103 . consider @xmath104 defined by @xmath105 behavior near the origin.,width=264 ] it is easy to see that @xmath106 has the shape of a cusp triangle without the vertex @xmath107 . in fact the vertex @xmath107 are cusp points at the boundary of each of these sets . the fact that @xmath108 together with equation ( [ l ] ) imply that @xmath106 are uniformly compressed in the @xmath109-direction . clearly each segment @xmath110 is taken by @xmath111 to another segment @xmath112 as sketched in figure [ l3dcusp ] . @xmath113 takes @xmath114 to @xmath52.,width=226 ] to imitate the random turns of a regular orbit around the origin and obtain a butterfly shape for our flow , as it is in the original lorenz flow depicted at figure [ fig - lorenz3d ] , we proceed as follows . recall that the equilibrium @xmath115 at the origin is hyperbolic and so its stable @xmath116 and unstable @xmath117 manifolds are well defined , @xcite . observe that @xmath117 has dimension one and so , it has two branches , @xmath118 , and @xmath119 . the sets @xmath106 should return to the cross section @xmath52 through a flow described by a suitable composition of a rotation @xmath120 , an expansion @xmath121 and a translation @xmath122 . the rotation @xmath120 has axis parallel to the @xmath109-direction . more precisely is such that @xmath123 , then @xmath124 the expansion occurs only along the @xmath1-direction , so , the matrix of @xmath125 is given by @xmath126 with @xmath127 and @xmath128 . the first condition is to ensure that the image of the resulting map is contained in @xmath52 , the second condition makes a certain one dimensional induced map to be piecewise expanding . this point will be discussed below . @xmath129 is chosen such that the unstable direction starting from the origin is sent to the boundary of @xmath51 and the image of both @xmath114 are disjoint . these transformations @xmath130 take line segments @xmath131 into line segments @xmath132 as sketched in figure [ l3d ] , and so does the composition @xmath133 . this composition of linear maps describes a vector field in a region outside @xmath84 ^ 3 $ ] in the sense that one can use the above matrices to define a vector field @xmath134 such that the time one map of the associated flow realizes @xmath133 as a map @xmath135 . this will not be explicit here , since the choice of the vector field is not really important for our purposes ( provided the return time is integrable ) . the above construction allow to describe for each @xmath136 the orbit @xmath137 of each point @xmath138 : the orbit will start following the linear field until @xmath114 and then it will follow @xmath134 coming back to @xmath52 and so on . let us denote with @xmath139 the set where this flow acts . the geometric lorenz flow is then the couple @xmath140 defined in this way . the poincar first return map will be hence defined by @xmath141 as @xmath142 the combined effects of @xmath113 and @xmath143 on lines implies that the foliation @xmath144 of @xmath52 given by the lines @xmath145 is invariant under the return map . in another words , we have @xmath146 _ for any given leaf @xmath70 of @xmath144 , its image @xmath147 is contained in a leaf of @xmath144_. combining equations ( [ l ] ) with the effect of the rotation composed with the expansion and the translation , we obtain that @xmath49 must have the form @xmath148 where @xmath149 and @xmath150 are given by @xmath151 @xmath152 where @xmath153 and @xmath154 are suitable affine maps . here @xmath155 , @xmath156 . now , to find an expression for @xmath157 we proceed as follows . recall @xmath158 , @xmath111 is as in ( [ l ] ) , @xmath159 is as in ( [ derivadader ] ) . given @xmath160 with @xmath161 , we have @xmath162 restricting the rotation and the other linear maps to @xmath114 and composing the resulting matrices we get @xmath163 the expression for @xmath157 at @xmath164 with @xmath165 is similar . observe that by construction @xmath166 in equation ( [ f ] ) is piecewise @xmath167 . moreover , equation ( [ eq : derivadaf ] ) implies the following bounds on its partial derivatives : 1 . for all @xmath168 , we have @xmath169 . as @xmath170 , @xmath171 , there is @xmath172 such that @xmath173 . 2 . for all @xmath174 , we have @xmath175 . as @xmath176 and @xmath171 , we get @xmath177 is uniformly contracting on the leaves of the foliation @xmath144 : there is @xmath178 such that , if @xmath70 is a leaf of @xmath144 and @xmath179 then @xmath180 where @xmath181 can be chosen as the one given by equation ( [ gy ] ) . the lorenz map @xmath182.,width=151 ] now let us outline the main properties of @xmath182 . we recall that we chosen @xmath183 such that @xmath128 . the following properties are easily implied from the construction of @xmath71 : 1 . by equation ( [ flo ] ) and the way @xmath122 is defined , @xmath182 is discontinuous at @xmath184 . the lateral limits @xmath185 do exist , @xmath186 , 2 . @xmath182 is @xmath187 on @xmath188 . by the choice of @xmath183 it holds @xmath189 . by the convexity properties of @xmath182 we then obtain that @xmath190 3 . the limits of @xmath191 at @xmath184 are @xmath192 . we obtain that @xmath182 is a piecewise expanding map . moreover @xmath182 has a dense orbit , which in its turn implies that the closure of the maximal invariant set by @xmath182 is the whole interval @xmath193 , see ( * ? ? ? * lemma 2.11 ) . now recall that the _ variation _ @xmath194 of a function @xmath195\to { { \mathbb r}}$ ] is defined by @xmath196 where the supremum is taken over all finite partitions @xmath197 , @xmath198 , of @xmath199 $ ] . the variation @xmath200 of @xmath201 over an arbitrary interval @xmath202 $ ] is defined by a similar expression , with the supremum taken over all the @xmath203 , with @xmath204 . one says that @xmath201 has _ bounded variation _ , or @xmath201 is bv for short , if @xmath205 . the one dimensional map has the following property , which is important to obtain the existence of an srb invariant measure and its statistical properties . [ boundedvariation ] let @xmath71 a @xmath187 geometric lorenz flow as before and @xmath182 be the one - dimensional map associated to @xmath71 . then @xmath206 is bv . each branch of @xmath182 is the composition of an affine map with @xmath207 then it is a convex function . hence , the derivative @xmath208 is monotonic on each branch , implying that @xmath209 is also monotonic . on the other hand , @xmath210 is bounded because @xmath211 . thus @xmath210 is monotonic and bounded and hence is bv . we have seen that @xmath182 is a topologically transitive piecewise expanding map with @xmath206 bv . the statistical properties of such maps are well known . next we state a result about it , which will be used later : [ prop : densidadebvdef](@xcite , prop.3.8 ) the one - dimensional @xmath182 admits a unique invariant probability @xmath212 which is absolutely continuous with respect to lebesgue measure @xmath213 , it is ergodic and so a srb measure for the map . moreover @xmath214 is a bv function and in particular it is bounded . furthermore @xmath182 has exponential decay of correlations for @xmath215 and bv observables and any a.c.i.m . converges exponentially fast to the invariant measure : there are constants @xmath216 and @xmath217 , depending on the system such that for each @xmath218 and observables @xmath219:@xmath220 summarizing , for what it was said above , the study of the @xmath69-flow can be reduced to the study of a bi - dimensional map @xmath49 . moreover , the dynamics of this map can be further reduced to a one - dimensional map , @xmath182 called one dimensional _ lorenz map_. figure [ l1d ] shows the graph of this one - dimensional transformation , and figure [ l2d ] sketches @xmath221 . in this section , following @xcite we construct a physical measure for a flow @xmath71 which satisfies the assumptions 1a), ... ,1d),2 ) in the introduction . as noticed in the previous section , these assumptions 1a), ... ,1d ) are satisfied by the geometric lorenz system . properties 1a), ... ,1d ) implies that the flow poincar map has an invariant foliation and the one dimensional induced map @xmath45 is piecewise expanding . piecewise expanding maps ( see proposition [ prop : densidadebvdef ] ) admits a unique invariant probability measure @xmath222 which is absolutely continuous with respect to lebesgue measure @xmath213 . from @xmath222 we may construct a srb measure @xmath223 , for the first return map @xmath49 through the following general procedure ( @xcite ) . since @xmath222 is defined on the interval @xmath193 which can be identified to the space of leaves of the contracting foliation @xmath144 , we may also think of it as a measure on the @xmath224-algebra of borel subsets of @xmath52 which are union of entire leaves of @xmath144 . using the fact that @xmath49 is uniformly contracting on leaves of @xmath144 we conclude that the sequence @xmath225 of push - forwards of @xmath222 under @xmath49 is weak*-cauchy : given any continuous @xmath226 @xmath227 is a cauchy sequence in @xmath228 , see @xcite . define @xmath223 to be the weak*-limit of this sequence , that is , @xmath229 for each continuous @xmath230 . then @xmath223 is invariant under @xmath49 , and it is an ergodic physical measure for @xmath49 . the last statement follows from the fact that @xmath222 is an ergodic physical measure for @xmath45 , together with the fact that asymptotic time - averages of continuous functions @xmath231 are constant on the leaves of @xmath144 . given any point @xmath1 whose orbit sooner or later will cross @xmath52 we denote with @xmath232 the first strictly positive time such that @xmath233 ( the _ return time _ of @xmath1 to @xmath52 ) . coherently with the geometric lorenz system , we will denote by @xmath234 the ( full measure , by the assumption 1 in the introduction ) subset of @xmath51 where @xmath94 is defined . now we show how to construct an physical invariant measure for the flow , _ when the return time is integrable _ : @xmath235 denote by @xmath236 the equivalence relation on @xmath237 given by @xmath238 let @xmath239 and @xmath240 , where @xmath241 is the quotient map and @xmath242 is a lebesgue measure in @xmath228 . equation ( [ eq : integraldotempo - limitado ] ) gives that @xmath243 is a finite measure . let @xmath244 be defined by @xmath245 . let @xmath246 . the measure @xmath247 is a physical for the flow @xmath71 : @xmath248 for every continuous function @xmath249 , and lebesgue almost every point @xmath250 we end the subsection remarking that the geometric lorenz flow has integrable return time , hence the above construction for the invariant measure can be applied to it . as before denote by @xmath251 the return time to @xmath52 . then , recalling equation ( [ tempo ] ) there are @xmath252 such that @xmath253 combining this with the definition of @xmath223 and the remark made above that @xmath254 is a bounded function , we conclude that the return time is integrable @xmath255 let us recall the definition of local dimension and fix some notations for what follows . let @xmath256 be a metric space and assume that @xmath8 is a borel probability measure on @xmath257 . given @xmath258 , let @xmath259 be the ball centered at @xmath1 with radius @xmath4 . the _ local dimension _ of @xmath8 at @xmath258 is defined by @xmath260 if this limit exists . in this case @xmath261 . this notion characterizes the local geometric structure of an invariant measure with respect to the metric in the phase space of the system , see @xcite and @xcite . we can always define the _ upper _ and the _ lower _ local dimension at @xmath1 as @xmath262 if @xmath263 almost everywhere the system is called _ exact dimensional_. in this case many properties of dimension of a measure coincide . in particular , @xmath18 is equal to the infimum hausdorff dimension of full measure sets : @xmath264 this happens in a large class of systems , for example , in @xmath265 diffeomorphisms having non zero lyapunov exponents almost everywhere , @xcite . let us establish a relation between @xmath266 and @xmath267 which will be used in the following . [ dimensao ] let @xmath268 and @xmath269 be the projection on @xmath51 given by @xmath270 if @xmath1 is on the orbit of @xmath271 and the orbit from @xmath109 to @xmath1 does not cross @xmath52 ( if @xmath138 then @xmath272 ) . for all regular points @xmath273 @xmath274 first observe that for product measures as @xmath275 , where @xmath242 is the lebesgue measure at the line , the formula is trivially verified . but , by construction @xmath276 , where @xmath277 is a local bi - lipschitz map at each regular point . since the local dimension is invariant by local bi - lipschitz maps , it follows the required equation ( [ relacao ] ) . in this section we estimate the decay of correlations for a class of lorenz like maps containing the first return map of the geometric lorenz system described above . inspired by a remark of r. s. mckay ( see @xcite , p. 8) , this will be done by estimating the speed of approaching of iterates of suitable measures ( corresponding to lipschitz observables ) to the invariant measure . for this purpose we will consider the space of measures on @xmath51 as a metric space , endowed with the wasserstein - kantorovich distance , whose basic properties we are going to describe . * notations . * let us introduce some notations : we will consider the @xmath278 distance on @xmath279^{2}$ ] , so that the diameter , @xmath280 . this choice is not essential , but will avoid the presence of many multiplicative constants in the following making notations cleaner . as before , the square @xmath51 will be foliate by stable , vertical leaves . we will denote the leaf with @xmath1 coordinate by @xmath281 or , with a small abuse of notation , when no confusion is possible we will denote both the leaf and its coordinate with @xmath37 . let @xmath282 be the measure @xmath283 such that @xmath284 . moreover , let us sometime for short denote the integral by @xmath285 let @xmath286 a measure on @xmath51 . in the following , such measures on @xmath51 will be often disintegrated in the following way : for each borel set @xmath287@xmath288with @xmath289 being probability measures on the leaves @xmath37 and @xmath290 is the marginal on the @xmath1 axis which will be an absolutely continuous probability measure . we will also denote by @xmath291 its density . let us consider the projection @xmath292 on the @xmath109 coordinate . let us denote the `` restriction '' of @xmath286 on the leaf @xmath37 by @xmath293this is a measure on @xmath193 and it is not normalized . we remark that @xmath294 the set of borel probability measures on @xmath295 . let us finally denote by @xmath296 be the best lipschitz constant of @xmath297 @xmath298 and set @xmath299 let us consider a bounded metric space @xmath295 and let us consider the following notion of distance between measures : given two probability measures @xmath300 and @xmath301 on @xmath295 @xmath302 where @xmath303 is the space of @xmath304-lipschitz functions on @xmath305 we remark that adding a constant to the test function @xmath306 does not change the above difference @xmath307 the above defined @xmath308 has moreover the following basic properties . * prop 7.1.5)[prop : ambros ] the following properties hold 1 . @xmath308 is a distance and if @xmath295 is separable and complete , then @xmath309 with this distance is a separable and complete metric space . 2 . a sequence is convergent for the @xmath308 metrics if and only if it is convergent for the weak topology . [ conv](distance and convex combinations ) if @xmath310 then@xmath311indeed@xmath312@xmath313@xmath314@xmath315we also remark that the same kind of estimation can be done if the convex combination has more than 2 summands . if @xmath306 is @xmath316-lipschitz and @xmath317 are probability measures then @xmath318 we give some general facts on the relation between @xmath308 distance and decay of correlations . let @xmath319 be a dynamical system on a metric space with invariant probability measure @xmath286 . the transfer operator associated to @xmath49 will be indicated with @xmath320 . [ prop:1]let @xmath321 and @xmath322 , @xmath323 . let @xmath283 be a probability measure which is absolutely continuous with respect to @xmath324 and @xmath325 . then @xmath326 dividing by @xmath296 we can suppose @xmath327 . as @xmath328 then the decay of correlations between @xmath329 and @xmath306 can be estimated in function of the distance between @xmath330 and @xmath286 as@xmath331@xmath332 @xmath333 conversely , [ prop:2]if for each @xmath334 @xmath323 and @xmath321 it holds @xmath335then taking @xmath336 it holds@xmath337 consider @xmath338 . hence@xmath339since this hold for each @xmath306 hence @xmath340 we will consider maps having an invariant foliation , as we have seen in the lorenz map . the invariant measure will then be disintegrated as in equation ( [ dis ] ) into a family of measures @xmath289 on almost each stable leaf @xmath37 and an absolutely continuous measure @xmath290 on the unstable direction . if @xmath341 and @xmath342 are two disintegrated measures as above , their @xmath308 distance can be estimated in function of some distance between their respective marginals on the @xmath1 axis and measures on the leaves : [ prod]let @xmath341 , @xmath342 be measures on @xmath51 as above , such that for each borel set @xmath287 * @xmath343 * @xmath344 with @xmath345 absolutely continuous with respect to the lebesgue measure , moreover let us suppose 1 . for almost each vertical leaf @xmath37 , @xmath346 and 2 . @xmath347 then @xmath348 considering the @xmath308 distance and disintegrating @xmath341 and @xmath342 : @xmath349 @xmath350 adding and subtracting @xmath351 the last expression is equivalent to @xmath352this becomes@xmath353@xmath354@xmath355since @xmath338 and @xmath280 ( on the square we consider the sup distance ) , then by adding a constant to @xmath306 ( which does not change @xmath356 ) we can suppose without loss of generality that @xmath357 and then for almost each @xmath37 it holds @xmath358 . hence , by assumption ( 2 ) the statement is proved . now we are ready to prove that a lorenz like two dimensional map @xmath49 has exponential decay of correlations with respect to its srb measure @xmath286 . we recall ( see proposition [ prop : densidadebvdef ] ) that for a piecewise expanding map of the interval @xmath45 , there are constants @xmath216 and @xmath217 , depending on the system such that , if @xmath306 and @xmath329 are respectively @xmath359 and bv ( bounded variation ) observables on @xmath193 for each @xmath218 it holds:@xmath360(recall that @xmath213 is the lebesgue measure above ) . this will be used in the proof of the following theorem [ resuno]let @xmath27 a borel function such that @xmath28 . let @xmath286 be an invariant measure for @xmath49 with marginal @xmath290 on the @xmath1-axis ( which is invariant for @xmath361 ) . let us suppose that 1 . @xmath362 satisfies the above equation [ l1bv ] and @xmath363 is finite for each @xmath364 . @xmath49 is a contraction on each vertical leaf : @xmath365 is @xmath35-lipschitz in @xmath109 with @xmath36 for each @xmath1 . @xmath286 is regular enough that for each @xmath316-lipschitz function @xmath366 the projection @xmath367 has bounded variation density @xmath368 , ] with@xmath369 where @xmath370 is not depending on @xmath329 . then @xmath371 has exponential decay of correlation ( with respect to lipschitz and @xmath215 observables as in equation [ decwass ] ) . we already saw that the first two points in the above proposition are satisfied by the first return map of the geometric lorenz system . in the appendix i we will prove that also the above item 3 is satisfied by the family of systems described in the introduction , containing the geometric lorenz one . we point out that this is the hard part of the proof that lorenz like maps have exponential decay of correlations and this will be done by a sort of lasota - yorke inequality . putting together all the necessary assumptions , this prove theorem a in the introduction . before the proof of theorem [ resuno ] we make the following remark which is a simple but important fact implied by the uniform contraction on stable leaves [ remark]under the above assumptions , let us consider a leaf @xmath37 and two probability measures @xmath286 , @xmath372 on it . then @xmath373 this is because the map is uniformly contracting on each leaf . if @xmath306 is @xmath304-lipschitz on @xmath374 then @xmath375 is @xmath35-lipschitz on @xmath37 . this implies that @xmath376finishing the proof . ( of theorem [ resuno ] ) let us consider @xmath377 with @xmath323 being @xmath378lipschitz and @xmath379 ( remark that this implies @xmath380 ) . the strategy is to use proposition prod and find exponentially decreasing bounds for @xmath381 and @xmath382 so that we can estimate the wasserstein distance between @xmath286 and iterates of @xmath282 and then apply proposition [ prop:1 ] to deduce decay of correlations from the distance . let us consider the leaf @xmath281 with coordinate @xmath1 . the density @xmath368 , by item 3 has bounded variation and @xmath383 . let @xmath384 the measure on the @xmath1-axis with density @xmath368 ( as before @xmath213 is the lebesgue measure ) . let us consider the base map @xmath45 . let @xmath385)$ ] . since @xmath386 , by equation ( [ l1bv])@xmath387implying that @xmath388 and hence we see that item ( 2 ) at proposition [ prod ] is satisfied with an exponential bound depending on the lipschitz constant @xmath316 of @xmath329 . let us consider @xmath389 again . since , as said before the map @xmath49 sends vertical leaves into vertical ones then there is a family of probability measures @xmath390 on vertical leaves such that@xmath391to satisfy item ( 1 ) at proposition [ prod ] and hence conclude the statement we only have to prove that there are @xmath392 s.t.@xmath393 this is because of uniform contraction on stable leaves . indeed , by remark [ remark ] , if @xmath394 and @xmath395 are the two probability measures on the leaf @xmath37 then the measures @xmath396 on the contracting leaf @xmath374 are such that@xmath397now let us consider @xmath398 and apply the above inequality to estimate the distance of iterates of the measure on the leaves . for simplicity let us show the case where the pre - image of a leaf consists of two leaves as it happen in the geometric lorenz system , the case where the pre - image consists of more leaves is analogous : let hence @xmath399 , after one iteration of @xmath320 on @xmath372 and @xmath286 the `` new '' measures @xmath400 and @xmath401 ( which is equal to @xmath402 because @xmath286 is invariant ) on the leaf @xmath37 will be a convex combination of the images of the `` old '' measures on @xmath403 and @xmath404@xmath405@xmath406with @xmath407 ( the second equality is again because @xmath286 is invariant ) . by the triangle inequality ( remark [ conv ] ) @xmath408 and by remark [ remark ] @xmath409hence @xmath410the same can be done in the case when the pre - image @xmath411 is only one leaf or more than two , hence by induction @xmath412 , and the exponential bound on the distance of iterates on the leaves ( item 1 of proposition [ prod ] ) is provided . we now consider again a lorenz like flow , with integrable return time , i.e. a flow @xmath71 having a transversal section @xmath51 whose first return map satisfies the assumptions of theorem [ resuno ] and the return time is integrable , as before . as before @xmath413 is the first return map associated . let @xmath414 and @xmath415 be the time needed for the @xmath416-orbit of a point @xmath1 to enter for the _ first time _ in a ball @xmath14 . the number @xmath417 is the _ hitting time associated to _ the flow @xmath71 and @xmath2 . if @xmath418 and @xmath419 , we define @xmath420 the _ hitting time associated to _ the discrete system @xmath49 . given any @xmath1 we recall that we denoted with @xmath232 the first strictly positive time , such that @xmath233 ( the _ return time _ of @xmath1 to @xmath52 ) . a relation between @xmath421 and @xmath422 is given by [ hittingsecaoelivre ] under the above assumptions , if @xmath423 , then , there is @xmath424 and a set @xmath425 having full @xmath426 measure such that for each @xmath427 , @xmath428 @xmath429with @xmath430 as @xmath431 . let us assume that @xmath432 @xmath433 @xmath51 , @xmath434 and @xmath435 . since the flow can not hit the section near @xmath3 without entering in a small ball of the space centered at @xmath3 before , then @xmath436 and @xmath417 are related by @xmath437 moreover , since the section is supposed to be transversal to the flow , there is a @xmath370 such that @xmath438.\ ] ] the last inequality follows by the fact that if the flow at some time crosses the ball centered at @xmath3 then after a time @xmath439 it will cross the section at a distance less than @xmath440 , where @xmath370 depends on the angle between the flow and the section ( when @xmath4 is small approximate locally the flow by a constant one ) . the above sums are birkhoff sums of the observable @xmath94 on the @xmath49-orbit of @xmath1 and @xmath54 is ergodic . then there is a full measure set @xmath425 ( and @xmath441 ) such that @xmath442for @xmath428 . hence @xmath443 for @xmath428 . thus we get that for each @xmath428 @xmath444with @xmath430 as @xmath431 . combining equations ( [ sum1],[sum2 ] ) and ( [ eq : finale ] ) we get ( [ eq : relationhitting ] ) . let @xmath445 be the projection on @xmath51 defined in proposition [ dimensao ] . the above statement implies the following [ c : relacaodimensao ] there is a full measure set @xmath446 ( for the flow invariant measure ) such that if @xmath447 is regular and @xmath448 it holds ( provided the limits exist ) @xmath449 the above proposition implies that if @xmath450 and @xmath428 then @xmath451 if @xmath452 is a regular point , the flow @xmath416 induces a bilipschitz homeomophism from a neighborhood of @xmath453 to a neighborhood of @xmath13 . hence there is @xmath454 such that @xmath455 where @xmath456 represents the time which is needed to go from @xmath457 to @xmath13 by the flow . this is also true for each @xmath458 . extracting logarithms and taking the limits we get the required result . we recall that ( see section [ sec : sbrfluxo ] ) the assumption @xmath459 is verified for the geometric lorenz flow . hence these results applies for this example . in this section we give the main result for the behavior of the hitting time on lorenz like flows . first let us recall a result on discrete time systems . let @xmath460 be a measure preserving ( discrete time ) dynamical system . we say that @xmath461 has super - polynomial decay of correlations with respect to lipschitz observables if @xmath462 where @xmath463 for all @xmath464 and @xmath465 is the lipschitz norm . in @xcite the following fact is proved for discrete time systems : [ qui ] let @xmath466 a measure preserving transformation having superpolynomial decay of correlations as above . for each @xmath467 such that @xmath468 is defined @xmath469 for @xmath8-almost each @xmath470 . applying this to the 2-dimensional system @xmath471 ( which satisfies the assumptions of theorem [ resuno ] since ans hence has exponential decay of correlations ) . we conclude the following [ th : stefano ] let @xmath472 be a map with an invariant measure @xmath473 satisfying the assumptions of theorem [ resuno ] . for each @xmath474 such that @xmath475 exists then @xmath476 for @xmath223-almost @xmath477 . now , if we consider a flow having such a map as its poincar section and integrable return time , we can construct as in section [ sec : sbrfluxo ] an srb invariant measure @xmath247 for the flow . by proposition [ c : relacaodimensao ] , corollary [ th : stefano ] and proposition [ dimensao ] we can estimate the hitting time to balls for the flow by the corresponding estimation for the poincar map and we get our main result , which corresponds to theorem b in the introduction ( where a set of sufficient assumptions on the map are listed ) : [ main1 ] if @xmath71 is a lorenz like flow , that is a flow having a transversal section , with a poincar map satisfying the assumptions of proposition [ resuno ] and integrable return time , then for each regular @xmath478 such that @xmath479 exists , it holds @xmath480 for @xmath247-almost each @xmath481 [ sec : recurrlorenz ] we now recall a general result proved by saussol in @xcite about quantitative recurrence in order to apply it to a lorenz like flow . the result shows that the power law behavior of the return time in small balls can be estimated by function of the local dimension if the system has fast enough decay of correlations . given a set @xmath287 , we denote the boundary of @xmath287 as @xmath482 . * thm 4 , lemma 13 ) . [ th : saussol ] let @xmath460 be a measure preserving dynamical system , where @xmath295 is a borel subset of some euclidean space . assume that the entropy @xmath483 and @xmath45 is such that there exists a partition @xmath484 ( modulo @xmath286 ) into open sets such that for each @xmath485 the map @xmath486 is lipschitz with constant @xmath487 . furthermore , suppose that 1 . the set @xmath488 is such that there are constants @xmath489 and @xmath490 so that @xmath491 2 . the average lipschitz exponent @xmath492 is finite , 3 . the decay of correlation of @xmath45 is super - polynomial with respect to lipschitz observables . then @xmath493 let us first show that the above theorem can be applied to the geometric lorenz system . the first return map @xmath494 of the geometric lorenz system ( described in section [ sec : lorenzmodel ] ) satisfies the hypothesis of theorem [ th : saussol ] above . since we have proved that the system @xmath495 is exponentially mixing , item ( 3 ) at theorem [ th : saussol ] is satisfied . the partition @xmath496 with @xmath497\times \mathring{i},\ i\in { \mathbb n}^+\]]where @xmath498 denotes the interior of @xmath193 , satisfies ( 1 ) and ( 2 ) at theorem [ th : saussol ] . here we note that @xmath49 is not globally lipschitz , but from eq . ( [ eq : derivadaf ] ) we get @xmath499 , with @xmath500 and @xmath501 . moreover , the fact that @xmath223 has a bounded density marginal ( the density will be denoted by @xmath502 as before ) on the @xmath1 direction implies that the measure of the sets @xmath503 can be estimated by @xmath504 thus , @xmath505 this finishes the proof . in the same way , replacing equation [ eq : derivadaf ] with assumption 3 ) in the introduction it can be proved that the above theorem applies to lorenz like flows : [ l : fsaussol ] if the system @xmath495 is the first return map of a flow satisfying assumptions 1.a), ... 1.d),2),3 ) of the introduction , then it satisfies the hypothesis of theorem [ th : saussol ] . applying theorem [ th : saussol ] to such system , then we get [ co : fsuassol ] for the system @xmath495 it holds @xmath506 finally , remarking that regular points have full measure , with the same arguments as in proposition [ c : relacaodimensao ] by proposition [ dimensao ] , we get [ co : xsaussol ] for the geometric lorenz flow and for lorenz like flows as above it holds @xmath507 where @xmath508 is the recurrence time for the flow , as defined in the introduction . this is the content of theorem c in the introduction . in this section we are going to prove that the srb measure of a lorenz like map satisfies item 3 of theorem [ resuno ] . we remark that this is a kind of regularity assumption for the measure @xmath54 ( a certain projection is bv ) . the proof is done in several steps and it will be completed at the end of the section . the statement we are going to prove is : [ resdue]let @xmath28 be a map preserving the vertical foliation , such that : 1 . there is @xmath29 and @xmath30 such that , if @xmath31 are such that @xmath510 $ ] then @xmath511@xmath512 2 . @xmath34 is @xmath35-lipschitz with @xmath36 ( hence is uniformly contracting ) on each leaf @xmath37 . @xmath39 is onto and , piecewise monotonic , with two @xmath40 increasing branches on the intervals @xmath41,@xmath42 $ ] and @xmath43 where it is defined . 4 . @xmath515 has bounded variation . then @xmath516 has an unique invariant srb measure which satisfies item 3 of theorem [ resuno ] . we recall that the existence and the uniqueness of the srb measure can be obtained by the general arguments explained in section [ sec : sbrfluxo ] . to proceed to prove the above statement , we need to introduce some concepts . to deal with non normalized measures as the measures @xmath517 on the leaves are , we consider the following modification of the wasserstein distance : let @xmath518 be the set of 1-lipschitz functions on @xmath193 having @xmath519 norm less or equal than @xmath304 ( @xmath520 ) . let us consider two finite measures @xmath521 on @xmath193 and the distance@xmath522 [ cont]we remark that choosing @xmath523 we obtain @xmath524 . let us consider the space @xmath525 of borel finite measures over @xmath193 with the distance @xmath526 . given a function @xmath527 we define the variation of @xmath365 as follows : let @xmath528 be an increasing finite sequence in @xmath193 ( which induces a subdivision in small intervals ) let @xmath529 be the set of such subdivisions . we define the variation of @xmath365 as:@xmath530 we will consider the lebesgue measure on the section @xmath51 and its iterates by @xmath49 . the strategy is to disintegrate along stable leaves and estimate the variation of the induced function @xmath531 proving that this is uniformly bounded . let us precise this point : if @xmath286 is a finite measure on @xmath51 , by disintegration this induces a function @xmath532 defined almost everywhere by@xmath533suppose that @xmath534 is defined everywhere . the bv norm of @xmath534 will be an estimation of the regularity of @xmath286 on the @xmath1-axis . for example , the lebesgue measure on the square @xmath51 induces a function @xmath535 which is constant everywhere and its value is the lebesgue measure on the interval . the variation in this case is obviously null . we remark that each iterate of the lebesgue measure by @xmath320 induces a @xmath536 which is defined everywhere ( see eq . [ iter ] ) . we will give an estimation of the variation for these iterates in our system . we say that a probability measure @xmath286 on @xmath51 is @xmath370-good if the function @xmath537 with @xmath538 as above is well defined and s.t . @xmath539 . [ [ some - preliminary - lemmata - and - remarks ] ] some preliminary lemmata and remarks [ 11]we remark that if @xmath286 is @xmath370-good then latexmath:[$\sup_{\gamma } ( \mu since @xmath286 is a probability measure then for some @xmath541 @xmath542 then by remark [ cont ] this would contradict @xmath539 . this elementary remark about real sequences will be used in the following . [ sequ]if a real sequence @xmath543 is such that @xmath544 for some @xmath545 , then@xmath546 if for some @xmath213 , @xmath547 then there is @xmath548 such that @xmath549 . hence , @xmath550 . similarly @xmath551 . the following is analogous to remark [ remark ] for the distance @xmath552 , and also follows by uniform contraction on stable leaves . [ remark2]let @xmath49 be @xmath35 contracting as above . let us consider a leaf @xmath37 and two finite ( non necessarily normalized ) measures @xmath286 , @xmath372 on it . then @xmath553 if @xmath306 is in @xmath554 on @xmath374 then @xmath375 is @xmath35-lipschitz on @xmath37 , moreover since @xmath555 then @xmath556@xmath557@xmath558 now we are ready to prove the main technical lemma estimating the regularity of the iterates @xmath559 . we will explicit the assumptions we need on @xmath49 . [ x1]let @xmath28 be a measurable map preserving the vertical foliation such that : 1 . there is @xmath29 and @xmath30 such that , if @xmath31 are such that @xmath510 $ ] then @xmath560 2 . @xmath34 is @xmath35-lipschitz with @xmath36 uniformly on each vertical leaf @xmath37 . let @xmath403 and @xmath404 two close leaves with @xmath561 and suppose that @xmath562 is defined at the points @xmath563 and @xmath564 and at these points @xmath565 . let @xmath566be a probability measure on @xmath51 such that @xmath567 is defined for each @xmath70 and @xmath568for a bounded density function @xmath569 . then @xmath570 let @xmath571 be the restriction of @xmath572 to the leaf @xmath37 . remark that@xmath573 where @xmath574 is given by @xmath575 and@xmath576 with similar notation for @xmath577 . now the remaining part of the proof is a ( long ) straightforward calculation:@xmath578and@xmath579let us estimate these two terms:@xmath580and similarly@xmath581hence @xmath582 @xmath583 @xmath584 to estimate the last expression by the triangle inequality , let us add and subtract @xmath585obtaining @xmath586where@xmath587and@xmath588**estimation of a. * * now let us estimate @xmath287:@xmath589let us analyze the first term in the sum ( the estimation of the other term is similar)@xmath590@xmath591adding and subtracting @xmath592 we obtain @xmath593@xmath594now , since @xmath569 is bounded @xmath595 and then @xmath596the other summand is @xmath597by assumption ( 1 ) then @xmath598summarizing @xmath599 considering in the same way the summand @xmath600 in the expression of a , this gives @xmath601@xmath602*estimation of b. * the upper bound on b follows by contraction on stable leaves . indeed , @xmath603now , since @xmath49 contracts all the leaves by a factor at least @xmath35 by remark [ remark2 ] it holds@xmath604summarizing , @xmath605 @xmath606finishing the proof . we remark that this last step in the proof ( equations [ above ] and following ) is the only one where the expansivity of @xmath45 is explicitly used . in fact , an equivalent result can be obtained with the weaker assumption @xmath607 , instead of @xmath608 . a similar lemma holds for the case where the pre - image of @xmath403 and @xmath404 is only one leaf . the proof is similar to the previous one . [ x2]let @xmath27 be as above , satisfying points ( 1)(3 ) of lemma [ x1 ] . let @xmath403and @xmath404 be two leaves and suppose that @xmath609 let us consider a probability measure @xmath566 on @xmath51 such that @xmath610 for a bounded function @xmath569 , then@xmath611 the above lemmata give the following result , in the spirit of the lasota yorke inequality ( see in the following proof , eq . [ ly ] and compare with @xcite , @xcite e.g. ) giving an upper bound on the variation of iterates @xmath612 . we recall that by the classical lasota - yorke inequalities , for piecewise expanding maps of the interval , iterating a bounded variation density @xmath613 we get a sequence of uniformly bounded variation densities,@xmath614where @xmath615 depends on @xmath613 and on the dynamics @xmath45 . [ bon]let @xmath27 be as above , satisfying assumptions ( 1)(4 ) of lemma [ resdue ] . let @xmath616 where @xmath566 is @xmath370-good and has bv density on the @xmath1 axis , @xmath569 . then , each @xmath617 is @xmath618-good , where @xmath619 . let us consider a subdivision @xmath620 made of small intervals and set @xmath621 . if we are in the case of lemma [ x1 ] @xmath622 consists of two small intervals , if we are in the case of lemma [ x2 ] @xmath622 consists of one small interval and in the remaining case we have one small interval and an interval of the type @xmath623 or @xmath624 ( this can happen only in two intervals of the subdivision containing the points @xmath625 and @xmath626 ) . the endpoints of all these pre - image intervals @xmath627 constitute another subdivision @xmath628 of @xmath193 . let us estimate the variation of @xmath629 on the subdivision @xmath620 . let us suppose that the intervals of @xmath620 which are of the third type are @xmath630 and @xmath631 . in this case we bound trivially from above the variation : @xmath632 ( for @xmath633 ) . lemma [ x1 ] and lemma [ x2 ] imply@xmath634@xmath635@xmath636hence@xmath637and we conclude that@xmath638if @xmath639are the marginals of @xmath640 on the @xmath1-axis , then as recalled before @xmath641 . this allows to iterate the above inequality and obtain , by lemma [ sequ ] @xmath642(remark that @xmath643 ) finishing the proof . if @xmath286 is a good measure , the measure @xmath282 associated to a lipschitz observable @xmath329 is also a good measure : [ 798]if @xmath617 is a sequence of @xmath644good measures on @xmath51 and @xmath645 with @xmath329 be @xmath316-lipschitz and @xmath646 then each @xmath647 is a @xmath648-good measure . let @xmath403 and @xmath404 be two close leaves@xmath649we recall that @xmath555 , hence@xmath650@xmath651hence @xmath652 [ 123]if @xmath617 is @xmath370-good for each @xmath218 and @xmath653 is the marginal , such that@xmath654since @xmath655 then it holds @xmath656for each @xmath218 . [ 456]if @xmath657 and @xmath645 , @xmath658 with @xmath329 be @xmath316-lipschitz then @xmath659(this is easily obtained because @xmath660 , for each continuous @xmath661 since @xmath662 is continuous ) . we are finally ready to end the proof of the main proposition of the section . ( of lemma [ resdue ] ) we prove that @xmath368 as defined at item 3 of theorem [ resuno ] has bounded variation and @xmath663 , where @xmath316 is the lipschitz constant of @xmath329 and @xmath618 does not depend on @xmath329 . let @xmath664 be the sequence of iterates of the lebesgue measure @xmath213 . by theorem [ bon ] these are @xmath618-good . by proposition [ prod ] , equation [ l1bv ] and uniform contraction on unstable leaves it follows @xmath657 in the weak topology . then for each continuous @xmath661 it holds @xmath665 . in particular this holds for the functions which are constant on each contracting leaf . let @xmath661 be such a function . then @xmath666 where @xmath667 are the densities of @xmath668 on the @xmath1 axis as in remark [ 123 ] . let @xmath329 be @xmath316-lipschitz , @xmath645 and @xmath377 as required by lemma [ resdue ] . since @xmath661 is constant along the leaves , again @xmath669 and @xmath670 where @xmath671 as above . by remark [ 456 ] @xmath672hence @xmath673 we have to prove that @xmath368 is bv . by lemma [ 798 ] the measures @xmath647 are @xmath674-good . now by remark [ 123 ] , @xmath675 . by the helly theorem there is a sub - sequence @xmath676 converging in the @xmath359 norm to some bounded variation function @xmath677 such that @xmath678 . hence @xmath679 for each @xmath680 as above and so @xmath681 for each continuous @xmath661 and then this implies that they coincide a.e .. hence @xmath368 can be supposed to be bv and having @xmath682 . [ sec : muexata ] in several of the above results we used the local dimension of the system at certain points . in this section we recall a result of of steinberger ( @xcite ) about the local dimension of lorenz like systems and prove that for the geometric lorenz system the local dimension is defined at almost every point . let us recall the assumptions used in @xcite . 1 . @xmath685\to [ 0,1]$ ] is piecewise monotonic . this means that there are @xmath686 $ ] for @xmath687 with @xmath688 such that @xmath689 is continuous and monotone for @xmath690 . furthermore , for @xmath691 , @xmath692 is @xmath10 and that @xmath693 holds where @xmath694\setminus \cup_{0\leq i < n}c_i$ ] . 2 . @xmath695 ^ 2\to ( 0,1)$ ] is @xmath10 on @xmath696 $ ] . furthermore , @xmath697 , @xmath698 and @xmath699 for @xmath700 $ ] . @xmath701)\cap f((c_i , c_{i+1})\times [ 0,1])=\emptyset$ ] for distinct @xmath702 with @xmath703 . now consider the projection @xmath704 , set @xmath705 and @xmath706 , which is a partition of @xmath707 into open intervals . for @xmath708 let @xmath709 be the unique element of @xmath710 which contains @xmath1 . we say that @xmath711 _ is a generator _ if the length of the intervals @xmath709 tends to zero for @xmath712 for any given @xmath1 . set @xmath713 ( * ? ? ? * theorem 1 ) [ th : steinberg ] let @xmath49 be a two - dimensional map as above and @xmath8 an ergodic , @xmath49-invariant probability measure on @xmath714 with the entropy @xmath715 . suppose @xmath711 is a generator , @xmath716 and @xmath717 . if the maps @xmath718 are uniformly equicontinuous for @xmath719 and @xmath720 is bv then @xmath721 for @xmath8-almost all @xmath722 . now we verify that the lorenz geometric system as defined in section [ sec : sbrfluxo ] is exact dimensional . first we observe that for the first return map @xmath723 associated to the lorenz geometric flow its entropy @xmath724 , see @xcite . next , equations ( [ gy ] ) , ( [ gx ] ) , and the properties of @xmath182 described in subsections [ asegundacoordenada ] and [ sec : propert - one - dimens ] guaranty that @xmath725 is a two - dimensional transformation satisfying the above points ( 13 ) . so , all we need to prove that @xmath726 is exact dimensional is to verify that @xmath727 satisfies the hypothesis of theorem [ th : steinberg ] . for this , let @xmath728 then the following result holds : _ proof of ( 1 ) : _ by the expression above we have @xmath737 and so @xmath738 does not depend on @xmath109 . since the measure @xmath223 is constant at each leaf @xmath739 and the projection of @xmath223 on the @xmath1-axis , @xmath212 , is absolutely continuous with respect to lebesgue measure ( and even has a finite density ) , see proposition [ prop : densidadebvdef ] , we immediately conclude that @xmath740 proving ( 1 ) . bonatti c. , daz l. j. and viana m. , _ dynamics beyond uniform hyperbolicity : a global geometric and probabilistic perspective _ , encyclopedia of mathematical sciences * 102 * , springer - verlag , berlin , 2005 liverani c. invariant measures and their properties . a functional analytic point of view _ in dynamical systems . part ii : topological geometrical and ergodic properties of dynamics . _ pubblicazioni della classe di scienze , scuola normale superiore , pisa . centro di ricerca matematica ennio de giorgi ( 2004 ) .

we report on changes of the cyclotron resonance energies of the recurrent transient pulsar , x0331 + 53 ( v0332 + 53 ) . the whole _ rxte _ data acquired in the 2004 - 2005 outburst were utilized . the kev source luminosity varied between and erg s , assuming a distance of 7 kpc . we confirmed that the fundamental cyclotron resonance energy changed from to kev in a clear anti - correlation to the source luminosity , and without any hysteresis effects between the rising and declining phases of the outburst . in contrast , the second harmonic energy changed from to kev , implying a weaker fractional change as a function of the luminosity . as a result , the observed resonance energy ratio between the second harmonic and the fundamental was when the source was most luminous , whereas the ratio decreased to the nominal value of 2.0 at the least luminous state . although the significance of this effect is model dependent , these results suggest that the fundamental and second harmonic resonances represent different heights in the accretion column , depending on the mass accretion rate .

1001.1612

magnetic field strengths on the surface of accreting x - ray pulsars can be measured accurately by observing cyclotron resonant scattering feature ( crsf ) , because the fundamental electron cyclotron resonance energy @xmath9 is described as @xmath10 . here , @xmath11 is the magnetic field strength in units of @xmath12 gauss , and @xmath13 is the gravitational redshift . applying this relation to x - ray detected crsfs , the magnetic field strengths of @xmath1415 x - ray pulsars have been accurately measured ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) . the results are clustered in a relatively narrow range of @xmath15 g @xcite . since the cyclotron resonant scattering occurs near the pulsar surface , the resonance energies were believed to be constant and intrinsic to each pulsar . however , an unexpected crsf energy change was found from the recurrent transient pulsar 4u 0115 + 63 with the _ ginga _ observations performed in 1990 and 1991 @xcite ; in the 1991 minor outburst when the luminosity was @xmath16 of that in the 1990 and other typical outbursts of this object , the value of @xmath9 was @xmath17 times higher than the so far reported @xmath9@xmath18 kev . in order to interpret the change in @xmath9 , mihara et al . ( 2004 ) proposed that the height of cyclotron scattering region in the accretion column changes depending on the x - ray luminosity . a further study of the luminosity - dependent crsfs change was conducted by nakajima ( 2006 ) , nakajima et al . ( 2006a , b ) , and tsygankov et al . ( 2007 ) , using _ rossi x - ray timing explorer _ ( _ rxte _ ) data of 4u 0115 + 63 acquired in 1999 and 2004 . according to these results , @xmath9 increased from @xmath19 to @xmath20 kev as the source luminosity decreased across a narrow range of @xmath21 erg s@xmath3 ( at an assumed distance of 7 kpc ; negueruela & okazaki 2001 ) . in addition , the second harmonics observed at @xmath22 kev disappeared as @xmath9 started to change . this luminosity - dependent change in the crsf energy has been found from another source , x0331 + 53 ( v0332 + 53 ) , with _ ginga _ , _ integral _ and _ rxte _ observations ( mihara et al . 1998 ; mowlavi et al . 2006 ; nakajima 2006 ; tsygankov et al . 2006 , 2009 ) . this source is a recurrent transient pulsar , and exhibits a very prominent fundamental crsf at @xmath23 kev of the x - ray spectrum , as first discovered with _ @xcite after a suggestion with _ temma _ @xcite . furthermore , its two higher harmonics have been discovered at energies of @xmath24 and @xmath25 kev @xcite . thus , x0331 + 53 is the second object which has three or more crsfs in the spectrum , following 4u 0115 + 63 @xcite . however , compared with the results from 4u 0115 + 63 , this object exhibited two intriguing differences ( mowlavi et al . 2006 ; nakajima 2006 ; tsygankov et al . 2006 , 2009 ) . one is that the change in @xmath9 started at a higher luminosity , @xmath26 erg s@xmath3 , at an assumed distance of 7 kpc @xcite . the other is that the second harmonic absorption feature persisted in the x - ray spectra through an outburst . including the above cases of 4u 0115 + 63 and x0331 + 53 , the behavior of the fundamental crsf has been extensively studied so far . in contrast , details of the second harmonic , including its possible luminosity dependence , are less understood , primarily because of lower data quality at higher energies . with this in mind , we analyzed the whole _ ( asm , pca and hexte ) data of x0331 + 53 , acquired in a 2004 - 2005 outburst . as a result , we have discovered that the second crsf energy @xmath27 possibly depends more weakly on the luminosity than @xmath9 . the @xmath27/@xmath9 ratio was @xmath8 when the source was luminous , while it approached the nominal value of 2.0 toward lower luminosities . these results suggest that the two resonances take place at different heights in the same accretion columns , and that @xmath27 , formed at a lower height , provides a more reliable estimate of the surface field strength . in order to study the luminosity - dependent changes of @xmath9 and @xmath27 of x0331 + 53 , here we utilized all the _ rxte _ data sets of this transient pulsar acquired in the 2004 - 2005 outburst . figure [ f1]a shows the whole light curve of this outburst acquired with the all sky monitor ( asm ; levine et al . 1996 ) onboard _ rxte_. an abrupt brightening was detected on 2004 november 25 @xcite , and the @xmath28 kev flux continued to increase up to 1.1 crab @xcite . as revealed by the asm , the x - ray intensity declined toward middle of 2005 february , and exhibited a small recovery at the beginning of 2005 march , presumably synchronized with the 34.25 day orbital period ( * ? ? ? * and references therein ) . at the outburst peak , a number of target - of - opportunity observations were performed by _ integral _ , _ swift _ , and _ rxte_. from 2004 november 27 through 2005 march 27 , 108 pointing observations were made with the proportional counter array ( pca ; jahoda et al . 2006 ) and the high energy x - ray timing experiment ( hexte ; rothschild et al . 1998 ) on board _ rxte_. based on the inspection of the operation status of the five proportional counter units ( pcu0@xmath14pcu4 ) , we have selected the data sets acquired with pcu2 , which worked throughout this outburst . as reported by pottschmidt et al . ( 2005 ) , hexte cluster a did not function from 2004 december 13 to 2005 january 14 . thus , we utilize only the data acquired with hexte cluster b. using part of the _ rxte _ data sets that were acquired in the descent phase , the behavior of the fundamental crsf was already studied @xcite . in the present paper , we complement these studies of the descent phase , by analyzing those datasets which were left unused . in addition , we analyze the ascent - phase datasets to examine whether the crsf energy exhibits any hysteresis effects similar to those suggested by the 4u 0115 + 63 data @xcite . during the 4 months of observation , several energetic background events , such as large solar flares and precipitations of high - energy particles trapped in the radiation belts , affected some of the data sets . in order to exclude such low - quality data , we inspected the electron - rate light curves , and have selected 86 data sets with good quality as listed in table [ t1 ] . to analyze them , we utilized the heasoft version 6.0.5 . we assume 1% systematic error for all energy bins of the pca data ( same as in nakajima et al . all of the errors presented in this paper are 90% confidence levels . as shown in figure [ f1 ] , the intensity of x0331 + 53 changed largely in the asm , pca and hexte energy bands during this outburst . in addition , the hardness - ratio exhibited an anti - correlation to the source count rates as shown in figure [ f1]d . in order to investigate the luminosity related crsfs changes through this outburst , we need to establish a unified spectral modeling which can be applied consistently to the data sets at all luminosity levels . so , we first examined pulse - phase - averaged pca and hexte spectra from several representative data sets . since many of the data sets have rather short exposure , we selected four representative ones sets with relatively long exposures , denoted in table [ t1 ] as dec 2a , dec 24 , jan 20 , and feb 13b ; these represent the beginning , the peak , the declining phase , and the end of the outburst , respectively . figure [ f2]a shows the background - subtracted pcu2 ( @xmath29 kev ) and hexte b ( @xmath30 kev ) spectra obtained on these four occasions . the pca background spectra were all estimated with the bright - background model @xcite , while the hexte backgrounds were extracted from the off - source position which is offset from the source by 1.5 degrees @xcite . the four spectra , even in their raw forms , clearly reveal the prominent fundamental crsf at @xmath31 kev , which was observed in the previous ( makishima et al . 1990a , b ) and the present ( kreykenbohm et al . 2005 ; pottschmidt et al . 2005 ; mowlavi et al . 2006 ; nakajima 2006 ; tsygankov et al . 2006 , 2009 ) outburst of this object . furthermore , the brighter two spectra clearly show the second harmonic resonance at @xmath32 kev . from early days ( e.g. white et al . 1983 ) , continuum spectra of accretion powered pulsars were approximated by a power - law modified by exponential cutoff . in this paper , we employ npex ( negative and positive power - law with exponential ) model @xcite , which is described as @xmath33 where @xmath34 is the x - ray energy in units of kev , @xmath35 and @xmath36 are the normalization and photon index of the negative power - law , respectively , @xmath37 is the normalization of the positive power - law , and @xmath38 represents the cutoff - energy in units of kev . this model successfully reproduced the continuum spectra of 4u 0115 + 63 ( mihara et al . 2004 ; nakajima 2006 ; nakajima et al . 2006ab ) at various luminosity levels . as already reported by several authors ( e.g. makishima et al . 1999 ; coburn et al . 2002 ) , the determinations of the cyclotron line parameters are affected by the continuum and the line modeling . employing this model , we can compare the results of this work with the previous ones @xcite . therefore , we have selected this model . we first attempted to fit the 4 spectra with the npex model , with @xmath35 , @xmath37 , @xmath36 , and @xmath38 all left free . since the pca background estimations are only good to a few percent , we further adjusted the background normalizations down to a level of @xmath141% , following the method described in the _ rxte cook book_. however , as expected , the npex model gave acceptable fits to none of the four spectra . as revealed clearly by the data - to - model ratio in figure [ f2]b , all the four spectra exhibit the fundamental crsf strongly at @xmath31 kev . the second harmonic crsf is observed not only in the brighter two spectra , but also in the two fainter ones . in addition , the spectrum on dec 24 exhibits evidence of the third harmonic crsf at @xmath25 kev , as already reported by coburn et al . ( 2005 ) , kreykenbohm et al . ( 2005 ) , and pottschmidt et al . below , we concentrate on the fundamental and the second harmonic . the fluorescent fe k@xmath39 lines at 6.4 kev with a width of @xmath40 kev are also taken into account @xcite . in order to evaluate the fundamental and second crsf parameters , we next introduce a cyclotron absorption model . although some authors used gaussian - shaped absorption cross section @xcite to reproduce the crsf , we employ the cyclotron absorption ( cyab ) factor which has been used successfully ( clark et al . 1990 ; makishima et al . 1990a , 1999 ; mihara 1995 ; mihara et al . 1998 ; nakajima 2006 ; nakajima et al . 2006ab ; tsygankov et al . 2006 , 2007 , 2009 ) . the cyab factor is described as @xmath41 where @xmath42 is the resonance energy , @xmath43 is the resonance width , @xmath44 is the resonance depth , and @xmath45 is the harmonic number ( with @xmath46 fundamental ) . using high - quality _ suzaku _ data , enoto et al . ( 2008 ) confirmed that eq.[e2 ] can successfully reproduce the 36 kev crsf of the x - ray pulsar her x-1 , whereas the gaussian - shaped absorption cross - section is less successful . since we already know that x0331 + 53 has the multiple crsfs , we attempted to fit the data with an npex model multiplied by two cyab factors ( hereafter npex@xmath47cyab2 model ) . the values of @xmath9 and @xmath27 were both left to vary independently , rather than constrained as @xmath27@xmath48@xmath9 . as a result , the fits were much improved compared to the npex fit , as shown by the data - to - model ratios in figure [ f2]c . due to the complex shape of the fundamental crsfs , however , the npex@xmath47cyab2 model is not yet fully successful , with @xmath49 . this fact was already reported from previous studies @xcite . in order to better describe the fundamental crsf , kreykenbohm et al . ( 2005 ) and pottschmidt et al . ( 2005 ) introduced an additional gaussian absorption ( gabs ) factor with a similar energy but a different width , thus constructing a nested double - gaussian absorption profile . therefore , let us try to explain away the fit residuals employing an additional gabs factor , which is described as @xmath50 } , \label{e3}\ ] ] where @xmath51 is the line - center energy , @xmath52 is the line width , and @xmath53 is the optical depth at the line center . although a more sophisticated crsf modeling has been introduced by schnherr et al . ( 2007 ) , we employ the cyab and gabs models in the present work , so that the results can be directly compared with previous studies of the luminosity dependence of crsfs ( makishima et al . 1999 ; mihara et al . 2004 ; nakajima et al . 2006 ) . we fitted the four spectra by the npex@xmath47cyab2@xmath47gabs model , namely the same npex@xmath47cyab2 model but further multiplied by eq.[e3 ] , in which the three gabs parameters ( @xmath51 , @xmath52 , @xmath53 ) were left free but @xmath51 was given an initial value close to @xmath9 . due to low statistics in the higher energies , we fixed @xmath27 at 2@xmath9 in fitting the dec 2a and feb 13b data , while we left @xmath9 and @xmath27 both free and independent for the brighter two data sets . then , as illustrated in figure [ f2]d , the four spectra have been described successfully with this model . the derived best - fit parameters are summarized in table [ t2 ] . to see the configuration of the nested cyclotron models for the fundamental resonance , the spectrum observed on dec 24 is presented in figure [ f3 ] in its @xmath54 form . thus , the cyab factor explains the gross shape of the fundamental crsf , while the remaining narrow core is represented by the additional gabs factor . the value of @xmath51 agrees , within a few percent , with @xmath55 , or @xmath56@xmath9 in the present case , where the cyab factor becomes deepest @xcite . for comparison , kreykenbohm et al . ( 2005 ) and pottschmidt et al . ( 2005 ) applied nested two gabs factors to the fundamental crsf and obtained two gaussian centroids which differs by @xmath57% . since the continuum and the crsf factor couple strongly , the 90%-confidence errors of individual fit parameters , given in table 2 , might be significantly underestimated . to examine this concern , we present in figure [ f4 ] confidence contours between several pairs of the model parameters , obtained from the dec 24 data . thus , @xmath38 and the resonance energies are almost uncorrelated . although the npex @xmath36 and the cyab @xmath9 exhibit some correlations , the single - parameter 90% error ranges are confirmed to adequately represent the two - dimensional confidence ranges , without being under - estimated . from these spectral analyses , we have established the unified model , which can reproduce the complex shape of the x0331 + 53 spectra regardless of the source luminosity . hereafter , we use this model , namely , npex@xmath47cyab2@xmath47gabs . although our analysis was performed with the heasoft version 6.0.5 , the pca response generator has been updated in the latest heasoft version 6.8 , and a new pca caldb has been released on 2009 december 2 . to examine possible effects of these software and calibration updates , we analyzed the dec 24 data with the heasoft version 6.8 and the latest pca caldb . as a consequence , the cyclotron resonance energies changed only @xmath141.0% , which is within statistical errors . thus , we retain our results obtained with the heasoft version 6.0.5 . just for comparison , we attempted to replace the cyab modeling of the two crsfs with that employing three gabs factors , with nested two for the fundamental and the other for the second harmonic . the three parameters of the three gabs factors were all left free . as a result , the model gave , for example , an acceptable fit ( @xmath58 ) to the dec 24 data . the obtained crsf parameters are , @xmath59 , @xmath60 , @xmath61 , @xmath62 , @xmath63 , @xmath64 , @xmath65 , @xmath66 , and @xmath67 . here , the subscript of @xmath68 and @xmath69 specify the parameters of the nested two gabs model for the fundamental resonance . however , we retain our original modeling using the cyab model , because it gives generally better fits : for example , the difference in @xmath70 is 28 ( for @xmath71 ) in the case of the dec 24 spectrum . we applied the model established in [ sec3.1 ] to all the daily - averaged spectra , and studied the time evolution of the crsf parameters . since some data sets have insufficient statistics , the gabs model parameters did not converge . following the obtained results in the previous subsection , we therefore fixed @xmath72 in all cases . the other procedures of the spectral analyses are the same as in [ sec3.1 ] . as a result of this analysis , our model has given acceptable fits to all the data acquired in the major outburst spanning 2004 november through 2005 february . while the normalization parameters of the npex model varied in correlation with the luminosity , the npex @xmath38 , which is thought to give a measure of the electron temperature in the emission region @xcite , stayed rather constant like in the previous results on 4u 0115 + 63 @xcite . although the fundamental crsf parameters are well determined , the second crsf parameters in some data sets became unconstrained due to insufficient data statistics in higher energies . in such cases , we fixed @xmath27 at 2@xmath9 . in contrast to the november - february outburst data , the remaining three data sets were observed in march during the minor outburst recovery ( figure [ f1 ] ) . since the source was rather faint on these occasions , the three spectra were reproduced with the npex model multiplied by a single cyab factor describing the fundamental crsf . therefore , we do not discuss these data sets acquired in 2005 march . figure [ f5]a shows the derived values of @xmath9 , together with 90%-confidence errors , as a function of the @xmath0 kev source luminosity @xmath73 . as already reported by mowlavi et al . ( 2006 ) , nakajima ( 2006 ) , and tsygankov et al . ( 2006 , 2009 ) , @xmath9 thus changed from @xmath4 to @xmath5 kev as @xmath73 varied between @xmath74 and @xmath2 erg s@xmath3 . when the data points in [ f5]a are fitted by a linear function of @xmath73 having a form of @xmath75 its two parameters ( with 90% confidence errors ) were obtained as @xmath76 kev and @xmath77 erg s@xmath3 , with @xmath78 ( @xmath79 ) . in addition , we found no hysteresis effects in the variation of @xmath9 between the ascent and descent phases . this inference is consistent with a recent report by tsygankov et al . ( 2009 ) , who used the same data set . although the values of @xmath9 derived here are slightly discrepant with those measured from the same outburst by other authors @xcite , the difference is within @xmath19% , and can be attributed to the different modelings of the fundamental crsf and the continuum . as shown in figure [ f5]b , the second resonance energy @xmath27 , when determined independently , showed a considerably weaker dependence on @xmath73 . when these data points are fitted by another linear function of @xmath73 as @xmath80 we obtain @xmath81 kev and @xmath82 erg s@xmath3 , together with @xmath83 ( @xmath84 ) . thus , @xmath85 is consistent with @xmath86 , while the luminosity dependence differs significantly between @xmath9 and @xmath27 ( i.e. , @xmath87 beyond their errors ) . as a result of the above two correlations , the @xmath27/@xmath9 ratio , shown in figure [ f5]c , increases from @xmath88 to @xmath8 as @xmath89 changes from @xmath90 erg s@xmath3 to @xmath2 erg s@xmath3 . since these ratios give @xmath91 ( @xmath92 ) when fitted with a constant value , they are inconsistent with being constant . in contrast , the fit became satisfactory with @xmath93 ( @xmath84 ) , when we employ yet another linear function as @xmath94 the two parameters were obtained as @xmath95 and @xmath96 erg s@xmath3 . thus , the value of @xmath97 is consistent with 2.0 , while @xmath98 remains finite . in other words , the @xmath27/@xmath9 ratios exhibit a statistically significant dependence on the 380 kev luminosity . in 3.3 and 4.2 , we discuss possible artifacts on these results introduced by our choice of the crsf modeling . in order to investigate the behavior of the second crsf under better statistics , we next carried out the flux - sorted analysis which was already performed successfully on 4u 0115 + 63 ( nakajima 2006 ; nakajima et al . 2006a , b ) . specifically , we sorted the pcu2 and hexte cluster b data into 8 flux levels in reference to figure [ f6 ] , and co - added those data which fall in the same flux range . the flux was calculated every 64 sec , so that the flux sorting can catch up with short - term intrinsic variations , but not affected by photon - counting statistics . in addition , the whole data were divided into outburst ascent and descent phases ( referring to figure [ f6 ] ) , to examine the crsf energy changes for possible hysteresis effects . the procedure of the spectral analysis is the same as in [ sec3.2 ] , but @xmath51 was left free , because of the improved statistics . figure [ f7]a and [ f8]a show the flux - sorted spectra of the ascent- and descent - phases , respectively . we again applied the npex@xmath47cyab2@xmath47gabs model to these data , and found that the model gives acceptable fits as shown in figure [ f7]b and [ f8]b . as summarized in table [ t3 ] , this analysis allowed us to accurately determine the parameters of the fundamental and second crsfs . figure [ f9]a and b show the values of @xmath9 and @xmath27 , respectively , against @xmath89 . thus , @xmath9 changed from @xmath4 to @xmath5 kev , while @xmath27 from @xmath6 to @xmath7 kev , thus reconfirming the essential features of figure [ f5 ] with higher confidence . in fact , like in the case of the date - sorted spectra ( 3.2 ) , the fit with eq.[e4 ] to figure [ f9]a gave @xmath99 kev , @xmath100 erg s@xmath3 , and @xmath101 ( @xmath102 ) , while that with eq.[e5 ] to figure [ f9]b resulted in @xmath103 kev , @xmath104 erg s@xmath3 , and @xmath105 ( @xmath102 ) . the errors associated with these fit parameters are larger than those obtained in the date - sorted analysis , because here we left @xmath51 free while we previously fixed it at 1.2@xmath9 . figure [ f9]c reconfirms that the resonance energy ratio @xmath27/@xmath9 changed from @xmath8 to the nominal value @xmath88 as the source luminosity decreased . the fit with eq.[e6 ] indicates @xmath106 , @xmath107 erg s@xmath3 , and @xmath108 ( @xmath102 ) . through the date - sorted and flux - sorted analyses , we confirmed that the luminosity - dependent changes in @xmath27 , @xmath9 and the @xmath27/@xmath9 ratio are statistically significant . then , does this conclusion remain unaffected even if considering systematic errors ? an immediate concern , namely spurious effects due to coupling between the continuum and crsf parameters , can be ruled out because the error estimates are appropriate as indicated by figure [ f4 ] . another problem is that the values of @xmath9 and @xmath27 do not pick up the deepest position of the crsfs . accordingly , we examined the behavior of the energy @xmath109 ( @xmath110 ) at which the absorption is expected to become deepest @xcite . figure [ f10 ] shows luminosity dependence of @xmath111 . thus , the values of @xmath111 also exhibit a clear luminosity dependence . by fitting these data points with eq.[e4 ] and eq.[e5 ] , we obtained the characteristic luminosities as @xmath112 erg s@xmath3 and @xmath113 erg s@xmath3 . we hence reconfirm that the second resonance energy depends less steeply on the luminosity than the fundamental energy , although the difference between @xmath114 and @xmath115 is less significant than that between @xmath116 and @xmath117 , and the average @xmath118 ratio , 1.82.0 is smaller than 2.0 in contrast to the case of the @xmath27/@xmath9 ratios . the largest systematic effect that may possibly affect the @xmath27/@xmath9 ratio would be the choice of the continuum and crsf models . since we have already tried in [ sec3.1 ] replacing the crsf model with the gabs model , here we examine the choice of the continuum . in order not to be affected by the crsf modeling , we fitted the lowest - energy ( 3 - 13 kev ) and highest - energy ( 6080 kev ) ends of the flux - sorted fd1 and fd6 spectra , using a single npex continuum , and then normalized the overall spectrum to the best - fit model that was determined by each spectrum . the results , shown in figure [ f11]a , reveal the absorption features in a relatively model - free manner , with the continuum approximately removed . there , we have adjusted the energy scale and the ratio scale between the two spectra , so that the fundamental crsfs best overlap . thus , the fd1 ratio spectrum exhibits the 2nd harmonic at a significantly higher energy than the fd6 ratio . this result visualize the luminosity - dependent change in the @xmath27/@xmath9 ratio . to examine the overall behavior of @xmath27 , @xmath9 , and their ratios under a different continuum model , we employed a power - law cutoff ( plcut ) model ( coburn et al . 2002 ) , namely a power - law multiplied by an exponential cutoff factor . figure [ f11]b represents the same analysis as figure [ f11]a conducted with this continuum model . thus , the result again visualize that fd1 has a higher @xmath27/@xmath9 ratio than fd6 . we further fitted the fd1 and fd6 spectra by multiplying these plcut continua with the same absorption factors as used so far , and obtained the results summarized in table [ t4 ] . thus , the fit goodness is comparable to those obtained with the npex continuum , and the values of @xmath9 and @xmath27 remain unchanged within errors , although the luminosity - dependent change in the @xmath27/@xmath9 ratio became less significant . as our further confirmation of the luminosity - dependent changes in the crsf energies and their ratios , we fitted the fd1 and fd6 spectra simultaneously , with the npex@xmath47cyab2 model . while the model parameters were generally allowed to differ between the two spectra , we imposed 4 stepwise constraints on the resonance energies . in the first step , we required @xmath9 to be the same between fd1 and fd6 , and constrained @xmath27 to be twice @xmath9 . however , the fit was not acceptable , with @xmath119 . next , we allowed fd1 and fd6 to have different values of @xmath9 , but retained the 1:2 harmonic constraint on both spectra . this gave an improved fit with @xmath120 , implying that the two spectra have different resonance energies . third , we allowed the two spectra to have different values of @xmath9 , but required them to have the same @xmath27/@xmath9 ratio . the fit was then improved to @xmath121 , together with @xmath27/@xmath9@xmath122 which deviate from the the nominal value 2.0 . finally , we left all the four resonance energies free , and obtained a improved fit with @xmath123 . this means that the @xmath27/@xmath9 ratio is different not only from the harmonic condition , but also between the two spectra . ( the fit is still unacceptable because we did not include the gabs factor to model the narrow core of the fundamental resonance . ) finally , we repeated the same 4-step analyses , but using the plcut continuum . then , the fit goodness from the 4 steps were @xmath124 , @xmath125 , @xmath126 , and @xmath127 . the implication is essentially the same as that derived with the npex continuum . since the flux - sorted spectra have rather good statistics , we examined them , up to 110 kev , for evidence of the third harmonic crsf which was already reported by kreykenbohm et al . ( 2005 ) and pottschmidt et al . ( 2005 ) . for this purpose , we further multiplied the npex@xmath47cyab2@xmath47gabs model by an additional cyab factor centered at @xmath25 kev , and applied it to the flux sorted spectra . then , two data sets , fa1 and fd2 , gave statistical significant ( at @xmath1490% ) evidence for the third crsf , while the fundamental and second - harmonic crsf parameters remained essentially the same as before . as a representative case , the third resonance parameters derived from the fa1 data are @xmath128@xmath129 kev , @xmath130 kev , and @xmath131 . these parameters are comparable to the previous results derived from the same _ rxte _ data @xcite . the third resonance energy was found to change no less than by @xmath1323% , as the @xmath0 kev luminosity varied from @xmath133 to @xmath134 erg s@xmath3 . over this luminosity range , the @xmath128/@xmath27 ratio stayed at an average value of 1.46 within a typical uncertainty of @xmath135 . in contrast , the @xmath128/@xmath9 ratio is systematically higher , 3.2 . therefore , @xmath128 and @xmath27 are consistent with being in the 2:3 harmonic ratio , while @xmath9 is inferred to be systematically lower , and depend significantly on the luminosity . in addition to the investigation of the resonance energy changes , we examined whether the depths of the cyclotron resonance , @xmath136 ( @xmath110 ) and @xmath53 , depend on the luminosity . figure [ f12 ] shows these depth parameters , against the @xmath0 kev luminosity . both the fundamental and second crsf depths , @xmath137 and @xmath138 respectively were found to decrease toward higher luminosities . although @xmath53 has large errors , it also exhibits the same tendency . these luminosity dependence in @xmath137 and @xmath138 are statistically significant , because fitting the @xmath137 and @xmath138 measurements with constant values gave reduced chi - square of 2.11 ( @xmath139 ) , 5.88 ( @xmath139 ) , respectively . we analyzed the whole set of _ rxte _ pca and hexte data of x0331 + 53 , covering the 2004@xmath1402005 outburst , with two objectives in mind . one is to examine whether the luminosity - anticorrelated changes in the fundamental crsf energy , observed during the outburst descent phase ( mowlavi et al . 2006 ; nakajima 2006 ; tsygankov et al . 2006 ) , is also present during the outburst rising phase . this agrees with the conclusion of tsygankov et al . the other is to examine the reported second harmonic resonance of this pulsar ( coburn et al . 2005 ; kreykenbohm et al . 2005 ; pottschmidt et al . 2005 ) for possible luminosity - dependent changes , and compare the results with those of the fundamental crsf . following our experience ( e.g. , mihara 1995 ; makishima et al . 1999 ) , we adopted the npex model of eq.[e1 ] to represent the underlying 3@xmath14080 kev pca and hexte continua , while two cyab factors , eq.[e2 ] , to account for the fundamental and second harmonic cyclotron features . furthermore , to express the complex shape of the deep fundamental cyclotron feature as revealed in the previous studies @xcite , we introduced a fine - tuning gabs factor , eq.[e3 ] , which is nested with the fundamental cyab factor . this npex@xmath47cyab2@xmath47gabs model has given acceptable fits to all the daily - averaged and flux - sorted spectra . furthermore , some of the spectra exhibited statistically significant evidence for the third harmonic resonance , as already reported by coburn et al . ( 2005 ) , kreykenbohm et al . ( 2005 ) , and pottschmidt et al . ( 2005 ) . as exemplified in figure [ f12 ] and table [ t3 ] , our fits generally imply that the fundamental and second harmonic resonances both have optical depths of @xmath141 , with comparable cyab widths of @xmath142 kev . the fundamental crsf is explained mainly by the cyab model , with the additional gabs factor carrying @xmath143 of the optical depth when the source is luminous . replacing the two cyab models with two gabs models made the fits significantly worse ( e.g. @xmath144@xmath145 in fa2 spectrum fitting ) , mainly because the observed flux falls considerably below the npex continuum , even at the @xmath146 kev range which is in between the two cyclotron troughs ( see figure [ f3 ] ) . this effect can not be reproduced adequately by the gabs model which is symmetric between the red and blue sides , while it can be successfully accounted for by the cyab factor of which the wing is more extended in the blue side due to the @xmath147 factor in eq.[e2 ] . by applying the npex@xmath47cyab2@xmath47gabs model to the daily - averaged and flux - sorted pca@xmath148hexte spectra , we reconfirmed the reports by mowlavi et al . ( 2006 ) , nakajima ( 2006 ) , and tsygankov et al . ( 2006 , 2009 ) , that the fundamental crsf energy @xmath9 decreases as the source gets more luminous ( figure [ f9 ] ) . thus , x0331 + 53 becomes a second binary x - ray pulsar , after 4u 0115 + 63 , of which @xmath9 correlates negatively with the source luminosity . like the case of 4u 0115 + 63 , this effect observed from x0331 + 53 can be explained by presuming that the cyclotron resonance `` photosphere '' gets higher in the accretion column , as the source luminosity increases and hence the column becomes taller @xcite . these cases make a contrast to the behavior of her x-1 , in which @xmath9 is reported to have been varying in a positive correlation with the luminosity @xcite , at least while the source is in a sub - eddington regime . on the other hand , the recent works of her x-1 @xcite reported that the anti - correlation behavior might exist when the source was in the super - eddington regime . in 4u 0115 + 63 , the luminosity dependence of @xmath9 showed some hysteresis effects between the ascent and descent phases of the outburst ( nakajima et al . in contrast , we found , in agreement with tsygankov et al . ( 2009 ) , no such effects from x0331 + 53 ; in this source , the value of @xmath9 can be considered as a single - valued function of the source luminosity . as the most important result of the present work , our npex@xmath47cyab2@xmath47gabs modeling leads to an inference that the second resonance energy @xmath27 also correlates negatively with the source luminosity , but significantly more weakly than @xmath9 does . the two resonance energies decrease by @xmath149 and @xmath150 , as the @xmath0 kev luminosity increases from @xmath90 to @xmath2 erg s@xmath3 . consequently , the @xmath27/@xmath9 ratio is consistent with the nominal value of 2.0 when the source is dim , while it increases to @xmath8 toward the outburst peak ( figure [ f9]c ) . these effects are somewhat model dependent , and become less significant when the absorption profile and/or continuum shape are modeled in different ways . although these effects were already noticed by tsygankov et al . ( 2006 ) , their spectra had rather low statistics in higher energies , and hence the second crsf parameters were not well constrained . while the relativistic effects @xcite predicts the @xmath27/@xmath9 ratios to be lower than the harmonic value of 2 , actual measurements do not necessarily agree . the ratio exceeds 2.0 in some cases , and fall in the range @xmath151 ; for example , 4u 1907 + 09 showed @xmath9 @xmath152 kev and @xmath27 @xmath153 kev @xcite ; vela x-1 showed @xmath9 @xmath154 kev and @xmath27 @xmath155 kev @xcite ; and a0535 + 26 @xmath9 @xmath24 kev and @xmath27 @xmath156 kev @xcite . in contrast , the ratios lower than 2.0 are reported in other cases , including the present _ rxte _ data of x0331 + 53 themselves analyzed by some other authors ( e.g. , pottschmidt et al . 2006 ) who generally took the deepest positions of the crsfs as the resonance energies ( [ sec3.3 ] ) . a dominant origin of the reported scatter in the @xmath27/@xmath9 ratio is presumably the use of different ( e.g. , cyab vs. gabs ) modeling of the crsf profile . in this sense , our result of @xmath27/@xmath9@xmath157 is not free from this systematic problem . in fact , we find @xmath27/@xmath9@xmath1582 if instead using @xmath159/@xmath160 . nevertheless , our result has some pieces of supporting evidence . one is that the cyab modeling is more successful than that with the gabs factor ( [ sec3.1 ] ) , when combined with the npex continuum . another is that the third to second resonance energy ratio , @xmath128/@xmath27 , is consistent with 1.5 , while @xmath128/@xmath9 is higher than the nominal value of 3.0 . yet another support is provided by a physically self - consistent interpretation to be described in the next subsection . even admitting that the absolute values of @xmath9 and @xmath27 might be subject to the systematic modeling uncertainly , the luminosity - dependent changes in the resonance energy ratio is considered to be robust as already shown in [ sec3.3 ] . let us consider a possible explanation of the luminosity - dependent changes in @xmath9 , @xmath27 , and their ratios . according to mihara et al . ( 2004 ) and nakajima et al . ( 2006a ) , the observed changes in @xmath9 of 4u 0115 + 63 can be explained in terms of those in the height of cyclotron resonance `` photosphere '' , in combination with a dipole field geometry . in this scenario , the photosphere height above the neutron star surface , @xmath161 , can be estimated as @xmath162 where @xmath163 is the neutron star radius , @xmath164 is the resonance energy to be observed on the neutron - star surface , and @xmath45 is the harmonic number . substituting the @xmath9 and @xmath27 values derived from the flux - sorted analysis ( [ sec3.3 ] ) into eq.[e7 ] , @xmath161 of the fundamental and second resonances have been estimated as shown in figure [ f13 ] . here , we employed @xmath165 kev and @xmath166 kev , referring to the lowest - luminosity state in figure [ f9 ] . also a typical value of @xmath167 km was employed . by translating the measurements into the photospheric heights , the figure thus yields two important implications . one is that the photosphere of the fundamental crsf gets higher as the source luminosity increases , up to 800 m , or @xmath142% of @xmath163 . the other is that the second resonance photosphere , though increasing with the luminosity as well , is located closer to the surface than that of the fundamental , just reflecting the deviation of the @xmath27/@xmath9 ratio from 2.0 . the height difference between the two photospheres reach @xmath14500 m when the source is most luminous . like the estimation of @xmath161 described above , we also calculated the resonance heights using the asymptotic energies @xmath168 which are derived from @xmath169 . assume @xmath170 kev and @xmath171 kev , the luminosity - dependent height variations turned out to be @xmath172 m for @xmath160 and @xmath173 m for @xmath159 . the altitude difference between the two photospheres can be explained qualitatively in the following manner . theoretically ( * ? ? ? * ; * ? ? ? * and references therein ) , the fundamental crsf is predicted to have a cross section which is @xmath19 times larger than that of the second resonance . then , we expect the fundamental photosphere to be formed nearly at the top of the accretion column . in contrast , the second harmonic photosphere will appear at lower altitudes , where the column density integrated along our line of sight is expected to become higher . these effects have already predicted theoretically by nishimura ( 2008 ) . then , a detailed modeling of the @xmath27/@xmath9 ratio may provide a valuable probe into the density distribution along the accretion column , and its dependence on the mass accretion rate , although such a work is beyond the scope of the present paper . assuming that the luminosity - dependent changes in the resonance energies are caused by the photospheric height variations , and not , e.g. , by relativistic effects ( [ sec4.2 ] ) , the above physical picture further gives an _ a posteriori _ support to our finding of @xmath27/@xmath9@xmath174 . in fact , if this ratio fell below 2.0 toward lower luminosities , we would have to conclude that the 2nd resonance photosphere is located higher than that of the fundamental , leading to a physically unrealistic condition . following our previous results on 4u 0115 + 63 @xcite , the present work provides valuable information as to the luminosity dependent changes of the crsf parameters in accreting x - ray pulsars . although the fundamental crsf of x0331 + 53 required the additional gabs factor ( which was not needed in 4u 0115 + 63 ) , the cyab and gabs centroids varied similarly as shown in figure [ f9 ] and [ f12 ] . we therefore consider that this slight difference in the crsf modeling does not hamper a direct comparison of the results from the two sources . let us fist compare the luminosity dependence of @xmath9 in the two objects . as shown in figure [ f14]a , the resonance energy of x0331 + 53 depends on the luminosity much less steeply than that of 4u 0115 + 63 , even though the sense of dependence is the same between them . specifically , the dependence is expressed as @xmath175 in x0331 + 53 , while @xmath176 in 4u 0115 + 63 . assuming that the change in @xmath9 is caused by variations in the height of cyclotron resonance `` photosphere '' , the above difference suggests that the accretion column height of 4u 0115 + 63 responds more sensitively to changes in the mass accretion rate . this inference is reinforced by figure [ f14]b , which directly relates the estimated photosphere heights with the luminosity . this difference , in turn , may be attributed to differences in the accretion column shape . indeed , as shown in figure [ f15 ] , the resonance depths @xmath177 of x0331 + 53 depend negatively on the resonance width @xmath178 , while the correlation is opposite in 4u 0115 + 63 . according to previous studies ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) , this is considered to indicate that the accretion column in 4u 0115 + 63 has a tall cylindrical shape , whereas that in x0331 + 53 a flat coin - like shape . the two sources are inferred to differ not only in the shape of the accretion column , but also in its area @xmath179 . we expect the observed luminosity to be expressed as @xmath180 @xcite , where @xmath181 is the stefan - boltzmann constant , and @xmath182 is the accretion column temperature which can be approximated by the observed npex @xmath38 @xcite . by substituting the derived parameters into this relation , we actually find x0331 + 53 to have a 1.43 times larger @xmath179 than 4u 0115 + 63 . the relatively shallow spin modulation in x0331 + 53 @xcite supports this idea . therefore , the accretion column in x0331 + 53 is considered to have a coin - like shape and a rather large area , so that its height depends only weakly on the mass accretion rate . let us finally discuss the behavior of the resonance depths . as already mentioned in the end of section [ sec3.3 ] , @xmath137 , @xmath138 , and @xmath183 of x0331 + 53 have all been found to decrease toward higher luminosity . this change is in the same sense as those in 4u 0115 + 63 @xcite and gx301 - 2 @xcite . one possible mechanism causing such a dependence may be related to the appearance of higher harmonic resonances . as argued by several authors @xcite , an increased luminosity may give rise to significant higher harmonic resonances , which tend to make the fundamental resonance shallower through so - called `` two - photon '' effects @xcite . in fact , the third crsf of x0331 + 53 was confirmed to appear when the source becomes luminous . in summary , we have quantified luminosity dependence of the fundamental and second resonance energies , and discovered clear luminosity - dependent changes in their ratio . in addition , we have discovered that the behavior of x0331 + 53 is qualitatively similar to that of 4u 0115 + 53 , but differs quantitatively . bildsten , l. , chakrabarty , d. , chiu , j. , finger , m. h. , koh , d. t. , nelson , r. w. , prince , t. a. , rubin , b. c. , scott , d. m. , stollberg , m. , vaughan , b. a. , wilson , c. a. , & wilson , r. b. 1997 , apjs , 113 , 367 kretschmar , p. , pottschmidt , k. , ferrigno , c. , kreykenbohm , i. , domingo , a. , wilms , j. , rothschild , r. , coburn , w. , kendziorra , e. , staubert , r. , schnherr , g. , santangelo , a. , & segreto , a. , 2006 , proceedings of the `` the x - ray universe 2005 '' , 273 makishima , k. , mihara , t. , ishida , m. , ohashi , t. , sakao , t. , tashiro , m. , tsuru , t. , kii , t. , makino , f. , murakami , t. , nagase , f. , tanaka , y. , kunieda , h. , tawara , y. , kitamoto , s. , miyamoto , s. , yoshida , a. , & turner , m. j. l. 1990a , , 365 , 59 makishima , k. , ohashi , t. , kawai , n. , matsuoka , m. , koyama , k. , kunieda , h. , tawara , y. , ushimaru , n. , corbet , r. h. d. , inoue , h. , kii , t. , makino , f. , mitsuda , k. , murakami , t. , nagase , f. , ogawara , y. , tanaka , y. , kitamoto , s. , miyamoto , s. , tsunemi , h. , & yamashita , k. 1990b , pasj , 42 , 295 l ccc cc l ccc nov 27 & 14:06/14:13 & @xmath184&@xmath185 & & & jan 11d & 14:48/15:52 & @xmath186&@xmath187 + dec 1 & 06:54/07:41 & @xmath188&@xmath189 & & & jan 14 & 08:43/15:59 & @xmath190&@xmath191 + dec 2a & 03:20/04:12 & @xmath192&@xmath193 & & & jan 15a & 00:28/04:33 & @xmath194&@xmath195 + dec 2b & 19:16/20:11 & @xmath196&@xmath197 & & & jan 15b & 08:19/15:14 & @xmath198&@xmath199 + dec 3 & 18:52/00:25 & @xmath200&@xmath201 & & & jan 15c & 20:57/21:17 & @xmath202&@xmath203 + dec 4a & 01:00/01:30 & @xmath204&@xmath205 & & & jan 15d & 22:28/03:59 & @xmath206&@xmath207 + dec 4b & 02:34/03:14 & @xmath208&@xmath209 & & & jan 16a & 09:30/14:29 & @xmath210&@xmath211 + dec 4c & 05:52/06:25 & @xmath212&@xmath213 & & & jan 16b & 20:33/02:12 & @xmath214&@xmath215 + dec 4d & 09:15/09:39 & @xmath216&@xmath217 & & & jan 17a & 02:49/03:36 & @xmath218&@xmath219 + dec 4e & 12:00/18:02 & @xmath220&@xmath221 & & & jan 17b & 07:31/13:00 & @xmath222&@xmath223 + dec 4f & 18:21/00:00 & @xmath224&@xmath225 & & & jan 17c & 20:10/00:13 & @xmath226&@xmath227 + dec 5a & 00:36/01:08 & @xmath228&@xmath229 & & & jan 18a & 00:50/01:55 & @xmath230&@xmath231 + dec 5b & 13:11/14:16 & @xmath232&@xmath233 & & & jan 18b & 07:08/12:58 & @xmath234&@xmath235 + dec 5c & 17:55/19:15 & @xmath236&@xmath237 & & & jan 18c & 21:17/01:24 & @xmath238&@xmath239 + dec 6 & 17:36/18:51 & @xmath240&@xmath241 & & & jan 19a & 02:01/02:46 & @xmath242&@xmath243 + dec 7 & 17:05/17:51 & @xmath244&@xmath245 & & & jan 19b & 03:39/08:41 & @xmath246&@xmath247 + dec 13 & 18:30/19:11 & @xmath248&@xmath249 & & & jan 20 & 07:53/11:30 & @xmath250&@xmath251 + dec 14 & 17:25/18:21 & @xmath252&@xmath253 & & & jan 21 & 07:42/08:21 & @xmath254&@xmath255 + dec 15a & 12:18/14:20 & @xmath256&@xmath257 & & & jan 23 & 05:07/06:32 & @xmath258&@xmath259 + dec 15b & 14:51/19:42 & @xmath260&@xmath261 & & & jan 24a & 06:23/13:49 & @xmath262&@xmath263 + dec 15c & 20:36/21:17 & @xmath264&@xmath265 & & & jan 24b & 14:13/15:07 & @xmath266&@xmath267 + dec 16 & 02:26/03:02 & @xmath268&@xmath269 & & & jan 25 & 07:26/10:36 & @xmath270&@xmath271 + dec 17 & 11:44/12:51 & @xmath272&@xmath273 & & & jan 28 & 10:57/11:43 & @xmath274&@xmath275 + dec 18 & 01:41/02:53 & @xmath276&@xmath277 & & & jan 29 & 14:25/14:51 & @xmath278&@xmath279 + dec 19a & 14:14/15:10 & @xmath280&@xmath281 & & & jan 31 & 08:48/09:13 & @xmath282&@xmath283 + dec 19b & 20:08/20:32 & @xmath284&@xmath285 & & & feb 2 & 08:09/08:23 & @xmath286&@xmath287 + dec 20 & 23:58/00:58 & @xmath288&@xmath289 & & & feb 4 & 05:39/05:54 & @xmath286&@xmath290 + dec 22 & 02:49/03:32 & @xmath291&@xmath292 & & & feb 6a & 13:34/16:19 & @xmath293&@xmath294 + dec 24 & 09:20/16:11 & @xmath295&@xmath296 & & & feb 6b & 18:16/18:45 & @xmath297&@xmath298 + dec 25 & 08:32/15:48 & @xmath299&@xmath300 & & & feb 6c & 23:00/23:36 & @xmath301&@xmath302 + dec 28 & 05:28/06:21 & @xmath303&@xmath304 & & & feb 8 & 06:46/07:34 & @xmath305&@xmath306 + dec 29a & 05:52/06:30 & @xmath307&@xmath308 & & & feb 9 & 18:02/18:14 & @xmath309&@xmath310 + dec 29b & 21:52/22:29 & @xmath311&@xmath312 & & & feb 10 & 14:52/15:26 & @xmath313&@xmath314 + dec 30 & 13:35/14:20 & @xmath315&@xmath316 & & & feb 12a & 01:23/04:10 & @xmath317&@xmath318 + jan 6a & 06:23/06:38 & @xmath319&@xmath320 & & & feb 12b & 17:04/21:06 & @xmath321&@xmath322 + jan 6b & 15:15/16:04 & @xmath323&@xmath324 & & & feb 13a & 00:58/01:45 & @xmath325&@xmath326 + jan 8a & 10:11/10:34 & @xmath327&@xmath328 & & & feb 13b & 04:12/08:18 & @xmath329&@xmath330 + jan 8b & 14:25/14:56 & @xmath331&@xmath332 & & & feb 13c & 11:45/14:10 & @xmath333&@xmath334 + jan 8c & 17:41/18:25 & @xmath335&@xmath336 & & & feb 13d & 18:01/19:01 & @xmath337&@xmath338 + jan 10 & 10:57/16:02 & @xmath339&@xmath340 & & & feb 15 & 12:07/14:56 & @xmath341&@xmath342 + jan 11a & 01:01/01:24 & @xmath343&@xmath344 & & & mar 7 & 07:07/08:06 & @xmath345&@xmath346 + jan 11b & 08:20/09:18 & @xmath347&@xmath348 & & & mar 9 & 07:54/08:10 & @xmath349&@xmath350 + jan 11c & 11:32/12:32 & @xmath351&@xmath352 & & & mar 18 & 01:34/01:50 & @xmath353&@xmath354 + lc cccc @xmath355 & & @xmath356&@xmath357&@xmath358&@xmath359 + @xmath36 & & @xmath360&@xmath361&@xmath362&@xmath363 + @xmath38 ( kev ) & & @xmath364&@xmath365&@xmath366&@xmath367 + @xmath368 ( kev ) & & @xmath369&@xmath370&@xmath371&@xmath372 + @xmath373 ( kev ) & & @xmath374&@xmath375&@xmath376&@xmath377 + @xmath378 & & @xmath379&@xmath380&@xmath381&@xmath382 + @xmath383 ( kev ) & & @xmath384&@xmath385&@xmath386&@xmath387 + @xmath388 ( kev ) & & @xmath389&@xmath390&@xmath391&@xmath392 + @xmath393 & & @xmath394&@xmath395&@xmath396&@xmath397 + @xmath398 ( kev ) & & @xmath399&@xmath400&@xmath401&@xmath402 + @xmath403 ( kev ) & & @xmath404&@xmath405&@xmath406&@xmath407 + @xmath183 & & @xmath408&@xmath409&@xmath410&@xmath411 + @xmath412 @xmath413 & & @xmath414&@xmath415&@xmath416&@xmath417 + @xmath418 & & @xmath419&@xmath420&@xmath421&@xmath422 + lcccccccccccccc ccccccc + fa1 & @xmath423&@xmath424 & & @xmath425&@xmath426&@xmath427&@xmath428&@xmath429&@xmath430 & & @xmath431&@xmath432&@xmath433 & 3.59 & 0.81 + fa2 & @xmath434&@xmath435 & & @xmath436&@xmath437&@xmath438&@xmath439&@xmath440&@xmath441 & & @xmath442&@xmath443&@xmath444 & 3.26 & 0.59 + fa3 & @xmath445&@xmath446 & & @xmath447&@xmath448&@xmath449&@xmath450&@xmath451&@xmath452 & & @xmath453&@xmath454&@xmath455 & 3.07 & 0.71 + fa5 & @xmath456&@xmath457 & & @xmath458&@xmath459&@xmath460&@xmath461&@xmath462&@xmath463 & & @xmath464&@xmath465&@xmath466 & 1.79 & 0.57 + fa6 & @xmath467&@xmath468 & & @xmath469&@xmath470&@xmath471&@xmath472&@xmath473&@xmath474 & & @xmath475&@xmath476&@xmath477 & 1.53 & 0.55 + fa7 & @xmath478&@xmath479 & & @xmath480&@xmath481&@xmath482&@xmath483&@xmath484&@xmath485 & & @xmath486&@xmath487&@xmath488 & 1.10 & 0.64 + + + fd1 & @xmath489&@xmath490 & & @xmath491&@xmath492&@xmath493&@xmath494&@xmath495&@xmath496 & & @xmath497&@xmath498&@xmath499 & 3.54 & 0.64 + fd2 & @xmath500&@xmath501 & & @xmath502&@xmath503&@xmath504&@xmath505&@xmath506&@xmath507 & & @xmath508&@xmath509&@xmath510 & 3.16 & 0.53 + fd3 & @xmath511&@xmath512 & & @xmath513&@xmath514&@xmath515&@xmath516&@xmath517&@xmath518 & & @xmath519&@xmath520&@xmath521 & 2.72 & 0.80 + fd4 & @xmath522&@xmath523 & & @xmath524&@xmath525&@xmath526&@xmath527&@xmath528&@xmath529 & & @xmath530&@xmath531&@xmath532 & 2.23 & 0.61 + fd5 & @xmath533&@xmath534 & & @xmath535&@xmath536&@xmath537&@xmath538&@xmath539&@xmath540 & & @xmath541&@xmath542&@xmath543 & 1.87 & 0.68 + fd6 & @xmath544&@xmath545 & & @xmath546&@xmath547&@xmath548&@xmath549&@xmath550&@xmath551 & & @xmath552&@xmath553&@xmath554 & 1.45 & 0.44 + fd7 & @xmath555&@xmath556 & & @xmath557&@xmath558&@xmath559&@xmath560&@xmath561&@xmath562 & & @xmath563&@xmath564&@xmath565 & 0.79 & 1.08 + fd8 & @xmath566&@xmath567 & & @xmath568&@xmath569&@xmath570&@xmath571&@xmath572&@xmath573 & & @xmath574&@xmath575&@xmath576 & 0.56 & 0.42 + lcccccccccccccc ccccccc fd1 & @xmath577&@xmath578&@xmath579 & & @xmath580&@xmath581&@xmath582&@xmath583&@xmath584&@xmath585 & & @xmath586&@xmath587&@xmath588 & 35.42 & 0.74 + fd6 & @xmath589&@xmath590&@xmath591 & & @xmath592&@xmath593&@xmath594&@xmath595&@xmath596&@xmath597 & & @xmath598&@xmath599&@xmath600 & 14.50 & 0.61 +

we present an independent estimate of the interstellar extinction law for the _ spitzer _ irac bands as well as a first attempt at extending the law to the 24 mips band . the source data for these measurements are observations of five nearby star - forming regions : the orion a cloud , ngc 2068/71 , ngc 2024/23 , serpens and ophiuchus . color excess ratios}$ ] were measured for stars without infrared excess dust emission from circumstellar disks / envelopes . for four of these five regions , the extinction laws are similar at all wavelengths and differ systematically from a previous determination of the extinction law , which was dominated by the diffuse ism , derived for the irac bands . this difference could be due to the difference in the dust properties of the dense molecular clouds observed here and those of the diffuse ism . the extinction law at longer wavelengths toward the ophiuchus region lies between that to the other four regions studied here and that for the ism . in addition , we extended our extinction law determination to for serpens and ngc using _ spitzer _ mips data . we compare these results against several iso extinction law determinations , although in each case there are assumptions which make absolute comparison uncertain . however , our work confirms a relatively flatter extinction curve from 4 - 8 m than the previously assumed standard , as noted by all of these recent studies . the extinction law at is consistent with previous measurements and models , although there are relatively large uncertainties .

astro-ph0703777

with the launch of the _ spitzer _ space telescope , identification of embedded young stellar objects ( ysos ) from their infrared excess emission at mid - infrared wavelengths ( @xmath4 using irac , the infrared array camera , and at 24@xmath3 m , using mips , the multiband imaging photometer ) has been facilitated by _ spitzer s _ unprecedented sensitivity . these wavelengths are sensitive to the emission from disk and envelope material surrounding ysos , making it easier to differentiate young stellar objects from foreground or background stars where the emission is primarily photospheric . @xcite and @xcite develop a classification scheme to differentiate young stellar objects based on their irac colors , and @xcite present a scheme utilizing 24@xmath3 m data in combination with irac data . @xcite has recently presented a robust classification scheme for ysos which utilizes de - reddened mid - ir fluxes of the cluster members in combination with ground - based photometry . these classification schemes are briefly compared in the recent review article by @xcite . @xcite also employ de - reddened irac observations to identify ysos in ic348 , using an alternate classification scheme . since these schemes involve de - reddening , it is critical to provide an accurate extinction law at irac and mips wavelengths that is appropriate for young clusters embedded in nascent molecular cloud material . examining the extinction to background stars beyond these embedded young clusters probes the column density of dust in the molecular cloud . the availability of 2mass has allowed for construction of large scale extincton maps in the near - infrared which trace a substantial range in column densities @xcite . with an appropriate extinction law , this technique could be extended to longer wavelengths where the extinction is lower and higher column densities can be probed . throughout much of the irac range of interest ( 0.7 - 5@xmath3 m ) , @xcite fit data existing at that time to a power - law @xmath5 , with @xmath6 = 1.75 , and extrapolated the power law to meet the beginning of a silicate contribution at @xmath77 @xmath3 m in his fig . 1 ; other early papers have found @xmath6 ranging from 1.61 @xcite to 1.80 @xcite . dust models @xcite employing size distributions of various grain components with known optical properties have been constructed to match observed extinction laws . measurements of the extinction law from @xmath8 to @xmath9 have been reported by many investigators using a variety of techniques ( from both ground - based and space observations ) along a number of lines of sight , and these were thought to be universal " , i.e. generally in agreement with each other @xcite , although recently some authors have questioned this universality @xcite . longward of @xmath10 the extinction is less certain ( see the range in @xmath6 above ) , and the most robust estimates are generally obtained with instruments on spaceborne telescopes . pre-@xmath11 studies have relied primarily on measurements of hydrogen recombination or hydrogen rotational and ro - vibrational lines @xcite to estimate the extinction law for @xmath12 . for example , @xcite used hydrogen recombination line emission toward the galactic center to probe the extinction from @xmath13 and found that the extinction law @xmath14/@xmath15 is flatter from @xmath16 than the power laws noted above , and approximately @xmath17 smaller than the value of the standard curve at 7@xmath3 m . @xcite notes that the lack of a deep minimum near 7@xmath3 m is in conflict with predictions assuming the standard mixture of silicate and graphite grains @xcite . @xcite and @xcite observed @xmath18 rotational and ro - vibrational lines toward omc-1 and assumed a power law ( @xmath19 extending from 2.4 to @xmath20 ) , and fit two silicate features to the extinction law beyond @xmath21 . the latter authors also determined extinctions from hydrogen recombination lines , and obtained smaller values of the relative extinction by this method than with @xmath18 lines . in that case @xmath22/@xmath15 was also relatively flatter from 3@xmath3 m - 7@xmath3 m , and from 1.5 - 2 times smaller than the comparison curve ( in this case the ir extinction law from @xcite with modifications to the 10@xmath3 m silicate feature strength : the cohen law is very similar otherwise to the one adopted by @xcite ) . a study using isogal , denis and 2mass data of intermediate agb and rgb stars in the inner galactic plane by @xcite determined values for @xmath23/@xmath24 and @xmath25 that are more comparable to those of @xcite than to earlier ` standard ' interstellar extinction curves in the solar neighborhood @xcite . using observations of stars they identified as reddened clump giant stars in the galactic plane obtained as part of the _ spitzer _ glimpse survey , @xcite directly measured the extinction law for the _ spitzer _ irac bands at 3.6 , 4.5 , 5.8 , and 8.0@xmath3 m . they observed the massive star - forming region rcw 49 at a distance exceeding 2 kpc , as well as two off - field regions very close to the galactic plane , and supplemented the _ data with 2mass observations of the same fields . for the off - field regions , their method determines the diffuse interstellar medium extinction , assuming that the red clump giants are uniformly distributed along the line of sight . reddened red clump giant stars were selected from their position in the @xmath26 vs. @xmath27 color - magnitude diagram , and a reddening band was identified which was then used to determine @xmath28 assuming a narrow distribution of intrinsic colors and luminosities of the clump giants . @xmath29 was then determined from @xmath30-k_s$ ] vs. @xmath27 color - color diagrams for these objects . the extinction curve for all three of these regions was nearly identical and also appeared to have a shallower slope for @xmath31 than the standard law . r. indebetouw kindly provided the data for these three regions so that we could process them using the same methods that we used to process our data , hence making possible a direct comparison of the extinction laws toward the young clusters studied here , and that along the lines - of - sight probed by @xcite . here we determine the extinction law from @xmath4 toward five young , nearby ( @xmath32 1 kpc ) cluster regions , including the orion a cloud , ngc 2068/71 , ngc 2024/23 , ophiuchus and serpens . we further determine the extinction at @xmath1 for two of the regions , ngc 2068/71 and serpens . in all cases we measure extinctions in relatively broad wavebands as determined by fixed , cold filters . by examining those stars without disk / envelope infrared excess emission in the cloud direction , we evaluate the selective extinction at mid - infrared wavelengths : the extinction derives primarily from their host molecular clouds , although extinction from the diffuse interstellar medium is also a factor . we compare our determination against prior determinations in the literature including the standard extinction law " such as that compiled by @xcite . it is important when comparing these determinations to keep in mind that spectroscopic measurements may differ from our broad - band measurements at these wavelengths . our separate extinction law study at the wavelengths of the irac bands is motivated by a discrepancy between slopes of the bands of reddened sources in the @xmath33 vs. @xmath34 $ ] color - color diagrams for our clusters , and slopes found using the glimpse data . reddened non - excess sources form a band on the @xmath33 vs. @xmath34 $ ] color - color diagrams extending from the intrinsic colors of the stars parallel to the reddening vector whose _ slope _ is the color - excess ratio @xmath0}$ ] @xmath35 @xmath36 . these color excess ratios , derived in section [ obs ] and given in table 2 for our data , and for the glimpse data provided by indebetouw , illustrate at a glance the discrepancy between selective extinction slopes toward our sources and theirs . @xcite also found a slope discrepancy between the selective extinction law toward isolated dense molecular cores observed in the _ spitzer _ cores to disks ( c2d ) legacy program @xcite , and toward those regions examined by @xcite . for these young clusters , and for the regions examined by @xcite , we utilize in our analysis those stars without a mid - infrared excess in the irac color - color diagrams @xcite , namely class iii cluster objects , and field stars . background dwarfs and red giants , including the red clump giants , are the primary probes of the extinction that originate in the clusters nascent molecular clouds . using complementary 2mass ground - based @xmath37 and @xmath38 data as well as @xmath11 data on these objects , we measure the color excess ratios @xmath0}$ ] from @xmath33 vs. @xmath34 $ ] color - color diagrams , and from these ultimately derive @xmath29 for an assumed value of @xmath39 . the same method allows extension to 24@xmath3 m for the serpens and ngc @xmath2 clusters , although these results are more uncertain because of the comparative paucity of detected sources . in section [ obs ] the observations and data reduction will be discussed , and section [ measure_ext ] will describe source selection and the derivation of the reddening law from these data . section [ results ] will report the results and discuss the differences between the extinction law derived here for our data and for the data used in the study by @xcite , as well as the extinction law derived using different methods and extinction probes by @xcite , with further reference to other determinations @xcite . we have not determined the total to selective extinction in this study , so shall concentrate on relative extinctions , assuming a value for @xmath39 in order to make comparisons with other work . because a non - universal extinction law may lead to variations in this assumed value , we consider how this assumption affects our results . observations of several young star - forming clusters were obtained with the irac camera @xcite aboard _ spitzer _ as part of several surveys concentrating on star - formation within @xmath40 kpc of the sun , including the young cluster survey " ( pid 6 ) , the orion survey " ( pid 43 ) from g. fazio s guaranteed observation time , as well as the cores to disks legacy program " ( pid 174 and 177 ) . the irac camera takes simultaneous images in four channels whose isophotal wavelengths are @xmath41 , @xmath42 , @xmath43 and @xmath44 respectively @xcite . some of the regions observed cover several square degrees so that multiple aors with the irac camera were required . ngc @xmath45 and ngc @xmath46 were imaged as part of the same region , while ngc @xmath47 and ngc @xmath48 were also observed together . observations of the orion a cloud included coverage of omc 2/3 , onc and l1641 . high dynamic range mode @xcite was used , resulting in a frame with a @xmath49 second exposure time and a frame with a @xmath50 second exposure time at each position . the maps for each of ngc 2024/23 , ngc 2068/71 and orion a were repeated twice . the maps of serpens and ophiuchus were repeated four times , with small offsets . the additional dithers provide both additional sensitivity to faint sources and greater ability to remove cosmic rays from the final mosaics . we have not corrected for the position dependent gain variation for point sources , which would result in an additional 5% uncertainty . for the orion fields individual frames were mosaiced using a custom idl program and from these mosaics , source selection and point source photometry were completed using gutermuth s photvis v@xmath51 @xcite . the photometry used an aperture of @xmath52 and a sky annulus with inner and outer radii of @xmath52 and @xmath53 . details on the data reduction and source selection in serpens and ophiuchus are described in @xcite and @xcite . @xcite discusses the calibration of the irac instrument . the calibration of the _ spitzer _ telescope was refined during the course of this work , and the resulting zero points are listed in table 1 . however , changes in the zero point do not change the results in this paper and are taken into account in our fitting technique . from the 2mass point source database we obtained corresponding @xmath54 and @xmath38 band magnitudes for each star . for the ngc @xmath2 , ngc @xmath55 and serpens regions , we required magnitude uncertainty @xmath56 in all seven bands for sources used to calculate the extinction law . for the orion a cloud and ophiuchus a detection in j band was not required , thereby increasing the number of highly reddened sources and providing a better determination of the extinction law . for the remaining 6 bands , a magnitude uncertainty of @xmath57 was imposed . if we do not require a j band detection for ngc @xmath55 , ngc @xmath2 or serpens , our derived values for the slope of the extinction vector are within @xmath58 of the values derived with the seven - band data . requiring sources to have this magnitude uncertainty reduced the spread of the sources on each color - color diagram , and minimized the uncertainty in the fit ( see section [ measure_ext ] ) . those sources for which only upper limits to 2mass magnitudes were available were not included in our analysis . most of the sources had much smaller magnitude uncertainties . the median magnitude uncertainties for ngc 2024/23 sources are @xmath59 for @xmath60 , and channels 1 through 4 of irac respectively . similar median magnitude uncertainties were obtained for the other regions studied here . loosening the requirement on the maximum uncertainty will include more sources with lower fluxes . it is possible that these sources have higher extinction and could provide a more accurate measure of the reddening law . however , increasing the uncertainty limit leads to the inclusion of more outliers that skew the slope and are difficult to remove with a simple color cut . a small number of fits did not have any outliers and for these fits including the sources with higher uncertainty does not change our results by more than the measured uncertainty in the color excess ratios . mips 24 @xmath3 m images were available for several of the regions studied . not all regions were suitable - there were insufficient non - excess sources in some regions to reliably derive an extinction law . we report on ngc 2068/71 and serpens observations with the multiband imaging photometer for _ spitzer _ ( mips , * ? ? ? we summarize the mips observations , with more detailed descriptions of the data reduction and photometric measurements for serpens and ngc 2068/71 in @xcite and @xcite respectively . the measurements were obtained in scan mode with half - array cross - section offsets . all three mips bands are observed simultaneously in this mode ; the total effective exposure time per point is about 80 seconds at 24 microns . photometry was determined using a psf fitting program in idl . background dwarfs and giants ( including red clump giants and rgb stars ) as well as embedded pre - main sequence stars can be used to measure the extinction due to the molecular cloud dust associated with the star - forming region as well as diffuse line - of - sight extinction . as mentioned above , the sources used to measure the extinction were chosen from those without infrared excesses from disk or envelope emission in the @xmath61-[4.5]$ ] vs. @xmath62-[8.0]$ ] color - color diagram toward each cluster region . eliminating infrared excess sources from consideration removes any ambiguities between stars with red colors caused by extinction vs. stars with circumstellar dust or envelope emission , including both pre - main sequence stars and agb stars . a non - excess source was defined as having colors @xmath61-[4.5 ] < 0.6 $ ] and @xmath62-[8.0 ] < 0.2 $ ] . this region of consideration was chosen so that it did not overlap with the expected position of pre - main sequence stars with disks according to @xcite and included the expected spread in colors due to interstellar dust reddening , which is larger at the shorter wavelengths . this definition of non - excess sources is consistent with the theoretical models of @xcite , and the observations of @xcite and @xcite . an upper limit on the @xmath61-[4.5]$ ] color is used to eliminate unresolved outflow sources misidentified as stars . outflows have emission features in the 4.5 @xmath3 m band , leading to @xmath61-[4.5]$ ] colors which could be confused with those from reddened non - excess stars . we extend our selection criteria to attempt to further remove foreground stars , and stars whose colors place them outside the reddening band . we remove foreground giants by excluding any star brighter than a k2iii giant with @xmath63 @xcite at the distance of each cluster . we also exclude stars with @xmath64 , which are mostly foreground dwarfs ( see discussion below ) . stars with @xmath65<0 $ ] do not fit into the reddening band and are removed as well as stars that appear to have an excess in the @xmath33 vs @xmath66 $ ] diagram ( e.g. @xmath67)-0.2 $ ] for serpens ) . we have not attempted to use only clump red giants , as was done by @xcite , because of the difficulty in distinguishing these stars from highly reddened dwarfs and pre - main sequence stars without disks . our reddest stars will be dominated by red giants and embedded pre - main sequence stars because of the large intrinsic luminosity needed to detect these stars at such high extinction although at lower extinction we may have a mix of reddened dwarfs as well as giants and pre - main sequence stars . for orion a , a number of galaxies contaminated the field and their removal from consideration is discussed in section [ oriona ] . @xcite point out that stars lying to bluer colors than the red clump giant distribution in a @xmath26 or @xmath68 vs. @xmath69 color - magnitude diagram for a given line of sight are primarily dwarfs , while stars lying to redder colors are predominantly m giants and agb stars . the red clump giant distribution is almost vertical around a median @xmath69 = 0.75 @xmath70 0.2 for off - cloud lines of sight at galactic latitudes with absolute value @xmath71 6@xmath72 . for fields probing greater extinction from either molecular clouds or along lines of sight closer to the galactic plane , the red clump giant distribution shifts to redder @xmath69 , fainter magnitudes and the distribution is broader . all of our clusters have galactic latitudes exceeding 6@xmath72 , with the exception of serpens , at @xmath73=5.4@xmath72 . to illustrate the above points for one of our clusters , we show a @xmath26 vs. @xmath27 diagram for ngc 2068/71 ( a region of high extinction ) as well as an off - cloud region overplotted ( fig . [ jvj - k_2068_off ] ) . the off - cloud field is centered at l=210@xmath72 , b=-8@xmath72 , a region that was chosen because it did not show any significant co emission indicative of a dense molecular cloud @xcite . in that diagram , the sources coincident with the cluster utilized in measuring the color - excess ratios are plotted with blue symbols , and red symbols illustrate cluster sources rejected by the @xmath33 cutoff . off - cluster sources are plotted with black symbols . in the black double - peaked distribution , dwarfs at various distances along the line of sight occupy the bluer peak , and the red clump sources the redder , with median color @xmath74 = 0.75 , as noted above . the effect of the molecular cloud s reddening can be seen by noting that most red clump extinction probes to ngc 2068/71 form a clump centered at @xmath27=1.05 at @xmath26=14.5 . the off - cloud giants have nearly constant @xmath27 color , indicating little extinction from the diffuse ism . for this reason we expect the contribution of the diffuse ism to our measured extinction law for the orion fields to be small , although this may not be equally valid for serpens and ophiuchus . many sources redder than the red clump giants are sampled , and in particular , for the cluster region ( red and blue points ) in figure [ jvj - k_2068_off ] there is no distinct red clump giant locus . in the cluster region , the highly variable extinction makes it impossible to isolate red clump giants , reddened dwarfs and embedded pre - main sequence stars . we are likely dominated by red clump giants , and perhaps pre - main sequence stars , in the densest parts of the cluster , because they are bright enough to be observed at large reddening . however , we do not attempt to use only red clump giants . we now move to discussion of reddening law determination . a straight line was fit to background sources in the ` reddening ' band on an @xmath33 vs. @xmath34 $ ] color - color diagram . after all of the restrictions described above were imposed , there were 415 , 670 , 1830 and 1225 background sources for ngc @xmath75 , ngc @xmath2 , orion a and serpens respectively . ophiuchus had 120 background sources and the implications of this small sample size are discussed further in section [ results ] . the slope of the line fit to the ` reddening ' band gives the color - excess ratio @xmath0}$ ] which is then used to determine the relative extinction law @xmath29 , where we assume @xmath39 = @xmath76 ( the value derived by @xcite ) . the effect of a different choice of @xmath39 is discussed in section [ iso ] . the line fit to the reddening band data in each of the @xmath33 vs. @xmath34 $ ] color - color diagrams minimized the chi - squared , weighted by the uncertainty in each color . @xmath77 where @xmath78 and @xmath73 are the y - intercept and slope , respectively , of the best fit line and @xmath79 and @xmath80 are the uncertainties in the x and y - values respectively . the algorithm _ amoeba _ in idl was used to minimize the chi - squared . determining the best fit - line to the reddening band data using equation [ chisq ] not only accounts for the magnitude uncertainties , but also provides a method for estimating the uncertainty in the slope , which leads directly to @xmath29 , the quantity of interest . @xcite presented a detailed analysis of the adopted fitting technique as well as the characterization of its uncertainty , which we summarize here for the convenience of the reader . at the best - fit points the @xmath81 is a minimum , but if the values of @xmath78 and @xmath73 are perturbed away from the best - fit values @xmath81 increases . a constant value of @xmath82 gives a confidence region within parameter space , which depending on the size of @xmath83 , could enclose a @xmath84 , @xmath85 , etc . confidence interval . the uncertainty in @xmath78 and @xmath73 is then the projection of this confidence region onto the @xmath78 and @xmath73 axis . finding @xmath86 where @xmath83 has a specific value is determined analytically with a taylor expansion when @xmath81 is linear in @xmath73 , but must be determined numerically here because @xmath81 is non - linear in @xmath73 . this determination is performed by varying @xmath73 , minimizing @xmath78 at each step , until @xmath83 equals the desired value . there are two values of @xmath73 , one greater and one less than the best - fit value , for which @xmath83 equals the desired value . the distances of these two new values of @xmath73 from the best fit slope are combined to determine the uncertainty in the slope . for the majority of our analysis @xmath83 was chosen to be 1 , which corresponds to a @xmath84 confidence level @xcite . increasing the confidence level to @xmath87 requires @xmath88 , which results in an increased uncertainty in the slope . both a @xmath84 and @xmath87 confidence interval were determined and are included in table [ slopes ] . the uncertainties are @xmath89 times bigger for the larger confidence interval than for the smaller confidence interval , but lead to at the most a @xmath90 uncertainty in the fit . the process that was used to derive the extinction law for the irac bands can be applied to the @xmath1 channel of mips . there were only sufficient numbers of non - excess sources with corresponding detections in 2mass , irac and mips in the serpens and ngc 2068/71 regions . the non - excess sources were defined as having @xmath61-[4.5]<0.6 $ ] , @xmath62-[8.0]<0.2 $ ] and @xmath91-[24]<0.5 $ ] , with the last color cut used to remove sources that had an excess only at 24 . for both ngc @xmath2 and serpens fewer than 50 sources were used for the final fit , making sure to exclude outliers and sources with @xmath92 , so we are working with quite small numbers of sources . for this reason , we view these results with extreme caution . while the process for deriving the extinction law from serpens , ophiuchus , ngc 2068/71 and ngc 2024/23 data was straightforward , the number and variety of sources toward orion a required more careful attention . the orion a cloud covers @xmath93 square degrees , a much larger area than the other four regions observed , increasing the number of sources used when determining the reddening law but also increasing the number of outliers . these outliers are most obvious on the @xmath33 vs. @xmath66 $ ] color - color diagram on which there is a ` branch ' with @xmath94 which could skew the fit . this ` branch ' is not as dominant in any of the other @xmath33 vs. @xmath34 $ ] color - color - diagrams and has nearly constant @xmath94 color as well as @xmath95 . the objects in this branch are too blue in @xmath33 color to be infrared excess sources but they are too red in @xmath66 $ ] color to be main - sequence stars , red giants , or class iii sources . the most likely explanation is that these sources are galaxies . their blue colors in the near infrared likely derive from main - sequence stars dominating the light at short wavelengths . polycyclic aromatic hydrocarbon ( pah ) emission from the dust within the galaxies is the likely cause of the red @xmath66 $ ] color , since strong pah feature emission falls within the @xmath91 $ ] band of irac . interpreting these sources as galaxies is consistent with their fairly uniform spread in position throughout the orion a cloud . they are also not present in the areas with the highest concentration of young stellar objects because these areas have a large density of nebulosity which decreases the sensitivity for detection of background objects , and a high column density of dust obscuring faint background sources . it is also possible that foreground sources contaminated by extended pah emission are contaminants : however , this does not explain the lack of outliers in the regions of largest pah emission near the young stars . regardless of the physical nature of these outliers , we remove them from our fitting procedure . a first attempt at removing these outliers was performed by changing the selection criteria for non - excess sources . this did not effectively remove the outliers without also removing a signifcant number of sources in the reddening band which are needed to accurately measure the extinction law . removing these outliers , without removing a substantial number of sources in the reddening band , was completed by first fitting a line in the @xmath33 vs. @xmath66 $ ] color - color diagram to all of the non - excess sources , including the outliers , and then using the slope of this fit to define a boundary line ( @xmath96)-0.2 $ ] ) , below which were assumed outliers and above which were the sources used for a more accurate measure of the extinction . the y - intercept of this line was chosen so that a majority of the outliers was removed without removing a significant portion of the reddening band . the initial fit in the @xmath33 vs. @xmath66 $ ] color - color diagram had a slope of @xmath97 and the final fit had a slope of @xmath98 with @xmath99 decreasing from @xmath100 for the initial fit to @xmath101 for the final fit . while this is a small correction , the final fit is a more accurate measurement of the extinction law at @xmath21 . the slopes of the linear fits to the @xmath33 vs. @xmath34 $ ] color - color diagrams for our five clusters are listed in table [ slopes ] and plotted in figures [ ngc2024_red ] , [ ngc2068_red ] , [ serp_red ] , [ oph_red ] , and [ oriona_red ] . it is apparent that our derived slopes ( color excess ratios ) are reasonably consistent from region to region , but in a few instances , are more than 1-@xmath102 different from each other at the longer wavelengths . in particular , the selective color excess ratio toward the ophiuchus cloud is lower than toward the other four clouds at 8.0@xmath103 . since ophiuchus has the smallest number of non - excess sources ( 120 vs. @xmath104400 for the other four regions)we assess the possibility that the difference is due to the small number of non - excess sources . we randomly select 120 non - excess sources from the orion a field and measure the color - excess ratios in the four irac bands for the subsample , repeating this process for 1000 subsamples . we can estimate how likely it is to measure the ophiuchus color - excess ratios based on the distribution of color - excess ratios in each of the four bands for the subsamples . for the 5.8 and 8 bands the average color - excess ratio for the subsamples is 1.115 and 1.122 , with a standard deviation of 0.06 for both bands . the ophiuchus color - excess ratios are approximately 2@xmath102 from these mean color - excess ratios at both 5.8 and 8 , only a marginally significant deviation . the fraction of subsamples with color - excess ratios at or below the ophiuchus values are 3% and 1% for the 5.8 and 8 bands respectively . thus sample size is unlikely to be the cause of the lower selective color excess ratio at 8.0@xmath103 for ophiuchus . the color excess ratios for the @xmath105 off - cloud line - of - sight glimpse data were derived using the same source selection criteria and methods as used for the five star - forming regions discussed here . in contrast to the general agreement among the clusters , the slopes from the glimpse data listed in table [ slopes ] are generally different from the cluster slopes . for completeness , we also quote the color excess ratios from @xcite for the @xmath106 off - cloud line of sight . since they directly measured @xmath107 - k_s}/e_{j - k_s}$ ] , we form @xmath0}$ ] from those ratios in table [ slopes ] . those values differ at 3.6 and 4.5 @xmath3 m from values determined by our analysis . the question remains whether the larger sample size employed by @xcite in determining the extinction law is the cause of the discrepancy , as was suspected for ophiuchus . their method uses red clump giant stars selected by their @xmath27 colors and only requires a detection in the three bands ( @xmath108 $ ] ) that are used when measuring the color - excess ratios . this means that more sources are available for determining the color - excess ratios . to avoid this , and any other possible biases due to differing methodologies , we have reanalyzed the @xcite data using the same methods that were applied to our five clusters as noted above the @xmath105 off - cloud line - of - sight was studied in detail because it had the largest number of background sources and its extinction law was representative of the other two lines of sight . there were 1590 background sources that met our criteria , comparable to the number of background sources used for four of our five clusters . @xcite find 3560 red clump giants along this same line of sight when requiring a detection in j and @xmath38 band . this includes stars that have an infrared excess due to circumstellar dust and stars that are not detected by irac . the actual number of sources used by @xcite when measuring the color - excess ratios was smaller ( @xmath109 ) after requiring an irac detection and removing stars with an infrared excess . we examine the effect of differing sample sizes by considering the orion a field and varying the limit on the magnitude uncertainty . when limiting the magnitude uncertainty to 0.1 , there were 1825 background sources , while changing the limit from 0.05 to 0.25 results in sample sizes ranging from 1265 to 2360 . the color - excess ratios for the extreme cases differ from the case where the magnitude uncertainty is less than 0.1 , and from each other , by less than the measured uncertainty in the color - excess ratios . as a result we do not expect differing sample sizes to provide a source of systematic error when comparing our results to those of @xcite . figure [ excess_wav ] gives the color excess ratio @xmath0}$ ] versus @xmath110 for the five regions observed here as well as for the glimpse @xmath111 off - cloud line - of - sight data analysed by us , showing the systematic separation between the color excess ratios derived for the glimpse line of sight , and for the five young clusters . our values are consistently 1-@xmath102 , or more , higher than those rederived from @xcite data with the exception of the @xmath112 band . the irac non - excess stars used when measuring the color - excess ratios can include class iii pre - main sequence stars , main sequence stars and red giants ( clump giants , and rgb stars ) , all of which have different intrinsic colors . our constraint to exclude unreddened stars and bright probes should minimize contribution from foreground objects . the proximity of serpens and ophiuchus reduces the number of foreground contaminants which must be removed . @xcite restrict themselves to red clump giant stars and find they are well - separated from main sequence stars on a @xmath26 vs. @xmath27 color - magnitude diagram for the regions they studied . we examine the dependence on extinction probe by calculating residuals which are defined as the difference between @xmath113 for a given @xmath34 $ ] and the predicted @xmath114 using the @xcite determination of @xmath115 . figure [ residual ] , where the residuals are plotted versus @xmath33 color , shows the deviation of our extinction law determination for ngc @xmath2 , using non - excess stars , from that of @xcite using red clump giants . as noted in @xmath116 3 , the parameterization used by @xcite to pick out red clump giants will not apply to our cluster lines of sight . as can be seen in figure [ jvj - k_2068_off ] our extinction probes in ngc @xmath2 exhibit varying levels of extinction . if using only red clump giant probes produced a systematically different extinction law than we measured , we would expect a locus of sources to follow the @xcite law at all four bands . no such locus is observed in figure [ residual ] , suggesting that extinction probes are not the main cause of the systematic difference . sources with @xmath117 , which are mostly red giants , show a substantial deviation from the @xcite extinction law . since both methods referred measurements to the @xmath38-band rather than to the @xmath68-band , a correction for different photometric systems is not necessary . rederived glimpse color excess ratios are systematically separated from the color excess ratios for the five young clusters observed here . while there are differences in the results derived from the glimpse data using the different methodologies , especially at 3.6 and 4.5 , there is still a significant separation between our color excess ratios and those from @xcite . this suggests that the separation is not primarily caused by the differing methods and/or extinction probes . the systematic separation may instead be due to differences in the extinction from dust in the molecular clouds and the diffuse interstellar medium . this possibility will be further discussed in @xmath116 [ cloudvism ] we derive the extinction law @xmath29 for the five clusters studied here as well as for the selected glimpse data , using the color excess ratios determined above . @xmath29 is related to the color excess ratios , and the assumed value of @xmath118 by : @xmath119 since we have used precisely the same methods to establish the extinction law for both data sets , we have removed uncertainties in the comparison that are method - dependent . in both cases we assume @xmath120 = 1.55 : values for @xmath29 are listed in table [ ext ] . in addition , we also give the extinction law for the @xmath106 off - cloud line of sight as derived by @xcite . different choices for @xmath39 will affect our derived values of @xmath29 . for ngc @xmath55 , for example , if @xmath121 ( @xmath122 ) was chosen then @xmath29 would be @xmath123 for the four irac bands . if instead @xmath39 of 1.73 ( @xmath6=1.99 ) from @xcite was chosen then @xmath29 would be @xmath124 for the four irac bands . the difference in these two determinations of the relative extinction law is larger than the statistical uncertainties in @xmath29 derived from the uncertainties in the selective color excess ratios , which are @xmath125 . the uncertainties reported in table [ ext ] include the uncertainties in the measured color excess ratios but do not include the uncertainty in @xmath39 . the values of @xmath29 for each of the five regions observed here , subject to the assumption of @xmath126 , are plotted in figure [ ext_wav ] . the rederived extinction law for the @xmath106 off - cloud line of sight from @xcite is also plotted in figure [ ext_wav ] . extinction law measurements by @xcite , included as squares in figure [ ext_wav ] , are more consistent with @xcite , although some of their measurments overlap with our results . the measurement at @xmath127 from @xcite is marked with a circle in figure [ ext_wav ] , and the spread in their results for many different sight lines encompass both our results and the results from @xcite . our extinction law deviates from a power law , as is also seen in these previous studies . @xcite compared the observed to expected fluxes for 15 hydrogen recombination lines toward the galactic center from iso data and from this determined the extinction ratios at the wavelengths of the emission lines . theoretical estimates of the expected fluxes assumed t@xmath128 = 7000k , n@xmath128=3000 @xmath129 , case b recombination ( which they verify from their observations ) and refer directly to the calculations by @xcite . the recombination line results also depend on the assumed helium abundance , [ heii]/[heiii ] , and the assumption that line emissions emanate from the same volume . a crucial part of their analysis is the scaling of their relative extinction law : they deduce @xmath14/@xmath130 ( br@xmath6 at 2.626@xmath3 m is the shortest - wave emission line they observe ) and use a @xmath131 scaling law to refer to @xmath68-band . we will use a @xmath132 law which corresponds to our assumed @xmath120 = 1.55 , allowing for a more direct comparison with our values of @xmath29 ( see discussion below ) . numerical data on @xmath22/@xmath15 were provided courtesy of d. lutz : here we deduce from these data @xmath14/@xmath130 , and using a factor of @xmath24/@xmath130=1.38 ( assuming the power law index @xmath6=1.65 ) , deduce @xmath14/@xmath24 for their data . since we use 2mass @xmath133 data , we have taken the wavelength for @xmath38-band to be 2.16@xmath3 m . power law indices of 1.61 and 1.8 lead to at most a 2% error in the extinction ratio . this is considerably more certain than variations in @xmath134 , which can be as much as 30% @xcite , so we choose not to compute @xmath135 for our data . @xcite combined iso observations at 7 and 15@xmath103 of the inner galactic plane with 2mass and denis near - infrared observation to determine the mid - infrared extinction law and its variation along different sightlines . they assumed a constant value of @xmath136 , @xmath137)_o$ ] and @xmath138)_o$ ] for early agb and rgb - tip stars and measured the color excess of these stars along various sightlines . assuming @xmath139 from @xcite they find @xmath140 and @xmath141 ( marked as circles in figure [ ext_wav ] ) . although @xcite did not measure the extinction law with the 2mass photometric system , @xcite found @xmath28 in the 2mass photometric system to be similar to the value of @xmath139 measured by @xcite . the results from @xcite do not change significantly when using @xmath28 from @xcite as opposed to @xmath139 from @xcite . at 7@xmath103 they find the extinction law measurements have a gaussian distribution whose one sigma size encompasses the measurements presented here , as well as those of @xcite and @xcite . there is more uncertainty at @xmath142 but there also appears to be evidence for a gaussian distribution of extinction values , @xmath143 . we chose @xmath39 from @xcite to provide a direct comparison of all the mid - infrared selective extinction law measurements discussed here . the extinction law measurements by @xcite use a similar normalization to our assumed @xmath120 , and the measurements of @xcite were normalized using a power law form that is consistent with @xmath126 . different assumptions of the near - infrared extinction law will affect our derived @xmath29 as discussed earlier , but we choose a single consistent normalization to better compare the various determinations of the mid - infrared extinction law . a plausible explanation for the separation between our extinction law and the extinction law of @xcite is a variation in the extinction law between that through the diffuse ism and through molecular clouds . variation in dust optical properties between molecular cloud dust and that in the ism has been invoked by @xcite , who , from observations of cloud cores obtained as part of the c2d _ spitzer _ legacy program , have also found extinction law slopes ( color excess ratios ) which deviate from those of @xcite . the cold ism includes diffuse interstellar clouds ( @xmath144 0.1 ) , poorly shielded regions of molecular clouds ( @xmath144 3 ) , and dense molecular cloud cores ( @xmath145 3 ) . @xcite and references therein , have explored grain growth in dense clouds , and have reported differing extinction laws for the diffuse ism and dense clouds . for reference , @xmath15=3 is shown as a vertical dashed line in figure [ residual ] : most of the deviation of our results from the @xcite extinction law occurs at higher extinctions . changes in the extinction law originating in these regions are likely due to grain growth , either by coagulation of small grains or the growth of icy mantles , both of which will affect the continuum level of extinction . small grains are thought to coagulate to form larger grains ( e.g. @xmath146 oph , @xcite ) : grain emissivity is dependent on the optical constituent properties , and for the same constituents is proportional to a power(powers ) of _ a_/@xmath147 where _ a _ is the grain size . grain growth will not only remove small grains as potential absorbers , but also as emitters . the lack of emission from small grains has been observed along lines of sight with high column densities @xcite suggesting grain growth leading to porous grains with enhanced emissivity within these clouds . @xcite have found a decrease in the depth of the 9.7 @xmath3 m silicate feature with increasing @xmath15 , indicating silicate grain growth . @xmath148 3 is the approximate extinction threshold for the detection of ice mantles , and in the taurus dark cloud , icy mantle growth is implicated @xcite . dust in molecular cloud cores is protected both from the uv interstellar radiation field and from cosmic ray processing ; those dust grains may therefore have larger mantles consisting of water and organic ices @xcite . absorption features between 3 and 8 due to ices are often seen along lines of sight intersecting dense molecular clouds @xcite while they are not seen in the diffuse ism @xcite . on the other hand , grain processing and fragmentation because of radiative shock propagation , and turbulence , may also be involved in grain evolution in both molecular clouds and in the diffuse ism . our extinction law for the young clusters studied here derives mainly from dust associated with their molecular clouds , while that of @xcite derives primarily from diffuse interstellar dust , especially for the @xmath106 off - cloud line of sight . there is a greater contribution from diffuse interstellar dust for the closest clusters in our sample , ophiuchus and serpens , partially because they are in the inner galaxy , and partially because the extinction probes ( not excluded by the @xmath149 brightness cutoff ) lie well behind the cluster , so that their light passes through more of the ism before reaching the cluster . the extinction curve toward ophiuchus has been termed anomalous : measurements of the uv / optical extinction curve and linear polarization @xcite toward ophiuchus indicate a depletion of small grains and the presence of large grains within the molecular cloud . studies of other dark clouds have produced similarly anomalous extinction as compared with the ism : we are unaware of any comprehensive study comparing optical / uv extinction in the dark clouds studied here . @xcite observe h@xmath150o and co ices in the taurus and ophiuchus dark clouds , and concur with previous studies that different physical conditions obtain in those two clouds . although column densities of the h@xmath150o ices and the visual extinction are correlated in each case , the regression line is different . there are several ophiuchus sources which follow the taurus regression line rather than that for ophiuchus , suggesting physical properties of the grains can also vary within the cloud . @xcite and @xcite measure the emissivity spectral index @xmath6 ( @xmath151 ) in the far - infrared and find variations between orion , ophiuchus and serpens . this indicates that the grain properties differ between these three clouds , but the anticorrelation between spectral index and temperature within a cloud suggests that there can be significant variations in grain properties along different lines of sight through a single cloud . consequently , we can not determine here the extent to which the molecular cloud mid - ir extinction toward ophiuchus is anomalous compared with other clouds in our sample . in contrast to our clusters , all within 500 pc of the sun , the lines of sight for the iso measurement may include a larger contribution from a mixture of extinction from the diffuse ism and molecular clouds . the @xcite extinction law to the galactic center ionized gas is dominated by dust in the diffuse ism along the line of sight , although some of the extinction is due to molecular cloud material @xcite . broad ice absorption features at 3 and 6 are seen toward sgr a * and are attributed to foreground molecular cloud material @xcite . the @xcite extinction law is for a mix of diffuse ism and dense molecular cloud dust . given the uncertainties in the derived laws of @xcite and @xcite , we can not distinguish whether any agreement , or lack of agreement with either our law or that of @xcite depends on whether the material causing extinction is part of the diffuse ism or a dense cloud . the differing dust properties between diffuse ism and dense molecular clouds could lead to differences in the extinction law . we directly measure @xmath0}$ ] , which relates @xmath152}/a_{k_s}$ ] and the assumed value of @xmath39 . a variation in the near - infrared extinction law , the mid - infrared extinction law , or both , could lead to the observed systematic separation in the selective color excess ratios between our observations through molecular clouds and @xcite s observations through the diffuse ism . if a larger value of @xmath39 was more appropriate for the extinction law toward molecular clouds , then @xmath152}/a_{k_s}$ ] would be smaller and the _ mid - infrared _ portion of the extinction law deduced by @xcite and us could be made to agree . values of @xmath39 between 1.6 and 1.65 , depending on the region , would bring our mid - infrared extinction law into agreement with the average extinction law reported by @xcite , assuming @xmath126 still applied to the @xcite extinction law . it may also be the case that the near - infrared extinction law does not appreciably vary between the diffuse ism and molecular clouds and there is instead a variation in the mid - infrared extinction law . we now examine the near- and mid - infrared data separately in order to attempt to break this degeneracy . the possibility of variations in @xmath39 is suggested by recent observations . although @xcite and @xcite found the near - infrared extinction law to be universal , more recent studies have produced mixed results . one of the difficulties in evaluating the variation in the near - infrared extinction law along different lines of sight is that different photometric systems ( e.g. @xmath68 vs. @xmath38 ) can have a substantial effect on the measured extinction law @xcite . @xcite and @xcite observed the nearby star - forming regions @xmath146 oph and cha i , respectively , and determined the near - infrared reddening law using the same technique , but with different photometric systems , and find a substantial difference in the reddening law between these two lines of sight . @xcite reobserved @xmath146 oph and cha i with one photometric system and used the same technique , the same off - cloud regions , and the same limiting magnitudes as the previous studies by @xcite and @xcite but found no variation in the near - infrared reddening law between these two molecular clouds . @xcite measured the extinction law toward the galactic center using the same photometric system as @xcite and their results are consistent with the reddening law being universal between the galactic center and the @xmath146 oph and cha i star - forming regions . evidence for a variation in the near - infrared reddening law comes from @xcite who observed the coalsack globule 2 using the same photometric system as @xcite , and used the same technique to determine the reddening law , but found a significant difference between the reddening laws along these two lines of sight . to isolate the influence of the near - infrared extinction law on our measured color - excess ratios we have measured @xmath153 in a manner similar to the previous color - excess ratios . since we did not require a j band detection for the orion a and ophiuchus regions , we excluded these regions . ngc 2024/23 , ngc 2068/71 and serpens have color excess ratios of @xmath154 and @xmath155 . l_=@xmath156 off - cloud region has a color excess ratio of @xmath157 . no significant difference exists in the near infrared color excess ratios among these lines of sight within uncertainties , indicating that there is no significant variation in the near - infrared extinction law through these three molecular clouds and through the interstellar medium . we examine the mid - infrared portion of the extinction law by forming color - excess ratios that include only irac bands from the already measured color excess ratios . from the definition of the color excess ratio @xmath158}}=\frac{a_h - a_{k_s}}{a_{k_s}-a_{\lambda}}\ ] ] we form the following quantity : @xmath159 by choosing @xmath160 and @xmath161 we have a measure of the irac - only color - excess ratio . by combining color - excess ratios rather than using the extinction law measurements , @xmath29 , we reduce the uncertainty in the irac - only color - excess ratio . the results are listed in table [ irac_ratio ] for all five young clusters as well as for the @xmath162=@xmath156 off - cloud line of sight . while the uncertainties are large , there is clear evidence for a separation between the extinction law measured toward the molecular clouds and the extinction law for only the diffuse ism . the tabulated values of this ratio demonstrate that the variations between our color - excess ratios and those of @xcite can not be explained solely by a change in the value of @xmath118 , and that the form of the interstellar extinction law must differ between the diffuse ism , ophiuchus , and the remaining four molecular clouds . it is also apparent that the ophiuchus cloud reddening law differs from the reddening law for the other four young clusters . extending our analysis out to 24 we measure @xmath163}=0.98\pm0.04 $ ] and @xmath164 , @xmath165 and @xmath166 , assuming @xmath126 , for serpens and ngc 2068/71 respectively . the large uncertainty for both regions color excess ratios reflects the small number of sources used in the fitting procedure . figure [ 24_red ] shows the fit to the reddening bands at 24@xmath103 , similar to figures [ ngc2024_red ] - [ oriona_red ] . figure [ ext_wav ] illustrates that our 24@xmath103 determinations are in reasonable accord with those of @xcite shown as squares in figure [ ext_wav ] , given the large uncertainties , as well as with the measurement at 15@xmath103 by @xcite , @xmath167 , shown as a circle in figure [ ext_wav ] . @xcite also measured the extinction law at @xmath1 and found it to be consistent with our results . @xcite measured the extinction law from @xmath168 toward omc-1 using the relative intensities of @xmath18 rotation and ro - vibration lines . they fit a functional form assuming a power law for @xmath169 and included the silicate features at @xmath170 and @xmath171 , as well as a water ice feature at @xmath172 . the strength of the @xmath170 silicate feature was determined from a fit to their observations , while the relative strength of the @xmath171 and @xmath170 silicate feature was taken to be @xmath173 , an average of previous results from @xcite , @xcite and @xcite . the shape of the 9.7 feature was chosen to match the trapezium emissivity profile , which is more appropriate for extinction due to dense molecular clouds rather than the narrower @xmath3 cephei profile @xcite . while the assumption of a power law from @xmath4 is clearly not valid , the extinction law from @xmath174 , as measured here and by @xcite and @xcite , is very well fit by a law dominated by silicate emission with the adopted strengths of the two silicate features given by @xcite , as can be seen in figure [ ext_wav ] . the @xmath170 silicate feature has been found to vary along different lines of sight @xcite , particularly on the long - wavelength side of the feature , although there is insufficient data here to draw any conclusions about the universality of the extinction law from @xmath175 . we measured the extinction law from @xmath176 toward five nearby star - forming regions and found it to be relatively internally consistent among different cluster - forming molecular clouds although there is evidence that the ophiuchus extinction law deviates from the other four regions at 5.8 and 8 . the extinction laws we present here differ systematically from the extinction at irac wavelengths derived by @xcite , as seen in both the color excess ratios and the relative extinction ratios . to exclude any possibility that this difference could be due to the differing methodologies used to derive the extinction law , we have also analysed data provided by r. indebetouw using our methods , and continue to find systematic variation . the result could reflect a physical difference in the extinction law between the diffuse ism and molecular clouds , or the differing extinction probes utilized in the two studies . our study of the probes leads us to believe that the major cause of the discrepancy is variation in the diffuse ism vs. molecular cloud extinction laws . the observed difference in the extinction law towards ophiuchus could be caused by diffuse ism behind the molecular cloud that the extinction probes must pass through or differing dust properties between the ophiuchus cloud and the other four clouds studied . the extinction law derived by @xcite agrees more closely with the re - derived @xcite extinction law at wavelengths @xmath177 m . at 3@xmath3 m @xmath178 m , the opposite is true . at @xmath127 the relative extinction measured by @xcite agrees with the cluster results . the extension of our extinction law to @xmath1 was also determined from analysis of _ spitzer _ images of serpens and ngc 2068/71 , but further observations at this wavelength are required to provide statistically significant conclusions . however , our results at @xmath1 , along with the results of @xcite and @xcite , match well the functional form of @xcite which includes extinction dominated by silicate features at 9.7 and 18@xmath103 . future measurements of the extinction law for @xmath174 will help to better constrain this fit , as well as to examine any variation among different lines of sight . allen , l.e . , et al . 2004 , , 154 , 363 allen , l.e . 2007a in protostars and planets v , ed b. reipurth , d. jewitt , & k. keil ( tucson : univ . arizona press ) , 361 allen , l.e . 2007b in prep bertoldi , f. , timmermann , r. , rosenthal , d. , dropatz , s. & wright , c.m . 1999 , , 346 , 267 bessel , m.s . & brett , j.m . 1988 , , 100 , 1134 bowey , j.e . , adamson , a.t . & whittet , d.c.b . 1998 , , 298 , 131 cambresy , l. , jarrett , t.h . , & beichman , c.a . 2005 , , 435 , 131 cardelli , j.a . , clayton , g.c . & mathis , j.s . 1989 , 345 , 245 chiar , j.e . , tielens , a.g.g.m . , whitter , d.c.b . , shuttle , w.a . , boogert , a.c.a . , lutz , d. , van dishoeck , e.f . & bernsteain , m.p . 2000 , , 154 , 363 cohen , m. 1993 , 105 , 1860 del burgo , c. , laurejis , r.j . , brahm , p. & kiss , cs . 2003 , , 346 , 403 draine , b.t . 1989 proc . 22nd eslab symposium on infrared spectroscopy , esa sp-290 , 93 . draine , b.t . & lee h. 1984 , , 285 , 89 dupac , x. , et al . , 2003 , , 404 , l11 evans , n.j . , ii , et al . 2003 , , 115 , 965 fazio , g.g . , et al . 2004 , , 154 , 10 gibb , e.l . , whittet , d.c.b . , boogert , a.c.a . & tielens , a.g.g.m . 2004 , , 151 , 35 giles , k. 1977 , , 180 , 57p glass , i. 1999 , handbook of infrared astronomy ( cambridge university press ) gomez , m. & kenyon , s. 2001 , , 121 , 974 green , j.c . , snow , t.p . , cook , t.a , cash , w.c . & poplawski , o. 1992 , , 395 , 289 gutermuth , r.a . 2005 , phd thesis . gutermuth , r.a . , et al . 2004 , 154 , 374 huard , t. , et al . 2007 in prep indebetouw , r. , et al . 2005 , , 619,931 jiang , b.w . , gao , j. , omont , a. , schuller , f. & simon , g. 2006 , , 446 , 551 kenyon , s. , lada , e. & barsony , m. , 1998 , , 115 , 252 kim , s .- h . & martin , p.g . 1996 , , 462 , 296 knez , c. , et al . 2005 , , 635 , 145 lada , c.j . , et al . 2006 , , 131 , 1574 lada , c.j . , alves , j.f . & lombardi , m. 2007 , in protostars and planets v , ed . b reipurth , d. jewitt , & k. keil ( tucson : univ . arizona press ) , 3 lombardi , m. , alves , j.f . & lada , c.j . 2006 , , 454 , 781 lpez - corredoira , m. cabrera - lavers , a. , garzn , f. & hammersley , p.l . 2002 , , 394 , 883 lutz , d. , et al . 1996 , , 315 , l269 lutz , d. 1999 , in the universe as seen by _ iso _ , ed . p. cox & m.f.kessler ( esa sp-427;noordwijk : esa ) , 623 martin , p.g . & whittet , d.c.b . 1990 , , 357 , 113 megeath , s. t. , gutermuth , r. a. , allen , l.e . , pipher , j. l. , myers , p. c. & fazio , g. g. , 2004 , , 154 , 367 muzerolle , j. , et al . 2004 , , 154 , 379 muzerolle , j. , et al . 2007 in prep nishiyama , s. , et al . 2006 , , 638 , 839 naoi , t. , et al . 2006 , , 640 , 373 pegourie , b. & papoular , r. 1985 , , 142 , 451 pendleton , y. , et al . 2006 , , 208.4915 press , w.h . , teukolsky , s.a . , vetterling , w.t . & flannery , b.p . 1992 `` numerical recipes in c : the art of scientific computing '' , ( cambridge : cambridge university press ) racca , g. , gomez , m. & kenyon , s. , 2002 , 124 , 2178 reach , w.t . , 2005 , , 117 , 978 . rieke , g.h . , et al . 2004 , , 154 , 25 rieke , g.h . & lebofsky , m.j . 1985 , , 288 , 618 rosenthal , d. , bertoldi , f. & drapatz , s. 2000 , , 356 , 705 schnee , s.l . , ridge , n.a . , goodman , a.a . & li , j.g . 2005 , , 634 , 442 stepnik , b. , et al . 2003 , , 398 , 551 storey , j. & hummer , d. 1995 , 272 , 21 teixeira , t. c. & emerson , j.p . 1999 , 351 , 292 volk , k. & kwok , s. 1988 , , 331 , 435 weingartner , j.c . & draine , b.t . 2001 , , 548 , 296 wilson , b.a . , dame , t.m . , masheder , m.r.w . & thaddeus , p. 2005 , , 430 , 523 winston , e. , et al . 2007 in prep whittet , d.c.b . , martin , p.g . , fitzpatrix , e.l . , massa , d. 1993 , , 408 , 573 whittet , d.c.b . , et al . 1997 , , 490 , 729 whittet , d.c.b . , gerakines , p.a . , hough , j.h . & shenoy , s.s . 2001 , , 547 , 872 whitney , b. , wood , k. , bjorkman , j.e . & cohen , m. 2003 , , 598 , 1079 lcccc ngc @xmath55&43&s9.5.0&21.93,21.26,19.08,19.44&12 + ngc @xmath2&43&s9.5.0&21.93,21.26,19.08,19.44&12 + orion a&43&s9.5.0&21.93,21.26,19.08,19.44&12 + serpens&6,174,177&s11.4&21.99,21.26,19.07,19.44&12 + ophiuchus&6,174,177&s11.4&21.99,21.26,19.07,19.44&12 + lcccc ngc @xmath55&@xmath179&@xmath180&@xmath181&@xmath182 + ngc @xmath2&@xmath183&@xmath184&@xmath185&@xmath186 + serpens&@xmath179&@xmath180&@xmath187&@xmath188 + orion a&@xmath189&@xmath190&@xmath191&@xmath192 + ophiuchus&@xmath193&@xmath194&@xmath195&@xmath195 + ind@xmath196data&@xmath197&@xmath198&@xmath199&@xmath200 + ind@xmath196paper&@xmath201&@xmath202&@xmath203&@xmath204 + ngc @xmath55&@xmath205&@xmath206&@xmath207&@xmath208 + ngc @xmath2&@xmath205&@xmath209&@xmath210&@xmath211 + serpens&@xmath212&@xmath206&@xmath207&@xmath207 + orion a&@xmath213&@xmath214&@xmath215&@xmath216 + ophiuchus&@xmath217&@xmath206&@xmath218&@xmath218 + ind@xmath196 data&@xmath219&@xmath208&@xmath220&@xmath221 + ind@xmath196 paper&@xmath222&@xmath223&@xmath224&@xmath225 +

results are presented for the nonequilibrium dynamics of a quantum-spin chain whose spins are initially arranged in a domain wall profile via the application of a magnetic field in the-direction which is spatially varying along the chain . the system is driven out of equilibrium in two ways : a ) . by rapidly turning off the magnetic field , b ) . by rapidly quenching the interactions at the same time as the magnetic field is turned off . the time - evolution of the domain wall profile as well as various two - point spin correlation functions are studied by the exact solution of the fermionic problem for the chain and via a bosonization approach and a mean - field approach for the chain . at long times the magnetization is found to equilibrate ( reach the ground state value ) , while the two - point correlation functions in general do not . in particular , for quenches within the gapless phase , the spin correlation function transverse to the direction acquires a spatially inhomogeneous structure at long times whose details depend on the initial domain wall profile . the spatial inhomogeneity is also recovered for the case of classical spins initially arranged in a domain wall profile and shows that the inhomogeneities arise due to the dephasing of transverse spin components as the domain wall broadens . a generalized gibbs ensemble approach is found to be inadequate in capturing this spatially inhomogeneous state .

1002.4446

recent years have seen remarkable progress in manipulating cold - atom systems @xcite providing us with almost ideal realizations of strongly correlated many - particle systems . cold atom systems also have the unique feature that the interactions between particles and the external potentials that they are subjected to are highly tunable and can be changed rapidly in time , thus driving these systems into highly nonequilibrium states @xcite . this property has in turn motivated a lot of theoretical activity that revolves around studying nonequilibrium time - evolution of strongly correlated systems arising due to a sudden change in some parameter of the hamiltonian referred to as a `` quantum quench '' . the general consensus is that if the strongly correlated system is integrable , its time - evolution from some arbitrary initial state is highly constrained by the initial conditions so that these systems do not reach the ground state but instead reach interesting nonequilibrium time - dependent or time - independent steady states . in this paper we study quenched dynamics in one such integrable model , namely the @xmath0 spin chain . this model shows rich behavior even in its ground state @xcite . for example , it exhibits a quantum critical point at @xmath3 where @xmath4 is the exchange interactions for the @xmath5 components of the spins , and @xmath6 is the exchange interaction for the @xmath1 component of the spins . for @xmath7 , the ground state is an @xmath2 phase characterized by gapless excitations , while for @xmath8 , the spins are in an ising phase where the excitation spectrum has a gap . note that the special point in the gapless phase corresponding to @xmath9 will be referred to as the xx - model , while the term gapless @xmath2 phase will refer to the regime @xmath7 . nonequilibrium dynamics of the @xmath0 model arising due to quenches has been a very active area of research ( see for example @xcite ) . while most previous work has studied quenched dynamics in spatially homogeneous systems , in this work we consider the nonequilibrium time - evolution arising from quenching from an initial state which is spatially inhomogeneous . such an initial state is created by the application of an external magnetic field in the @xmath1 direction that varies in magnitude along the chain , changing its sign at some point so that the spins are aligned in a domain wall pattern . ( see @xcite for an experimental realization of such a setup ) . we study nonequilibrium dynamics that arise in the following two ways : a ) . by a sudden quench of the magnetic field to zero ; b ) . by a sudden quench of the magnetic field to zero along with a quench in the magnitude of the exchange interactions . note that the above are situations where the time evolution is due to a final hamiltonian that has a spatially homogeneous ground state . we would like to understand how the system evolves in time after the quench , and whether the initial inhomogeneity affects the properties of the system at long times . some literature already exists which studies time evolution of a spin chain initially arranged in a domain wall profile . antal et al @xcite study the time evolution of a sharp domain wall in the exactly solvable @xmath2 model and find ballistic broadening of the domain wall width . similar studies for the time - evolution of a domain wall has been carried out for the xx model @xcite and the critical transverse spin ising model @xcite . the numerical method time - dependent density matrix renormalization group ( t - dmrg ) has been used to study the time - evolution of a domain wall both in the gapless phase as well as the gapped ising phase of an @xmath0 chain @xcite . their study reveals qualitatively different behavior in the two phases , with ballistic broadening of the domain wall in the gapless phase , and more complicated non - ballistic behavior in the ising phase . in addition conformal field theory methods @xcite have been used to study both the time - evolution of the domain wall profile as well as two - point correlation functions for the ising chain . studies also exist on the time - evolution of other quantities such as the entanglement entropy after a quench from a spatially inhomogeneous initial state @xcite . thus what is lacking in the literature is the study of the time evolution of two - point correlation functions for the @xmath0 chain from an initial inhomogeneous state . the advantage of studying these quantities is that they are often more sensitive to nonequilibrium initial conditions than averages of local quantities such as the magnetization . in this paper we extend previous results by studying the time evolution of _ both _ the local magnetization as well as two point correlation functions related to the longitudinal ( along @xmath1 direction ) and the transverse spin correlation function for the @xmath0 chain . in addition we use a variety of theoretical methods such as bosonization , exact solution of the fermionic problem as well as a classical treatment of spins which allows us to compare the relative merits of these approximations . consistent with previous results , we find that the magnetization always equilibrates ( i.e , reaches the ground state value ) . in addition the details of how the domain wall profile evolves in time depend on whether the system after the quench is in the gapless phase or the gapped ising phase @xcite . for example in the gapless phase the domain wall is found to spread out ballistically , however in the gapped phase the dynamics is much more complicated showing oscillations and revivals at short time scales . in studying the time - evolution of two - point correlation functions we obtain the interesting new result that the transverse spin - correlation function shows a lack of equilibration reaching a nonequilibrium steady state that is also spatially inhomogeneous . the length scale of the spatial inhomogeneity is found to depend on the details of the initial domain wall profile and is found to arise due to the dephasing of transverse spin components as the domain wall broadens . this result is qualitatively different from those obtained for the ising chain initially arranged in a domain wall profile where no residual inhomogeneity was observed @xcite . the paper is organized as follows . in section [ xxfermions ] we study the xx model which is initially subjected to a spatially varying magnetic field so that the spins are arranged in a domain wall profile . we study the nonequilibrium dynamics that arises when this magnetic - field is suddenly switched off . these results are obtained for two cases . one for an initial external magnetic field that varies linearly in position and is equivalent to an effective electric field on the fermions . the second case is that of an initial magnetic field which has a sharp step - function profile . in section [ magxxzbos ] the effect of quenching only a spatially varying magnetic field in the xx model is studied via a bosonization approach . the results of section [ xxfermions ] and [ magxxzbos ] are found to be in qualitative agreement and in particular recover ballistic domain wall motion and spatial inhomogeneities in the transverse spin correlation function . a physical explanation for this inhomogeneity is provided in section [ cleom ] where the problem of quenching a spatially varying magnetic field is studied classically and a spin - wave pattern is obtained as a long time solution of a landau - lifshitz equation . in section [ magintxxz ] , the effect of simultaneously quenching the magnetic field and interactions is studied using a bosonization approach . in section [ magintxx ] results are presented for the case when the quench is entirely within the gapless xx phase . the spatial inhomogeneities in the correlation functions are recovered along with the nonequilibrium exponents obtained earlier in a purely homogeneous interaction quench @xcite . in section [ magintsc ] the effect of quenching into the gapped ising phase is studied via a semiclassical approach . here we find the domain wall dynamics to be qualitatively different , however at long times we find that all inhomogeneities eventually decay away . the latter result may very well be an artifact of the semiclassical approach which neglects creation of solitons . the results of the semiclassical approach is complemented by a mean - field treatment in section [ meanfield ] where we address the question of whether ising order can develop for a system initially in a domain wall state . finally we conclude in section [ conclusions ] . in this section we will study the xx model with a magnetic field in the @xmath1 direction that varies linearly in position and changes sign at the center of the spin chain , so that the ground state is a domain wall configuration . in subsection [ dwstate ] we will study the properties of this domain wall state , and in subsequent subsections study the nonequilibrium time - evolution arising when the the spatially varying magnetic field is suddenly switched off . our starting point is the @xmath2 hamiltonian in a magnetic field , @xmath10,\ ] ] where @xmath11 , @xmath12 are pauli matrices , @xmath13 is an external magnetic field aligned along the @xmath14 direction , and @xmath4 is the exchange energy . for a uniform magnetic field one commonly employs the jordan - wigner transformation @xcite to map the spin system to a system of spin - less non - interacting fermions which can be easily diagonalized ( for details see appendix [ diagonol ] ) . to create an inhomogeneous state we will consider a linearly varying field @xmath15 , where @xmath16 is a lattice spacing . note that the ground state properties of spin - chains in various spatially varying magnetic fields has been studied in @xcite . after the jordan - wigner transformation eq . [ eq : xyham ] for @xmath15 maps to the wannier - stark problem @xcite of electrons in a lattice subjected to a constant force ( or electric field ) @xmath17 , @xmath18 + \sum_{j}j \mathcal{f}ac_{j}^{\dagger}c_{j } \label{hxxjw}\ ] ] this hamiltonian can be diagonalized as follows , @xmath19 where @xmath20 and @xmath21c^{\dagger}_{k } \label{eq : etaj1}\end{aligned}\ ] ] above @xmath22 , @xmath23 is the total number of sites and @xmath24 is a bessel function of the first kind . we construct the ground state @xmath25 by including all negative energy states , @xmath26 this is a domain - wall state with a characteristic width @xmath27 to the left ( right ) of which all spins are up ( down ) . this can be seen by evaluating the magnetization at the position @xmath28 @xmath29 which on rewriting @xmath30 in terms of the @xmath31 operators and evaluating the expectation value is found to be @xmath32 upon noting that the support of @xmath24 is restricted to @xcite @xmath33 , and using the identity @xcite @xmath34 , it is easy to see that this indeed describes a domain - wall state of width @xmath35 . . [ mz ] is also plotted in fig . [ domainwall ] . ( color online ) plot of the initial magnetization for various values of the domain wall width @xmath36.,width=340 ] it is worth noting that the case of @xmath37 corresponds to an infinite `` electric field '' @xmath38 that forces a sharp wall of zero width . this case was investigated by antal _ et al_@xcite . we will extend some of their results to more general domain walls , and also study spin - spin correlation functions . the basic spin correlation functions that we will compute are , @xmath39 where we will refer to @xmath40 ( @xmath41 ) as the longitudinal ( transverse ) spin correlation function . these may be expressed in terms of majorana operators @xcite @xmath42 , and @xmath43 as follows , @xmath44 and @xmath45 the above correlations can be evaluated by rewriting @xmath46 in terms of @xmath47 and applying wick s theorem . the basic contractions are @xcite @xmath48 , \\ & & \left\langle a_{j}a_{j+n } \right\rangle = - \left\langle b_{j}b_{j+n } \right\rangle = \delta_{n=0}.\end{aligned}\ ] ] using some bessel function identities , the mixed contraction can be simplified to @xmath49.\end{aligned}\ ] ] since we encounter no contractions of the form @xmath50 , the function @xmath51 may be written@xcite as a toeplitz determinant which is a computationally cheap task . in the next sub - section when we evaluate the correlation functions after a quench of the magnetic field , we will find @xmath52 in the nonequilibrium state . in this case , the evaluation of @xmath41 will require computing the square root of a determinant @xcite . the correlation functions @xmath40 and @xmath41 for @xmath53 are given in fig . [ fig : initczz ] and fig . [ fig : initcxx ] respectively . the main point to notice here is that close to the center of the domain wall where the external magnetic field is small , the correlations tend to mimic those in the ground state of the homogeneous ( @xmath54 ) @xmath2 model . for example , the nearest neighbor longitudinal spin correlation function in the absence of a magnetic field is @xcite @xmath55 . this is precisely the value that the @xmath56 correlation function takes in fig . [ fig : initczz ] at the center of the domain wall . ( color online ) plot of @xmath57 for @xmath58 and @xmath59.,width=340 ] ( color online ) plot of @xmath51 for @xmath58 and @xmath60.,width=340 ] we will now explore the dynamics that arises when the spatially varying magnetic field is suddenly switched off at @xmath61 _ i.e. _ , @xmath62 . thus at @xmath63 the many - body wavefunction of the system is eq . [ eq : initstate ] , while for @xmath64 , as shown in appendix [ diagonol ] , the wave - function evolves according to the hamiltonian , @xmath65 where @xmath66 . the time - dependent magnetization is given by @xmath67 the time - evolution of @xmath68 in terms of the @xmath31 operators is given in eqns . [ cjtim1 ] , [ cjtim2 ] from which we obtain @xmath69 where @xcite @xmath70 evaluating the integral , we find @xmath71 this corresponds to a wall whose width @xmath72 increases linearly and hence ballistically in time with a velocity of @xmath73 for times @xmath74 . setting @xmath37 , we recover the result of antal _ et al _ @xcite . note that the time - evolution of the entanglement for general @xmath35 has been studied in @xcite . we now turn to the evaluation of the longitudinal and transverse correlation functions given in eqns . ( [ eq : czzm ] ) and ( [ eq : cxxm ] ) at a time @xmath75 after the quench . the basic contractions that we need are @xmath76 and @xmath77 for which we find the following expressions @xmath78,\end{aligned}\ ] ] @xmath79,\end{aligned}\ ] ] with @xmath80 and @xmath81 . as before , we have @xmath82 and @xmath83 . the transverse correlation function @xmath84 is given by @xmath85 where @xmath86 is the anti - symmetric matrix @xmath87 with the @xmath88 sub - matrices defined by @xmath89 note that both @xmath90 and @xmath91 are anti - symmetric , and thus defined by the elements with @xmath92 , while no such restriction is placed on @xmath93 . the numerical evaluation of the above expression for @xmath94 is shown in fig . [ fig : finalxcorr ] for three different times : @xmath61 , an intermediate time and for long times where one finds the appearance of a spatially oscillating pattern at the scale of the fermi - wave vector @xmath95 . in order to get some insight into the long time behavior of the system , we consider the limit of @xmath96 while imposing the condition @xmath97 on the two position indices of the correlation functions . this limit corresponds to a very broad domain wall which may arise either due to a very weak electric field @xmath98 at arbitrary times @xmath75 ( and hence arbitrary @xmath99 ) or could arise on waiting long enough after a quench so that an initial narrow domain wall has had enough time to broaden . the condition on the position indices imply that we are looking at correlations in the vicinity of the center of the domain wall where the magnetization has locally equilibrated . using the asymptotic expansion for the bessel functions with large arguments @xcite , we find @xmath100 substituting the asymptotic expansions in eq . [ asm1 ] , eq . [ asm2 ] we find , @xmath101 where @xmath102 is the equilibrium result ( _ i.e. _ the result in the ground state of @xmath103 ) for the @xmath104 correlation function @xcite . employing the asymptotic expansions in the evaluation of @xmath105 , we have verified through @xmath106 that @xmath107 where @xmath108 is the equilibrium value of the @xmath109-correlation function @xcite which has the asymptotic form @xcite , @xmath110 with @xmath111 . ( color online ) time evolution of @xmath112 correlation function for @xmath113 . solid line is the equilibrium correlation function @xmath114.,width=340 ] the appearance of these oscillations at a wave - length of @xmath115 where @xmath116 is indeed intriguing . to understand better what sets this length scale , we study more general initial domain walls . the one considered so far has the maximum possible positive ( negative ) polarization of @xmath117 on the left ( right ) ends of the domain wall , with a width of the domain wall controlled by @xmath118 . antal et al @xcite have shown how to construct domain walls of zero width ( @xmath119 ) but arbitrary polarization @xmath120 at the two ends . starting with this initial state _ i.e , _ @xmath121 we study the time - evolution under the @xmath2 model and extend the results of antal _ et al _ to the study of the transverse spin correlation function . details of the computation are given in appendix [ m0vary ] . in the long time limit of @xmath122 , analytic expressions for contractions in eq . [ asm1 ] , [ asm2 ] generalize to eq . [ asmgen ] . these can then be used to construct the determinant of @xmath123 ( eq . [ detc ] ) required for the computation of @xmath41 . the results for @xmath41 in the long time limit for several different domain wall heights @xmath124 are shown in fig . [ fig : cxxperiod ] . indeed we see the appearance of a spatial oscillation where the wavelength depends on the height @xmath124 of the domain wall as follows , @xmath125 thus the main results of this section on the effect of quenching a spatially varying magnetic field in a xx model are as follows : a ) . an initial domain wall spreads out ballistically after the magnetic field quench so that the magnetization equilibrates to its homogeneous ground state value of @xmath126 . for a sharp domain wall , ( @xmath127 ) , the magnetization at a distance @xmath128 from the center of the domain wall begins to equilibrate after a time @xmath129 . this is an example of the horizon effect @xcite , namely the minimal time required for ballistically propagating excitations from the center of the domain wall to reach the observed point ; b ) . another consequence of the horizon effect is that any two point correlation function @xmath130 is found to change significantly only after times @xcite @xmath131 for @xmath132 ; c ) . the @xmath133 correlation function eventually equilibrates ( _ i.e. _ reaches the ground state value ) at long times where the approach to equilibrium is a power - law of @xmath134 ; d ) . the @xmath105 correlation function reaches a nonequilibrium steady state at times @xmath135 to a value given by eq . [ cxxansgen ] which is basically the ground state correlation function with oscillations at a wave - length @xmath136 superimposed on it , where @xmath124 is the height of the domain wall , _ i.e. _ , the magnitude of the maximum polarization at its two ends . note that the effect of quenching from an initial domain wall state was also studied in @xcite for the critical transverse spin ising model ( @xmath137 ) using the methods of boundary conformal field theory . our results for the @xmath2 model differ from those of @xcite in two important ways . one is that at long times we recover a spatial spin - wave pattern that was not captured in @xcite . secondly at long times after the quench , our two - point correlation functions continue to be critical ( i.e. , have a power - law decay ) whereas the results of @xcite point to thermal behavior with exponential decay in the two - point correlation function . the differences between these results are due to the very different universality classes and conservation laws of the @xmath2 model and the transverse spin ising model . for example , while the total magnetization in the z - direction is conserved in the @xmath2 model , it is not conserved in the ising model . as a result the magnetization has a nontrivial time evolution in the ising model even for points outside the domain wall horizon , which eventually gives rise to a thermal behavior . in addition , as we shall show in section [ cleom ] , the appearance of the spin - wave pattern in the transverse spin correlation functions in the @xmath2 model is due to precession of spins in an effective magnetic field created by the initial domain wall . this precessional physics is absent for ising spins . we would also like to point out that the spatial oscillations found by us have also been observed in a different physical setting by rigol _ @xcite who have studied inhomogeneous quantum quenches for one - dimensional hard - core bosons . via a jordan - wigner transformation hard - core bosons in one dimension can be mapped onto a system of spinless fermions . the physical situation considered by rigol _ et al _ was one where an initial strong confining potential produces a local region of high density . ( in the spin language this would correspond to creating a pile up of magnetization in the center of a spin - chain ) . the time evolution of the system when this confining potential was suddenly switched off was studied . the initial inhomogeneity was found to move out ballistically in two opposite directions . moreover , the analog of the transverse spin correlation function which in the boson language corresponds to the off - diagonal matrix element was found to show spatial oscillations in the two out - going lobes . ( this is somewhat different from our set - up where the spatial oscillations appear at the center of the chain and extend over longer portions of the chain as the domain wall broadens ) . for an initial tight confinement , rigol _ et al _ found that the oscillations in the lobes appeared at precisely the same wave - length of @xmath115 that we find here for a domain wall of height @xmath138 . for the case of bosons , these oscillations had the physical interpretation of the appearance of quasi - condensates at the wave - vector @xmath139 . a similar observation that an initial spatial inhomogeneity for a system of hard core bosons can lead to correlations in momentum space was also made by gangardt _ @xcite . in subsequent sections we will study the effect of quenching a spatially varying magnetic field with and without an interaction quench in a xxz chain using a bosonization approach . for quenches within the gapless xx phase , results of this section , namely critical behavior in the two - point correlation function even after the quench , and the presence of spatial inhomogeneities will be recovered . in addition we will find that the wavelength of the inhomogeneity depends on the strength of the exchange coupling @xmath6 ( and hence the luttinger interaction parameter ) . ( color online ) long time behavior of @xmath140 for different values of the domain wall height @xmath141 defined as @xmath142 . the solid line is the equilibrium correlation function @xmath114 . , width=340 ] in this section we will study a general xxz chain which is subjected to a time - dependent spatially varying magnetic field . the most convenient way to study this problem is via the bosonization approach . we first review some of the notation . the hamiltonian for the @xmath0-spin chain in a magnetic field is @xmath143,\ ] ] the above can be mapped onto a luttinger liquid with a back - scattering potential @xcite , @xmath144 \nonumber \\ & & + \frac{1}{\pi}\int dx h(x)\nabla\phi- \frac{2g}{(2\pi\alpha)^{2}}\int dx \cos(4\phi(x)),\end{aligned}\ ] ] where @xmath145 , and the fields @xmath146 and @xmath147 are defined in terms of the boson creation / annihilation operators as @xmath148 where @xmath149 , and @xmath150 denotes normal ordering of operators . in the continuum limit , we interpret @xmath151 for some integer @xmath152 and lattice spacing @xmath16 . the coupling - parameter relationships are @xmath153.\end{aligned}\ ] ] for an @xmath154 anti - ferromagnet ( @xmath155 ) , the mapping from spin language to boson language is given by @xcite @xmath156,\\ \hat{s}^{+}(x ) & = & \frac{1}{\sqrt{2\pi\alpha}}\exp[-i\theta(x)]\left[(-1)^{x } + \cos[2\phi(x)]\right].\end{aligned}\ ] ] in equilibrium and zero magnetic fields , the magnetization and basic physical correlations in the gapless anti - ferromagnetic @xmath2 phase are @xmath157 where the @xmath158 are non - universal constants . the antiferromagnetic hamiltonian may be transformed to a ferromagnetic hamiltonian by the transformation @xmath159 , resulting in ferromagnetic operators @xmath160,\\ & & \hat{s}^{+}(x)_{\mbox{\scriptsize f.m . } } = \frac{1}{\sqrt{2\pi\alpha } } \exp[-i\theta(x)]\left[1 + ( -1)^{x}\cos[2\phi(x)]\right]\nonumber \\\end{aligned}\ ] ] as a consequence , the magnetization and spin correlations in the gapless ferromagnetic @xmath2 phase are , @xmath161 in this section we will study the time - dependent hamiltonian @xmath163 we choose a point in the @xmath2 phase corresponding to the @xmath2 model @xmath162 and @xmath164 . thus , @xmath165 so that the initial state at @xmath166 is the spatially inhomogeneous ground state of @xmath167 . the magnetic field @xmath168 is suddenly switched off at @xmath61 so that the system evolves according to @xmath169 since eq . [ hi1 ] basically represents harmonic oscillators subjected to an electric field , it may be easily diagonalized by the shift , @xmath170 where , @xmath171 and @xmath172 is the fourier transform of @xmath168 . taking @xmath173 ( corresponding to a magnetic field that is antisymmetric in space allowing for a domain wall solution ) , this gives @xmath174 for some time @xmath75 after the quench , it is straightforward to show that @xmath175 where @xmath176 since we will be computing expectation values with respect to an initial state which is the ground state of the @xmath177 operators , @xmath178 will just return the equilibrium results . thus the effect of the shift and hence the initial magnetic field is contained entirely in @xmath179 \cos[px]h_{p } , \label{dphi}\\ \delta\theta(x , t ) & = & \frac{-i}{\pi v_{f}}\int_{0}^{\infty}\frac{dp}{p}\sin[pv_{f}t]\sin[px]h_{p}. \label{dtheta}\end{aligned}\ ] ] for any antisymmetric @xmath168 the above reduces to , @xmath180\right . \nonumber \\ & & \left.+\mbox{sgn}[x'+z_{+}]+\mbox{sgn}[x'-z_{+}]+\mbox{sgn}[x'-z_{-}]\right],\\ & & \delta\theta(x , t)=\frac{1}{8v_{f}}\int_{-\infty}^{\infty}dx'h(x ' ) \left[\mbox{sgn}[x'+z_{+}]\right.\nonumber \\ & & \left.+\mbox{sgn}[x'-z_{+}]-\mbox{sgn}[x'+z_{-}]-\mbox{sgn}[x'-z_{-}]\right],\end{aligned}\ ] ] where @xmath181\equiv x/|x|$ ] , and @xmath182 . the above implies that the magnetization before the quench is @xmath183 thus within the bosonization approach , the magnetization follows the magnetic - field profile . the magnetization at some time @xmath75 after the magnetic - field quench is @xmath184 as anticipated , the domain wall spreads out with speed @xmath185 in both directions . this result is consistent with the ballistic time - evolution found in section [ xxfermions ] from the exact solution of the fermionic problem . next we turn to the evaluation of the @xmath41 correlation function which from the exact solution of the fermionic model was found to be @xmath186 where @xmath187 was the equilibrium correlation function , and the wavelength of the oscillation was found to be @xmath188 , @xmath124 being the height of the domain wall ( or the magnitude of the maximal magnetization at the ends of the domain wall ) . in the bosonization approach , @xmath189 \frac{1}{\sqrt{n}}.\end{aligned}\ ] ] let us suppose @xmath190 implying @xmath191 then we find , @xmath192\cosh[(x+n+v_ft)/\xi ] } { \cosh[(x+n - v_ft)/\xi]\cosh[(x+v_ft)/\xi]}\right]\nonumber \\ & & \xrightarrow{v_ft , |v_ft \pm x| , |v_f t \pm ( x+n)|\gg \xi } -\frac{\xi h_{0}}{2v_{f}}\left[\frac{|x - v_ft|}{\xi } \right . \nonumber \\ & & \left . + \frac{|x+n+v_ft|}{\xi } -\frac{|x+n - v_ft|}{\xi}-\frac{|x+v_ft|}{\xi}\right]\end{aligned}\ ] ] for @xmath193 , _ long after the moving domain wall front has crossed the two observation points at @xmath194 and @xmath195 , we find @xmath196 so that @xmath197 thus we find that for an initial domain wall that has the profile @xmath198 the transverse spin correlation function at long times do not equilibrate but acquire a spatial oscillation at the wavelength @xmath199 this result is identical to that obtained from the exact solution of the fermionic problem ( eqns . [ cxxansgen ] , [ lamgen ] ) . note that the wavelength of the spatial oscillation is set by the height of the domain wall @xmath124 and is independent of the width @xmath200 ( or @xmath201 in the fermionic problem ) . for the case of a linearly varying magnetic field studied in the previous section , the domain wall magnetization approached its maximum possible value @xmath202 at its two ends . this in the boson language corresponds to @xmath203 so that writing @xmath204 in eq . [ s+-bos ] , where @xmath205 is an integer , @xmath206 giving us precisely the observed behavior of eq . [ cxxans ] . for the evaluation of the @xmath207 correlation function , we need to evaluate @xmath208 $ ] . we find , @xmath209 \cosh\left[\frac{x+n - v_ft}{\xi}\right]}{\cosh\left[\frac{x+v_ft}{\xi}\right ] \cosh\left[\frac{x - v_ft}{\xi}\right]}\right]\\ & & \xrightarrow{v_ft\gg \xi , |x|,|x+n| } 0 \nonumber\end{aligned}\ ] ] the above result together with the fact that @xmath210 implies that for @xmath211 , the @xmath40 correlation function equilibrates which is in agreement with the results in section [ xxfermions ] . to summarize the results of this section : a ) . within the bosonization approach , an applied magnetic - field produces a spin pattern that follows the local magnetic field . b ) . after a quench , the domain wall spreads out ballistically . the magnetization equilibrates ( in this case vanishes to zero everywhere ) . the @xmath40 correlation function equilibrates , whereas the @xmath41 correlation function acquires spatial oscillations at wavelength @xmath188 where @xmath124 is the height of the initial domain wall defined as @xmath212 . these results are in agreement with the exact results of section [ xxfermions ] obtained from solving the fermion lattice problem . thus bosonization is good at capturing the behavior at long distances and times , even for the nonequilibrium problem . as expected it misses some of the details of the short distance physics both in the static properties ( as shown in fig . [ fig : comparison ] ) , as well as during the time - evolution such as the existence of a complex internal structure in the propagating domain wall front @xcite . ( color online ) plot of the magnetization obtained from the bosonization approach with @xmath213 , and from the fermionic lattice model with @xmath214.,width=340 ] in the next section we provide an explanation for the inhomogeneities observed in the transverse spin correlation function by solving the classical equations of motion for the spins . the hamiltonian in eq . [ eq : xyham ] implies the following equations of motion for the spins @xmath215\nonumber\\ \label{szeom}\\ & & \frac{d s^x_j}{dt } = -j s^z_j\left(s^y_{j+1 } + s^y_{j-1}\right ) - h_j s^y_j \label{sxeom}\\ & & \frac{d s^y_j}{dt } = j s^z_j\left(s^x_{j+1 } + s^x_{j-1}\right ) + h_j s^x_j \label{syeom}\end{aligned}\ ] ] we will treat the spins as classical variables , and solve the above equations for a sudden quench of a spatially inhomogeneous magnetic field , @xmath216 . first notice that for @xmath217 , after performing the course - graining @xmath218 , the following spin profile satisfy the equations of motion , @xmath219 with the above as an initial condition , the dynamics for @xmath64 can be easily studied numerically and we find a ballistic spreading out of the domain wall profile ( see fig . [ semicl ] ) , and the appearance of a spin - wave pattern . the latter result , _ i.e. _ , the appearance of an inhomogeneous pattern at long times can be verified analytically rather easily as follows . suppose , @xmath220 . then for @xmath64 , if we assume a ballistic broadening of the domain wall , @xmath221 \label{szguess}\end{aligned}\ ] ] writing eq . [ sxeom ] , [ syeom ] in terms of @xmath222 , and performing a course - graining , one obtains the following equation for @xmath223 @xmath224 using eq . [ szguess ] , eq . [ s+eom ] may be easily solved to obtain @xmath225\end{aligned}\ ] ] thus for @xmath226 we find the spin - wave pattern @xmath227 in agreement with the results of the previous sections . thus the appearance of the spatial inhomogeneity may be understood as follows . right after the magnetic field is switched off , the spins in the xy plane begin to precess due to a non - zero magnetization in the z - direction . however this precession only lasts for as long as it takes the domain wall profile to flatten out to zero . since this time is different for spins located at different spatial positions , the spins locally dephase with respect to each other and arrange themselves in a spin - wave pattern . ( color online ) thick line : time evolution of the magnetization from the solution of the classical landau - lifshitz equation assuming that @xmath228 is given by eq . thin lines : time evolution from the exact solution of the quantum problem.,width=340 ] we will now turn our attention to the case where as the magnetic field is switched off at @xmath61 , the interactions are also turned on so that the luttinger parameter is @xmath229 . we will first consider the case where the ground state both before and after the quench corresponds to a gapless xx phase , so that the @xmath230 term can be neglected . we wish to begin with a non - interacting system in an inhomogeneous magnetic field , @xmath231 and quench to zero field , while turning on interactions so that @xmath232 the @xmath233 operators have a simple time evolution and may be related to the @xmath234 operators by a bogoliubov rotation @xcite , @xmath235 where @xmath236 . we define , @xmath237 and @xmath238 . performing the shift @xmath239 as before , we write the @xmath240 and @xmath241 fields as @xmath242 , where @xmath243 the @xmath244 arise due to the spatially varying magnetic field and are given by , @xmath245 where @xmath246 are given in eq . [ dphi ] and [ dtheta ] respectively . it is straightforward to show that the magnetization at a time @xmath75 after the quench is @xmath247\nonumber\\\end{aligned}\ ] ] thus even with the interaction quench , the domain wall spreads out ballistically , but with the renormalized velocity @xmath248 . to compute the correlation functions , some basic expressions we need are @xmath249^{2}\right\rangle\\ g_{\theta\theta } = \left\langle [ \theta^{a}_{\beta}(x , t ) - \theta^{a}_{\beta}(x',t)]^{2}\right\rangle\end{aligned}\ ] ] note that @xmath250 is related to @xmath251 by @xmath252 the computation of which is identical to that already presented in @xcite . for completeness we present the results below , @xmath253\nonumber \\ & & \times ( 1-\cos[p(x - x ' ) ] ) e^{-\alpha p}\end{aligned}\ ] ] which eventually gives @xmath254^{2 } [ \alpha^{2}+x^{2}]^{2}}{\alpha^{4}[\alpha^{2}+(2ut + x)^{2}][\alpha^{2 } + ( x-2ut)^{2}]}\right ] \\ & & g_{\theta\theta } = g_{\theta\theta}^{(0 ) } + \left(\frac{k^{-2}-1}{8}\right ) \times \nonumber \\ & & \log\left[\frac{[\alpha^{2}+(2ut)^{2}]^{2 } [ \alpha^{2}+x^{2}]^{2}}{\alpha^{4}[\alpha^{2}+(2ut + x)^{2}][\alpha^{2 } + ( x-2ut)^{2}]}\right].\end{aligned}\ ] ] where @xmath255 $ ] , @xmath256 $ ] . taking the @xmath257 limit , we can easily obtain the correlator asymptotics . the memory of the initial domain wall profile appears as oscillatory pre - factors of the form @xmath258 $ ] and terms such as @xmath259 . for the specific case of @xmath260 we find at long times , @xmath261 where @xmath262 the case of the pure interaction quench ( @xmath263 ) was studied in @xcite where the above anomalous power - laws corresponding to an algebraic decay which is faster than in equilibrium ( since @xmath264 for all real @xmath265 ) were obtained . from eq . [ s+-int ] we find that the effect of an initial domain wall profile is to superimpose spatial oscillations onto this decay , where the wavelength of the oscillations depends on the interaction parameter @xmath265 and is found to be @xmath266 as expected , increasing the exchange interaction @xmath6 increases the effective magnetic field seen by the precessing spins , thus decreasing the wave - length of the spin wave pattern . rigol _ et al _ have made the interesting proposal that the long - time properties of out of equilibrium integrable systems may be captured by a generalized gibbs ensemble that enforces the constraints imposed by the integrals of motion @xcite . this approach has been applied with success by iucci and cazalilla @xcite who studied homogeneous interaction quenches in models similar to those studied in this paper . it has also been successfully applied to models of free bosons @xcite , and for some general family of integrable models @xcite . an interesting question has been to identify situations in which the gibbs ensemble argument might break down @xcite . we find that for an inhomogeneous quantum quench starting from an initial domain wall state , the gibbs ensemble argument can not capture the spatial inhomogeneities at long times after the quench . in this section we review the gibbs ensemble argument and show why this does not work for our case . consider a quantum quench of the form @xmath267 , so that for @xmath166 the system is in the ground state of @xmath167 , whereas at @xmath64 , the wave - function begins to evolve according to @xmath103 . let us further assume that both @xmath167 and @xmath103 can be diagonalized @xmath268 and that the two hilbert spaces may be related to each other via a canonical transformation . for the case of bosons a general canonical transformation is of the form , @xmath269 where @xmath270 is a linear shift and @xmath271 is a rotation angle . the essential idea behind predicting the long - time behavior is that an integrable model such as @xmath103 has a conserved quantity for every degree of freedom . for the example given above , the conserved quantity is trivially given by @xmath272 . thus during the time - evolution @xmath273 should be conserved . since the initial state is the ground state of @xmath167 for which @xmath274 for @xmath275 , it follows that the gibbs ensemble should be characterized by the distribution function @xmath276 for the particular models studied here , we find the initial spatial inhomogeneity is captured by the linear shift @xmath270 ( given by eq . [ lambda ] ) , while the angle @xmath271 captures the homogeneous changes in the interaction parameter . since in our example @xmath277 vanishes in the thermodynamic limit , any effect of the initial spatial inhomogeneity is lost . thus while the gibbs ensemble captures the effect of the rotation , and therefore correctly predicts the new power law exponents of @xcite , it misses the spin - wave pattern at long times . in particular the gibbs ensemble prediction for the transverse spin correlation function is given by @xmath278 which differs from the correct answer ( eq . [ s+-int ] ) by the absence of the spatially oscillating factors @xmath279 . in this section we will study the effect of quenching the external magnetic field and the interactions into the massive ising phase where the @xmath280 term is strongly relevant . in order to study the time evolution , we will employ a semi - classical approximation which corresponds to replacing @xmath281 . thus for this case the initial wave - function is the ground state of @xmath282,\\ & = & \sum_p v_f |p| a_{p}^{\dagger } a_p\end{aligned}\ ] ] while for @xmath64 the wave - function evolves according to @xmath283 \\ & = & \sum_p \omega_p \eta_p^{\dagger } \eta_p\end{aligned}\ ] ] where @xmath284 and @xmath285 . the calculations mirror that of the previous section . writing @xmath286 , we find for a time @xmath75 after the quench @xmath287 , \\ & & \theta^a(x , t)= \frac{i\pi}{l}\sum_{p\neq0}\left(\frac{l|p|}{2\pi}\right)^{1/2 } \nonumber \\ & & \times \frac{1}{|p|}e^{-\alpha|p|/2-ipx}\left[a_{p}^{\dagger } \left(\cos(\omega_p t ) + i\frac{\omega_p}{ku|p| } \sin(\omega_p t)\right ) \right . \nonumber \\ & & \left . - a_{-p } \left(\cos(\omega_p t ) - i \frac{\omega_p}{k u |p| } \sin(\omega_pt)\right ) \right]\end{aligned}\ ] ] and @xmath288\cos[px]h_{p } , \label{dphi1}\\ \delta\theta(x , t ) & = & \frac{-i}{\pi k v_{f}}\int_{0}^{\infty}\frac{dp}{p}\sin[\omega_p t ] \sin[px]\left(\frac{\omega_p}{u|p|}\right)h_{p } \nonumber \\\label{dtheta1}\end{aligned}\ ] ] the magnetization at a time @xmath75 after the quench is given by @xmath289 . due to the presence of infra - red divergences , we find that for all times after the quench , @xmath290 so that ising order never develops . this is consistent with the results of @xcite and may be an artifact of the semi - classical approximation . in section [ meanfield ] we will compare this result with a mean - field treatment for the order - parameter dynamics . thus , within the validity of the semiclassical approximation we find the following time evolution for the initial domain wall profile , @xmath291 the above is plotted in fig . [ sztgap ] for several different masses . the domain wall profile shows a more complex time evolution than for the quench in the gapless phase where a simple ballistic broadening of the domain wall occurs . in the ising phase instead we find oscillations of the local magnetization at the frequency of the mass . in addition the domain wall spreads out less and less as the mass increases , a result which is in agreement with a t - dmrg study of the xxz chain @xcite . note however that we find oscillations of the magnetization outside the light - cone at the time - scale of @xmath292 . this is due to the nature of the klein - gordon equation @xmath293 which has solutions of the form @xmath294 . the t - dmrg results on the other hand found oscillations only within the light cone . we also find that the magnetization eventually equilibrates at long times . [ sztgap1 ] shows the time evolution for the magnetization for a given point in position space . at long times eq . [ szgap ] may be evaluated in a stationary phase approximation which gives , @xmath295 showing that the magnetization decays locally as a power - law . contour plot for @xmath296 after a `` semi - classical '' quench for several different masses.,width=340 ] ( color online ) the magnetization decays as @xmath297 with oscillations of period @xmath298 . here , @xmath299.,width=264 ] here we give results for some typical spin correlation functions . again , due to the presence of infra - red divergences , the transverse - spin correlation function is found to vanish at all times after the quench , @xmath300 on the other hand the @xmath301 correlation function is found to be @xmath302 for a purely interaction quench ( @xmath303 ) we find a power law decay in the ising phase with an exponent which is half of that in the initial gapless phase , a result which was also found by @xcite for homogeneous quenches . the initial domain profile superimposes the additional structure @xmath304 } \end{aligned}\ ] ] thus at long times , the spatial oscillations due to the initial domain - wall eventually decay to zero . in this section we will use a mean - field argument to explore whether after quenching both a magnetic field as well as interactions , local anti - ferromagnetic ( af ) ising order can develop . thus our initial wave - function corresponds to the ground state of the following hamiltonian @xmath305\end{aligned}\ ] ] with @xmath306 , and we will assume @xmath15 . while , as a result of the quench the wave - function evolves in time due to @xmath307 performing a jordan - wigner transformation followed by the change of variables @xmath308 , @xmath167 becomes identical to the hamiltonian ( eq . [ hxxjw ] ) already studied in section [ dwstate ] , whereas the hamiltonian after the quench is , @xmath309 thus the initial state which is the ground state of @xmath167 ( given by eq . [ eq : initstate ] ) evolves in time according to eq . [ hfaf ] . in what follows we will study this time evolution using a mean - field approximation . defining the af order - parameter at a time @xmath75 , @xmath310 as @xmath311 , eq . [ hfaf ] within a mean - field approximation becomes @xmath312 where @xmath313 , and @xmath310 is determined from the self - consistency condition @xmath314 of being in the adiabatic eigenstates of @xmath315 , the self - consistency condition becomes @xmath316 \label{deltamf}\end{aligned}\ ] ] where @xmath317 and the coefficients @xmath318 satisfy the equations of motion , @xmath319 e^{\pm i\int_0^t dx ( e_{k+}(x ) - e_{k-}(x))}\nonumber \\\end{aligned}\ ] ] in an adiabatic approximation @xmath320 , we can approximate @xmath318 by its value at @xmath61 . in this case . [ deltamf ] becomes @xmath321 \label{deltamfad}\end{aligned}\ ] ] where @xmath322 denote the following expectation values in the initial state before the quench , @xmath323 since in the initial domain wall state , @xmath324 , this implies that @xmath325 , and hence anti - ferromagnetic order does not develop within a mean - field and adiabatic approximation . this result can also be understood from the fact that the initial state is a highly excited state of the anti - ferromagnetic model as it corresponds to introducing a domain wall in the staggered magnetization at almost every site @xcite . the above conclusion also holds if we relax the adiabatic approximation . to see this note that at every @xmath326 , the mean - field hamiltonian eq . [ hfafmf ] corresponds to a pseudo - spin @xmath327 @xmath328 where @xcite @xmath329 , @xmath330 and @xmath331 . this pseudo - spin is subjected to a magnetic field that has to be determined self - consistently at each time @xmath75 where for @xmath61 @xmath332 . since for the initial domain wall state @xmath333 , it implies @xmath334 to all orders in @xmath335 , so that anti - ferromagnetic order ( @xmath336 ) will not develop . in summary we have studied quenched dynamics in a xxz chain starting from an initial inhomogeneous state where the spins are arranged in a domain wall profile . the dynamics of the domain wall is found to be qualitatively different within the gapless xx phase and the gapped ising phase . in the former the domain wall broadens ballistically , while in the latter the domain wall spreads out less and less the deeper one is in the ising phase . the magnetization is locally found to oscillate , and eventually decays as a power - law . although the results in the ising phase were obtained within a semiclassical treatment , they are in qualitative agreement with t - dmrg simulations of the xxz chain @xcite . we have also presented results for the time - evolution of two - point correlation functions . for quenches within the gapless xx phase we find that the longitudinal spin correlation function ( @xmath40 ) reaches a steady state which is independent of the initial domain wall . on the other hand , the transverse spin correlation function ( @xmath41 ) reaches a nonequilibrium steady state which retains memory of the initial domain wall . this memory appears as an inhomogeneous spin wave pattern with a wavelength which is inversely related to the height of the initial domain wall . this result was obtained by using three different methods : exact solution of the fermionic problem ( eqns . [ cxxansgen ] , [ lamgen ] ) , bosonization ( eqns [ s+-int ] , [ lamint ] ) and the solution of the classical equations of motions for spins ( eq . [ s+class ] ) . the spatial oscillations are found to arise due to the dephasing of transverse spin components as the domain wall broadens . we also find that a gibbs ensemble argument is not adequate in capturing the spin wave pattern . for quenches into the gapped ising phase , all inhomogeneities both in the local magnetization and in the two - point correlation functions are found to eventually decay away . this may be an artifact of the semiclassical approximation which neglects soliton creation . @xcite we would also like to point out a recent preprint @xcite that studies time - evolution of an initial domain wall in the gapped ising phase using algebraic bethe ansatz . the authors study the loschmidt echo and find a lack of thermalization . finally we perform a mean - field treatment for the time - evolution of the anti - ferromagnetic ising gap and find that for an initial state corresponding to a domain wall profile , anti - ferromagnetic order never develops . this should be expected on the grounds that the initial domain wall profile corresponds to a highly excited state of the anti - ferromagnetic ising model . the authors are particularly indebted to eugene demler and ulrich schollwck for helpful discussions . this work was supported by nsf - dmr ( award no . 0705584 ) . the hamiltonian for the @xmath2 model is @xmath337.\ ] ] the jordan - wigner transformation maps the spin hamiltonian to that for spin - less fermions @xmath338 and @xmath339 which are defined in terms of the spin raising and lowering operators @xmath340 as follows , @xmath341c_{j},\\ s_{j}^{+ } & = & \exp\left[i\pi\sum_{n =- n/2 + 1}^{j-1}c^{\dagger}_{n}c_{n}\right]c^{\dagger}_{j},\end{aligned}\ ] ] the hamiltonian is diagonal in momentum space @xmath342 where , @xmath343 thus the operators @xmath344 , have the trivial time - evolution @xmath345 . in terms of the @xmath346 quasi - particles that diagonalize the wanner - stark problem ( eq . [ hxxjw ] ) , the transformation given by eqns . ( [ eq : etaj ] , [ eq : etaj1 ] ) yields @xmath347 antal _ et al . _ have considered more general domain walls @xcite in which the homogeneous magnetization on either side of the wall is taken to be @xmath348 , where @xmath349 . they have studied the time evolution of the domain wall under the influence of the @xmath2 hamiltonian . in this section we will extend their results by studying how the transverse spin correlation function evolves in time . a domain wall of height @xmath124 may be constructed as follows @xcite , @xmath350 where @xmath351 and @xmath352 . due to the xx hamiltonian @xmath353 , the operators @xmath338 evolve as @xmath354 in order to compute any correlation function , the contraction we need is @xmath355 where , @xmath356 } { \pi(s - m ) } & \mbox{if } s , m < 0\\ \frac{\sin\left[\left(\frac{\pi}{2}-\pi m_{0}\right)\left(s - m\right)\right]}{\pi(s - m ) } & \mbox{if } s , m > 0 \end{array}\right.\ ] ] thus , @xmath357}{\pi(m - s ) } \nonumber\\ & & + \sum_{m , s>0}i^{s - m - n}j_{m - j}(jt)j_{s - j - n}(jt)\nonumber \\ & & \times \frac{\sin\left[\left(\frac{\pi}{2}-\pi m_{0}\right)(m - s)\right]}{\pi(m - s)}.\end{aligned}\ ] ] we write both terms as sums over positive @xmath358 , @xmath359 , ( i.e. , letting @xmath360 in the second term ) , and rewrite @xmath361 with @xmath362 , and use the identity @xmath363 next we employ the asymptotic expansion for large arguments , @xmath364 $ ] to find the following nonequilibrium steady state expressions , @xmath365\sin\left[\frac{\pi}{2}\left(l - n\right)\right]}{\pi^{2}l(l - n ) } \right . \nonumber\\ & & \left . + i^{l+n}\frac{\sin\left[\left(\frac{\pi}{2}+\pi m_{0}\right)l\right]\sin\left[\frac{\pi}{2 } \left(l+n\right)\right]}{\pi^{2}l(l+n)}\right]\label{asmgen}.\end{aligned}\ ] ] examining the limit of @xmath366 , we recover exactly the values in eq . [ asm1 ] , [ asm2 ] . we use eq . [ asmgen ] in the evaluation of the determinant in eq . [ detc ] required for computing the transverse spin correlation function . we recover a spatial period of oscillation in @xmath367 that increases with smaller @xmath141 ( see fig . [ fig : cxxperiod ] ) . i. bloch , j. dalibard and w. zwerger , _ rev . _ * 80 * , 885 ( 2008 ) . t. kinoshita , t. wenger , and d. s. weiss , _ nature _ * 440 * 900 ( 2006 ) . t. giamarchi , _ quantum physics in one dimension _ , ( oxford university press , oxford , 2004 ) . k. sengupta , s. powell and s. sachdev , _ phys . a _ * 69 * , 053616 ( 2004 ) . p. barmettler , m. punk , v. gritsev , e. demler and e. altman , _ phys . lett . _ * 102 * , 130603 ( 2009 ) . d. m. weld , p. medley , h. miyake , d. hucul , d. e. pritchard , and w. ketterle _ phys . lett . _ * 103 * , 245301 ( 2009 ) . t. antal , z. rcz , a. rakos , g. m.schutz , _ phys . e _ * 59 * , 4912 ( 1999 ) . d. karevski , _ eur . j _ , b 27 , 147 ( 2002 ) . y. ogata , _ phys e _ * 66 * , 066123 ( 2002 ) . t. platini and d. karevski , _ j. phys . a _ * 40 * , 1711 ( 2007 ) . t. platini and d. karevski , _ eur . phys . j. _ b 48 , 225 ( 2005 ) . d. gobert , c. kollath , u. schollwck , and g. schtz , _ phys . e _ * 71 * , 036102 ( 2005 ) . s. langer , f. heidrich - meisner , j. gemmer , i. p. mcculloch , and u. schollwck , _ phys . rev . b _ * 79 * , 214409 ( 2009 ) . p. calabrese , c. hagendorf and p. le doussal , _ _ , p07013 ( 2008 ) . v. eisler , f. igloi and i. peschel , _ j. stat . mech _ , p02011 ( 2009 ) . v. eisler , t. karevksi , t. platini , and i. peschel , _ j. stat . _ , p01023 ( 2008 ) ; p. calabrese and j. cardy , _ j. stat . p10004 ( 2007 ) ; v. eisler and i. peschel , _ j. stat . _ , p06005 ( 2007 ) . m. a. cazalilla , _ phys . lett . _ * 97 * , 156403 ( 2006 ) . a. iucci and m. a. cazalilla , _ phys . a _ * 80 * , 063619 ( 2009 ) ; _ ibid _ , arxiv:1003.5167 . e. lieb , t. schultz , and d. mattis , _ ann . ( n.y . ) * 16 * , 407 ( 1961 ) . e. barouch , b. mccoy , and m. dresden , _ phys . a _ * 2 * , 1075 - 1091 ( 1970 ) . e. barouch and b. mccoy , _ phys . rev . a _ * 3 * , 786 - 804 ( 1971 ) . e. barouch and b. mccoy , _ phys . rev . a _ * 3 * , 2137 - 2140 ( 1971 ) . t. platini , d. karevski and l. turban , _ j. phys . a : math . theor . _ * 40 * , 1467 ( 2007 ) ; m. collura , d. karevski and l. turban , _ j. stat . _ , p08007 ( 2009 ) ; m. campostrini and e. vicari , _ phys . rev . a _ , * 81 * , 023606 ( 2010 ) ; m. campostrini and e. vicari , arxiv:1003.3334 . e. r. smith , physica * 53 * , 289 ( 1971 ) . k. m. case and c. w. lau , j. math . , * 14 * , 720 ( 1973 ) . t. hartmann , f. keck , h. j. korsch , and s. mossmann , _ new j. phys . _ * 6 * , 2 ( 2004 ) . , edited by m. abramowitz and i. a. stegun ( dover : new york , 1965 ) . note that in the limit of a large number of sites , the result is insensitive to whether the @xmath368 mode is included in the sum or not . a. a. ovchinnikov , _ phys . a _ * 366 * , 357 ( 2007 ) . p. calabrese and j. cardy , _ mech _ , p04010 ( 2005 ) ; p. calabrese and j. cardy , _ phys . rev lett . _ * 96 * , 136801 ( 2006 ) . s. sotiriadis and j. cardy , _ j. stat . _ , p11003 ( 2008 ) . m. rigol and a. muramatsu , _ phys . * 93 * , 230404 ( 2004 ) ; _ ibid _ , _ mod . b _ * 19 * , 861 ( 2005 ) . d. m. gangardt and m. pustilnik , _ phys . rev . a _ * 77 * , 041604(r ) ( 2008 ) v. hunyadi , z. rcz , and l. sasvri , _ phys . e _ * 69 * , 066103 ( 2004 ) . m. rigol , v. dunjko , v. yurovsky , and m. olshanii , _ phys . lett . _ * 98 * , 050405 ( 2007 ) . p. calabrese and j. cardy , _ p06008 ( 2007 ) . d. fioretto and g. mussardo , _ new j. phys . _ , * 12 * 055015 ( 2010 ) . t. barthel and u. schollwck , _ phys * 100 * , 100601 ( 2008 ) . g. biroli , c. kollath and a. laeuchli , arxiv:0907.3731 . peter barmettler , matthias punk , vladimir gritsev , eugene demler , ehud altman , _ new j. phys . _ * 12 * 055017 ( 2010 ) . j. mossel and j. s. caux , _ new j. phys . _ * 12 * , 055028 ( 2010 ) .

in manifold learning , algorithms based on graph laplacian constructed from data have received considerable attention both in practical applications and theoretical analysis . much of the existing work has been done under the assumption that the data is sampled from a manifold without boundaries and singularities or that the functions of interest are evaluated away from such points . at the same time , it can be argued that singularities and boundaries are an important aspect of the geometry of realistic data . boundaries occur whenever the process generating data has a bounding constraint ; while singularities appear when two different manifolds intersect or if a process undergoes a `` phase transition '' , changing non - smoothly as a function of a parameter . in this paper we consider the behavior of graph laplacians at points at or near boundaries and two main types of other singularities : _ intersections _ , where different manifolds come together and sharp _ `` edges '' _ , where a manifold sharply changes direction . we show that the behavior of graph laplacian near these singularities is quite different from that in the interior of the manifolds . in fact , a phenomenon somewhat reminiscent of the gibbs effect in the analysis of fourier series , can be observed in the behavior of graph laplacian near such points . unlike in the interior of the domain , where graph laplacian converges to the laplace - beltrami operator , near singularities graph laplacian tends to a first - order differential operator , which exhibits different scaling behavior as a function of the kernel width . one important implication is that while points near the singularities occupy only a small part of the total volume , the difference in scaling results in a disproportionately large contribution to the total behavior . another significant finding is that while the scaling behavior of the operator is the same near different types of singularities , they are very distinct at a more refined level of analysis . we believe that a comprehensive understanding of these structures in addition to the standard case of a smooth manifold can take us a long way toward better methods for analysis of complex non - linear data and can lead to significant progress in algorithm design . -.5 in graph laplacian , singularities , limit analysis

1211.6727

dealing with high - dimensional non - linear data is one of the key challenges in modern data analysis . in recent years a class of methods based on the mathematical notion of a manifold has become popular in machine learning , starting with the papers @xcite . the underlying intuition is that a process with a small number of parameters will generate a low - dimensional surface in the potentially very high - dimensional space of features and that this situation is ubiquitous in real - world data . this idea is captured nicely by the technical notion of a smooth embedded riemannian manifold , which provides the first realistic model for general non - linear data . still it does not reflect certain important aspects of real data , which can be mathematically understood as singularities and boundaries . the most basic and , arguably , the most important singularity in real data is an _ intersection _ , where two different manifolds come together in a region of space . this often happens in classification where two classes with presumably different structure can give rise to similar objects ( consider , e.g. , the similarity of mnist digits `` 7 '' and `` 1 '' on the right ) . another important type of singularity is a ( co - dimension one ) `` _ edge _ '' , which is ubiquitous in computer graphics ( think of an edge of the surface of a table ) and happens whenever the behavior of an underlying process changes rapidly beyond a certain point ( a `` phase transition '' ) . finally , _ boundaries _ occur when there are bounding constraints on the underlying process ( think how poses of the human body are constrained by the range of motion of the joints ) or on the representation in the ambient space ( e.g. , pixels can not have negative intensity ) . in this paper we provide an analysis of these three cases for laplacian - based learning algorithms . it turns out that all of these singularities result in a behavior near the singularity markedly different from that inside the manifold , a phenomenon somewhat reminiscent of the gibbs effect in fourier analysis . in particular , the scaling of the operator is different near singularities . as a result of this scaling behavior , singularities can not be ignored , even globally , despite the fact that relatively few points are located near them , since each of these points contributes disproportionately to the total operator . we also find that at a finer level of analysis these singularities have quite different effects , which are discussed below . we believe that a comprehensive understanding of these structures in addition to the standard case of a smooth manifold can take us a long way toward better methods for analysis of complex non - linear data and can lead to significant progress in algorithm design . * related work . * methods based on graph laplacians constructed from data have gained acceptance for a range of inference tasks including clustering @xcite , semi - supervised learning ( e.g. @xcite ) and dimensionality reduction @xcite , as well as others . an analysis in @xcite provided a mathematical framework for many of these methods by connecting graph laplacian constructed from data to the laplace - beltrami operator of the underlying manifold based on the relation of the laplacian and the heat equation . that analysis was extended and generalized in @xcite providing a detailed understanding of graph laplacians obtained from manifold data . while boundary effects have been studied in non - parametric kernel smoothing ( see , e.g. , ( * ? ? ? * chapter 4.4 ) ) , they have not been considered in the laplacian - related literature , although we note that boundary behavior for the infinite data case can be derived from the taylor series expansion in lemma 9 of @xcite . to the best of our knowledge , other singularities are not considered in that literature . ( in fact , the reason causing the different scaling behavior for intersection and edge types of singularities is somewhat different from that for the boundary case . ) in related developments , a considerable amount of recent work has been aimed at understanding the case of intersecting linear spaces using various techniques from algebraic geometry to spectral clustering ( e.g. , @xcite ) both in terms of algorithms and theoretical guarantees . in our setting , this is a special case of a singular manifold with intersection singularities . another related work is @xcite , where the authors analyze a model based on a mixture of manifolds in the context of semi - supervised learning . we also note that there has been much recent interest in reconstructing topological invariants of manifolds and other spaces ( see , e.g , @xcite ) . the line of work most related to our results is on learning stratified spaces , where multiple submanifolds ( strata ) are `` glued '' nicely together @xcite , which provides an even more general model for a singular space . * summary of results . * consider the ( appropriately scaled , see section 2 ) graph laplacian @xmath0 constructed from @xmath1 data points , using the gaussian kernel with bandwidth @xmath2 . it can be shown ( see the references above ) that as @xmath2 tends to zero and @xmath1 tends to infinity at an appropriate rate , @xmath0 converges to the laplace - beltrami operator @xmath3 ( a second order differential operator ) inside the manifold . in this paper , we show that for singular manifolds @xmath0 exhibits a very different behavior near the singularity set . specifically , for a sufficiently smooth function @xmath4 and a small @xmath2 , within distance @xmath5 of the singularity set , @xmath6 is approximated by @xmath7 , where @xmath8 is some first order derivative operator that will be described explicitly . the difference in scaling becomes crucial when @xmath2 is small as the @xmath9 term becomes dominant . to simplify the discussion , we will first assume that the data is infinite , and thus @xmath6 can be replaced by @xmath10 . finite sample bounds and rates will be discussed later . the types of singularities considered in the paper are as follows : * boundary . * at a point @xmath11 near manifold boundary , @xmath12 can be approximated by @xmath13 , where @xmath14 is the distance from @xmath11 to the boundary and @xmath15 is the unit vector toward the nearest boundary point ( outward normal at the boundary ) . @xmath16 denotes the directional derivative , while @xmath17 is a scalar function of the form @xmath18 . + * intersection and edge . * the situation is more complicated for other types of singularities , where two different manifolds @xmath19 and @xmath20 intersect or come together . given a point @xmath21 consider its projection @xmath22 onto @xmath20 and its nearest neighbor @xmath23 in the singularity . similarly to the boundary case , let @xmath24 and @xmath25 , be the directions to @xmath23 from @xmath26 and @xmath22 respectively and let @xmath27 and @xmath28 be the corresponding distances . @xmath29 can be approximated by @xmath30 . importantly , the form of the scalar functions @xmath31 is different for different types of singularities : for intersection - type singularity , we have : @xmath32 , while for edge - type we have s for edge singularities by removing some higher order terms ( see section 3.3 ) . ] : @xmath33 . here the coefficients @xmath34 , @xmath35 and @xmath36 depend on the angle between the manifolds and can be written explicitly . * significant points and observations : * * 1 . scaling . * within @xmath37 distance of the singularity , @xmath12 is dominated by a differential operator different from the laplace - beltrami operator . away from the singularity , the higher - valued ( @xmath38 ) term fades and @xmath12 becomes an approximation to the laplace - beltrami operator @xmath3 . * 2 . contributions of singular points . * let the dimension of the manifold be @xmath39 . the volume of points within @xmath5 of the singularity is approximately @xmath40 , where @xmath41 is the @xmath42 dimensional volume of the singular points is always @xmath42 dimensional ( co - dimension one ) , for the boundary and `` edge '' singularities ( although not necessarily for the intersection ) . ] . fix a function @xmath4 . as @xmath43 , the integral @xmath10 over the points within @xmath5 of the singularity does not disappear and , in fact , will converge to a constant , depending on the function and the @xmath42 dimensional volume of the singularity @xmath41 . moreover , because of the scaling , the contribution of those near - singular points to the @xmath44 norm will tend to infinity . we remark that , while different from a technical point of view , this phenomenon is reminiscent to the gibbs effect in fourier series , where the effects of discontinuity do not vanish as approximation becomes more precise . * shape near the boundary . * while the @xmath45 scaling is common for all singularities , the type of singularity is reflected in the shape of the function @xmath46 , which is very different for different types of singularities . somewhat over - simplifying matters , one can think of @xmath47 for the boundary case , @xmath48 for the intersection singularity and @xmath49 for the edge ( see figures above ) . these differences lead to quite distinctive patterns , when the operator is applied to a fixed function . for example , the directional derivative terms disappear _ at _ the intersection - type singular points . in fact for these points @xmath50 converges to the usual laplace - beltrami operator . however , at points within @xmath5 distance of these points we expect to see @xmath51 take both positive and negative values of magnitude @xmath5 . we think that this difference may be a key to future algorithms . * 4 . rates and finite sample bounds . * convergence rates for @xmath2 as well as finite sample bounds are provided in section [ sec : convergencerate ] . interestingly , and perhaps counter - intuitively , the scaling implies that fewer data points are required to ensure that @xmath52 is an accurate ( relative ) estimate of @xmath53 near the boundary than inside the domain . * 5 . impact on eigenfunctions*. eigenfunctions of the graph laplacian associated to data are used in a large number of applications . while the question of convergence is very subtle ( see @xcite for the proof in the case of a smooth manifold ) , our results for fixed functions provide an indication of the expected answer assuming the convergence holds . we provide a brief discussion of it in section [ sec : implication ] . it turns out that the boundary case automatically leads to neumann boundary conditions ( even though the operator does not `` see '' the boundary explicitly ) . quite surprisingly , the edge singularity appears to have no effect on the limit behavior of eigenfunctions , at least if the singular manifold is isometric to a smooth one ( think of bending a sheet of paper ) . the case of the intersection is more complex , but intersections of co - dimension two or more seem to have no effect on the limit behavior of eigenfunctions ( in @xmath44 norm ) . * 6 . normalized laplacians . multiple manifolds . * while we do not provide a detailed discussion due to space constraints , our analysis can be fairly straightforwardly generalized to other versions of graph laplacians , including the random walk normalized , symmetric normalized , and twice normalized graph laplacian discussed in @xcite . interestingly , the popular symmetric normalized laplacian does not lead to neumann boundary conditions as a density term appears inside the directional derivative , i.e. , @xmath54 $ ] . the resulting expression involves derivatives of the density and does not appear to be a natural boundary condition . the analysis can also be extended to the case of multiple manifolds . we restrict our discussion to the case of two manifolds to simplify the exposition . finally , some experimental results can be found in appendix [ appendix : sec : exp ] . let @xmath55 be the set of @xmath56 smooth compact riemannian submanifolds of intrinsic dimension @xmath39 embedded in @xmath57 . for each smooth component @xmath58 , let @xmath59 denote its interior and @xmath60 the boundary of @xmath59 . we assume the boundaries @xmath60 satisfy the _ necessary smoothness conditions _ is a @xmath61-manifold ] . let @xmath62 be the union of @xmath59 , and @xmath63 be the union of @xmath64 . @xmath58 and @xmath65 may not be disjoint , and we will consider the following two ways that they can associate with each other : they intersect in their interior ( i.e , @xmath66 ) or they are `` glued '' along ( part of ) their boundaries . more precisely , given a point @xmath67 , we say that @xmath11 is _ regular _ if it falls in the interior of _ exactly one _ submanifold @xmath59 ; that is , @xmath68 . otherwise , @xmath11 is a _ singular point_. there are several possible singularities , and this paper considers the following _ three general classes _ boundary - type * : boundary points that belong to exactly one smooth submanifold @xmath59 : that is , @xmath69 ; + * 2 . intersection - type * : points at the intersection of two smooth submanifolds . for simplicity of exposition , we assume that the intersection happens between the interior of the submanifolds ; that is , @xmath70 ; + * 3 . edge - type * : boundary points that belong to the boundaries of two smooth submanifolds . that is , @xmath71 . -0.01 in for technical reasons , we will assume that the set of singular points is also a smooth manifold of lower dimension , at least locally . moreover , for edge and intersection singularities the tangent space at each point of the singular manifold is the intersection of the tangent spaces to each `` piece '' . we will call a union of smooth manifolds @xmath55 with the above types of singularities a _ singular manifold_. notice that for a regular point @xmath11 , the manifold is smooth at @xmath11 , and the tangent space at @xmath11 is homeomorphic to @xmath72 . for a singular manifold @xmath55 , we consider the following _ piecewise - smooth function @xmath73_. in particular , set @xmath74 , @xmath75 , be @xmath4 restricted to the submanifold @xmath58 . we require that each @xmath76 is @xmath61-continuous on interior points . let @xmath77 be a piecewise smooth probability density function on @xmath55 and we assume that @xmath78 . * graph laplacian . * given @xmath1 random samples @xmath79 drawn i.i.d . from a distribution with density @xmath77 on @xmath55 , we can build a weighted graph @xmath80 by mapping each sample point @xmath81 to vertex @xmath82 and assigning a weight @xmath83 to edge @xmath84 . one typical weight function is the gaussian , which is used in this paper and defined as follows : @xmath85 notice that in this gaussian weight function , the euclidean distance is used . the normalization by @xmath86 is for the convenience of limit analysis . let @xmath87 be the edge weight matrix of graph @xmath88 with @xmath89 , and @xmath90 be a diagonal matrix such that @xmath91 , then the unnormalized graph laplacian is defined as the @xmath92 matrix @xmath93 now for a fixed smooth function @xmath94 , and any point @xmath67 , the graph laplacian applied to this function @xmath4 is thus @xmath95.\ ] ] * limit behavior of graph laplacian . * the limit study of graph laplacians primarily involves the limits of two parameters , sample size @xmath1 and weight function bandwidth @xmath2 . as @xmath1 increases , one typically decreases @xmath2 to let the graph laplacian capture progressively a finer local structure . with a proper rate as a function of @xmath1 and @xmath2 , the limit of @xmath0 and its various aspects at regular points , including the finite sample analysis , are studied in @xcite . the basic result is that the limit of @xmath96 at a regular point @xmath11 is @xmath97\}$ ] , where @xmath98 is called the _ weighted laplacian _ , see @xcite . see @xcite for the limit analysis of other versions of graph laplacians at regular points . the limit analysis of graph laplacians typically involves limits of two parameters , the sample size @xmath1 and the ( gaussian ) kernel bandwidth @xmath2 . this is usually done in two steps , first analyzing the limit @xmath99 for infinite data and then obtaining finite sample results and rates as @xmath100 using concentration inequalities . in this section we analyze the behavior of the infinite graph laplacian ( i.e. , @xmath101 ) , @xmath102 , when @xmath11 is on or near a singular point , @xmath2 is small and the function @xmath103 , is fixed . finite sample results and rates are given in section [ sec : convergencerate ] . for a fixed @xmath2 we define @xmath12 as the limit of @xmath0 as the amount of data tends to infinity : @xmath104 } f(x)=\frac{1}{t}\int_{{\overline{\omega}}}k_t(x , y)(f(x)-f(y))p(y)dy.\ ] ] * _ local coordinate system _ * : the above integral is defined on the manifold @xmath55 . in order to study the behavior of this integral for a small @xmath2 , we need to introduce a local coordinate systems at @xmath11 . the most convenient coordinate system for our purposes is obtained by a local projection from the manifold to its tangent space @xmath105 . this projection is one - to - one and smooth for points sufficiently close to @xmath11 . hence its inverse , @xmath106 exists as well . note that on an intersection or edge singularity the projection is onto the union of tangent spaces . however , it is still one - to - one as long as we restrict the projection of points from each piece to their respective tangent space . in what follows , we often use change of variables to convert an integral over a manifold to an integral over the tangent space at a specific point . the following bounds are used throughout the paper . for two points @xmath107 , let @xmath108 be the projection of @xmath109 in the tangent space @xmath110 of @xmath58 at @xmath11 . let @xmath111 ( resp . @xmath112 ) denote the jacobian of the map @xmath113 at point @xmath114 ( resp . of the inverse map @xmath115 at @xmath116 ) . for @xmath109 sufficiently close to @xmath11 , we have ( e.g. , @xcite)@xmath117 we first analyze the limit of @xmath102 when @xmath11 is near a _ smooth _ boundary of a single smooth manifold @xmath55 as the results for the boundary case are simpler to state . [ thm : bnds ] let @xmath118 , @xmath119 with @xmath120 , and @xmath121 be the smooth boundary of @xmath122 . given a point @xmath11 near the boundary , let @xmath23 be its nearest neighbor on the boundary @xmath63 , and @xmath123 the inward normal direction of @xmath122 at @xmath23 . put @xmath124 . for @xmath2 sufficiently small we have @xmath125 consequently , if @xmath11 is on the boundary , i.e. , @xmath126 , then @xmath127 and @xmath128 the proof of the theorem can be found in appendix [ appendix : sec : bnd ] . for comparison , the corresponding result for interior points is as follows : @xmath129 hence the graph laplacian has a different behavior on or near a boundary point from interior points . the values of @xmath130 are of different order , @xmath131 near the boundary as opposed to @xmath132 in the interior . intuitively , the scaling difference stems from integrating over a half - plane for boundary points versus integrating over a ( high - dimensional ) plane for points away from boundary , which causes the integration of the first - order derivative terms non - vanishing in the former case . in practice , we do not know the boundary , and apply the same _ global _ normalization for all @xmath67 . the result is that the large values of @xmath133 are likely to correspond to points near the boundary ( or other singular set , see below ) . we think that these observation could lead to useful techniques for data analysis and algorithms that respect singularities . see appendix [ appendix : sec : expdetection ] for some experiments on mnist dataset . we now present results on the behavior of the graph laplacian near the intersection of two @xmath39-manifolds @xmath134 and @xmath135 embedded in @xmath57 with @xmath136 . note that we do not assume that the intersection is of codimension one . we also remark that while the boundary behavior is caused by the integration over a half - disk ( which causes asymmetry in the integration of first - order terms ) for boundary points , this is not the case for points around intersection singularity and edge - type singularity . the behavior around these latter two types of singularities are more involved and are more subtle to analyze . [ thm : intersection ] let @xmath134 and @xmath135 be two @xmath39-dimensional smooth manifolds in @xmath57 potentially with boundaries , and their intersection @xmath137 is a smooth manifold of dimension @xmath138 . let @xmath4 be a continuous function over @xmath139 whose restriction @xmath140 on @xmath58 , @xmath141 , is @xmath61-continuous . given a point @xmath142 near the intersection , let @xmath23 be its nearest neighbor in @xmath137 , and @xmath26 ( resp . @xmath22 ) be its projection in the tangent space of @xmath134 at @xmath23 ( resp . in the tangent space of @xmath135 at @xmath23 ) . put @xmath143 . for a sufficiently small @xmath2 , we have @xmath144 where @xmath145 and @xmath146 are the unit vectors in the direction of @xmath147 and @xmath148 , respectively , and @xmath149 is the angle between @xmath145 and @xmath146 . * proof . * given @xmath142 and its nearest neighbor @xmath23 in @xmath150 , recall that @xmath26 and @xmath22 are the orthogonal projection of @xmath11 onto the tangent space @xmath151 of @xmath152 at @xmath23 , and onto the tangent space @xmath153 of @xmath154 at @xmath23 , respectively . for any @xmath155 , let @xmath156 denote its orthogonal projection onto the tangent space @xmath153 . see the right figure for an illustration . by the definition of @xmath12 , we have @xmath157 . since @xmath158 , this can then be decomposed as @xmath159 the first term above , which is the integral over @xmath134 , is exactly the graph laplacian of @xmath160 at an interior point @xmath11 of @xmath134 . thus , it is bounded by @xmath132 . we now focus on the second integral over @xmath135 . let @xmath161 denote the ball in @xmath57 center around @xmath11 with radius @xmath162 for a sufficiently small constant @xmath163 . in the derivation below , we will approximate the integral over the manifold @xmath135 by an integral over the region @xmath164 . the error term induced is @xmath165 for small @xmath2 . using the fact that @xmath166 , we have : @xmath167 since @xmath168 , it follows from eqn ( [ equ : basicapprox1 ] ) that @xmath169 . it can then be shown that @xmath170 and @xmath171 . for simplicity , let @xmath172 denote @xmath173 for a sufficiently small positive constant @xmath174 . for a point @xmath175 and its projection @xmath156 in @xmath153 , we have that @xmath176 and @xmath177 . this implies that @xmath178 and @xmath179 by eqn ( [ equ : basicapprox1 ] ) . using these distance bounds , we have : @xmath180 now consider the first term in eqn ( [ eqn : a1 ] ) . let @xmath181 denote the projection map from @xmath154 onto @xmath153 and @xmath182 its inverse . denote by @xmath183 to be the projection of @xmath184 on the tangent space @xmath153 . by applying taylor expansion to @xmath94 at @xmath23 and @xmath185 at @xmath23 , and combining eqn [ equ : basicapprox1 ] and [ eqn : jacobianapprox ] , we have that : @xmath186 j \pi^{-1}|_{y ' } dy ' \nonumber\\ = & \frac{1}{t^{\frac{d+1}{2}}}(r \partial_{\n_1 } f_1 ( x_0)+\tilde{o}((t^{\frac{1}{2}}))e^{-r^2\sin^2{\theta}}(1+\tilde{o}((t^{\frac{1}{2}}))(p(x_0)+\tilde{o}((t^{\frac{1}{2}}))\int_{b}e^{-\|x_2-y'\|^2/t}dy ' \nonumber \\ = & \frac{1}{t^{\frac{d+1}{2}}}(r\partial_{\n_1 } f_1 ( x_0)+\tilde{o}(t^{\frac{1}{2}}))e^{-r^2\sin^2{\theta}}(1+\tilde{o}(t^{\frac{1}{2}}))(p(x_0)+\tilde{o}(t^{\frac{1}{2}}))(\int _ { { t_{x_0,{\omega}_2}}}e^{-\frac{\|x_2-y'\|^2}{t}}dy'+o(1))\nonumber\\ = & \frac{1}{\sqrt{t } } c_3 p(x_0 ) r e^{-r^2\sin^2{\theta}}\partial_{\n_1 } f_1 ( x_0 ) + o\left(\frac{1}{\sqrt{t}}\right ) , \label{eqn : term1}\end{aligned}\ ] ] where @xmath187 . note that we use the fact that @xmath188 in the above derivation . from the first line to the second line we perform a change of variable . from the third line to the fourth line in the above derivation , we relax @xmath189 to be @xmath190 based on the result from appendix b in @xcite . for the second item in eqn ( [ eqn : a1 ] ) , we have @xmath191 \int_{\b(x)\cap { \overline{\omega}}_2}k_t(x , y ) dy + \frac{1}{t } \int_{\b(x)\cap { \overline{\omega}}_2}k_t(x , y)(f(x_2 ) - f(y))dy \nonumber \\ = & \frac{1}{\sqrt{t}}c_3 p(x_0 ) r\cos{\theta } \cdot e^{-r^2 \sin^2{\theta}}\partial_{\n_2 } f_2 ( x_0 ) + o\left(\frac{1}{\sqrt{t}}\right ) \label{eqn : term2}\end{aligned}\ ] ] intuitively , the first term in the third line is bounded by the quantity in the last line by a similar argument as the one carried out in eqn ( [ eqn : term1 ] ) above . for the second term in the third line , observe that if we replace the kernel @xmath192 by @xmath193 , then @xmath194 roughly corresponds to the standard weighted graph laplacian of function @xmath195 at an interior point @xmath22 in @xmath135 . furthermore , by a derivation similarly to the one in eqn ( [ eqn : kernelapprox ] ) , replacing @xmath192 by @xmath196 only changes the integral by a factor of @xmath197 . hence the second term can be bounded by @xmath198 . see appendix [ appendix : sec : int ] for details . is in @xmath153 . in eqn ( [ eqn : term2 ] ) , to illustrate the intuition behind the derivation , we abuse the notation slightly and use @xmath199 to refer to @xmath200 , where @xmath201 is the point from @xmath135 whose projection in @xmath153 is @xmath22 . ] the theorem then follows from eqn [ eqn : term1 ] and [ eqn : term2 ] . from the theorem , we can see that for a point @xmath11 on the intersection or near intersection , @xmath133 is of the form @xmath202 where the coefficient @xmath36 is determined by the derivatives of @xmath4 and the position of @xmath11 . furthermore , for a point @xmath11 on the intersection , we have @xmath126 . hence it follows that the order of @xmath133 _ at _ an intersection point is _ the same _ as those at regular points , i.e. , @xmath132 , instead of order @xmath203 as for points _ near _ singularities . we now consider the case where two manifolds are glued along ( a part of ) their boundaries . the behavior of graph laplacian for edge - type points is in some sense the combination of that for boundary - type and for intersection - type points : locally , a point can both see " the boundary ( thus will have terms arised from boundary effect ) and the other manifold ( which has effects similar to those produced by eqn ( [ eqn : a1 ] ) ) . see appendix [ appendix : sec : edge ] for the proof of the following theorem . [ thm : edge ] let @xmath58 ( @xmath204 ) be two @xmath39-dimensional smooth manifolds in @xmath57 with interior @xmath59 and nonempty boundary @xmath64 . assume their shared boundary @xmath205 is a @xmath42-dimensional smooth manifold . let @xmath4 be a continuous function over @xmath206 whose restriction @xmath207 is @xmath61-continuous . given a point @xmath208 near the glued boundary , let @xmath23 be its nearest neighbor in @xmath205 , and put @xmath209 . then for a sufficiently small @xmath2 , we have that @xmath210+o\left(\frac{1}{\sqrt{t}}\right),\ ] ] where @xmath149 is the angle between the two inward boundary normal @xmath145 and @xmath146 of @xmath211 and @xmath212 at @xmath23 respectively , + @xmath213 , + @xmath214 , + and @xmath215 is the cumulative distribution function of the standard normal distribution . furthermore , if @xmath11 is on the edge , i.e. , @xmath126 , then we have @xmath216+o\left(\frac{1}{\sqrt{t}}\right).\ ] ] the theorem indicates that for a point @xmath11 near @xmath205 , the limit of @xmath130 is close to the weighted sum of two normal gradients . as @xmath11 moves to the set @xmath205 , the weights for the two normal gradients will be equal . finally , for a point @xmath11 on the boundary @xmath205 , the behavior @xmath130 is equivalent to the addition of the two boundary effects from @xmath134 and @xmath135 together . [ [ intersection - type - singularity - again . ] ] intersection - type singularity again . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + notice that for the case of two manifolds @xmath152 and @xmath154 intersect at @xmath137 of co - dimension @xmath217 , we can actually regard this scenario as four pieces of manifolds , @xmath218 and @xmath219 , glued together by @xmath220 . here , we use @xmath221 and @xmath222 to denote the two pieces of @xmath59 but on the different side of @xmath150 . the advantage of taking this view is that we allow the functions on manifolds @xmath152 and @xmath154 to be only @xmath223-continuous at the intersection @xmath137 , instead of @xmath61-continuous as required in theorem [ thm : intersection ] . ( however , note that theorem [ thm : intersection ] also holds for the cases when the co - dimension of the intersection is higher than @xmath217 . ) using the same argument as in the proof of theorem [ thm : edge ] to this case of gluing 4 @xmath39-manifolds along a common @xmath42-boundary , we obtain the following corollary . [ thm : intersection1 ] let @xmath134 and @xmath135 be two @xmath39-dimensional smooth manifolds in @xmath57 potentially with boundaries , and their intersection @xmath137 is a smooth manifold of dimension @xmath217 . let @xmath4 be a continuous function over @xmath139 whose restriction @xmath140 on @xmath58 , @xmath141 , is @xmath61-continuous at any regular point and @xmath223-continuous at points in @xmath137 . given a point @xmath142 near the intersection , let @xmath23 be its nearest neighbor in @xmath137 , and @xmath26 ( resp . @xmath22 ) be its projection in the tangent space of @xmath134 at @xmath23 ( resp . in the tangent space of @xmath135 at @xmath23 ) . for a sufficiently small @xmath2 , we have @xmath224 + o\left(\frac{1}{\sqrt{t}}\right)\ ] ] where @xmath225 are the directional derivatives of @xmath160 on two sides of @xmath226 , same for @xmath227 . interestingly , it appears the eigenfunctions for a singular manifold with intersection - type singularity can be only @xmath223-continuous across the intersection ( instead of @xmath61-continuous ) within each piece of manifold . hence the above corollary will be useful in analysing the eigenfunctions around intersection singularities ( or co - dimension 1 ) . [ thm : complexity ] let @xmath228 , be i.i.d . random variables in @xmath57 sampled from a probability distribution on @xmath55 with the intrinsic dimension @xmath39 with density @xmath77 , @xmath229 , defined on the manifold on @xmath55 with the intrinsic dimension @xmath39 . let @xmath4 be a bounded function , @xmath230 . then for any @xmath231 we have @xmath232 where @xmath233 and @xmath234 are constants depending on the manifold . the proof is based on an application of the bernstein inequality and the union bound and can be found in appendix [ appendix : sec : convergencerate ] . the immediate corollary is that to get an asymptotic convergence for an arbitrary fixed point ( that is an error of the order of @xmath235 ) , we need to select @xmath2 so that @xmath236 . that is , we can choose @xmath237 , where @xmath238 is an arbitrary function such that @xmath239 . on the other hand , for points near the singular set the scaling of the operator @xmath0 changes . while inside the domain @xmath240 for a smooth ( fixed ) function @xmath4 , for @xmath11 on the singular set or sufficiently close to it ( within @xmath5 distance ) , @xmath241 . thus an accurate estimate for the appropriately rescaled operator requires that @xmath242 leading to @xmath243 . this may seem counter - intuitive as fewer points are required on the singularity than inside the domain to obtain an accurate estimate . one explanation is that near a singularity we are effectively estimating a degree one differential operator , while inside the domain we are estimating the laplace - beltrami operator , which is degree two . eigenfunctions of graph laplacians obtained from data play an important role in a variety of applications from spectral clustering to dimensionality reduction and semi - supervised learning . while full proof of their convergence is likely to be very subtle and is beyond the scope of this paper ( see @xcite for the proof of eigenfunction convergence for the smooth case ) , we would like to discuss the implications of our results for eigenfunctions under the assumption that such convergence takes place . let @xmath244 be an eigenfunction of @xmath245 ( normalized to norm @xmath217 in @xmath44 ) . we put @xmath246 and @xmath247 ( with the functions converging in @xmath44 norm ) . we will also assume that the necessary derivatives exist . below , we briefly discuss the impact of each type of singularity . some numerical experiments are provided in appendix [ appendix : sec : eigenexp ] . for simplicity , we will assume the samples to be infinite and the multiplicity of each eigenfunction to be one . [ [ boundary - singularity ] ] boundary singularity : + + + + + + + + + + + + + + + + + + + + + for a point @xmath11 on or near ( within @xmath5 of ) the boundary our results indicate @xmath248 where @xmath36 is a constant independent of @xmath2 . thus , for small values of @xmath2 , and @xmath249 , which is clearly impossible unless @xmath250 . passing to the limit are ignored . ] we see that for @xmath11 on the boundary , @xmath251 , meaning that the limit eigenfunctions of the graph laplacian automatically satisfy the neumann boundary conditions . the experimental results in appendix [ appendix : sec : eigenexp ] are consistent with this finding . [ [ edge - singularity ] ] edge singularity : + + + + + + + + + + + + + + + + + consider now a point @xmath11 at the edge where two `` sheets '' @xmath152 and @xmath154 come together . from theorem [ thm : edge ] we have @xmath252+o\left(\frac{1}{\sqrt{t}}\right).\ ] ] where @xmath145 and @xmath146 are the two directions normal to the singular set at @xmath11 . by an argument following the one above , we have @xmath253 and hence @xmath254 in other words the directional derivatives for the limit eigenfunction must cancel each other . this is quite surprising for the following reason : suppose , that the edge singularity is obtained by folding a smooth manifold ( imagine folding a sheet of paper , creating an edge ) . then the above condition means that the eigenfunctions of the original surface are invariant under this `` folding '' , despite the fact that the graph laplacian operator is very different near the singular set ! in other words if a singular surface is isometric to a smooth manifolds , the spectral structure appears to be preserved , even though the distances computed for points near the edge are very far from the intrinsic distances on smooth and the operator near the edge is not the laplace - beltrami operator . some numerical illustrations of this phenomenon are given in the appendix [ appendix : sec : eigenexp ] . [ [ intersection - singularity ] ] intersection singularity : + + + + + + + + + + + + + + + + + + + + + + + + + for the intersection singularity the analysis follows that of the edge case , but the implications for the shape of eigenfunctions are not completely clear , and warrant further investigation . one intriguing observation is that if the co - dimension of the singularity is greater than one , it has only a local effect on the shape of eigenfunctions . the situation is analogous to the disjoint union of manifolds . the reason is easy to see the volume of @xmath5-neighborhood @xmath255 around the singular set of co - dimension at least two , will be at most @xmath256 . recall that an eigenfunction minimizes the quantity @xmath257 under orthogonality conditions . we see that the contribution of the points around the singular set is bounded by @xmath258 and vanishes as @xmath259 . 29 [ 1]#1 [ 1]`#1 ` urlstyle [ 1]doi : # 1 m. aanjaneya , f. chazal , d. chen , m. glisse , l. guibas , and d. morozov . metric graph reconstruction from noisy data . in _ proc . 27th sympos . _ , pages 3746 , 2011 . m. belkin . _ problems of learning on manifold_. phd thesis , university of chicago , 2003 . m. belkin and p. niyogi . laplacian eigenmaps for dimensionality reduction and data representation . _ neural comp _ , 150 ( 6):0 13731396 , 2003 . m. belkin and p. niyogi . towards a theoretical foundation for laplacian - based manifold methods . _ journal of computer and system sciences _ , 740 ( 8):0 12891308 , 2008 . mikhail belkin and partha niyogi . convergence of laplacian eigenmaps . _ preprint , short version appeared in nips 2007 _ , 2008 . paul bendich , david cohen - steiner , herbert edelsbrunner , john harer , and dmitriy morozov . inferring local homology from sampled stratified spaces . in _ proc . ieee sympos . foundat . comp . _ , pages 536546 , 2007 . paul bendich , bei wang , and sayan mukherjee . local homology transfer and stratification learning . in _ proceedings of acm - siam symposium on discrete algorithms _ , 2012 . o. chapelle , b. schlkopf , and a. zien . _ semi - supervised learning_. mit press , 2006 . frdric chazal and steve oudot . towards persistence - based reconstruction in euclidean spaces . in _ proc . 24th acm sympos . on comput . _ , pages 232241 , 2008 . frdric chazal , david cohen - steiner , and andr lieutier . a sampling theory for compact sets in euclidean space . _ discrete & computational geometry _ , 410 ( 3):0 461479 , 2009 . g. chen and g. lerman . foundations of a multi - way spectral clustering framework for hybrid linear modeling . _ foundations of computational mathematics _ , 90 ( 5):0 517558 , 2009 . r. r. coifman and s. lafon . diffusion maps . _ applied and computational harmonic analysis _ , 210 ( 1):0 530 , 2006 . evarist gin and vladimir koltchinskii . empirical graph laplacian approximation of laplace - beltrami operators : large sample results . 51:0 238259 , 2006 . a. goldberg , x. zhu , a. singh , z. xu , , and r. nowak . multi - manifold semi - supervised learning . in _ artificial intelligence and statistics , aistats _ , 2009 . a. grigoryan . heat kernels on weighted manifolds and applications . _ , 398:0 93191 , 2006 . wolfgang hrdle . _ applied nonparametric regression_. cambridge univeristy press , 1992 . gloria haro , gregory randall , and guillermo sapiro . translated poisson mixture model for stratification learning . _ international journal of computer vision _ , 800 ( 3):0 358374 , 2008 . m. hein . _ geometrical aspects of statistical learning theory_. phd thesis , wissenschaftlicher mitarbeiter am max - planck - institut fr biologische kybernetik in tbingen in der abteilung , 2005 . matthias hein , jean yves audibert , and ulrike von luxburg . graph laplacians and their convergence on random neighborhood graphs . _ journal of machine learning research _ , 8:0 13251368 , 2007 . s. lafon . _ diffusion maps and geodesic harmonics_. phd thesis , yale university , 2004 . gilad lerman and teng zhang . probabilistic recovery of multiple subspaces in point clouds by geometric lp minimization , 2010 . partha niyogi , stephen smale , and shmuel weinberger . finding the homology of submanifolds with high confidence from random samples . _ discrete & computational geometry _ , 390 ( 1 - 3):0 419441 , 2008 . partha niyogi , stephen smale , and shmuel weinberger . a topological view of unsupervised learning from noisy data . _ siam j. comput . _ , 400 ( 3):0 646663 , 2011 . s. t. roweis and l. k. saul . nonlinear dimensionality reduction by locally linear embedding . _ science _ , 2900 ( 5500):0 23232326 , 2000 . a. singer . from graph to manifold laplacian : the convergence rate . comput . harmon . _ , 21:0 128134 , 2006 . j. b. tenenbaum , v. de silva , and j. c. langford . a global geometric framework for nonlinear dimensionality reduction . _ science _ , 2900 ( 5500):0 23192323 , 2000 . r. vidal , y. ma , and s. sastry . generalized principal component analysis ( gpca ) . _ ieee transactions on pattern analysis and machine intelligence _ , pages 19451959 , 2005 . u. von luxburg . a tutorial on spectral clustering . in _ statistics and computing _ , volume 17 , pages 395416 , 2007 . x. zhu . . _ computer science , university of wisconsin - madison _ , 2006 . for a sufficiently small @xmath2 , let @xmath260 be the set of points that are within distance @xmath261 from the boundary @xmath262 ( a thin layer of `` shell '' ) , where @xmath174 is any sufficiently small positive constant . set @xmath263 . we will show that for a small @xmath2 , @xmath264 is approximated by two different terms on @xmath265 and @xmath266 , and more importantly they have different orders of @xmath2 . [ [ points - away - from - boundary . ] ] points away from boundary . + + + + + + + + + + + + + + + + + + + + + + + + + + given a point @xmath267 , let @xmath161 denote the ball of radius @xmath261 centered at @xmath11 , where @xmath174 is any sufficiently small positive constant . first , we approximate the integral in eqn ( [ eqn : aimintegral ] ) when the integral is taken inside the ball @xmath161 . by results from appendix b of @xcite , the error induced by constraining the integral to within @xmath268 is only @xmath269 . now given a point @xmath270 , let @xmath271 be the projection of @xmath109 onto the tangent space @xmath272 of @xmath55 at @xmath11 . choose @xmath11 to be the origin of @xmath272 . we have the following approximations for each of the three terms in the integral in eqn ( [ eqn : aimintegral ] ) ( see e.g , ( * ? ? ? * chapter 4.2 ) and ( * ? ? ? * appendix b ) ) . here @xmath273 denotes @xmath274 . @xmath275 and @xmath39 are the dimension of the ambient space and of the tangent space , respectively . @xmath276 where @xmath277 is the hessian of @xmath94 at @xmath11 . now for simplicity let @xmath278 denote @xmath173 for any sufficiently small positive constant @xmath279 . let @xmath280 be the projection from @xmath55 onto the tangent space @xmath272 , and set @xmath281 . obviously , @xmath282 . combing the above approximations together with the fact that @xmath283 and thus @xmath284 , and using a change of variable @xmath285 , we obtain the following : @xmath286t^{d/2}j \pi^{-1}|_{u}du + \frac{1}{t } \cdot \tilde{o}(t^{1/2})\\ = & -\int_{{t_{x,{\omega } } } } e^{-\|w\|^2_{\r^d } } \{\frac{1}{\sqrt{t}}[p(x)(w^t\nabla f(x))]+\\ & \qquad \qquad \qquad \qquad \quad [ w^t \nabla f(x)\times w^t \nabla p(x)+\frac{1}{2}p(x)w^th(x)w]\ } dw + o(\frac{1}{\sqrt{t } } ) \end{array}\ ] ] from the first line to the second line , we replace the integral over @xmath55 with ball @xmath161 , generating an exponentially small error for sufficiently small @xmath2 ( * ? ? ? * appendix b ) . from the second line to the third line , by changing the variable from @xmath109 to @xmath287 , this integral can be rewritten as an integral over the region @xmath282 . in the fourth line , we relax this integral over @xmath14 to be the integral over the entire tangent space @xmath272 , plus another exponentially small error @xmath269 which is consumed by @xmath198 . for an interior point @xmath288 , @xmath272 is a @xmath39-dimensional linear space , and the function @xmath289 is an even function on @xmath272 . when taking the integral , the first term in the last integral in eqn ( [ equ : bd : decomposition ] ) with order @xmath290 is odd and therefore vanishes . the remaining three terms in the last line of eqn ( [ equ : bd : decomposition ] ) are of order @xmath132 inside the integral , and they are exactly the weighted laplacian at @xmath11 as previously studied in the literature , and hence @xmath291 . the kernel is however not symmetric for points on or near the boundary , which are studied next . near the boundary . ] [ [ points - near - boundary . ] ] points near boundary . + + + + + + + + + + + + + + + + + + + + + we now consider @xmath292 ( the `` shell '' ) near the boundary . see figure [ fig : bd : integral ] , where we show the local neighborhood around @xmath11 projected in @xmath272 . let @xmath23 be the nearest boundary point on @xmath293 to @xmath11 . choose a local coordinate system of @xmath272 with @xmath11 at the origin , and assume the projection of @xmath23 in @xmath272 is at @xmath294 . let @xmath123 the unit direction along @xmath295 . when @xmath11 is close to the boundary , the kernel @xmath192 is no longer symmetric . in particular , consider an orthonormal coordinate system around @xmath11 , the gaussian kernel is symmetric in all coordinate - axis other than along @xmath295 . along the direction @xmath123 , the gaussian convolution is from @xmath296 ( as we have changed variable from @xmath297 ) to @xmath298 , which is not symmetric . therefore , @xmath299 is not an even function in the normal direction @xmath123 , and the integral of the highest order term ( i.e , the first term of @xmath290 ) in the last integral of eqn ( [ equ : bd : decomposition ] ) will not vanish . specifically , let @xmath300 be the coordinate of @xmath271 in the aforementioned coordinate system of @xmath272 . assume that @xmath301 is the coordinate along the normal direction @xmath123 . recall that we have applied the change of variable @xmath302 . so let @xmath303 be the coordinate of @xmath304 . since we have assumed in the theorem that @xmath305 , we have that @xmath306 . when we integrate the last integral in eqn ( [ equ : bd : decomposition ] ) , all the odd terms of @xmath307 still vanish in all directions except the normal direction @xmath123 , and the most important point is that the leading term along the normal direction is of order @xmath290 , different from @xmath132 for the interior points . in particular , we have : @xmath308 the theorem follows from further performing the taylor expansion of @xmath94 at @xmath23 . our goal is to bound the integral @xmath309 as in eqn ( [ eqn : term2 ] ) . first , recall that @xmath310 is the projection map from @xmath135 to the tangent space of @xmath135 at @xmath23 . for points on @xmath135 sufficiently close to @xmath23 , this map is in fact a bijection and its inverse @xmath182 exsits . hence for @xmath2 sufficiently small ( and thus @xmath22 is close enough to @xmath11 ) , @xmath201 exists and we set it to be @xmath311 . since @xmath312 , we have @xmath313 by eqn ( [ equ : basicapprox1 ] ) . it follows from triangle inequality that @xmath314 . we can then apply a similar derivation as in eqn ( [ eqn : kernelapprox ] ) to show that @xmath315 where @xmath156 is the projection of @xmath109 on @xmath153 as defined in section [ sec : inter ] . we now have @xmath316 following the same derivation as in eqn ( [ eqn : term1 ] ) , and combining with @xmath317 , the first term can be bound by : @xmath318 by using eqn ( [ eqn : kernelapprox2 ] ) , the second term in eqn ( [ eqn : newterm2 ] ) is : @xmath319 now observe that since the radius of @xmath161 is @xmath320 for a sufficiently small positive @xmath321 , and since the distance from @xmath11 to @xmath322 is @xmath323 , we have that there exists some ball @xmath324 of radius still in the asymptotic order of @xmath320 around @xmath322 such that @xmath324 is contained inside @xmath161 . this implies that @xmath325 . hence by results from appendix b of @xcite , @xmath326 on the other hand , @xmath327 is simply an approximation of the functional laplacian @xmath328 of the function @xmath195 at an interior point @xmath322 over a manifold @xmath135 . hence by eqn ( [ eqn : interior ] ) , @xmath329 it then follows from eqn ( [ eqn : newterm2 - 2 ] ) that @xmath330 putting eqn ( [ eqn : newterm2 ] ) , ( [ eqn : newterm2 - 1 ] ) and ( [ eqn : newterm2 - 3 ] ) together , we conclude that @xmath331 which is what we claimed in eqn ( [ eqn : term2 ] ) . suppose we have two manifolds forming the `` edge '' , denote them by @xmath134 and @xmath135 . we use the same notation as in the proof for the intersection case ; see the illustration in figure [ fig : lcorner ] . in particular , let @xmath23 denote the closest point from @xmath332 to a point @xmath333 . let @xmath26 and @xmath22 be the projection of @xmath11 on the tangent space @xmath151 and @xmath153 at @xmath23 , respectively . note that @xmath124 . let @xmath334 be the restriction of @xmath4 on @xmath152 and @xmath154 respectively ; we know that @xmath335 . again , @xmath161 denotes the ball centered at @xmath11 with radius @xmath162 . then we have @xmath336 and @xmath337 the first integral in eqn ( [ eqn : c ] ) on @xmath338 is exactly what we had before for the boundary singularity ( recall the second line in eqn ( [ equ : bd : decomposition ] ) ) , which we have shown to be the following in appendix [ appendix : sec : bnd ] : @xmath339 now consider the second integral in eqn ( [ eqn : c ] ) on @xmath340 : @xmath341 let @xmath113 denote the projection from @xmath135 to @xmath153 , and set @xmath342 for any point @xmath343 . since the notation is the same with intersection case , we have by eqn ( [ eqn : kernelapprox ] ) @xmath344 set @xmath345 as the image of @xmath346 in the tangent space @xmath153 . let @xmath347 denote the half - space of the tangent space @xmath153 that contains @xmath14 . combing the bounds in eqn ( [ equ : basicapprox1 ] ) and ( [ eqn : jacobianapprox ] ) , for the first integral in eqn ( [ eqn : d ] ) , we have @xmath348 where @xmath349 , and @xmath350 is the @xmath39-th entry of coordinate of @xmath109 in the @xmath39-dimensional tangent space . the last step above also involves applying the following formula : @xmath351 now for the second integral in eqn ( [ eqn : d ] ) , we have : @xmath352 the second term from the above equation is exactly the same as if we were considering a point @xmath22 nears the boundary of @xmath135 . hence using the results from appendix [ appendix : sec : bnd ] , we have that @xmath353 for the first term from eqn ( [ eqn : yusua ] ) , we have : @xmath354 where @xmath355 is defined earlier . let @xmath356 be the projection of @xmath26 on the tangent space @xmath153 . the last step in eqn ( [ eqn : yusuc ] ) uses the fact that @xmath357 . combining all the results in ( [ eqn : int1 ] ) , ( [ eqn : int2 ] ) , ( [ eqn : yusua ] ) , ( [ eqn : yusub ] ) , and ( [ eqn : yusuc ] ) , we prove theorem [ thm : edge ] . [ lemma : convergencerate ] let @xmath358 , z be i.i.d . random variables of @xmath57 from density @xmath119 , @xmath229 , @xmath192 be a gaussian weight function and @xmath359 , then for any @xmath67 @xmath360\right|>\epsilon\right)\le 2\exp{\left(-\frac{nt^{d/2}\epsilon^2 } { 2c_v+2c_m\epsilon/3}\right)}\ ] ] where @xmath233 and @xmath234 only depend on @xmath39 and @xmath361 , @xmath362 , @xmath363 . [ lemma ] let @xmath364 , and @xmath365 $ ] . we have @xmath366=0 $ ] , and @xmath367|\\ = & |k_t(x , x_i)f(x_i)|+|\int_{{\overline{\omega } } } k_t(x , y)f(y)p(y)dy|\\ \le & \frac{1}{t^{d/2 } } m + mb \int_{{\overline{\omega } } } k_t(x , y)dy\\ \le & \frac{m}{t^{d/2}}+mb c_g=(c_m+mb c_g t^{d/2})/t^{d/2 } \end{array}\ ] ] where @xmath368 , and @xmath369 . for var@xmath370 $ ] , @xmath371=&{\mathbb{e}}_{x_i}[w_i^2(x)]-\{{\mathbb{e}}_{x_i}[w_i(x)]\}^2\\ \le&\int_{{\overline{\omega } } } k_t^2(x , y)f^2(y)p(y)dy + [ \int_{{\overline{\omega } } } k_t(x , y)f(y)p(y)dy]^2\\ \le & \frac{1}{t^{d/2}}m^2 b \int_{{\overline{\omega } } } k_t(x , y)dy + m^2b^2c_g^2\\ \le & \frac{1}{t^{d/2}}m^2 b c_g + m^2b^2c_g^2=(c_v+m^2b^2c_g^2t^{d/2})/t^{d/2}\\ \end{array}\ ] ] where @xmath372 . by the bernstein s inequality @xmath373 _ proof of theorem [ thm : complexity ] _ : for @xmath374 , we put @xmath375 into lemma [ lemma : convergencerate ] , obtaining : @xmath376\right|>t\epsilon \right)\\ \le & 2\exp{\left(-\frac{nt^{(d+4)/2}\epsilon^2 } { 2c_v+2c_mt\epsilon/3}\right ) } \end{split}\ ] ] the proof of theorem [ thm : complexity ] is done . below we provide some numerical results validating theoretical analysis , some experiments with real datasets and several numerical experiments in support of our conjectures on the eigenfunction behavior . in this section , we illustrate the behavior of graph laplacian on or near the singularity points through a simple numerical example . in order to explore different cases , we consider the union of 1-dimensional manifolds @xmath55 as shown in figure [ fig : scalebehavior ] ( a ) , which is a combination of three linear intervals , @xmath134 , @xmath135 , and @xmath377 in @xmath378 . we take 2500 point uniformly spaced over the singular manifold and choose @xmath379 ( restricted to the manifold ) . * shape and scaling behavior : * the value of @xmath380 over all points in @xmath134 is shown in figure [ fig : scalebehavior ] ( b ) . we can see that @xmath380 near the boundary is approximately half a gaussian function , while the function near the intersection is similar to a function of the form @xmath381 , as predicted by theorem [ thm : intersection ] . the shape of @xmath380 near the corner is also what we expect from theorem [ thm : edge ] . to explore the scaling effects , in figure [ fig : scalebehavior ] ( c ) we plot values of @xmath382 against @xmath383 for different values of @xmath2 . each graph corresponds to a fixed point @xmath11 chosen near a boundary , intersection or edge singularity . our theoretical results predict a linear curve with slope @xmath384 in @xmath385-@xmath385 coordinates ( corresponding to @xmath386 , which is consistent with our experimental results as shown in ( c ) . in the following example , we use the union of the stanford bunny dataset ( in blue ) and a plane @xmath387 ( in red ) , which intersects the bunny in the middle as shown in panel ( a ) . we have take @xmath388 points for both the bunny and the plane in the point cloud , and choose the function @xmath389 . the values of @xmath6 on the bunny are shown in panel ( b ) . we can see that the values of @xmath6 at the intersection are @xmath390 and values near the intersection go up ( in yellow ) and down ( in blue ) as expected from the theoretical analysis of the intersection singularity . from the scaling behavior of the graph laplacian near singularities , it is potentially possible to detect certain such singularities on real world data sets . we use the example of the mnist digit images . the image of each digit can be thought of as a sample from a low - dimensional manifold embedded in the high - dimensional pixel space . we choose function @xmath94 to be the summation of the pixel intensity of each image . we take the images corresponding to the top @xmath391 of the highest value of @xmath392 as the `` near singularity / boundary '' images , which are compared to the `` average '' images of each digit . the results are shown in figure ( [ fig : bd : mnist ] ) . we see that the images on the right are significantly larger ( and systematically different ) from average , suggesting that they may belong to the part of the boundary , which can be detected through the sum - of - pixel - intensity function . in this example , we consider two manifolds : ( 1 ) the rectangular region @xmath152 , @xmath393 , in @xmath394 and ( 2 ) the `` folded '' rectangle @xmath154 , obtained by transforming the positive @xmath109 part of @xmath152 , i.e. @xmath395 , by applying the linear transformation given by the matrix @xmath396 and keeping the rest of @xmath152 fixed as showing in fig . [ fig : data_rect ] . we see @xmath152 and @xmath154 are intrinsically isometric , yet @xmath154 has an edge singularity . with this two data sets , we construct their graph laplace matrices respectively , and calculate their eigenvalues and eigenvectors using the gaussian kernel with @xmath398 . figure [ fig : eigen_rect ] shows the first 16 nontrivial eigenvectors of two graph laplace matrices . we can see that the corresponding eigenvectors match nearly perfectly . it can be easily verify that they also match eigenfunctions of the rectangle with neumann boundary conditions . to give some precise numerical results we measure the relative difference between the sets of eigenvalues . let @xmath399 and @xmath400 be the two vectors of first @xmath56 eigenvalues for @xmath152 and @xmath154 respectively . we define the ( relative ) difference as @xmath401 the table below shows the difference for various values of @xmath56 . we observe that the approximation appears quite precise with difference of the order of @xmath402 , thus providing further evidence in support of our conjecture on the behavior of eigenfunctions .

we study numerically the collapse of rotating , magnetized molecular cloud cores , focusing on rotation and magnetic braking during the main accretion phase of isolated star formation . motivated by previous numerical work and analytic considerations , we idealize the pre - collapse core as a magnetized singular isothermal toroid , with a constant rotational speed everywhere . the collapse starts from the center , and propagates outwards in an inside - out fashion , satisfying exact self - similarity in space and time . for rotation rates and field strengths typical of dense low - mass cores , the main feature remains the flattening of the mass distribution along field lines the formation of a pseudodisk , as in the nonrotating cases . the density distribution of the pseudodisk is little affected by rotation . on the other hand , the rotation rate is strongly modified by pseudodisk formation . most of the centrally accreted material reaches the vicinity of the protostar through the pseudodisk . the specific angular momentum can be greatly reduced on the way , by an order of magnitude or more , even when the pre - collapse field strength is substantially below the critical value for dominant cloud support . the efficient magnetic braking is due to the pinched geometry of the magnetic field in the pseudodisk , which strengthens the magnetic field and lengthens the level arm for braking . both effects enhance the magnetic transport of angular momentum from inside to outside . the excess angular momentum is carried away in a low - speed outflow that has , despite claims made by other workers , little in common with observed bipolar molecular outflows . we discuss the implications of our calculations for the formation of true disks that are supported against gravity by rotation .

astro-ph0311377

rotation is observed ubiquitously in star - forming cores . on the scale of @xmath00.1 pc , the detected rotation rate is small , with a typical value corresponding to a ratio of rotational to gravitational energy of @xmath1 ( goodman et al . 1993 ) . the rotation rate is much lower than expected if the core is condensed out of a low density background medium with angular momentum conserved . the slow core rotation is thought to be due to magnetic braking ( mestel 1985 ; mouschovias & ciolek 1999 ) , which is particularly efficient during the long , magnetically subcritical phase of cloud evolution driven by ambipolar diffusion ( basu & mouschovias 1994 ) . this assumes that the clouds are strongly magnetized to begin with , as envisioned in the standard picture of isolated low - mass star formation ( shu , adams & lizano 1987 ) . once a magnetically supercritical core develops , it begins to collapse dynamically . it is often assumed that magnetic braking then becomes inefficient , with most of the angular momentum of the core material carried by the collapsing flow into the central star - plus - disk or binary system . however , the typical value of specific angular momentum inferred for dense cores is an order of magnitude higher than the typical binary value , at least in the taurus cloud complex ( see fig . 8 of simon et al . 1995 and discussion in 8.5 of myers 1995 ) . this suggests that a large fraction of the core angular momentum may still be removed _ during collapse_. the problem of angular momentum redistribution in a collapsing magnetized core thus warrants a closer examination . the efficiency of magnetic braking depends sensitively on the magnetic field geometry ( mestel 1985 ) . the exact field geometry of a magnetized core on the verge of collapse is difficult to determine without following the cloud evolution in three dimensions . simplified numerical calculations of cloud evolution driven by ambipolar diffusion have shown , however , that by the time a central density cusp is formed the mass - to - flux ratio is approximately spatially constant in the supercritical core , and the ( volume ) density decreases with radius roughly as a power - law of @xmath2 ( lizano & shu 1989 ; basu & mouschovias 1994 ) . this self - similar behavior motivated li & shu ( 1996 ) to idealize the end state of the ambipolar diffusion - driven core formation as a scale - free configuration supported in part by a large scale magnetic field and in part by the thermal pressure gradient , with a fixed mass - to - flux ratio and an exact @xmath2 density profile . such a configuration turns out to be a singular isothermal toroid , for which the magnetic field geometry can be computed exactly ( see also baureis , ebert , & schmitz 1989 ) . the field geometry is distorted by fluid motions during the subsequent inside - out collapse . in a companion paper to this journal ( allen , shu & li 2003 ; hereafter paper i ) , we have studied the collapse of magnetized , singular isothermal toroids without rotation . the main features of the collapse solution are governed by the anisotropy in the magnetic forces , which induces in the toroid an evacuated region near the axis , where the plasma @xmath3 parameter ( the ratio of thermal to magnetic energy density ) drops below unity ( see fig . 4a of paper i ) . as the core collapses from inside out , matter in the magnetically dominated region slides along the field lines towards the equatorial region , creating a dense , flattened structure a pseudodisk ( galli & shu 1993a , b ) . the material in the pseudodisk collapses radially towards the central point mass in a magnetically diluted free - fall , dragging the footpoints of field lines along with it ( see fig . 4b of paper i ) . a highly pinched field configuration ( split monopole " ) is produced , with important implications for the magnetic transport of angular momentum from inside to outside . in the presence of rotation , the collapse would lead to a spinup of the pseudodisk , which is linked magnetically to a more slowly rotating envelope of infalling and nearly static material . we will show in this paper that the twisting of magnetic field lines actually drives a slow outflow during the core collapse , which can remove much of the angular momentum remaining in the infalling pseudodisk . this removal poses a substantial brake against the eventual formation of a centrifugally supported disk unless the trapped magnetic field is lost in the realistic situation by nonideal mhd effects . as the first step toward the dynamical calculations , we will generalize the magnetized , singular isothermal toroid solutions of li & shu ( 1996 ) to include rotation in [ initial ] . these magnetized rotating toroids are then induced to collapse from inside out in [ collapse ] , and the collapse is followed numerically in a manner similar to that described in paper i. the numerical solutions are discussed in [ discussion ] , with an emphasis on the efficiency of magnetic braking during core collapse and the implications for the difficulty of forming rotationally supported disks unless significant flux loss appears in the realistic problem at small scales , as is expected theoretically for high - density situations ( e.g. , nishi , nakano & umebayashi 1991 ; desch & mouschovias 2001 ) . we end with a brief summary and conclusion in [ conclusion ] . we consider a rotating , scale - free ( self - similar ) , self - gravitating cloud that is magnetized by a purely poloidal field and in mechanical force balance . the cloud is specified by the mass density @xmath4 , the magnetic flux function @xmath5 which determines the poloidal field through @xmath6 $ ] , and the rotational speed @xmath7 , in a spherical polar coordinate system @xmath8 . under the assumption of axisymmetry , the radial and angular dependences of these cloud quantities can be separated , on dimensional grounds , into @xmath9 @xmath10 @xmath11 where @xmath12 is the isothermal sound speed , which we take to be a constant , and the functions @xmath13 , @xmath14 and @xmath15 describe the angular distributions of the density , magnetic flux , and rotational speed . in order to be in a mechanical force balance , the cloud must satisfy the momentum equation , @xmath16= -\rho \nabla v - \nabla p + { 1\over 4\pi } ( \nabla\times { \bf b } ) \times { \bf b } , \label{e4}\ ] ] where @xmath17 is the thermal pressure , and the velocity @xmath18 has only a toroidal component . the gravitational potential @xmath19 is related to the density through poisson s equation @xmath20 it can be decomposed into a form ( toomre 1982 ) @xmath21 , \label{e6}\ ] ] where @xmath22 is the fractional over - density of the ( rotating ) magnetized configuration over the singular isothermal sphere that was introduced by li & shu ( 1996 ) . this overdensity arises from a combination of magnetization and rotation . the angular function @xmath23 is equivalent to the function @xmath24 of toomre ( 1982 ; see his equation [ 7 ] ) . substituting equations ( [ e1 ] ) , ( [ e2 ] ) , ( [ e3 ] ) and ( [ e6 ] ) into equations ( [ e4 ] ) and ( [ e5 ] ) and eliminating the function @xmath23 yield two coupled second - order ordinary differential equations ( odes ) for the angular distributions of density , @xmath13 , and magnetic flux , @xmath14 , @xmath25 = 2(r-1-h_0 ) \sin\theta , \label{e7}\ ] ] @xmath26 where the angular distribution of rotational speed @xmath15 remains to be determined . in the nonrotating limit that @xmath27 , these two coupled odes reduce to equations ( 12 ) and ( 13 ) of li & shu ( 1996 ) . the eliminated function @xmath23 that appears in the expression for gravitational potential , equation ( [ e6 ] ) , is related to other functions by @xmath28 . \label{e9}\ ] ] in the nonmagnetic limit @xmath29 , equation ( [ e8 ] ) demands that @xmath30 , i.e. , constant rotational speed everywhere . it is easy to verify the remaining two governing equations , ( [ e7 ] ) and ( [ e9 ] ) , reduce to equations ( 15 ) and ( 16 ) of toomre ( 1982 ; see also hayashi , narita & miyama 1982 ) , with slightly different notations . in this purely rotating case , toomre ( 1982 ) gave an exact solution for the angular distribution of density , which in our notation becomes @xmath31 , \label{e10}\ ] ] where the new variable @xmath32 is related to the polar angle @xmath33 through @xmath34 . equation ( [ e10 ] ) describes the so - called `` toomre - hayashi toroid '' . in the general case , both the flux function @xmath14 and the rotation speed @xmath15 are nonzero . this creates a potential problem , because fluid parcels along a given field line will generally wrap up the field , introducing a toroidal component that redistributes the fluid angular momentum , unless the angular speed @xmath35 is constant along the field line ( i.e. , matter and magnetic field corotate ) . the result follows from field freezing , i.e. , the induction equation of ideal mhd , which we have yet to consider , coupled with the condition @xmath36 . applied to an axisymmetric configuration of poloidal field lines , the presence of axial rotation , @xmath37 , will lead to a generation of a toroidal component at a rate : @xmath38 . \label{induceq}\ ] ] for there to be no toroidal field generation or change , the right - hand side has to be zero , i.e. , @xmath39 , which is the requirement of isorotation on field lines ( ferraro 1937 ) . in a self - similar toroid , isorotation requires @xmath40 where @xmath41 is a constant . unfortunately , self - similar magnetic field lines providing nontrivial cloud support extend infinitely far in cylindrical radius @xmath42 . if @xmath43 were to remain a nonzero constant as @xmath44 on a field line , then the linear rotational speed @xmath45 would increase without bound . the resulting divergent centrifugal force can not be balanced by any combination of the other self - similar forces in the problem . differential rotation along a field line is therefore unavoidable if the unbounded cloud is to start in mechanical force balance everywhere . alternatively , we may say that if @xmath43 has a nonzero constant value on a field line , then the quantity @xmath46 would yield an intrinsic length scale , and we would have to give up the simplifying assumption of self - similarity . we choose to retain exact self - similarity and forego instead isorotation on field lines . differential rotation along a field line , @xmath47 at the initial ( pivotal ) instant @xmath48 , extracts a price , and that price is an instantaneous wrapping of the magnetic field for @xmath49 via equation ( [ induceq ] ) . the spontaneous generation of toroidal field is an inevitable consequence of differential rotation in the problem , so one might wonder why we do not include a systematic toroidal field at the outset in the pivotal state . we do not do so for three reasons . first , it would introduce arbitrary extra parameters into a problem that is already sufficiently complex . second , axisymmetric models that contain toroidal fields which extend infinitely in the axial direction ( e.g. , fiege & pudritz 2000 ) generally have undesirable unclosed systems of currents that lead to large - scale separations of electric charge over time . third , observations of millimeter - wave and far infrared polarization patterns do not indicate substantive toroidal fields in the noncollapsing portions of molecular and dark clouds ( see , e.g. , ward - thompson et al . 2000 for the case of starless core l1544 ) . in our problem , the spatially limited generation of a finite level of @xmath50 slowly throws the cloud out of mechanical equilibrium in the poloidal directions . we regard this feature of the solutions to be less important than the fast poloidal motions that are generated because the pivotal state is unstable to inside - out dynamical collapse . thus , we shall focus our attention for @xmath51 on the magnetic torques exerted on exterior material connected to field lines dragged into a rotating , collapsing pseudodisk . we shall find the back reaction of those torques on the infalling material to be a formidable magnetic brake working against the formation of centrifugally supported disks . a simple way to introduce the necessary degree of differential rotation into the equilibrium toroid is to multiply the expression for @xmath15 in equation ( [ e11 ] ) by an extra factor of @xmath52 , so that @xmath53 we have checked by direct computation that models satisfying equation ( [ e12 ] ) have finite rotation speeds everywhere , with a moderate angular variation . the above choice is arbitrary , however , and a simpler , esthetically more appealing alternative is a constant rotational speed @xmath54 independent of @xmath33 . this choice for the pivotal state is suggested by the pre - pivotal evolutionary calculations of basu & mouschovias ( 1994 ) including both ambipolar diffusion and rotation . the choice also has the advantage of describing the toomre - hayashi toroid in the nonmagnetic limit . collapse calculations computed with pivotal states satisfying equation ( [ e12 ] ) yield results that are qualitatively similar to the cases with @xmath55 const . in addition , self - gravitating isothermal configurations with flat rotation curves have a long respected tradition in observational astronomy and theoretical astrophysics , with generalizations even possible into the relativistic regime ( cai & shu 2002 , 2003 ) and for nonaxisymmetric equilibria ( syer & tremaine 1996 , galli et al . 2001 ) . with @xmath54 = const @xmath56 , we can now cast the two coupled second - order odes ( [ e7 ] ) and ( [ e8 ] ) into a set of four first - order odes @xmath57 @xmath58 @xmath59 @xmath60 where the auxiliary functions @xmath61 and @xmath62 are defined from equations ( [ e15 ] ) and ( [ e16 ] ) . the above equations are supplemented by four boundary conditions , two each on the axis and at the equatorial plane . on the axis @xmath63 , we demand @xmath64 the first condition means that the magnetic flux enclosed by the polar axis vanishes , and the second that the polar axis contains no line mass ( see li & shu 1996 for a discussion ) . at the equatorial plane @xmath65 , we impose the conditions @xmath66 by symmetry . these translate to the requirements @xmath67 numerically , we integrate the set of first - order odes from small @xmath33 to larger values using a runge - kutta method . to initiate the integration , we use the following expansions at small angles that satisfy the boundary conditions near the polar axis listed in equation ( [ e17 ] ) : @xmath68 @xmath69 @xmath70 , \label{e22}\ ] ] @xmath71 where @xmath72 is the exponent of the dominant term in the expansion for the density @xmath13 . it reduces to the familiar form of @xmath73 for the nonrotating case ( li & shu 1996 ) . the constants @xmath74 and @xmath75 are free parameters of the expansions . they are determined , for any given pair of @xmath22 and @xmath76 , by shooting the solution towards the equator @xmath65 and matching the equatorial boundary conditions listed in equation ( [ e19 ] ) . other coefficients that appear in the expansions are related to these two parameters through @xmath77 @xmath78 @xmath79 equations ( [ e14 ] ) and ( [ e15 ] ) imply that the combination @xmath80 is a measure of the angle - averaged support provided by magnetic fields in comparison with the level provided by thermal pressure . when this combination vanishes , the magnetic field is zero , and the pivotal configuration becomes a hayashi - toomre toroid . when this combination is greater than zero , we may introduce a better measure of magnetization , the dimensionless mass - to - flux ratio defined in equation ( 37 ) of li & shu ( 1996 ) : @xmath81 a value of @xmath82 divides supercritical clouds from subcritical ones . since our models have no bounding surface pressures , all cases encountered in this paper are supercritical , @xmath83 . for example , the choice @xmath84 produces the angular distribution functions displayed in fig . [ fig : f1 ] . the integral of equation ( [ masstoflux ] ) then yields @xmath85 , a case considered by observers to be magnetically `` weak '' ( e.g. , crutcher 1999 ) . however , far from satisfying the intuition that magnetic fields are therefore ignorable in the subsequent collapse , the relatively weak " fields in the problem actually dominate many subsequent phenomena of crucial astrophysical interest e.g. , the formation of large pseudodisks , the transport of angular momentum , and the resulting size of centrifugally supported ( true " ) disks . numerically , we follow the inside - out collapse of the rotating toroid using zeus2d ( stone & norman 1992 ) , which solves the ideal mhd equations in three dimensions with an axis of symmetry . several modifications to the zeus2d code were needed to follow the toroid collapse ; these are discussed in paper i for the nonrotating case . we refer the reader to that paper for details . the presence of rotation makes it necessary to consider the toroidal component of magnetic field , which is already included in zeus2d . to illustrate the general features of the collapse solutions , we first consider as a standard for comparison the case with @xmath86 and @xmath87 . toroids with different combinations of @xmath22 and @xmath76 will be discussed in the next subsection . we first examine the time evolution of the mass accretion rate into the central cell ( point mass ) , which is displayed in fig . [ fig : massr ] . to facilitate comparisons with observations , we have plotted the abscissa in physical time units , adopting a fiducial value for the isothermal sound speed of @xmath88 km s@xmath89 . the mass accretion rate into the central cell is plotted in units of @xmath90 , and the asymptotic value , @xmath91 is presumably the correct self - similar solution ( for arbitrary values of @xmath12 and physical times @xmath92 0 ) if one could rid the calculation of numerical artifices . among such artifices is the possession of a strong initial peak in the accretion rate , reminiscent of the peak in the nonrotating case ( fig . 3d of paper i ) induced by an artificial point mass placed at the center to initiate the inside - out collapse . after a short period of adjustment , the central accretion rate quickly settles down to the more or less steady value of @xmath93 . the value is only slightly reduced ( by @xmath94% ) from that of the corresponding nonrotating case ( see paper i ) . evidently accretion into the central sink cell has not been significantly impeded by rotation . the most prominent feature of the nonrotating collapse solution is the formation of a dense , flattened pseudodisk in the equatorial region ( paper i ) , which collapses radially towards the center . inclusion of a modest amount of rotation does not significantly modify this basic density structure , as shown in fig . [ fig : n1r ] . in the figure , the initial state is plotted , along with the collapse solution at @xmath95 seconds after the initiation of collapse when artifacts from the starting conditions have essentially disappeared ( see fig . [ fig : massr ] ) . for the convenience of observers , we have again plotted the results in physical units using the canonical choice @xmath88 km s@xmath89 for low - mass cloud cores in taurus . ( scalings to different values for @xmath12 or to different times @xmath96 occur by the rules given in eq . [ 3 ] of paper i ; these rules continue to hold even when @xmath7 and @xmath50 are nonzero . ) the pseudodisk grows in the same way as it does in the nonrotating case . in fact , for the inner collapse region ( say within a reduced radius of @xmath97 or a dimensional radius of @xmath98 cm at a time @xmath99 s when @xmath100 km s@xmath89 ) , the density distributions of the rotating and nonrotating cases are practically indistinguishable from each other . the main difference lies in the rotation of the pseudodisk , which is of course nonzero for the rotating case . in panel ( c ) of fig . [ fig : n1r ] , we display contours of constant azimuthal speed . they show clearly that the pseudodisk , defined somewhat arbitrarily as the darkest region in the plot , has a significant rotation , with the ( linear ) rotational speed peaking in the middle section of the disk , in a `` ridge '' ( or a `` shell''in three dimensions ) that extends from the disk upwards into the envelope . the rotational speed has a maximum value @xmath101 above the sound speed . it falls off to subsonic values in regions both closer to the polar axis and further away . we note that , despite significant rotation , the motions inside the pseudodisk remain predominantly poloidal , as in the nonrotating case . another feature absent from the nonrotating case is the the twisting of magnetic field lines . in panel ( d ) of fig . [ fig : n1r ] , we plot contours of constant field pitch - angle , defined as @xmath102 , where @xmath50 and @xmath103 are the toroidal and poloidal field strength . clearly , ( differential ) rotation has distorted the initial , purely poloidal field configuration , although not by much . the maximum twist again occurs in a `` ridge '' , roughly where the rotation is fastest , with a pitch angle of order 20@xmath104 . in the polar region interior to the ridge , the plasma @xmath105 , and the field lines are essentially rigid against wrapping . outside the ridge , the collapsing flow has yet to be spun up enough to wrap the field lines appreciably ( even though the plasma @xmath3 is greater than unity and would allow large twists ) . a more graphical display of the 3d magnetic field is presented in fig . [ fig : btwist ] . the view obtained at 60@xmath104 inclination to the rotational and magnetic axis testifies to the relatively small amount of field wrapping that occurs in the problem because magnetic effects dominate those of rotation in the regions of interest . the pole - on view of the same magnetic field lines allows a clearer look at the top half of the twisted split - monopole configuration within the pseudodisk and permits a visualization of the magnetic torques ( @xmath106 , where @xmath107 is the cylindrical radius ) that lead to efficient angular momentum transfer and magnetic braking . the lack of significant field winding on scales of the outer pseudodisks and larger may explain why oh - maser polarization studies of magnetic - field directions in regions of massive star formation show surprising correlation with the large - scale magnetic field of the milky way galaxy on a scale of at least 2 kpc ( reid & silverstein 1990 , fish & reid 2003 ) . the implied accompanying loss of angular momentum in the gas during collapse may already have been detected in high - mass star - forming regions by keto , ho , & reid ( 1987 ) . in the absence of angular momentum transport , we would have expected the formation of a rotationally supported disk with approximate size @xmath108 , or @xmath109 cm at the time shown in fig . [ fig : n1r ] . no such disk has appeared in the actual calculation , perhaps because of the lack of adequate spatial resolution in the central regions , but more likely because of the efficiency of the magnetic braking , as we shall discuss in the next subsection . one might also question whether the assumption of field freezing is likely to hold at small scales where densities are high and ionization fractions become very low . we address this practical question in [ discussion ] ; here , we merely remark that a substantial layer of the pseudodisk is likely to remain suitably well coupled to magnetic fields almost right to the surface of the central protostar and makes equivocal any answer to whether or not one can expect the formation of true keplerian disks in the face of efficient magnetic braking . a somewhat unexpected feature of the collapse of the rotating toroid is the presence of an outflow region , part of which is already apparent in the upper - left corner of panel ( b ) of fig . [ fig : n1r ] from the velocity vectors . the highly distorted velocity contours ( particularly that of @xmath110 ) are another manifestation of its presence . in fig . [ fig : velr ] , we show the outflow more clearly over a larger region , including all four quadrants . the outflow has a relatively low speed , comparable to the sound speed @xmath12 . it is thus unrelated to the famous bipolar molecular outflows of many co observations . the same statement applies to the low - speed outflows obtained in tomisaka s ( 1998 , 2002 ) calculations , although tomisaka expresses contrary sentiments . nor are the motions to be confused with the fast - moving , hypersonic jet / wind that is thought to be driven from near the forming star ( knigl & pudritz 2000 ; shu et al . 2000 ) . some shaping of such winds , if they start out in a wide - angle configuration ( see shang et al . 2002 ) , might occur , if they , like their low - speed counterparts in fig . [ fig : velr ] , escape preferentially along the low - density , axial direction of the ambient medium . in any case , interaction between the fast - moving jet / wind of an embedded protostar with the transonic outflow / infalling envelope appears unavoidable , but we will postpone a treatment of this interaction to a future investigation . the slow outflow is driven by magnetic braking . as collapse proceeds , matter in the magnetically dominated polar region ( where the plasma @xmath111 ) first slides along field lines towards the equatorial plane to create the dense , pseudodisk . the pseudodisk is not an equilibrium structure ; its thermal and magnetic forces are too weak to balance the gravitational pull of the central object . as the material in the disk falls inward dynamically , it tends to spin up because of angular momentum conservation . however , the field lines threading the contracting pseudodisk also pass through the more slowly rotating envelope material higher up . the twisting of field lines in the pseudodisk generates a torsional alfvn wave , which propagates away from the equatorial plane . the wave removes angular momentum from the pseudodisk and deposits it in a small fraction of the envelope material near the @xmath112 line , which moves away as a low - speed wind . the outflow is concentrated in the `` ridge '' where the rotation is fastest and the field lines are most twisted ( see panels [ c ] and [ d ] of fig . [ fig : n1r ] ) . closer to the polar axis , the plasma @xmath113 and the wrapping of field lines unwinds almost instantaneously ; further away , the field lines are not twisted enough to reverse the collapsing inflow . we will show below that , as the field strength increases , the slow outflow grows wider and faster . the angular momentum redistribution due to magnetic braking is shown pictorially in fig . [ fig : n1rangle ] . in the figure , angular momentum is depleted from the depressed , `` valley '' in the equatorial region due to collapse . in the absence of magnetic braking , all of the depleted angular momentum should be deposited in the pseudodisk , where the `` mountain '' in the figure is . magnetic braking lowers the height of the `` mountain '' considerably , creating a `` ridge '' that runs more or less parallel to the polar axis . from the volumes beneath the `` mountain '' and the `` ridge '' , we estimate that more than @xmath114 of the angular momentum that would have accumulated in the pseudodisk is removed by magnetic braking in this particular case , although the fraction depends on the somewhat subjective choice of where the `` ridge '' ends and the `` mountain '' begins . for ideal mhd , magnetic braking is very efficient . it removes most of the angular momentum of the infalling material before it enters the central cell . as a result , no centrifugally supported , thin disk that we could resolve numerically forms throughout the run of the simulation . we believe that even with increased numerical resolution a rotationally supported disk will not form in the ideal mhd limit under the isothermal assumption . the reason is that in any ideal , self - similar , calculation that has been carried out correctly , the origin must contain a point mass that grows linearly with time , and this mass must trap a magnetic flux with a dimensionless mass - to - flux given by the the original ratio @xmath115 . the trapped flux emanates from the center as a split monopole ( galli & shu 1993b , li & shu 1996 ) , which causes the infalling matter to become more magnetically dominant and slowly rotating as it approaches the origin . the reason for magnetic domination is obvious . that for slow rotation is less so , and can be shown mathematically as follows . written in cylindrical coordinates , @xmath116 , the angular momentum equation for our problem reads @xmath117 , \label{angmomeq}\ ] ] where @xmath118 is the fluid specific angular momentum . if we integrate over a cylindrical volume of a small radius @xmath119 and infinite @xmath120 height , and if we approximate the fluid distribution to be confined to an infinitesimally thin pseudodisk of surface density @xmath121 , with @xmath122 , we obtain @xmath123 where @xmath124 equation ( [ totangmom ] ) states that the time rate of change of @xmath125 , the total fluid angular momentum contained within a circle of radius @xmath119 in the plane of the pseudodisk , plus the radial flux of fluid angular momentum across the circumference @xmath126 of the same circle , is equal to the torque exerted by the maxwell stress acting across a cylindrical surface of midplane circumference @xmath127 and infinite height in @xmath120 . for @xmath128 satisfying a split monopole geometry , the torque on the right - hand side of equation ( [ totangmom ] ) can be approximated as a negative constant times a representative mean value of @xmath129 on the important part of this cylindrical surface . equation ( [ induceq ] ) states , on the other hand , that under the assumption of strict field freezing , azimuthal field @xmath50 is generated at a rate @xmath130 if for small @xmath119 , @xmath43 were to approach anything resembling a keplerian field of differential rotation @xmath131 in the presence of a point mass @xmath132 at the origin , then @xmath50 would be generated at a highly divergent rate @xmath133 . this behavior would lead to divergent negative torques on the right - hand side of equation ( [ totangmom ] ) and rob the interior of angular momentum that it does not possess . the contradiction implies that centrifugally supported disks _ can not _ form in the presence of frozen - in magnetic fields of typical interstellar origin if they are dragged into the central regions of a pseudodisk configuration . in such circumstances , the angular velocity @xmath43 is more likely to reach a nearly constant value at small @xmath119 , saturating the spinup seen at large @xmath119 in the overall gravitational collapse . this combination of behaviors then explains why the _ linear _ velocity of rotation reaches a maximum value at some intermediate value of @xmath119 . since exterior gas must ultimately carry away the excess angular momentum transferred to them by magnetic torques , one way out of the dilemma posed above is to suppose that the magnetic fields on the outside can enforce strict isorotation on field lines , so that @xmath134 . in practice this enforcement is possible only in a near - vacuum environment where alfvn waves can propagate almost at infinite speed . in other words , we can expect braking to become less efficient once the toroidal alfvn wave breaks out of the cloud boundary , a development neither allowed by our simulation protocol nor theoretically attractive as a general solution to the surprising dilemma of _ too much _ outward angular - momentum transport . as we shall discuss in [ discussion ] , we believe a better alternative is ambipolar diffusion and/or magnetic reconnection to decouple eventually the pseudodisk from its attached magnetic field . this decoupling would diminish the braking efficiency and perhaps lead to a growing rotationally supported disk at some point during the gravitational collapse . the attractiveness of this proposal lies in the unified possibility of resolving two problems of long - standing in star formation without invoking additional physical processes : the loss of the magnetic flux associated with trapped interstellar fields , and the appropriate outward transport of angular momentum in disks and pseudodisks . to wit , an unmagnetized , rotating cloud will not collapse directly to a central point - like mass , whereas an ideally magnetized , nonrotating cloud can collapse to a central point - like mass without a rotationally supported disk . with more realistic physics ( see [ nonideal ] ) , a magnetized , rotating cloud may collapse to a protostar while forming a keplerian disk that could eventually produce a planetary system . alternatively , if the disk or pseudodisk becomes sufficiently massive in the process , nonaxisymmetric gravitational instabilities may take over to yield binary and/or multiple star systems ( see , e.g. , the speculations of galli et al . 2001 ) . we now explore , for a fixed value @xmath86 for the over - density parameter , a range of values for the dimensionless rotation rate : @xmath135 and @xmath136 . the corresponding dimensionless mass - to - flux ratios @xmath115 of the sequence are 4.51 , 4.61 , 4.94 , and 7.15 . thus , the sequence not only represents the effects of increasing rates of rotation on the collapse , but also those of a decreasing magnetic field ( both as measured against the support provided by thermal pressure ) . the self - similar results , all obtained by running the simulations long enough to eliminate unwanted transients , are shown in fig . [ fig : allss ] . to begin the discussion , we note that rotation has only a modest effect on the density distribution of the collapse solution . the pseudodisk is slightly thicker for a faster rotation , partially because the infall is retarded more by rotation , and partially because of a more vigorous outflow ( still very slow by bipolar outflow standards , not to mention stellar jets ) , which lifts up some of the disk material ( see panel [ d ] ) . the increasing strength of outflow as @xmath76 increases is most evident in the shape of the constant ( total ) velocity contour of @xmath137 , which gets more distorted for a larger @xmath76 . the distortion comes mostly from the toroidal component of the velocity , which can be close to the sound speed or larger in the magneto - rotationally driven outflow . an interesting feature , related to the strength of the outflow and the strength of the pre - collapse magnetic field , is that the magnetically dominated region enclosed by the @xmath112 line shrinks as @xmath76 increases . this effect is seen most clearly in the fastest rotating , most weakly magnetized @xmath138 case , where the enhanced density in the outflow has apparently pushed the @xmath112 line close to the polar axis . note also the pronounced distortions of the density contours and field lines caused by the outflow in this case . rotation decreases somewhat the rate of mass accretion onto the central point mass , a trend also present in the calculations of tomisaka ( 2002 ) and krasnopolsky & knigl ( 2002 ) . in steady state after the initial transient peak , the accretion rate is reduced , respectively , by about @xmath139 , @xmath140 and @xmath141 for @xmath142 and @xmath136 , in comparison with the nonrotating case @xmath143 . the reduction may be due to a combination of slower infall and mass removal in the outflow . however , the numerical outward diffusion of magnetic field exacerbated by artificial viscosity mediating shock waves at the boundaries of the central cell could also help to lower effective mass - accretion rates . in any case , for values of @xmath76 approaching the upper limit @xmath144 that would make the associated magnetic field in the model go to zero , we expect the central mass accretion rate to go to zero in any axisymmetric calculation that also suppresses gravitational torques . by explicit calculation , we verified the formal vanishing of @xmath145 into the central cell for nonmagnetic calculations in which @xmath76 had a large enough value ( e.g. , 0.5 ) to allow the numerical resolution of a centrifugally supported disk . for a nonmagnetic toomre - hayashi toroid , @xmath146 . the axisymmetric collapse of such a toroid should not directly produce a point mass at the center in principle . in practice , the numerical grid is finite , and the central cell will experience some mass accretion , particularly when the rotational speed @xmath76 is small . sooner or later , the numerical artifacts disappear from our simulations as the outward expansion of infall occupies more and more computational cells . the `` central '' accretion diminishes drastically , when most of the infall material lands on a spatially resolved , rotationally supported , disk of approximate size @xmath147 , in dimensionless units of @xmath148 . in an axisymmetric simulation , the accreted material simply accumulates in the disk because all processes of angular momentum removal or redistribution are suppressed in the absence of any magnetic and gravitational torques . eventually the disks become sufficiently massive that they are gravitationally unstable to axisymmetric transformation into rings ( toomre s 1964 instability ; see also goldreich & lynden - bell 1965 ) , a condition that plagues the long - time behavior of some of our computed nonmagnetic results . in a perfect self - similar collapse simulation , some inner part of the disk would always be unstable by toomre s criterion , but the difficulty does not manifest itself practically until the calculation has proceeded long enough for these regions to become sufficiently spatially resolved in a finite - difference scheme . the magnetized , rotating collapse is qualitatively different due to magnetic braking . angular momentum removal , particularly in the magnetically dominated region close to the origin ( where @xmath111 ) , allows infall material to accrete all the way onto the central cell , provided that ideal mhd holds . to begin our exploration of these effects of magnetic fields , we fix the rotational speed @xmath76 at the fiducial value of @xmath149 , and consider four different values of @xmath150 ( @xmath151 ) , 0.125 ( @xmath152 ) , 0.25 ( @xmath153 ) , and 0.5 ( @xmath154 ) . the @xmath150 case satisfies the nonmagnetic relation @xmath146 , so the mass - to - flux ratio is infinite . this rotating toroid is initially close to being spherical , containing only @xmath155 3% more mass than a singular isothermal sphere . it has a ratio of rotational to gravitational energy of @xmath156 , which is typical of molecular cloud cores ( goodman et al . 1993 ) . its collapse is followed with zeus2d , with the magnetic field turned off . the results are shown in panel ( a ) of fig . [ fig : allss2 ] . the collapse solution resembles that of the singular isothermal sphere shown in fig . 6a of paper i , except very close to the origin , where flattening of the mass distribution because of rotation becomes more prominent . the expected size of the rotationally supported disk , @xmath157 is too small to be resolved by our grid , or the scale of the figure , in this particular example . however , by keeping track of the mass and angular momentum that flowed into the central cell , we verified that angular momentum conservation holds to high accuracy in our modified version of the zeus2d code . as @xmath22 increases well beyond @xmath158 , most of the extra mass over the singular isothermal sphere is supported by magnetic fields rather than rotation . this is the case for @xmath159 and @xmath136 , whose collapse solutions are displayed in panels ( b)-(d ) of fig . [ fig : allss2 ] . these panels are to be compared with panels ( b)-(d ) of fig . 6 of paper i , where the corresponding nonrotating collapse solutions are plotted . in all cases , the density distribution shows a prominent pseudodisk , and the velocity field contains a slow outflow component . the former is basically a magnetic structure ; it increases in size and becomes more flattened as the field strength ( or @xmath22 ) increases , as expected . the latter is due to a combination of magnetic fields and rotation . we find that , for a given @xmath76 , as the field strength increases , the outflow becomes faster and wider . it also starts at a larger height above the pseudodisk , most likely as a result of a faster propagation speed for the torsional alfvn wave in the more strongly magnetized case . in each of the three magnetized cases , the slow outflow removes the bulk of the angular momentum originally associated with the accumulated mass in the central cell . determining the exact fraction of the angular momentum removed is somewhat uncertain . as a rough estimate , we divided the average specific angular momentum of the central - cell mass by that in the nonmagnetic case . the ratio typically turns out to be of order @xmath160 or smaller in our best resolved examples , and smaller than 0.1 in almost all cases . but in some sense the exercise is pointless , since in a perfectly resolved calculation , where the central object becomes a point mass , its angular momentum has to go to zero . the physically meaningful number is not how much angular momentum has to be transferred for matter to reach the center , but what is the mass - accretion rate achieved in accomplishing this feat , with and without magnetization in the problem . complicating a definitive answer to the latter question is the numerical instability described already in paper i that leads to oscillations of the magnetic field configuration as it becomes severely pinched and relaxes by numerically unpinching . such oscillations become more severe at late times in the simulation ( defined by when the innermost parts of the simulation become well resolved ) , and interfere thus in a fundamental way with a determination of the time - steady value of the central @xmath145 . in the interim , we can examine the effects of magnetic braking on rotation a little more directly . in fig . [ fig : ang ] , we plot the contours of constant rotational speed for the nonmagnetic and magnetic cases . note that the nonmagnetic cloud rotates fastest on small scales , and this rotation is braked significantly in all of the three magnetized cases , with the pseudodisk in the most strongly magnetized case rotating most slowly , as expected . a feature common to all magnetized cases is the strong differential rotation in and around the pseudodisk , with the rotational speed peaking in the middle sector where the outflowing `` ridge '' originates . the much smaller rotational speed in the region close to the axis ( interior to the ridge ) compared with that in the nonmagnetic case provides clear evidence that most of angular momentum is indeed stripped before mass enters the central cell . formation of centrifugally supported circumstellar disks is a central problem in star and planet formation . the initial distribution of angular momentum in the dense core , which collapses to form the star and disk , and the angular momentum redistribution during the collapse and disk accretion both affect the final outcome . we have shown that almost all of the core angular momentum can be removed in a low - speed outflow driven by magnetic braking during the collapse , provided that ( 1 ) the core is significantly magnetized and ( 2 ) the ideal mhd approximation holds . we now examine these two conditions from a practical point of view . the degree of magnetization of dense cores of molecular clouds prior to collapse is somewhat uncertain . in the standard scenario for isolated low - mass star formation ( shu et al . 1997 ; mouschovias & ciolek 1999 ) , the cores are envisioned to form out of a magnetically subcritical background cloud ( bounded at its surface by significant thermal or ram or turbulent pressure ) , with the evolution driven by ambipolar diffusion . the cores remain well magnetized , even during the supercritical phase of evolution , when their dimensionless mass - to - flux ratio @xmath115 is about 2 ( basu & mouschovias 1994 ; nakamura & li 2003 ) . this ratio , corresponding to @xmath161 ( li & shu 1996 , this paper ) , is consistent with the field strength obtained for the well studied cores l1544 ( crutcher & troland 2000 ) and b1 ( goodman et al . 1989 ) from zeeman measurements , after likely geometric corrections ( crutcher et al . 1994 ; shu et al . 1999 ; ciolek & basu 2000 ) . polarization has been detected in the dust continuum emission from a number of cores ( e.g. , ward - thompson et al . 2000 ; matthews & wilson 2002 ) , which also points to a significant degree of magnetization . in addition , the cores are known to be elongated in general , with a typical aspect ratio of 2:1 ( myers et al . 1991 ) ; the elongation is too large to be induced by rotation , and is consistent with the presence of a dynamically important magnetic field ( li & shu 1996 ) . even if the molecular clouds are magnetically supercritical to begin with and cores form in them dynamically , it would still be difficult to avoid having an @xmath80 at least as high as 0.2 to 0.3 , which corresponds to a weak field of @xmath162 g threading a slab of interstellar medium of @xmath163 . based on these arguments , we believe that the dense cores formed in present - day star - forming clouds should trap enough magnetic flux for magnetic braking to operate efficiently during gravitational collapse , as long as the field remains well coupled to the bulk of the neutral core matter . good coupling between the magnetic fields and neutral matter is maintained through collisions with ions in relatively low - density regions of a dark , molecular cloud core . in such regions , the primary source of ionization , and the time scale for ambipolar diffusion is typically an order of magnitude longer than the dynamic time scale , which means that there is little time for magnetic flux to be redistributed in a dynamically collapsing flow . thus , ideal mhd is a reasonable first approximation in collapse calculations . nishi et al . ( 1991 ; see also desch & mouschovias 2001 ) claim that good coupling is lost above a volume density of order @xmath164 @xmath165 . magnetic decoupling occurs for densities higher than this critical value , which reduces the efficiency of the magnetic braking . the decoupling density well exceeds the density in most of the pseudodisk , which is typically less than @xmath166 @xmath165 ( see the density contours in fig . [ fig : n1r ] for an example ) , except in one or two zones right next to the central cell . therefore , most of the angular momentum of the collapsing core material is already removed by magnetic braking before the braking is rendered ineffective by decoupling . let us now examine the supposition that the gas volume density will eventually exceed the critical value for decoupling . while true in rotationally supported protoplanetary disks with similar physical properties to the so - called `` minimum solar nebula , '' it is not necessarily the case for the dynamically collapsing pseudodisks , whose surface densities are much lower and provide less shielding against cosmic ray ionization . it is an easy matter to estimate the surface density @xmath121 of a flattened configuration accreting axisymmertically in quasi - steady state @xmath145 at radial speeds that approach free - fall values @xmath167 onto a central mass @xmath168 : @xmath169 within a factor of 2 or so , the inner regions of pseudodisks obtained by numerical simulation in this paper satisfy the estimate of equation ( [ surfdenff ] ) . galactic cosmic rays can keep a layer with a surface density of about 100 g @xmath170 sufficiently ionized as to couple well to magnetic fields ( nishi et al . thus , in the absence of a magnetized yso wind or jet that can sweep out cosmic rays before they penetrate pseudodisks , galactic cosmic rays will make a pseudodisk with a surface density profile given by equation ( [ surfdenff ] ) sufficiently well - ionized all the way to the surface of a protostar with characteristic radius @xmath171 au . even after a yso jet or wind turns on , if the event is accompanied by stellar magnetic activity and the emission of x - rays , the observed levels of x - rays in low - mass protostars are sufficient to keep a layer with the surface - density profile of equation ( [ surfdenff ] ) marginally well - coupled ( glassgold , najita , & igea 1997 ) . why then , should centrifugally supported disk ever arise ? because the coupling to magnetic fields in the pseudodisks , while good , is not perfect . nonideal mhd effects of two types have been described in the literature that can substantially modify the discussions considered so far in this paper . li & mckee ( 1996 ) focused on slippage of magnetic flux in the radial direction by the effects of ambipolar diffusion . they proposed that the magnetic flux decoupled this way from the mass that enters the star can drive a hydromagnetic shock against the collapsing inflow . behind the shock , the accretion flow can be slowed down by a large factor , and become magnetically dominated . it is conceivable that magnetic braking would still remove a large fraction of the remaining angular momentum in the slowly - contracting post - shock region ( krasnopolsky & knigl 2002 ) . galli & shu ( 1993b ) focused on the vertical stratification produced by the highly pinched magnetic field in the pseudodisk as a condition whereby magnetic flux can be lost by reconnection . as we also noted in this paper and paper i , infall in the pseudodisk tends to produce a split - monopole geometry with reversed poloidal fields across the midplane . in an ideal mhd calculation , the reversed magnetic fields across the midplane are supported by current sheets . in reality , the large electric current density near the midplane will dissipate by the presence of nonzero electrical resistivity . the dissipation of these currents corresponds to annihilation of the reversed fields above and below the midplane . the higher magnetic pressures of the surrounding regions will then press toward the middle , leading to further field annihilation . except for the continued introduction of new interstellar flux by infall into the pseudodisk , the entire magnetic flux threading the pseudodisk could eventually be lost by this reconnection process . in quasi - steady state , we therefore expect a `` sandwich - like '' structure to develop . the `` bread '' of the sandwich is the well - coupled , magnetized , regions where angular momentum can be removed at a sufficient pace as to allow the matter to spiral essentially unimpeded at dynamical rates toward the central protostar . this bread is the radially inward extension of the original pseudodisk , and it may already have been detected in the case of irs l1489 by the observations of boogert , hogerheijde , & blake ( 2002 ) . the `` meat '' of the sandwich is the part which is sufficiently shielded by the bread against external ionizing radiation as to become essentially decoupled from any magnetic fields . this decoupling causes eventual spinup of the inwardly spiraling material , because of the lack of magnetic braking , and the eventual formation of a centrifugally supported disk . this `` true '' disk is bounded above and below by the dynamically infalling remnant of the pseudodisk . how massive the meat becomes relative to the bread is , at this point , entirely a matter of speculation . it could vary considerably between regions of isolated star formation , such as taurus - auriga , and regions of crowded star formation , such as orion , because of the difference of ionizing radiation ( particulates and photons ) associated with star - birth activity . like many examples in astrophysics , moreover , such as the sun or other magnetically active stars , the reconnection process may occur not quiescently but explosively and intermittently in flares . it is interesting to speculate whether such intermittent behavior has anything to do with fu orionis outbursts . other effects may enter to further complicate the situation . the flow could become turbulent because of kelvin - helmholtz instabilities that arise from the velocity shear between the bread and meat parts of the sandwich , or because of interchange instabilities in 3d that manifest themselves in the pinched magnetic configuration of the split - monopole . under laminar conditions , resistive effects become important in the same density regime where ambipolar diffusion becomes competitive with the dynamics ( nishi et al . 1991 ) , so a combination of the effects discussed by galli & shu ( 1993b ) and li & mckee ( 1996 ) may be operative . all these effects operating together in lightly ionized media , under turbulent conditions , may considerably increase the size scales over which nonideal effects must be considered ( zweibel & brandenburg 1997a , b ) . challenging nonideal 3d simulations with adaptive mesh refinement may be needed to quantify such possibilities . magnetic braking in star formation has been studied previously by many authors , in connection with the long - standing `` angular momentum problem '' . most of the early studies were concerned with the relatively low - density , initial stage of star formation using semi - analytic treatments ( e.g. , mestel & paris 1984 ; nakano 1989 ; see mouschovias 1990 for a review ) . recent numerical calculations have extended the treatment to higher densities associated with the runaway core contraction either prior to star formation ( basu & mouschovias 1994 ) or slightly beyond the formation of a central object ( tomisaka 2002 ) . these calculations showed that , during the short runaway phase when the central density increases by several orders of magnitude , the angular momentum is nearly conserved , because the density increase involves little positional change of fluid elements in both radial and azimuthal directions ( and thus little pinching and twisting of field lines ) . our work deals exclusively with magnetic braking in the infall or accretion phase , _ after _ the formation of a compact central object . one may naively think that , once a molecular - cloud core starts to collapse dynamically , there should be little time for magnetic braking to operate . and yet we find that the bulk of the angular momentum , if not the total , can be removed during the dynamical collapse , even with a magnetic field of relatively moderate strength . why is there this difference between previous expectations and current calculations ? there are two key ingredients behind the efficient braking during the accretion phase : ( 1 ) the domination of magnetic energy over the rotational energy , and ( 2 ) the distinct geometry of the magnetic field . the foot points of the magnetic field lines associated with the matter already landed on the central object are pinned near the origin by accretion , creating a split monopole for the magnetic geometry ( see fig . [ fig : n1r ] ) . the split - monopole field lines push on the field lines outside ( see also li & shu 1997 ) , bending them into an hour - glass shape that is conducive to efficient magnetic braking . the enhancement comes from both the greater field strengths due to pinching and the radial fan - out ( as opposed to vertical collimation ) which lengthens the level arm for magnetic braking . these two effects are not taken into account in previous simple estimates . they are analogous to magneto - centrifugally driven winds ( knigl & pudritz 2000 ; shu et al . 2000 ) . indeed , the slowly moving outflow we find in the collapse solution is fundamentally similar to these winds , except for scale and for the ultimate origin of the field ( interstellar , disk dynamo , or stellar dynamo ) . magneto - rotationally driven outflows were also found in the collapse calculations of tomisaka ( 1998 , 2002 ) . these calculations start from infinitely long cylinders with magnetic field lines parallel to the axis . the magnetized cylinder is induced to break up into dense cores by a sinusoidal density perturbation along its length . the subsequent core evolution prior to the formation of a central object is more dynamic than that driven by ambipolar diffusion . it produces a pivotal state at time @xmath48 ( when a _ hydrostatic _ core starts to form in the calculation ) that differs substantially from the singular toroids that we adopted based on ambipolar diffusion calculations . nevertheless , soon after the appearance of the hydrostatic core , an outflow develops , which tomisaka attributed to core formation . a more precise cause for the outflow , we believe , is the trapping of a finite amount of magnetic flux ( rather than mass ) in a small ( core ) region , which gives rise to the central magnetic monopole responsible for the pinched magnetic geometry that is conducive to outflow driving . for numerical reasons , tomisaka s calculations were stopped a short time into the accretion phase , typically @xmath172 times the free - fall time at the center of the initial cylinder beyond the moment of hydrostatic core formation ( t=0 ) . during this brief period of time , the outflow appears to have removed most of the angular momentum of the material close to the origin ( tomisaka 2000 ) . in particular , no rotationally supported disk is able to form , consistent with our findings which , because of their self - similar nature , apply to all times @xmath51 . formation of rotationally supported disks during the collapse of dense cores formed in strongly magnetized molecular clouds was recently examined by krasnopolsky & knigl ( 2002 ) under thin - disk approximation . the material in the disk is connected through field lines to an external medium , which brakes the disk rotation with a parameterized efficiency . they find that , in the ideal mhd limit , a rotationally supported disk can form , provided that the braking efficiency is low enough ( see their fig . 3 ) . if the braking is more efficient , the disk angular momentum can be radiated away completely , as shown in their fig . our calculations and those of tomisaka ( 1998 , 2002 ) , which treat the dynamics of the material above and below the disk self - consistently , suggest that the braking efficiency is high enough to prevent the rotationally supported disk from forming in a strongly magnetized core , as long as ideal mhd holds . in the presence of ambipolar diffusion , krasnopolsky & knigl ( 2002 ) find that the bulk neutral material can retain a small rotation even in the strong braking case , because the neutral component must rotate relative to the charged component in order to feel the magnetic braking torque ( see their fig . whether the residual angular momentum is large enough in realistic situations to form a rotationally supported disk well outside the forming star remains to be seen . to conclude this subsection , we note an interesting feature of regarding numerical errors , which are always present , of course , in any simulation . simulations of self - similar configurations such as ours confer the advantage that although one does not know ( in analytically intractable situations ) what the exact details of the solution should be , one does know from dimensional considerations how each quantity should scale . this knowledge provides a powerful check on the present simulations , and therefore we have a good handle on the numerical reliability of the results quoted in this paper . rotation introduces a new feature into our collapse solution - the slow outflow ( see also tomisaka 1998 , 2002 ) . despite its crucial role in angular momentum redistribution , this outflow is unlikely to be observed soon : it is too easily masked by the faster - moving bipolar molecular outflows driven by a jet / wind from near the central object . another prominent feature of the collapse solution the flattened , high - density pseudodisk in the equatorial region should be less affected by the bipolar outflow . indeed , there are many examples of dense structures surrounding deeply embedded ysos that are elongated perpendicular to the outflows . they can be plausibly interpreted as the pseudodisks formed during the collapse of magnetized toroids or , on a larger scale , the more static toroids themselves , although detailed kinematic studies and direct measurements of the strength and morphology of magnetic fields are needed to strengthen the case . the best studied circumstellar structure of this type is perhaps the one around the source iram 04191 ( belloche et al . 2002 ) , which we discuss in some depth below . , motte , & bacmann 1999 ) . this class 0 source drives a well - developed molecular outflow , as all class 0 sources do . perpendicular to the outflow lies a flattened structure that is mapped in both dust continuum and several molecular lines ( belloche et al . the line study revealed that the flattened structure is infalling at a velocity between half and one isothermal sound speed , with a substantial amount of rotation . the ( linear ) rotational speed increases from @xmath175 at a radius of @xmath176 au to @xmath177 at @xmath178 au , where @xmath179 km s@xmath89 is the isothermal sound speed at 10 k ( see fig . 12c of belloche et al . 2002 ) . we identify the increase in rotational speed with the spinup in the outer part of the pseudodisk as a result of radial contraction under the approximate conservation of specific angular momentum , before significant magnetic braking sets in . in our model , the rotational speed peaks at some radius , and then decreases inward in a region of strongly pinched magnetic field that efficiently removes the angular momentum of the accreting flow , as shown in fig . [ fig : n1r]c . if the flattened structure of iram 04191 is indeed a pseudodisk , then its rotational speed should decrease inward inside @xmath0 3,500 au according to our model . this is consistent with , but not required by , the available spectral data , particularly the `` s''-shaped position - velocity diagrams of several molecular species . higher resolution observations are needed to test the model . the test may be hampered by the freeze - out of molecules onto dust grains in the high - density region near the protostar , as evidenced by the `` hole '' in the nh@xmath180 distribution around this source ( wootten et al . 2003 ) . interpreting the flattened structure of iram 04191 strictly in terms of the models of this paper is not without difficulties . in the fiducial case where @xmath87 ( comparable to the @xmath181 seen at 11,000 au in iram 04191 ) , the rotational speed achieves a peak near the observed value @xmath177 at a reduced radius of @xmath182 . if this corresponds to the radius of @xmath183 au for iram 04191 , then the collapse would have been initiated @xmath184 yrs ago , and the central star would have a mass of @xmath185 . but the dynamical time for the molecular outflow of iram 04191 is only @xmath186 yrs , much shorter than the above estimate . the low luminosity 0.15 @xmath187 of the central source also suggests a fairly small mass for the central star , accreting gas at not much higher than the standard rates for taurus ( see , e.g. , adams , lada , & shu 1987 ) . a flawed way to reconcile theory and observation , within the context of quasi - steady flow , is to postulate a significant temporal offset between the pivotal instant @xmath48 and the appearance of a powerful outflow in the system . such an offset requires , for example , some way to store infalling material ( from a pseudodisk ) in a rotationally supported disk without producing either much accretion luminosity or an outflow for a duration approaching @xmath188 yr . this task is not easy to do without inducing fierce nonaxisymmetric gravitational instabilities that would undermine the assumption of little mass accretion into the central regions . moreover , the observational limit claimed for the mass or size of any disk in iram 04191 is less than @xmath189 @xmath190 or 10 au . efficient magnetic braking has evidently prevented a more massive / larger disk from forming close to the star , and such braking works against the idea of appreciable mass storage in a disk . a better solution might be to suppose that , in addition to an offset between the pivotal instant and the onset of a wind , accretion onto the central source is time - variable , with fu orionis outbursts punctuating quiescent periods of relative calm ( hartmann , kenyon & hartigan 1993 ) . in such a picture , iram 04191 has just recently experienced such an outburst ( accounting for the associated powerful molecular outflow ) and has now faded considerably in its accretion luminosity ( accounting for the low luminosity currently ) . oscillatory inflow into the central regions in the presence of steady infall from the pseudodisk is a common bane of our simulations ( see the discussion in paper i ) . it remains to be seen whether the numerical artifact of field pinching and unpinching observed in the simulations might not have their physical analogues in more realistic calculations performed with nonideal mhd . another plausible option invokes a pre - existing contracting and rotating toroid to explain the observations of the regions at a few thousand au and beyond . subsonic cloud contraction prior to the formation of a formal density singularity for @xmath191 is consistent with the observed motions in both iram 04191 and the well - studied starless core l1544 ( tafalla et al . 1998 ) . as discussed already in paper i , subsonic contraction of the requisite magnitude might be initiated by the dissipation of turbulent support in dense molecular - cloud cores ( myers 1999 ) or resulted from ambipolar diffusion ( e.g. , ciolek & basu 2000 ) . we have performed simulations of the fiducial ( rotating and magnetized ) case with an initial uniform radial inflow @xmath192 equal to half the sound speed @xmath12 . as we found in paper i for the example without rotation , the results look like the configuration started without initial radial motion , except that the density profiles next to the axis are partially filled in by the additional inflow and the central accretion rate is increased by about a factor of 2 . to summarize , we have studied numerically the inside - out collapse of rotating , magnetized singular isothermal toroids . we find that a pseudodisk forms , just as in the nonrotating case , through which most of the core material is accreted . the pinched magnetic field in the pseudodisk removes most of the angular momentum of the accreted matter , and deposits it in a low - speed wind . the efficient magnetic braking during the protostellar accretion phase has important implications for the formation of rotationally supported disks that need to be pursued with nonideal mhd calculations . the complementary side to this phenomenon is the discovery of an extremely simple and efficient mechanism of angular momentum transport that allows direct accretion of extended rotating material onto relatively compact objects without the necessary concomitant formation of a standard viscous accretion disk . whether this mechanism has anything to do , for example , with the mini - spiral that seems to be feeding ionized gas , with low radiative efficiency , into the massive black hole at the center of our own galaxy ( lo & claussen 1983 ) remains to be seen . we have speculated in this paper that the process , with the inclusion of nonideal mhd effects , opens new vistas by which the two classical conundrums of star formation discussed long ago by mestel and spitzer the angular - momentum problem and the magnetic - flux problem - might be resolved , perhaps simultaneously , in the disks and pseudodisks surrounding newly formed stars .

we study motion of a charged particle confined to dirichlet layer of a fixed width placed into a homogeneous magnetic field . if the layer is planar and the field is perpendicular to it the spectrum consists of infinitely degenerate eigenvalues . we consider translationally invariant geometric perturbations and derive several sufficient conditions under which a magnetic transport is possible , that is , the spectrum , in its entirety or a part of it , becomes absolutely continuous .

1701.05714

a homogeneous magnetic field acting on charged particles has a localizing effect , both classically and quantum mechanically . since numerous physical effects are based on moving electrons between different places , mechanism that can produce transport in the presence of a magnetic field are of great interest . they typically require presence of an infinitely extended perturbation , a standard example being a barrier or a potential wall producing edges states , cf . @xcite and references therein . this is not the only possibility , though . in his classical paper @xcite iwatsuka demonstrated that a transport can be induced be a modification of the magnetic field itself under the assumption of a translational invariance , see also @xcite . the aim of the present paper is to show still another mechanism which can produce transport in a homogeneous field for particles confined to a layer with hard walls . as in the case of the iwatsuka model we will express the effect in spectral terms seeking perturbations that change the hamiltonian spectrum to absolutely continuous . our departing point is a flat layer of width @xmath0 to which a charged particle is confined and which is exposed to the homogeneous magnetic field perpendicular to the layer plane . the spectrum of such a system is easily found by separation of variables . it combines the landau levels with the dirichlet laplacian eigenvalues in the perpendicular direction , and needless to say that all the resulting eigenvalues are infinitely degenerate , see sec . [ sss : flatlayer ] below for more details . we are going to discuss _ geometric perturbations _ of such a system , in particular , deformations of the layer which are invariant with respect to translation in a fixed direction . such layers can be described , e.g. , as a set of points satisfying @xmath1 where @xmath2 is a surface obtained by shifting a smooth curve which can be parametrized by relation below . we are going to derive several conditions which ensure that the unperturbed pure point spectrum will change into an absolutely continuous one . more specifically , our main results can be summarized in the following assertion . let @xmath3 be the hamiltonian of a charged quantum particle confined to a layer @xmath4 of a constant width @xmath0 in @xmath5 built over a @xmath6-smooth , translationally invariant surface and exposed to a nonzero homogeneous magnetic field pointing in the @xmath7-direction . the spectrum of @xmath3 is purely absolutely continuous if together with technical assumptions [ eq : curv_bound ] and [ eq : diff_cond ] any of the following conditions is satisfied : a. @xmath4 is a one - sided - fold layer , @xmath8 or @xmath9 . furthermore , we suppose that the second part of [ eq : v_bound ] is fulfilled . b. @xmath4 is bent and asymptotically flat , @xmath10 for all large enough positive and negative @xmath11 , respectively , where @xmath12 $ ] . furthermore , one requires that @xmath13 and the halfwidth @xmath14 satisfies the bound described in lemma [ lem : ev_diff ] and remark [ rem : a_crit_bound ] below . moreover , for a fixed @xmath15 , the spectrum of @xmath3 below @xmath16 is absolutely continuous if @xmath4 is thin , i.e. the halfwidth @xmath14 is sufficiently small , and the generating surface satisfies one of the conditions specified in proposition [ prop : thin ] . the proof of the theorem will be given in sec . [ s : classes ] , before coming to it we will describe the geometry of the layer and explain the main steps of the argument . let us add a few remarks . first of all , in sec . [ sss : flatlayer ] we demonstrate that the perturbation must be geometrically nontrivial , because a mere tilt of the layer with respect to the field direction is not enough , with one notable exception . furthermore , except the claim ( i ) our condition impose restrictions on the layer thickness . on the other hand , the thinner the layer , the more general deformations we can treat . in particular , the last claim covers perturbations which are compact with respect to the @xmath17 variable , cf . proposition [ prop : thin ] . note also that the shift by @xmath18 in the last claim is needed ; without it the claim would be trivial because in a thin layer the spectral threshold is pushed up due to the dirichlet boundaries . the method used to treat thin layers is also useful with respect to the original iwatsuka model and its generalization including addition of a potential perturbation . recall , in particular , the conjecture stated in @xcite according to which _ any _ nontrivial translationally invariant magnetic perturbation gives rise to the purely absolutely continuous spectrum . despite a number of sufficient conditions derived after the original iwatsuka paper @xcite to which we add a new one in theorem [ theo : iwatsuka ] , the question in its generality remains open . in a similar vein , we are convinced that the sufficient conditions we find in this paper are by far not necessary . let @xmath2 be a surface in @xmath5 invariant with respect to translation in the @xmath19 direction and described by means of the following parametrization , @xmath20 with @xmath21 . the functions @xmath17 and @xmath7 here are assumed to be smooth enough , unless said otherwise we suppose they are @xmath22 , and such that @xmath23 , where the dot stands for the derivative with respect to @xmath11 . the last condition means that the curve @xmath24 in the @xmath25 plane is parametrized by its arc length measured from some reference point on the curve . therefore the ( signed ) curvature @xmath26 of @xmath27 may be fixed as @xmath28 and the corresponding unit normal vector to @xmath2 is @xmath29 let us stress that throughout the paper we assume that @xmath30 if we regard @xmath2 as a riemannian manifold , then the metric induced by the immersion @xmath31 is @xmath32 let @xmath33 and @xmath34 . we define the layer @xmath4 of width @xmath0 built over the surface @xmath2 as the image of @xmath35 we will always assume that @xmath36 under these conditions , @xmath37 is a diffeomorphism onto @xmath4 as one can see , e.g. , from the formula for the metric @xmath38 on @xmath4 , induced by the immersion @xmath37 , that reads @xmath39 where @xmath40 . the assumption [ eq : diff_cond ] implies , in particular , that @xmath41 holds for all @xmath42 . note that one can use @xmath43 as a natural transverse variable . the choice we made is suitable in situations when we want to discuss asymptotic properties of thin layers . the main object of our interest is the magnetic laplacian on @xmath4 subject to the dirichlet boundary condition , @xmath44 with a special choice of the vector potential , @xmath45 , that corresponds to the homogeneous magnetic field @xmath46 . using the unitary transform @xmath47 , @xmath48 , we may identify @xmath49 with the self - adjoint operator @xmath50 defined , in the form sense , on @xmath51 by @xmath52 where @xmath53 with @xmath54 by another unitary transform , @xmath55 , @xmath56 , we pass to the unitarily equivalent operator defined , again in the form sense , as @xmath57 where @xmath58 these formul are easy to derive , see e.g. @xcite or ( * ? ? ? remark that we needed @xmath6smoothness of @xmath2 to write @xmath59 down as an operator . nevertheless , @xmath60 may be introduced via its quadratic form for any @xmath61 surface . the translational invariance of @xmath4 makes it possible to pass finally to still another unitarily equivalent form of the operator by means of the fourier plancherel transform in the @xmath19 variable , @xmath62 which yields @xmath63 for a fixed @xmath64 , we define @xmath65:=-\partial_{s}f_a(s , u)^{-2}\partial_{s}+(\xi+\tilde{a}_{2}(s , u))^2-a^{-2}\partial_{u}^{2}+v(s , u),\ ] ] which allows us to write our hamiltonian in the form of a direct integral , @xmath66\,{\mathrm{d}}\xi,\ ] ] where @xmath67 is the momentum of the motion in the @xmath19 direction . note that since @xmath68 , the operator @xmath3 and its fiber @xmath69 $ ] act in @xmath70 and @xmath71 , respectively . as mentioned in the introduction we are interested in situations when the confinement causes a magnetic transport manifested through the absolute continuity of the spectrum . our aim is to describe several classes of layers @xmath4 for which the spectrum @xmath49 is purely absolutely continuous . since this operator is unitarily equivalent to the above described @xmath3 which in turn decomposes into a direct integral with fibers @xmath69 $ ] , by ( * ? ? ? * thm xiii.86 ) , it is sufficient to prove that a. the family @xmath72 \mid \xi\in{{\mathord{\mathbb r}}}\}$ ] is analytic with respect to @xmath67 in the sense of kato , [ suff_1 ] b. the resolvent of @xmath69 $ ] is compact for any @xmath64 , [ suff_2 ] c. no eigenvalue branch of @xmath69 $ ] is constant in the variable @xmath67 . [ suff_3 ] in the following subsections we address consecutively each of these points . [ rem : spec_prop ] in fact , to see that are sufficient conditions for the absolute continuity of @xmath3 , we have to modify ( * ? ? ? * thm xiii.86 ) slightly . in particular , the variable @xmath67 in the direct integral runs over a non - compact interval and the eigenvalues branches are not necessarily bounded . an alternative short proof is based on the main result of @xcite which says that under and , @xmath73 and @xmath74 may consist of isolated points without finite accumulation point only , and moreover , each one of these eigenvalues is of infinite multiplicity . since there are no finite accumulation points in @xmath74 and @xmath73 , we have @xmath75 , and consequently , it remains to check that @xmath76 . by ( * thm xiii.85 ) , @xmath77 is an eigenvalue of @xmath3 _ iff _ @xmath78)\}>0 $ ] , where @xmath79 stands for the one - dimensional lebesgue measure . since @xmath69 $ ] has compact resolvent by and it is unbounded but lower bounded , its spectrum consists of a sequence of eigenvalues of finite multiplicity which can accumulate only at @xmath80 . this together with implies that there are countably many eigenvalue branches . consequently , @xmath81 _ iff _ there is an eigenvalue branch , say @xmath82 $ ] , such that @xmath83=\lambda\}>0 $ ] . by , the function @xmath84-\lambda$ ] is real analytic , and by , it is non - constant . this means that the equation @xmath82=\lambda$ ] may have at most countable number of solutions and proves thus the claim . for any @xmath85 , we have in the form sense @xmath86=h[\xi_0]\dotplus p_{\xi},\ ] ] where the quadratic form @xmath87 is given by @xmath88 for any @xmath89 , one easily gets from here @xmath90\psi\rangle+\delta\langle\psi , v_{-}\psi\rangle , \end{split}\ ] ] where @xmath91 stands for the negative part of @xmath92 . if we assume that @xmath93,\ ] ] which is equivalent to the assumption that @xmath91 is relatively form bounded by @xmath94 $ ] , then @xmath69 $ ] is lower bounded and @xmath87 is infinitesimally form bounded by @xmath94 $ ] . in combination with ( [ eq : xi_pert ] ) this implies that @xmath69 $ ] forms an analytic family of type ( b ) , in particular , that @xmath69 $ ] is an analytic family in the sense of kato @xcite . assume now , in addition , that @xmath95 remark that under the second assumption , [ eq : neg_part_assump ] is trivially satisfied . by ( * ? * thm xiii.64 ) , the fiber @xmath69 $ ] has compact resolvent if and only if the set @xmath96,b}:=\ { \psi\in q(h[\xi ] ) \mid \|\psi\|\leq 1,\ , \langle\psi , h[\xi]\psi\rangle\leq b\}\ ] ] is compact for all @xmath97 . by [ eq : diff_cond ] we have @xmath98 , and therefore @xmath86\geq -d^{-2}\partial_{s}^{2}+(\xi+\tilde{a}_2)^2-a^{-2}\partial_{u}^{2}-\|v_{-}\|_\infty.\ ] ] moreover , there exists clearly an @xmath99 such that for all @xmath100 , we have @xmath101 if we introduce the constant @xmath102 then we can estimate @xmath69 $ ] from below as follows , @xmath86\geq -d^{-2}\partial_{s}^{2}+\left(\xi+\frac{b_0}{2}x(s)\right)^2-a^{-2}\partial_{u}^{2}-k-\|v_{-}\|_\infty=:h_{-}-k-\|v_{-}\|_\infty.\ ] ] the operator @xmath103 on the right - hand side decomposes into a sum of the one - dimensional harmonic oscillator hamiltonian and the dirichlet laplacian on @xmath104 , hence it has compact resolvent ( * ? ? ? * thm xiii.64 ) , and consequently , @xmath105 is compact for any @xmath97 . clearly , @xmath106,b}\subset\mathscr{c}_{h_{-},b+k+\|v_{-}\|_\infty}$ ] , which means that @xmath106,b}$ ] has to be precompact , but at the same time @xmath106,b}$ ] is closed , cf . the proof of ( * ? ? ? * thm xiii.64 ) , and thus compact . by ( * thm vii.4.3 ) , the fiber @xmath69 $ ] has compact resolvent either for all @xmath67 or for no @xmath67 at all . it is easy , however , to demonstrate the resolvent compactness directly for any fixed @xmath67 in the above described way . in the following section , we will demonstrate this property for several special but still rather wide classes of layers indicated in the introduction . specifically , we will be concerned with the following cases : a. _ a one - sided - fold layer : _ @xmath8 or @xmath9 , b. _ a bent , asymptotically flat layer : _ @xmath107 for all large enough positive @xmath11 and @xmath108 for all large enough negative @xmath11 , where @xmath12,\ , \alpha_+\neq\alpha_-$ ] , c. _ a thin non - planar layer : _ @xmath14 is sufficiently small . let us remark that there are basically two methods how to demonstrate non - constancy of the eigenvalues . the first relies on the feynman - hellmann formula that gives the derivative of an eigenvalue with respect to a parameter . it is useful in situations when the curvature is compactly supported . the other method is based on some type of a comparison argument that should give asymptotic behavior of the eigenvalues at @xmath109 . it is usually applied in situations when the curvature behaves differently at @xmath109 . recall that throughout the paper we assume [ eq : curv_bound ] and [ eq : diff_cond ] . in this section , we will suppose that [ eq : v_bound ] ( which in turn implies [ eq : neg_part_assump ] ) is always satisfied , too . for the sake of definiteness , assume that @xmath8 . then @xmath110 holds for all @xmath111 . recall that @xmath112 which makes it possible to estimate @xmath113 . this in turn implies that @xmath114 is compactly supported . for any @xmath115 we have @xmath116 and therefore @xmath86\geq -\partial_s f_{a}^{-2}\partial_s - a^{-2}\partial_{u}^{2}+\tilde{a}_{2}^{2}+\xi^2 - 2\xi\|(\tilde{a}_2)_{-}\|_\infty-\|v_{-}\|_\infty.\ ] ] the first three terms on the right - hand side are positive and for the remaining part , independent of @xmath11 and @xmath117 , we have @xmath118 thus to any @xmath119 there is a @xmath120 such that @xmath69>c$ ] holds for all @xmath121 , and consequently , no eigenvalue branch may be constant as a function of @xmath67 . before proceeding further we are going to show that a mere rotation of the layer around an axis perpendicular to the magnetic field is not sufficient with one notable exception to produce a transport in the considered system . the decisive quantity is the tilt angle between the field direction and the layer . let @xmath122 stands for the angle between the magnetic field @xmath123 and the normal vector to the layer @xmath124 . in the case of the unperturbed system , a planar layer with a perpendicular field ( @xmath125 ) , it is straightforward to see by separation of variables in the cylindrical coordinates that the spectrum of @xmath3 is purely point . the same is true if @xmath126 holds for all @xmath127 . indeed , each of the fiber operators @xmath86=-\partial_{s}^{2}+\left(\xi+b_0(s \cos{\gamma}-au\sin{\gamma})\right)^2-a^{-2}\partial_{u}^{2}\ ] ] is unitarily equivalent to @xmath128 $ ] , as one can verify employing a unitary transform @xmath129 using then ( * ? ? ? * thm xiii.85 ) in combination with the fact that @xmath128 $ ] has compact resolvent , we conclude that @xmath130)$ ] consists of infinitely degenerate eigenvalues . for @xmath125 , this yields an alternative way to determine the spectral character in the unperturbed case by observing that @xmath86=-\partial_{s}^{2}+(\xi+b_{0}s)^2-a^{-2}\partial_{u}^{2}\ ] ] decomposes then into the sum of the hamiltonian of the harmonic oscillator with the origin shifted by @xmath131 and the dirichlet hamiltonian on the line segment @xmath104 . for its eigenpairs , @xmath132,\psi_{m , n}[\xi])$ ] , we have @xmath133\equiv \lambda_{m , n}=b_{0}(2m+1)+\left(\frac{n\pi}{2a}\right)^2\\ & \psi_{m , n}[\xi](s , u)=\psi_{m}(s+\xi / b_{0})\chi_{n}(u),\end{aligned}\ ] ] with @xmath134 where @xmath135 stands for the @xmath136th hermite polynomial , and @xmath137 here @xmath138 and all the eigenfunctions are normalized to one in the respective hilbert spaces . since @xmath139 $ ] is independent of @xmath67 , all the eigenvalues of @xmath3 have infinite multiplicity . let us ask about additional degeneracies , that is , about the multiplicity of eigenvalues of @xmath69 $ ] . assume that @xmath140 for some @xmath141 and @xmath142 such that @xmath143 and @xmath144 . this means that @xmath145 where @xmath146 , which implies that @xmath147 is a positive rational , @xmath148 . conversely , if this is the case then @xmath149 , for some @xmath150 , and the equation @xmath151 has an infinite number of solutions @xmath152 satisfying @xmath153 in other words , every eigenvalue of @xmath69 $ ] has infinite multiplicity . on the other hand , in the case @xmath154 the spectrum @xmath69 $ ] is simple but it becomes ` denser ' as the energy increases . in other words , the eigenvalue gaps have no positive lower bound . indeed , by dirichlet s approximation theorem , for all @xmath155 there exist @xmath150 such that @xmath156 and @xmath157 . we will be concerned about large values of @xmath158 , so we may assume that @xmath159 . the equation ( [ eq : degen ] ) has a infinite number of solutions @xmath152 satisfying @xmath160 for any of these solutions we obtain @xmath161 we infer that for any @xmath162 there are infinitely many pairs of eigenvalues @xmath163 with the property that @xmath164 which proves our claim . the situation changes when the tilted layer has the right angle with the original one becoming thus parallel to the field direction . then we have @xmath165 , and therefore @xmath166=-\partial_{s}^{2}+\left(\xi - b_0 au\right)^2-a^{-2}\partial_{u}^{2}=\overline{t_1\otimes i + i\otimes t_2[\xi]},\ ] ] where @xmath167:=-a^{-2}\partial_{u}^{2}+\left(\xi - b_0 au\right)^2,\ ] ] and we decomposed @xmath168 . the other choice @xmath169 gives rise to a unitarily equivalent operator . since @xmath170 and @xmath171 $ ] has a positive simple pure point spectrum , @xmath172)=[\inf\sigma(t_2[\xi]),+\infty)$ ] for all @xmath64 . in fact , we are going to prove that every @xmath69 $ ] is purely absolutely continuous . let @xmath173 , and @xmath174 be spectral families of the operators @xmath175 $ ] and @xmath69 $ ] , respectively . then for all decomposable vectors @xmath176 we have @xmath177}(u+s){\mathrm{d}}_u\|e_1(u)f_1\|^2 { \mathrm{d}}_s\|e_2(s)f_2\|^2\\ = \|e_1(\cdot)f_1\|^2 * \|e_2(\cdot)f_2\|^2((-\infty , t]);\end{gathered}\ ] ] for the first equality we refer here to the proof of ( * ? ? ? * thm 8.34 ) . using the fubini theorem it is easy to check that the convolution of an absolutely continuous measure with another ( borel ) measure is also absolutely continuous . consequently , the absolutely continuous subspace of @xmath178 contains all finite linear combinations of the decomposable vectors . however , these functions form a dense subspace , and moreover , the absolutely continuous subspace is always closed . this allows us to infer that @xmath69 $ ] is purely absolutely continuous . the fact that the absolute continuity of one of the operators in the decomposition implies the absolute continuity of the ` full ' operator is a common folk knowledge usually referred to , not quite exactly , to standard sources , see e.g. @xcite ; we prefer to include the above simple proof . let @xmath179,\varphi_n[\xi]),\ , n\in{{\mathord{\mathbb n}}}$ ] , denote the eigenpairs of @xmath171 $ ] , where the eigenvalues are numbered in the ascending order . some important properties of the @xmath180 $ ] s are reviewed in @xcite . in particular , @xmath180 $ ] are even and strictly increasing for @xmath115 , thus they have the only stationary point at @xmath181 . since @xmath172)=[\mu_1[\xi],+\infty)$ ] , @xmath182,+\infty)$ ] . hence the bottom of @xmath183 is given by the first eigenvalue of the harmonic oscillator constrained to the line segment @xmath104 . this question was addressed repeatedly in the literature , see e.g. @xcite , and it is easy to see that the answer is given by the smallest solution of the equation @xmath184 with respect to @xmath185 . here @xmath186 stands for the kummer confluent hypergeometric function . if we denote this solution @xmath187 , then @xmath188 , hence it is sufficient to inspect the dependence of @xmath185 on one of the parameters . the solution to the spectral condition can not be written in a closed form but can be found numerically , see fig . [ fig : bottom ] . moreover using known asymptotic properties of the kummer functions @xcite one can find the behavior of @xmath189 , @xmath190 as @xmath191 . to derive the asymptotic behavior of @xmath189 at large values of @xmath192 we begin with the variational characterization of the lowest eigenvalue , @xmath193|_{a=1}=\inf_{\psi\in h_0 ^ 1(i),\|\psi\|_i=1}\left(\|\psi'\|^2_i+b_0 ^ 2\|u\psi\|^2_i\right).\ ] ] let @xmath194 be the ground state of the one - dimensional harmonic oscillator , see for its explicit normalized form , then clearly @xmath195 if we put @xmath196 then @xmath197 and @xmath198|_{a=1}\tilde\psi\rangle_i=\frac{1}{\|\tilde\psi\|_i^{2}}\langle\tilde\psi , b_0\tilde\psi+b_0\psi_0(1)-b_0 ^ 2\psi_0(1)u^2\rangle_i\\ \leq \frac{1}{\|\tilde\psi\|_i^{2}}\langle\tilde\psi , b_0\tilde\psi+b_0\psi_0(1)\rangle_i = b_0+b_0\psi_0(1)\frac{1}{\|\tilde\psi\|_i^{2}}\langle\tilde\psi,1\rangle_i\\ \leq b_0+b_0\psi_0(1)\frac{1}{\|\tilde\psi\|_i}|i|^{1/2}.\end{gathered}\ ] ] next , for any fixed @xmath199 we have @xmath200 provided @xmath192 is sufficiently large . we conclude that , for any @xmath201 , @xmath202 as @xmath203 . putting this together with we arrive at the expansion @xmath204 valid as @xmath205 . as a function of @xmath192 ( red solid line ) ; @xmath206 . the black loosely dotted line corresponds to the first dirichlet eigenvalue on @xmath207 , i.e. , @xmath208 . the black densely dotted line corresponds to the asymptotic expansion at @xmath209 and the black dashed line to the asymptotic expansion at @xmath210.,width=377 ] in turn , @xmath3 is also purely absolutely continuous by ( * ? ? ? * thm xiii.85 ) . the present situation is particular because @xmath3 is also also invariant with respect to @xmath11-translations , i.e. in the @xmath7 direction . using the partial fourier plancherel transform in @xmath11 , we find that @xmath211 is unitarily equivalent to the operator having the following direct integral decomposition @xmath212{\mathrm{d}}\eta:=\int_{{{\mathord{\mathbb r}}}}^{\oplus}(\eta^2+t[0])\ , { \mathrm{d}}\eta,\text { where } t[0]=\left(-i\partial_y - b_0 au\right)^2-a^{-2}\partial_{u}^{2}.\ ] ] the physical contents of this decomposition is obvious : @xmath213 $ ] is a purely absolutely continuous operator describing the edge - state - induced transport in a two - dimensional dirichlet strip due to the perpendicular magnetic field @xcite , and so is @xmath214 $ ] . the difference between @xmath214 $ ] and @xmath213 $ ] which ` adds ' to the absolute continuity of @xmath3 is the square of the momentum in the @xmath7 direction where the motion is free . the decomposition is also reflected in the unitary propagator for @xmath3 which is given by a direct integral of the unitary propagators for @xmath69,\,\xi\in{{\mathord{\mathbb r}}}$ ] . since each @xmath69 $ ] is separated , we have @xmath215)}=\overline{\exp{(-itt_1)}\otimes\exp{(-itt_2[\xi])}}\ ] ] by ( * ? ? ? * thm 8.35 ) . in the @xmath7 direction the evolution governed by the propagator with the well - known kernel @xmath216 while the second propagator in the tensor product decomposition can be expressed in terms of the eigenpairs of @xmath171 $ ] describing the edge states as @xmath217)}=\sum_{n\in{{\mathord{\mathbb n}}}}\mathrm{e}^{-it\mu_n[\xi]}\vert\varphi_n[\xi]\rangle\langle\varphi_n[\xi]\vert.\ ] ] the latter describes the well - known edge - current dynamics @xcite , the additional degree of freedom is a free motion of the wavepackets in the @xmath7 direction having the usual properties , in particular , the spreading with time ( * ? ? ? 9.3 ) . let us return now to geometrically nontrivial perturbations of the layer and assume they are localized at any fixed @xmath19 cut , i.e. , that the layer is flat for @xmath11 outside a compact interval . specifically , we suppose that @xmath218 holds for all @xmath11 large enough positive , @xmath219 for all @xmath11 large enough negative , and @xmath220 . hence there are numbers @xmath99 and @xmath221 such that @xmath222 holds for all @xmath223 and @xmath224 holds similarly for all @xmath225 . recall that in view of the chosen layer parametrization we have @xmath226 , their positivity means that we exclude the situation where the layer is asymptotically parallel with the magnetic field . with the future purpose in mind we also assume that _ the magnitudes of the asymptotic slopes are different _ , @xmath227 ; without loss of generality we may suppose that @xmath228 , since in the opposite case is is sufficient to change the layer parametrization replacing @xmath11 by @xmath229 . remark that we have chosen @xmath230 , that determine the signs of the asymptotic slopes , to be non - negative just for definiteness . changing the sign either of one or the both limits changes nothing in the considerations below . let us start with the first case assuming that the parameter @xmath11 is sufficiently large and positive . let @xmath231 be a solution of the equation @xmath232 obviously , for all @xmath67 sufficiently negative @xmath233 exists , being of course unique . in the following we restrict ourselves only to those values of @xmath67 . it is straightforward to check that @xmath234 and @xmath235 holds for all @xmath223 . consequently , the fiber @xmath69 $ ] acts for all @xmath223 in the same way as the following positive operator , @xmath236 next we introduce the unitary transform @xmath237 , then @xmath238 and @xmath239:=u_{s_1}^{-1}h[\xi]u_{s_1}$ ] acts as @xmath240= \begin{cases } h_{\alpha_{+ } } & \quad s\geq s_0-s_1\\[.7em ] \begin{aligned}&-\partial_{s}f_a(s+s_1,u)^{-2}\partial_{s}+(\xi+\tilde{a}_{2}(s+s_1,u))^2\\ & -a^{-2}\partial_{u}^{2}+v(s+s_1,u)\end{aligned } & \quad s < s_0-s_1 . \end{cases}\ ] ] [ lem : top_eq ] let @xmath67 and @xmath233 be as above . then , for all @xmath67 sufficiently negative , there exist constants @xmath241 and @xmath242 , the latter being independent of @xmath67 , such that @xmath243 , @xmath244\leq c_{+}(\xi)h_{\alpha_{+}}+k_{+},\ ] ] and the @xmath241 have finite positive limits as @xmath245 . for all @xmath246 we have @xmath247\leq-(1-a\|\kappa\|_\infty)^{-2}\partial_{s}^{2}+(\xi+\tilde{a}_{2}(s+s_1,u))^2-a^{-2}\partial_{u}^{2}+\|v\|_\infty.\ ] ] since @xmath59 is continuous and compactly supported , @xmath248 . given @xmath127 , put @xmath249 , then @xmath250 , @xmath251 on @xmath252 , @xmath253 on @xmath254 , and @xmath255 . hence for all @xmath67 large enough negative , i.e. for @xmath256 sufficiently large there exist @xmath257 such that @xmath258 and @xmath259 on @xmath260 . moreover , since @xmath261 , we conclude that for all @xmath67 sufficiently large negative there are @xmath262 such that @xmath263 , and @xmath264 holds on @xmath265 . a closer inspection shows that @xmath266 may be chosen in such a way that @xmath267 holds as @xmath268 . now we can proceed with estimate , @xmath269\leq -(1-a\|\kappa\|_\infty)^{-2}\partial_{s}^{2}+\hat c_{+}^{2}b_0 ^ 2(\alpha_{+ } s - au\sqrt{1-\alpha_{+}^2})^2-a^{-2}\partial_{u}^{2}+\|v\|_\infty\\ \leq \max\left\{(1-a\|\kappa\|_\infty)^{-2 } , \hat c_{+}^{2}\right\}h_{\alpha_{+}}+\|v\|_\infty,\end{gathered}\ ] ] for all @xmath246 . for @xmath270 , this bound holds trivially , too . in a similar manner one can estimate @xmath239 $ ] from below putting @xmath271 if @xmath67 is sufficiently large positive the argument is a simple modification of the above one . we define @xmath272 as the unique solution of the equation @xmath273 where @xmath221 was introduced in . clearly , @xmath274 and the operator @xmath275:=u_{\tilde s_1}^{-1}h[\xi]u_{\tilde s_1}$ ] acts as @xmath276= \begin{cases } h_{\alpha_{- } } & \quad s\leq \tilde s_0-\tilde s_1\\[.7em ] \begin{aligned}&-\partial_{s}f_a(s+\tilde s_1,u)^{-2}\partial_{s}+(\xi+\tilde{a}_{2}(s+\tilde s_1,u))^2\\ & -a^{-2}\partial_{u}^{2}+v(s+\tilde s_1,u)\end{aligned } & \quad s>\tilde s_0-\tilde s_1 . \end{cases}\ ] ] the operator pair @xmath275 $ ] and @xmath277 satisfies an estimate analogous to . now we are in position to prove an important convergence result . [ prop : res_conv ] let @xmath278 , then we have @xmath279+\mu)^{-1}-(h_{\alpha_{\pm}}+\mu)^{-1}\big\|=0.\ ] ] it is sufficient to apply ( * ? ? * thm 2.3 ) . to make the paper self - contained we reproduce this result here : let @xmath280|\ , \alpha\in(-\infty,+\infty]\}$ ] be a one parametric family of lower bounded selfadjoint operators on @xmath281 , where @xmath282 is open , with the following properties 1 . [ prop : core ] @xmath283 is a core of @xmath284 $ ] for all @xmath285 $ ] . [ prop : low_bound ] there exist @xmath119 and @xmath286 such that , for all @xmath287 , @xmath288+k\leq a[\alpha]$ ] . [ prop : conv ] for any compact set @xmath289 , there exists @xmath290 such that , for all @xmath291 , @xmath284|_{c_{0}^{\infty}(\mathcal{k})}=a[+\infty]|_{c_{0}^{\infty}(\mathcal{k})}$ ] . [ prop : comp ] @xmath292 $ ] has compact resolvent . then , for any @xmath293)$ ] and @xmath162 , there exists @xmath294 such that for all @xmath295 , @xmath296)$ ] and @xmath297-z)^{-1}-(a[+\infty]-z)^{-1}\big\|<\varepsilon.\ ] ] to deal with the limit @xmath245 we put @xmath298=h_{+}[-\xi],$ ] and @xmath292=h_{\alpha_{+}}$ ] . the properties ( i ) and ( ii ) above are direct consequences of lemma [ lem : top_eq ] , ( iii ) is obvious from the definition of @xmath239 $ ] , and ( iv ) was proved in sec . [ sec : comp ] . the limit @xmath299 is treated in a similar manner . let us denote the eigenvalues of @xmath300 , arranged in the ascending order with the multiplicity taken into account , by @xmath301 . since the norm - resolvent convergence implies the convergence of eigenvalues , we see that in any neighborhood of @xmath302 , there is exactly the same number of eigenvalues of @xmath69 $ ] as is the multiplicity of @xmath302 in the spectrum of @xmath300 , provided @xmath67 is chosen sufficiently large negative . similarly , in any neighborhood of @xmath303 , there is exactly the same number of eigenvalues of @xmath69 $ ] as is the multiplicity of @xmath303 in the spectrum of @xmath277 , provided @xmath67 is positive and sufficiently large . moreover , if we fix @xmath304 then for all @xmath302 less than @xmath174 we may choose the said neighborhoods to be disjoint and to prove that in the remaining gaps there are no eigenvalues of @xmath69 $ ] for all @xmath67 sufficiently large negative . again , a similar statement holds true for large positive values of @xmath67 . in general , it may occur that the eigenvalue branches of @xmath69 $ ] cross . it can not happen , however , that a non - constant eigenvalue branch crosses a constant branch . in fact , if there is a constant eigenvalue branch then it has to be isolated from the rest of the spectrum , cf . remark [ rem : spec_prop ] above , hence it makes sense to denote it as @xmath305 $ ] , since it is indeed the @xmath136th eigenvalue of @xmath69 $ ] , with the multiplicity taken into account , for some @xmath306 and all @xmath64 . if it is independent of @xmath67 we would have @xmath305=\sigma_m(\alpha_{+},b_0 ) = \sigma_m(\alpha_{-},b_0)$ ] ; our aim now is to find a sufficient condition under which this can not happen . to this aim , let us denote @xmath307 and fix an @xmath162 . then , since @xmath308 we get the inequalities @xmath309 this allows us to infer that @xmath310 where @xmath311 . with respect to the first terms on the right - hand sides of the above inequalities , recall that @xmath312 for some @xmath313 and @xmath314 , where @xmath315 here , @xmath316 is the @xmath317th ` transverse ' dirichlet eigenvalue for @xmath206 . furthermore , note the monotonicity with respect to the field : if @xmath318 , then @xmath319 holds for all @xmath306 . this follows from the fact that we have @xmath320 & & \leq \lambda_{\tilde m , n}(b)\leq e-(\tilde b - b)\}|,\end{aligned}\ ] ] for all @xmath304 and @xmath321 , where @xmath322 stands for the cardinality of a set . now using the minimax principle we obtain @xmath323 combining this with we arrive at the following claim . [ lem : ev_diff ] let @xmath324 and @xmath325 then we have @xmath326 for all positive @xmath327 . if , in addition , @xmath328 holds for some @xmath327 , then @xmath329 . [ rem : a_crit_bound ] one can not maximize the threshold @xmath330 with respect to @xmath331 analytically . however , it is possible to find a closed - form estimate by maximizing a lower bound . first of all , note that @xmath332 holds for all @xmath333 $ ] . hence @xmath334 , where @xmath335 , and @xmath336 for all @xmath337 . the bound is maximal for @xmath338 , so we arrive at @xmath339 by a _ reductio ad absurdum _ we can thus make the following conclusion . under the assumptions of lemma [ lem : ev_diff ] , there are no constant eigenvalue branches of @xmath69 $ ] . the result of the previous section involved already a restriction to the layer thickness possibly going beyond the assumption [ eq : diff_cond ] , the severity of which depended on how much the layer was ` broken ' . now we will go further and look what sufficient condition can be derived if the layer is very thin . to begin with , recall that it was proved in @xcite that if , in addition to [ eq : curv_bound ] and [ eq : diff_cond ] , @xmath340 then for any @xmath341 large enough , @xmath342 holds as @xmath343 , with @xmath344 acting on @xmath345 . recall that @xmath346 is given by . also remark that [ eq : curv_bound ] combined with [ eq : conv_ass ] yields @xmath347 , i.e. , the second part of [ eq : v_bound ] . since we assume that the curvature @xmath26 is bounded , @xmath348 is essentially self - adjoint on @xmath349 and self - adjoint on @xmath350 see @xcite . the vector potential in the operator @xmath351 is @xmath352 , and consequently , @xmath353 using the partial fourier plancherel transform in the @xmath19 variable , we turn the operator @xmath351 into @xmath354 it is self - adjoint on its definition domain , @xmath355 and decomposes into a direct integral , @xmath356{\mathrm{d}}\xi,\ ] ] where the fiber is given by @xmath357=-\partial_{s}^{2}+(\xi+b_{0}x(s))^2-\frac{1}{4}\kappa^{2}(s)\ ] ] as an operator on @xmath358 . using ( * ? ? ? * thm 6.3 ) , we obtain further @xmath359 as @xmath343 , where @xmath360 is the ` leading term ' , @xmath361 and @xmath341 has to be , of course , chosen large enough . in view of the unitarity of the fourier plancherel transform , this implies @xmath362 where @xmath363 is defined similarly as @xmath360 but now with the help of @xmath364 . since this operator also decomposes into a direct integral , @xmath365\otimes i+i\otimes ( -a^{-2}\partial_{u}^{2})\right)\,{\mathrm{d}}\xi=:\int^\oplus_{{\mathord{\mathbb r}}}h_0[\xi]{\mathrm{d}}\xi,\ ] ] we obtain the corresponding limiting relation for the fibers , @xmath366-a^{-2}e_1 + k)^{-1}-(h_0[\xi]-a^{-2}e_1 + k)^{-1}\|= \mathcal{o}(a)\ ] ] as @xmath343 . this follows from the fact that @xmath367\|=\operatorname*{ess\,sup}_{m}\|a[\xi]\|$ ] , cf . * thm xiii.83 ) , which also implies , in particular , that the error term on the right - hand side of is uniform in @xmath64 . assume that the operator @xmath368 $ ] has compact resolvent and all its eigenvalues are simple and analytic in @xmath67 . this is fulfilled if , for instance , in addition to [ eq : curv_bound ] , which is sufficient for analyticity , we have @xmath369 , cf . here we employ for the sake of brevity the notation @xmath370 for a given @xmath371 . we denote the eigenvalues of @xmath368 $ ] , arranged in the ascending order , as @xmath372,\ , m\in{{\mathord{\mathbb n}}}$ ] . assume that they are non - constant as functions of @xmath67 . a sufficient condition for this reads @xcite @xmath373 another one is obtained from by changing the @xmath374 indices to @xmath375 everywhere . our aim here is to derive another sufficient condition for the non - constancy which will be presented at the end of the section . under the stated assumptions on @xmath368 $ ] , the spectrum of @xmath376-a^{-2}e_1 $ ] consists of isolated eigenvalues @xmath377=\nu_m[\xi]+a^{-2}(e_n - e_1).\ ] ] by the minimax principle , @xmath372\geq -\frac14\,\|\kappa^2\|_\infty$ ] . we fix an energy value @xmath15 which we will refer to for brevity as threshold , then there exists an @xmath378 such that @xmath379 holds for all @xmath380 . consequently , for these values of @xmath14 , the spectrum of @xmath381-a^{-2}e_1 $ ] strictly below @xmath174 consists of the simple eigenvalues @xmath382=\nu_m[\xi],\ , m=1,2,\ldots , n[\xi]$ ] , only . note that @xmath383= : n_e<+\infty$ ] . there exist at least one compact interval @xmath384 and a @xmath385 such that @xmath386<e-3\delta_e$ ] holds for all @xmath387 , because @xmath386 $ ] is by assumption non - constant and analytic . for a fixed @xmath388 we can then construct a tubular neighborhood @xmath389+t|\ : \xi\in i_m,\ , t\in ( -\delta,\delta)\}$ ] with @xmath390 , where @xmath391 , \sigma(h_0[\xi]-a^{-2}e_1)\right)\ ] ] is strictly positive . furthermore , one can find an @xmath392 such that for all @xmath393 there is exactly one eigenvalue branch of @xmath394 passing through each of the neighborhoods @xmath395 , @xmath388 . since @xmath372 $ ] are non - constant , these eigenvalue branches must be non - constant as well , if we choose @xmath396 and consequently also @xmath397 small enough . assume that there is a constant eigenvalue branch of @xmath394 below @xmath398 . then it must be , in particular , constant in the interval @xmath384 , and thus it could not intersect with any of @xmath395 , provided we chose @xmath396 and @xmath14 as above . moreover , by an easy perturbation theory consideration there are no eigenvalues of @xmath69-a^{-2}e_1 $ ] in the remaining gaps whenever @xmath14 is small enough . from this we can conclude that for any fixed threshold @xmath174 , all the eigenvalue branches of @xmath3 that lie ( at least partially ) below @xmath399 are non - constant provided the layer halfwidth @xmath14 is sufficiently small . now we are going to derive the indicated new sufficient condition for non - constancy of the energy bands of the effective hamiltonian . let us assume that , in addition to [ eq : curv_bound ] , @xmath400 it is convenient to write @xmath401 , where @xmath402 with @xmath147 being the heaviside step function . clearly , @xmath403 and it is not identically zero . note that this includes any perturbation of the planar layer that is compact in the @xmath17 direction , since without loss of generality we may always suppose that such a perturbation is supported to the right of the origin . we have @xmath404=-\partial_{s}^{2}+(\xi+b_0 s)^2 + 2b_0(\xi+b_0 s)r(s)+b_0 ^ 2r^2(s)-\frac{1}{4}\kappa^2(s)\\ = h_{\text{ho}}[\xi]+\theta(s)\left(b_0 r(s)\left(2\xi+b_0(s+x(s))\right)-\frac{1}{4}\kappa^2(s)\right).\end{gathered}\ ] ] for any @xmath405 , @xmath368 $ ] is thus a non - positive perturbation of the shifted harmonic oscillator hamiltonian @xmath406 $ ] . let us estimate the eigenvalue @xmath407 $ ] . let @xmath408 be the @xmath409th eigenfunction of the harmonic oscillator hamiltonian @xmath410 $ ] given by and @xmath411 ; note that every function in @xmath412 is real analytic on @xmath413 . now , by the minimax principle , @xmath414=\min_{\substack{s\subset { \mathrm{dom}(h_{\mathrm{eff}}[0])}\\\dim s = m}}\ \max_{\substack{\psi\in s \\ \|\psi\|=1}}\langle\psi , h[0]\psi\rangle\leq \max_{\substack{\psi\in s_m \\ \|\psi\|=1}}\langle\psi , h[0]\psi\rangle\\ \leq \max_{\substack{\psi\in s_m \\ \|\psi\|=1}}\langle\psi , h_{\text{ho}}[0]\psi\rangle+\max_{\substack{\psi\in s_m \\ \|\psi\|=1}}\langle\psi,\theta(s)(b_0 ^ 2 r(s)(s+x(s))-\frac14\kappa^2(s))\psi\rangle\\ = b_0(2m+1)+\max_{\substack{\psi\in s_m \\ \|\psi\|=1}}\int_{0}^{+\infty}\big(b_0 ^ 2 r(s)(s+x(s))-\frac14\kappa^2(s)\big)|\psi(s)|^2{\mathrm{d}}s.\end{gathered}\ ] ] the last term on the right - hand side is negative , because the sub - integral function is non - positive everywhere and strictly negative on some interval , and the maximum of the integral is attained for some @xmath415 . indeed , if the maximum was zero then @xmath416 would be zero on the mentioned interval , and therefore due to the analyticity it would vanish on @xmath413 , which is a contradiction . we conclude that the sharp inequality @xmath407<b_0(2m+1)$ ] holds for all @xmath306 . on the other hand , we have @xmath417 = b_0(2m+1)$ ] . to prove this claim we start with the unitary transform @xmath418 and introduce @xmath419=u_\xi h_{\mathrm{eff}}[\xi ] u_{\xi}^{-1}=\begin{cases } -\partial_{s}^{2}+b_{0}^{2}s^2 & s\leq \xi / b_0\\[.7em ] \begin{aligned } & -\partial_{s}^{2}+(\xi+b_0 x(s-\xi / b_0))^2 \\ & -\frac{1}{4}\kappa^2(s-\xi / b_0)\end{aligned } & s>\xi / b_0 \end{cases}\ ] ] and put @xmath420:=h_{\text{ho}}[0]$ ] . now we may apply the result of ( * ? ? ? * thm 2.3 ) , which we have reproduced here as a part of the proof of proposition [ prop : res_conv ] , to the family @xmath421|\ : \xi\in(-\infty,+\infty]\}$ ] . let us focus on the assumption ( ii ) of the theorem . for all @xmath11 sufficiently large we have @xmath422 , and consequently , there is a @xmath423 such that for all @xmath424 and @xmath425 , @xmath426 using this estimate on the interval @xmath427 , we obtain @xmath428-\frac{1}{4}\|\kappa\|_{\infty}^{2}\leq \hat{h}_{\mathrm{eff}}[\xi].\ ] ] the remaining assumptions are easy to verify . this makes it possible to infer from ( * ? ? ? * thm 2.3 ) that @xmath429^{-1}-h_{\mathrm{ho}}[0]^{-1}\|=0,\ ] ] which in turn implies that @xmath417 $ ] is just the @xmath136th eigenvalue of @xmath410 $ ] . our findings are summarized in the following claim . [ prop : thin ] let the assumptions [ eq : curv_bound ] , [ eq : diff_cond ] , and [ eq : conv_ass ] hold together with either or . then to any @xmath15 one can find an @xmath378 such that no eigenvalue branch of the total hamiltonian @xmath3 that lies at least partially below @xmath399 can be constant as a function of @xmath67 whenever @xmath430 . while our main interest concerns magnetic transport in the dirichlet layers , the considerations at the end of the previous section , in particular , the decomposition of the type can be in combination with the minimax principle applied also to the classic iwatsuka model . we start with the two - dimensional hamiltonian @xmath431 where @xmath432 fix a @xmath433 and assume that @xmath434 with as in the seminal paper of iwatsuka @xcite we start with a direct integral decomposition into fiber operators @xmath450=-\partial^2_x+(\xi+a_y(x))^2+w(x)\ ] ] on @xmath451 . in the same manner as in @xcite we show that @xmath452 $ ] has compact resolvent and all its eigenvalues numbered in the ascending order as @xmath305,\ , m\in{{\mathord{\mathbb n}}},$ ] are simple and analytic on @xmath413 as functions of @xmath67 . to prove the absolute continuity of @xmath447 it suffices to demonstrate that no @xmath453 $ ] is constant . we have @xmath454=h_{\text{ho}}[\xi]+\theta(x)\big(b_0 r(x)\big(2\xi+b_0(2x+r(x))\big)+w(x)\big),\ ] ] where @xmath406 $ ] was introduced in @xmath455 and @xmath456 note that @xmath457 is in view of ( absolutely ) continuous . using , we find @xmath458 for all @xmath439 , where @xmath459 } r(x)\leq 0 $ ] . hence , if we set @xmath460 then @xmath461 holds for all @xmath439 . taking the non - positivity of @xmath462 and @xmath446 into the account , we conclude that the second term in is also non - positive . moreover , the assumption @xmath463 implies that it is strictly negative on some interval . mimicking the estimates in with @xmath412 being the span of first @xmath136 eigenfunctions of @xmath464 $ ] , we arrive at @xmath465<b_0(2m+1)$ ] . due to , the second assumption of ( * ? ? ? * thm 2.3 ) is fulfilled which finally implies that @xmath466=b_0(2m+1)$ ] . let us recall that the family of magnetic fields considered above has a non - empty intersection with all the families studied earlier in the papers @xcite which , with the exception of the last one , treat the classic iwatsuka model , @xmath467 . hence we obtain a nontrivial extension of the known results , with notably weak regularity assumptions comparing to the other sources . note also the assumption crucial for the use of the minimax principle does not mean that the perturbation @xmath97 of the constant magnetic field must be everywhere negative ; it may be sign - changing and negative on a compact set only . m. abramowitz , i.a . stegun : _ handbook of mathematical functions with formulas , graphs , and mathematical tables _ , dover publications , new york 1965 . j. blank , p. exner , m. havlek : _ hilbert space operators in quantum physics _ , 2nd extended edition , springer , dordrecht 2008 . p. briet , g. raikov , e. soccorsi : spectral properties of a magnetic quantum hamiltonian on a strip , _ asymp . * 56 * ( 2008 ) , 127155 . cycon , r.g . froese , w. kirsch , b. simon : _ schrdinger operators , with applications to quantum mechanics and global geometry _ , springer , berlin and heidelberg 1987 . m. damak : on the spectral theory of tensor product hamiltonians , _ operator theory _ * 55 * ( 2006 ) , 253268 . p. dean : the constrained quantum mechanical harmonic oscillator , _ proc . * 62 * ( 1966 ) , 277286 . n. dombrowski , f. germinet , g. raikov : quantization of edge currents along magnetic barriers and magnetic guides , _ ann . h. poincar _ * 12 * ( 2011 ) , 11691197 . n. dombrowski , p.d . hislop , e. soccorsi : edge currents and eigenvalue estimates for magnetic barrier schrdinger operators , _ asymptot . * 89 * ( 2014 ) 331363 . p. duclos , p. exner , d. krejik : bound states in curved quantum layers , _ commun . math * 223 * ( 2001 ) , 1328 . p. exner , h. kovak : magnetic strip waveguides , _ j. phys . a : math . _ * 33 * ( 2000 ) , 32973311 . p. exner , h. kovak : _ quantum waveguides _ , springer , cham 2015 . n. filonov , a.v . sobolev : absence of the singular continuous component in spectra of analytic direct integrals , _ j. math . sci . _ * 136 * ( 2006 ) , 38263831 . v. geiler , m. senatorov : structure of the spectrum of the schrdinger operator with magnetic field in a strip and infinite - gap potentials , _ sb . _ * 188 * ( 1997 ) , 657669 . hislop , n. popoff , n. raymond , m.p . sundqvist : band functions in the presence of magnetic steps , _ math . models & methods in applied science _ * 26 * ( 2016 ) 161184 . p. hislop , e. soccorsi : edge currents for quantum hall systems , i. one - edge , unbounded geometries , _ rev . phys . _ * 20 * ( 2008 ) , 71115 . p. hislop , e. soccorsi : edge currents for quantum hall systems , ii . two - edge , bounded and unbounded geometries , _ ann . h. poincar _ * 9 * ( 2008 ) , 11411175 . hislop , e. soccorsi : edge states induced by iwatsuka hamiltonians with positive magnetic fields , _ j. math . * 422 * ( 2015 ) , 594624 . a. iwatsuka : examples of absolutely continuous schrdinger operators in magnetic fields , _ publ . rims , kyoto univ . _ * 21 * ( 1985 ) , 385401 . t. kato : _ perturbation theory for linear operators _ , 3rd ed . , springer , 1995 . d. krejik , n. raymond , m. tuek : the magnetic laplacian in shrinking tubular neighbourhoods of hypersurfaces , _ j. geom . * 25 * ( 2015 ) , 25432564 . h. leinfelder , c.g . simader : schrdinger operators with singular magnetic potentials , _ math . * 176 * ( 1981 ) , 119 . m. mntoiu , r. purice : some propagation properties of the iwatsuka model , _ commun . * 188 * ( 1997 ) , 691708 . m. reed , b. simon : _ methods of modern mathematical physics iii _ , academic press , new york 1979 . m. reed , b. simon : _ methods of modern mathematical physics iv _ , academic press , new york 1978 . m. tuek : on an extension of the iwatsuka model , _ j. phys . a : math . theor . _ * 49 * ( 2016 ) , 365205 . j. weidmann : _ linear operators in hilbert spaces _ , springer - verlag , new york 1980 .

we discuss the uniqueness of quantum states compatible with given measurement results for a set of observables . for a given pure state , we consider two different types of uniqueness : ( 1 ) no other pure state is compatible with the same measurement results and ( 2 ) no other state , pure or mixed , is compatible with the same measurement results . for case ( 1 ) , it was known that for a-dimensional hilbert space , there exists a set of observables that uniquely determines any pure state . we show that for case ( 2 ) , observables suffice to uniquely determine any pure state . thus there is a gap between the results for ( 1 ) and ( 2 ) , and we give some examples to illustrate this . unique determination of a pure state by its reduced density matrices ( rdms ) , a special case of determination by observables , is also discussed . we improve the best known bound on local dimensions in which almost all pure states are uniquely determined by their rdms for case ( 2 ) . we further discuss circumstances where ( 1 ) can imply ( 2 ) . we use convexity of the numerical range of operators to show that when only two observables are measured , ( 1 ) always implies ( 2 ) . more generally , if there is a compact group of symmetries of the state space which has the span of the observables measured as the set of fixed points , then ( 1 ) implies ( 2 ) . we analyze the possible dimensions for the span of such observables . our results extend naturally to the case of low rank quantum states .

1212.3503

in a @xmath0-dimensional hilbert space @xmath3 , the description of any quantum state @xmath4 generated by a source can be obtained by quantum tomography . for any density matrix @xmath4 , which is hermitian and has trace @xmath5 , @xmath6 independent measurements are sufficient and necessary to uniquely specify @xmath4 . when @xmath7 is a pure state , one may not need as many measurements to uniquely determine @xmath8 . as we will see later , however , exactly what is meant by `` uniquely '' in this context needs to be specified . consider a set of @xmath9 linearly independent observables @xmath10 where each @xmath11 is hermitian . measurements on state @xmath4 with respect to these observables give the following average values @xmath12 we denote the set of these @xmath13 for all states @xmath4 as @xmath14 for a pure state @xmath8 , these values are given by @xmath15 and we denote the set of these values for all pure states @xmath8 as the joint numerical range @xmath16 in this work we consider two different kinds of `` unique determinedness '' for @xmath8 : 1 . we say @xmath8 is _ uniquely determined among pure states ( udp ) by measuring @xmath17 _ if there does not exist any other pure state which has the same measurement results as those of @xmath8 when measuring @xmath17 . we say @xmath8 is _ uniquely determined among all states ( uda ) by measuring @xmath17 _ if there does not exist any other state , pure or mixed , which has the same measurement results as those of @xmath8 when measuring @xmath17 . it is known that there exists a family of @xmath1 observables such that any pure state is udp , in contrast to the @xmath6 observables in the general case of quantum tomography @xcite . the physical meaning for this case is clear : it is useful for the purpose of quantum tomography to have the prior knowledge that the state to be reconstructed is pure or nearly pure . many other techniques for pure state tomography have been developed , and experiments have been performed to demonstrate the reduction of the number of measurements needed @xcite . when the state is udp , to make the tomography meaningful , one needs to make sure that the state is indeed pure . this is not in general practical , but one can readily generalize the above mentioned udp results to low rank states , where the physical constraints ( e.g. , low temperature , locality of interaction ) may ensure that the actual physical state ( which ideally supposed to be pure ) is indeed low rank . if the state is uda , however , in terms of tomography one do not need to bother with these physical assumptions , because in the event there is only a unique state compatible with the measurement results , which turns out to be pure ( or low rank ) . there is also another clear physical meaning for the states that are uda by measuring @xmath17 . consider a hamiltonian of the form @xmath18 then any unique ground state @xmath8 of @xmath19 is uda by measuring @xmath17 . this is easy to verify : if there is any other state @xmath4 that gives the same measurement results , then @xmath4 has the same energy as that of @xmath8 , which is the ground state energy . therefore , any pure state in the range of @xmath4 must also be a ground state , which contradicts the fact that @xmath8 is the unique ground state . in other words , uda is a necessary condition for @xmath8 to be a unique ground state of @xmath19 . it is in general not sufficient , but the exceptions are likely rare @xcite . the uniqueness properties for pure states , for both udp and uda , have also been studied extensively in the case of multipartite quantum systems , where the observables correspond to reduced density matrices ( rdms ) . that is , the observables are chosen to act nontrivially on only some subsystems . for an @xmath20-particle system and a constant @xmath21 , there are a total of @xmath22 @xmath23-rdms , and the corresponding measurements @xmath17 are those @xmath24-body operators . for example , for a three - qubit system and @xmath25 , one can choose @xmath17 as all the one and two - particle pauli operators . of course , one can also choose to look at some of the @xmath26-rdms , rather than all of them . for instance , for a three - particle system , one can look at @xmath27-rdms of particle pairs @xmath28 and @xmath29 . it is known that almost all three - qubit pure states are uda by their @xmath27-rdms @xcite . these authors also show that udp implies uda for three - qubit pure states , for @xmath27-rdms . this result can be further improved to @xmath27-rdms of particle pairs @xmath28 and @xmath29 @xcite . more generally one can consider a three - particle system of particles @xmath30 with hilbert spaces whose dimensions are @xmath31 , respectively . if @xmath32 , then almost all pure states are uda by their @xmath27-rdms of particle pairs @xmath28 and @xmath29 . in contrast , if @xmath33 , then almost all pure states are udp by their @xmath27-rdms of particle pairs @xmath28 and @xmath29 , as shown by diosi @xcite . for @xmath20-particle quantum systems with equal dimensional subsystems , almost all pure states are uda by their @xmath23-rdms of just over half of the parties ( i.e. , @xmath34 ) . furthermore , @xmath35 properly chosen rdms among all the @xmath26 @xmath23-rdms suffice @xcite . w - type states are uda by their @xmath27-rdms , and @xmath36 of those @xmath27-rdms are enough @xcite . general symmetric dicke states are uda by their @xmath27-rdms @xcite . it has been shown that the only @xmath20-particle pure states which can not be udp by their @xmath37-rdms are those ghz - type states , and the result is further improved to the case of uda @xcite . their results also show that udp implies uda for @xmath20-qubit pure states , for @xmath37-rdms . despite these many results , there is no systematic study of these two different types of uniqueness for pure states . this will be the focus of this paper , where we are interested in knowing for given measurements @xmath17 , whether udp and uda are the same , or are different . we will give a general argument that there is a gap between the number of observables needed for the two different cases . however , in many interesting circumstances , they can coincide . our discussions extend naturally to the case of low rank quantum states instead of just pure states . here one can also look at two kinds of uniqueness when measuring given observables @xmath17 : one is uniqueness among all low rank states , the other is among all states of any rank . we organize the paper as follows . in sec . ii , we first show that there is a set of @xmath2 observables that insures every pure state is uda ; which should be compared to the udp result @xmath1 . thus in general there is a gap between the optimal results for the udp and uda cases , and we illustrate this with some examples . sec . iii discusses the case of observables corresponding to rdms of a multipartite quantum state , where for the three particle case , we show that if @xmath38 , then almost all pure states are uda by their @xmath27-rdms of particle pairs @xmath28 and @xmath29 , improving the bounds given in @xcite . however this still leaves a gap with the diosi result for the case of udp in @xcite . we further discuss circumstances where udp can imply uda for all pure states . in sec . iv , we show that when there are only two independent measurements performed , then udp always implies uda , by making use of convexity of the numerical range of operators . in a more general case , if there is a compact group of symmetries of the state space which has the span of the operators measured as its set of fixed points , then udp implies uda for all pure states . we analyze the possible dimensions for those fixed point sets . a summary and some discussions are included in sec . in this section , we discuss the minimum number of observables needed to have all pure states be uda . we start by choosing a hermitian basis @xmath39 for the operators on @xmath3 . without loss of generality we choose @xmath40 , the identity operator on @xmath3 , which has trace @xmath0 . we further require that the @xmath41 s are orthogonal , in the sense that for @xmath42 , @xmath43 the @xmath44 hermitian matrices form a real inner product space with inner product @xmath45 , so such a basis @xmath39 exists for any dimension @xmath0 . for instance , for the qubit case ( @xmath46 ) , we can choose the pauli basis @xmath47 for the qutrit case ( @xmath48 ) , one can choose @xmath49 for @xmath50 , where @xmath51s are the gell - mann matrices given by @xmath52 for general @xmath0 , one can choose @xmath53 for @xmath50 , where @xmath51s are the generalized gell - man matrices . we can now write any density operator @xmath4 as @xmath54 where @xmath55 , and where @xmath56 has real entries . we have @xmath57 , therefore @xmath58 , and the equality holds if @xmath4 is a pure state . however , not every state satisfying @xmath59 is a pure state . indeed , @xmath4 is a pure state if and only if @xmath60 , which gives equations that @xmath61 needs to satisfy . if one of the observables is a multiple of the identity , then we can drop it from the list of observables without affecting uda and udp . if two states agree on an observable @xmath11 , then they agree on @xmath62 for any real scalar @xmath63 , so we can adjust each of the observables @xmath64 to have trace zero without affecting uda or udp . hence hereafter we assume all @xmath11 are traceless . for any observable @xmath11 , we can expand in terms of @xmath65 as @xmath66 then the average value of @xmath11 is given by @xmath67 where @xmath68 . to discuss the problem for any pure state to be uda , the constant @xmath5 and constant factor @xmath69 can be ignored , as these are the same constants for all states . therefore we have @xmath70 where @xmath71 means that the average value of @xmath11 for the state @xmath4 is geometrically equivalent to the projection of @xmath61 onto @xmath72 . alternatively , define @xmath73 by @xmath74 . let @xmath75 be the linear subspace of @xmath76 spanned by @xmath77 , and let @xmath78 be the orthogonal projection from @xmath76 onto @xmath75 . then @xmath78 and @xmath79 have the same kernel , namely @xmath80 . thus for states @xmath81 , we have @xmath82 if and only if @xmath83 , so in considering uda and udp we can treat @xmath79 as being the orthogonal projection onto @xmath75 . if we subtract the density matrix @xmath84 from all states , then the translated set of states sits in the real @xmath6 dimensional subspace of trace zero hermitian matrices . in this sense , we are actually working with real geometry in @xmath76 . all quantum states then sit inside the @xmath6-dimensional unit ball , with pure states corresponding to unit vectors , but not every vector on the unit @xmath85-dimensional sphere is a pure state . the observables span an @xmath9-dimensional subspace that all the quantum states will be projected onto . we will simply say the subspace is spanned by @xmath17 when no confusion arises , and we will no longer distinguish an operator @xmath11 from the corresponding vector @xmath86 . indeed we only consider the real span of @xmath17 , and we denote it by @xmath87 . for each @xmath87 , there is an orthogonal subspace in @xmath76 of dimension @xmath88 , which we denote by @xmath89 . here we are taking the orthogonal complement in the space of traceless hermitian matrices , so that every @xmath90 is traceless . we now are ready to state our first theorem . [ th : uda ] for a @xmath0-dimensional system ( @xmath91 ) , there exists a set of @xmath2 observables for which every pure state is uda . to see why this is the case , note that in the above - mentioned geometrical picture , it is clear that a pure state @xmath92 is uda by measuring @xmath17 if there does not exist any operator @xmath90 , such that @xmath93 is positive . one sufficient condition will then be that any operator @xmath90 has at least two positive and two negative eigenvalues . we will use this sufficient condition to construct a desired @xmath89 . in order to construct @xmath89 , we provide a set of @xmath94 linearly independent hermitian matrices @xmath95 explicitly , such that the hermitian matrix @xmath96 has at least two positive eigenvalues for any nonzero real vector @xmath97 . our construction is motivated by and similar to the diagonal filling technique used in ref . @xcite , but along the other direction of the diagonals . this then means that measuring @xmath98 observables is enough for any pure state to be uda , which proves the theorem . there are indeed technical details to be clarified that we leave to appendix a. if we compare our results with those given in @xcite , which shows that measuring @xmath1 observables are enough for any pure state to be udp , there exists an obvious gap . we claim that this gap indeed can not be closed in general . to see this , let us look at the simplest case of @xmath48 , where the results just compared state that @xmath99 observables are enough for any pure state to be udp but @xmath100 observables are enough for any pure state to be uda . if one can measure a particular set @xmath17 with @xmath99 observables and have all pure states be uda , then also every state also must be udp for measuring @xmath17 . according to @xcite , this only happens if @xmath89 contains a single invertible traceless operator @xmath101 , meaning @xmath101 is rank @xmath102 . without loss of generality we can assume the largest eigenvalue @xmath101 to be positive with an eigenstate @xmath8 . then @xmath8 is not uda by measuring @xmath17 since as observed in @xcite there exists a mixed state which also has the same average values as those of @xmath8 . therefore , one can not only measure @xmath99 observables for all pure states to be uda . for general @xmath0 , our construction needs @xmath2 observables . we do not know whether this is the optimal construction , but it is very unlikely one can get this down to @xmath1 . in other words , in general uda and udp for pure states should be indeed two different concepts and there should always be gaps between the number of observables needed to be measured for each case to uniquely determine any pure quantum state this is one exception though , which is for the qubit case ( i.e. , @xmath46 ) where it is shown in @xcite that for all pure states to be udp , one needs to measure @xmath103 variables , which then uniquely determine any quantum state among all states . finally , we remark that our results in theorem 1 naturally extend to the case of low rank states . that is , for a rank @xmath104 quantum state @xmath4 , we can similarly consider two different cases : ( 1 ) @xmath4 is uniquely determined by measuring @xmath17 among all rank @xmath105 states ( which was considered in @xcite ) ( 2 ) @xmath4 is uniquely determined by measuring @xmath17 among all quantum states of any rank . [ th : udaq ] for a @xmath0-dimensional system ( @xmath91 ) measuring @xmath106 observables is enough for a rank @xmath105 state to be uniquely determined among all states . compared to the results in @xcite , where @xmath107 observables are needed to uniquely determine any rank @xmath105 states among all rank @xmath105 states , when @xmath0 is large the difference in the leading term has a @xmath0 gap . the proof idea is similar to that of theorem [ th : uda ] , so we leave the details to appendix a. in this section we discuss the case where the hilbert space @xmath3 is a multipartite quantum system , where the observables correspond to the reduced density matrices ( rdms ) . that is , the observables are chosen to be acting nontrivially only on some subsystems . for instance , for a three - qubit system , the observables corresponding to the @xmath27-rdms of particle pairs @xmath28 can be chosen as @xmath108 where @xmath109 are pauli @xmath110 operators acting on the @xmath111th qubit . for simplicity in this section we consider only @xmath102-particle systems , labeled by @xmath30 , and each with hilbert space dimension @xmath112 , respectively . that is , @xmath113 and @xmath114 . nevertheless , our method naturally extends to systems of more than @xmath102-particles . recall that for a three particle system , it is known that almost all three - qubit pure states are uda by their @xmath27-rdms @xcite . this result can be further improved to @xmath27-rdms of particle pairs @xmath28 and @xmath29 @xcite . more generally , if @xmath32 , then almost all pure states are uda by their @xmath27-rdms of particle pairs @xmath28 and @xmath29 @xcite . in contrast , if @xmath115 , then almost every pure state is udp by its @xmath27-rdms of particle pairs @xmath28 and @xmath29 @xcite . we notice that different from the discussion in sec . ii , one no longer considers uniqueness for all pure states , but ` almost all ' of them . this means there exists a measure zero set of pure states which are not uniquely determined . for instance , for the three qubit case , any state which is local unitarily equivalent to the ghz type state @xmath116 can not be udp , as any state of the form @xmath117 has the same @xmath27-rdms as those of @xmath118 . this means that , for a three qubit pure state @xmath8 , it is either uda , or not udp . in other words , if any three qubit pure state @xmath8 is udp , then it is uda by its @xmath27-rdms of particle pairs @xmath28 and @xmath29 . in this sense , we say in this case udp implies uda for all pure states . however , for the general case of a three particle system , there is a gap between known results of uda and udp . our following result improves the bound for the uda case . [ th : rdm1 ] if @xmath119 , then almost every tripartite quantum state @xmath120 is uda by its @xmath27-rdms of particle pairs @xmath28 and @xmath29 , to see why this is the case , an arbitrary pure state @xmath121 of this system can be written as @xmath122 if there is another state @xmath4 which agrees with @xmath121 in its subsystems @xmath28 and @xmath29 , then we can find a pure state @xmath123 which agrees with @xmath4 on the subsystem @xmath124 and also agrees with @xmath125 in subsystems @xmath28 and @xmath29 . since the rank of the @xmath27-rdm of the subsystem @xmath28 is at most @xmath126 , the pure state @xmath127 can be written as a superposition of @xmath128 as follows . @xmath129 where @xmath130 for any @xmath131 . here @xmath132 will be vectors ( perhaps unnormalized ) in @xmath133 . the states @xmath132 can be chosen to be orthonormal vectors in the subsystem @xmath133 , and then for almost all states @xmath121 , the set of @xmath134 will be linearly independent . let us write @xmath135 . for any @xmath131 , we will have @xmath136 now let s consider the subsystem @xmath29 . since @xmath125 and @xmath127 have the same rdms for particles @xmath29 , this gives @xmath137 substituting eqs . and into eq . , and comparing each matrix element , results in the following equalities ( for all @xmath138 ) : @xmath139 now let us define @xmath140 . then eq . is a linear equation system with variables @xmath141 . it is not hard to verify that @xmath142 is a solution to the equation system , which corresponds to the state @xmath125 . now we need to show that when @xmath143 , eq . has only one solution which is given by eq . . it turns out that this is indeed the case which then proves theorem [ th : rdm1 ] . in fact , the linear equations above are generically linearly independent . to see this , let s fix @xmath144 and @xmath145 , the right - hand side of eq . is @xmath146 where @xmath147 . then the coefficient matrix can be written as the following : @xmath148 the @xmath149 entry in the above matrix is @xmath150 . if there are more than @xmath5 solutions , then the determinant of the above matrix should be zero . note that the determinant can be written as a polynomial of @xmath151 s and @xmath152 s . since @xmath153 appears only once in the polynomial , the determinant of the top @xmath154 by @xmath154 submatrix must be non - zero generically . therefore , @xmath155 linear equations are sufficient to determine @xmath156 variables . however , we do not know whether the sufficient condition given by theorem [ th : rdm1 ] for almost all three - particle pure state to be uda by its @xmath27-rdms of particle pairs @xmath28 and @xmath29 is also necessary . this still leaves a gap between the result of theorem [ th : rdm1 ] for uda , and the result for udp in @xcite . they both only coincide when @xmath157 , i.e. , the three qubit case . it remains open for other cases , whether udp can imply uda . following a similar discussion as in sec . ii , our result in this section also extends to uniqueness of low rank quantum states . in particular , we have the following theorem . [ th : rdm2 ] almost every tripartite density operator @xmath4 acting on the hilbert space @xmath158 with rank no more than @xmath159 can be uniquely determined among all states by its @xmath27-rdms of particle pairs @xmath28 and @xmath29 . this result is to our knowledge , the first one for uniqueness of mixed states with respect to rdms . the proof is a direct extension of that for theorem [ th : rdm1 ] , but with more lengthy details that we will include in appendix b. let us look at some consequences of theorem [ th : rdm2 ] . consider a four qubit system with qubits @xmath160 , and look at the qubits @xmath161 as a single systems @xmath162 . then theorem [ th : rdm2 ] says also that almost all four qubit states of rank @xmath27 are uda by their rdms of particles @xmath28 and @xmath163 , or one can say that almost all four qubit states of rank @xmath27 are uda by their @xmath102-rdms . this is indeed consistent with the multipartite result in @xcite which states that almost all four - qubit pure states are uda by their @xmath102-rdms , and our result is indeed stronger . this demonstrates that our analysis naturally extends to systems of more than @xmath102-particles . we also remark that the rank of a state @xmath4 which could be uda by its @xmath23-rdms needs to be relatively low , otherwise one can always find another state @xmath164 with lower rank which has the same @xmath23-rdms as those of @xmath4 @xcite . in sec . ii and sec . iii , we discussed the difference and coincidence between the two kinds of uniqueness for pure states , uda and udp , which in general are not the same thing . however , in certain interesting circumstances such as the three qubit case with respect to @xmath27-rdms , and in general the @xmath20-qubit case respect to @xmath37-rdms , they do coincide . starting from this section we would like to build some general understanding of the circumstances when udp implies uda for all pure states . we start from the simplest case of @xmath165 , where only two observables are measured , i.e. , @xmath166 . intuitively , in this extreme case almost no pure state can be uniquely determined , either uda or even udp . however there are also exceptions . for instance , if one of the observables , say @xmath167 , has a nondegenerate ground state @xmath8 , then @xmath8 is uda ( hence , of course , udp ) even by measuring @xmath167 only . one would hope this is the only exception , that is , for a pure state @xmath8 , either it is uda , or it is not udp , when only two observables are measured . we make this intuition rigorous by the following theorem . [ th : num ] when only two observables are measured , i.e. , @xmath166 , udp implies uda for any pure state @xmath8 , regardless of the dimension @xmath0 . to prove this theorem , recall that measuring @xmath17 ( i.e. , measuring every observable in @xmath17 ) for all quantum states @xmath4 returns the set @xmath168 given by eq we know that @xmath168 is a convex set , meaning for any @xmath169 , we have @xmath170 for any @xmath171 . for pure states , the corresponding set of average values is given by @xmath172 as defined in eq . . unlike @xmath168 , @xmath172 in general is not convex . nevertheless , it is easy to see that @xmath173 when @xmath172 is convex . for @xmath165 , the hausdorff toeplitz theorem @xcite gives convexity of the numerical range of any operator , which in turn shows that @xmath174 is convex . we explain it briefly here . for any operator @xmath175 acting on a hilbert space @xmath3 , the numerical range of @xmath175 is the set of all complex numbers @xmath176 , where @xmath8 ranges over all pure states in @xmath3 . note that one can always write @xmath175 as @xmath177\nonumber\\ & = & \frac{1}{2}[(b+b^{\dag})+i(-ib+ib^{\dag})].\end{aligned}\ ] ] if we define @xmath178 and @xmath179 then clearly both @xmath167 and @xmath180 are hermitian . then @xmath174 is nothing but the numerical range of @xmath181 and hence is convex . furthermore , by studying the properties of the numerical range , it was shown in @xcite ( using different terminology ) that if a pure state @xmath8 is udp , the point @xmath182 must be an extreme point of @xmath174 . here @xmath183 is an extreme point of the convex set @xmath174 if there do not exist @xmath184 , such that @xmath185 for some @xmath171 . because @xmath186 , @xmath183 is also an extreme point of @xmath187 . one can further show that for any extreme point @xmath183 of @xmath187 , and any quantum state @xmath4 with @xmath188 , any pure quantum state @xmath121 in the range of @xmath4 will also have @xmath189 . this then implies that if a pure state @xmath8 is udp by measuring @xmath17 , it must also be uda , which proves the theorem . again , all the technical details of the proof will be presented in appendix c. in an attempt to extend theorem [ th : num ] to the @xmath190 case , a natural question that one could ask is whether or not udp implies uda whenever @xmath172 is convex . unfortunately this is not the case , as demonstrated by the following example . for the qutrit case ( @xmath48 ) , consider the observables @xmath191 , where the @xmath51s are the gell - mann matrices given in eq . . these are the pauli operators embedded in the qutrit space . it is easily verified that in this case , @xmath172 is the bloch sphere together with its interior and is thus convex . nonetheless , the unique pure state compatible with measurement result @xmath192 is the state @xmath193 , even though there are many mixed states sharing this measurement result , such as @xmath194 . therefore , although the hausdorff toeplitz theorem @xcite is famous for showing the convexity of numerical range of any operator , there is indeed a deeper reason than just the convexity of the numerical range which governs the validity of theorem [ th : num ] . we leave the more detailed discussion to appendix c. in this section , we discuss some circumstances where udp implies uda in a more general context where more than two observables are measured , i.e. , @xmath195 . our focus is on the symmetry of the set of all quantum states . for a @xmath0-dimensional hilbert space @xmath3 we denote this set of states by @xmath196 , that is @xmath197 note that @xmath196 is convex , as we know that for any @xmath198 , @xmath199 for all @xmath171 . furthermore , the extreme points of @xmath196 are all the pure states . @xmath196 is also called the state space for all the operators acting on @xmath3 . we now explain the intuition . if @xmath196 has a certain symmetry , then two pure states @xmath200 and @xmath201 that are ` connected ' by the symmetry will give the same measurement results , and states @xmath8 fixed by the symmetry will also be fixed by the projection onto the space of observables . in this situation , udp implies uda for all pure states . to make this intuition concrete , let us first consider an example for @xmath46 , i.e. , the qubit case . we know that @xmath196 can be parameterized as in eq . , where for @xmath202 , @xmath203 are chosen as pauli matrices given in eq . . here @xmath196 is the bloch ball as shown in fig . [ fig : example ] . the bloch ball is clearly a convex set and the extreme points are those pure states on the boundary , which give the bloch sphere . we know that geometrically , measuring the observables in @xmath17 corresponds to the projection onto the plane spanned by @xmath17 . for example , if we measure the pauli @xmath204 and @xmath205 operators , then geometrically this corresponds to the projection of the bloch ball onto the @xmath206 plane . since the bloch ball has reflection symmetry with respect to the @xmath206 plane , two pure states ( e.g. points @xmath175 and @xmath207 ) connected by that symmetry will project onto the same measurement result @xmath208 , as will all mixtures of @xmath175 and @xmath207 . hence neither udp nor uda hold for such pure states for measuring @xmath204 and @xmath205 . on the other hand , pure states fixed by the reflection symmetry are also fixed by the projection onto the @xmath206 plane . these are precisely the points on the bloch sphere that are in the @xmath206 plane ( e.g. the points @xmath209 and @xmath210 in fig . [ fig : example ] ) , and for such pure states both udp and uda hold . therefore , for the observables @xmath211 we conclude that udp = uda . now let us look at another case where we only measure the pauli @xmath204 operator . consider the group of symmetries of the bloch ball consisting of rotation around the @xmath212 axis . ( rotation by angle @xmath213 , is shown in fig . [ fig : example ] . in that figure , point @xmath175 will become point @xmath214 after this particular rotation , and indeed both points @xmath175 and @xmath214 yield the same measurement result , which is represented by point @xmath208 on the @xmath212 axis . ) note that two points on the bloch sphere will project to the same measurement result on the @xmath212 axis if and only if they are in the same orbit under the rotation group . thus a measurement result will come from a single pure state exactly when that pure state is a fixed point , and hence either both or neither of udp and uda hold for each pure state . for example , the point @xmath209 is fixed by the rotation , and is uniquely determined by the measurement of @xmath204 among all states . @xmath209 corresponds to the @xmath215 eigenstate of the pauli @xmath204 operator . therefore , the rotational symmetry of the bloch ball along the @xmath212 axis gives udp = uda for any pure state when measuring the pauli @xmath204 operator , which corresponds to the @xmath212 axis . mathematically , a symmetry of @xmath196 is an affine automorphism of @xmath196 . if @xmath216 is unitary , the map taking @xmath4 to @xmath217 is such an affine automorphism ( which for @xmath218 will just be rotation around some axis of the bloch ball ) . for instance , the rotation symmetry along the @xmath212 axis by an angle @xmath213 is given by conjugation by the unitary operator @xmath219 . if @xmath101 is the conjugate linear map given by complex conjugation in the computational basis ( @xmath220 ) , then the map taking @xmath4 to @xmath221 is the transpose map . for @xmath218 , this map is reflection of the bloch ball in the xy - plane . recall that for a set of observables @xmath17 , we denote the real linear span by @xmath87 . when discussing the uniqueness problems , it makes no difference if we append the identity operator to @xmath17 . let us then assume @xmath222 . we are now ready to put our intuition into a theorem . [ th : symmetry ] assume there exists a compact group @xmath210 of affine automorphisms of @xmath196 whose fixed point set is @xmath223 . then each pure state acting on @xmath3 which is udp for measuring @xmath17 is also uda . in the first example above , the group for the reflection consists of the two element group generated by the reflection . in the rotation example , we can take the group to consist of all rotations around the given axis . we will leave the detailed mathematical proof of theorem [ th : symmetry ] to appendix d , where operator algebras are one ingredient of the proof . to motivate some further consequences of theorem [ th : symmetry ] , consider a simple example . if @xmath17 consists of a basis of diagonal matrices ( i.e. , a set of mutually commuting observables ) , then for any pure state , udp implies uda by theorem [ th : symmetry ] . here the group of symmetries can be taken to be conjugation by all diagonal unitaries . this group has fixed point set @xmath224 . in a more general case , if the complex span of @xmath225 is a * -subalgebra of the operators acting on @xmath3 , then udp = uda for all pure states for measuring @xmath17 . this is a natural corollary of theorem [ th : symmetry ] that we will also discuss in detail in appendix d. in this work , we have discussed the uniqueness of quantum states compatible with given results for measuring a set of observables . for a given pure state , we consider two different types of uniqueness , udp and uda . we have taken the first step to study their relationship systematically . in doing so we have established a number of results , but also leave with many open questions . first of all , although in general udp and uda are evidently different concepts , their difference is surprisingly ` not that large ' . specifically in the sense of general counting of the number of variables one needs to measure to uniquely determine all pure states in a @xmath0 dimensional hilbert space . compared to full quantum tomography which requires @xmath6 variables measured to uniquely determine any quantum state , the @xmath2 observables we have constructed to uniquely determine any pure state among all states is a significant improvement . it is indeed larger than the @xmath1 observables given in @xcite to uniquely determine any pure state among all pure states , but the difference is only linear in @xmath0 . we do not know whether there could be another construction for which we could further close the linear difference between uda and udp , to leave only a constant gap for large @xmath0 . when the hilbert space is a multipartite quantum system , and the observables correspond to the rdms , we focused on the situation when ` almost all pure states ' are uniquely determined . we considered a @xmath102-particle system with hilbert space @xmath226 , and showed that if @xmath38 , then almost all pure states are uda by their @xmath27-rdms of particle pairs @xmath227 and @xmath228 . this improves the results of @xcite , where @xmath32 is required ; however it still leaves a gap compared to the diosi udp result which states that for @xmath33 , almost all pure states are udp by their @xmath27-rdms of particle pairs @xmath227 and @xmath228 . because our proof only gives a sufficient condition for uda , we do not know whether it can be further improved . we also do not have an example showing there is indeed gap between uda and udp for almost all three - particle pure states to be uniquely determined by @xmath27-rdms of particle pairs @xmath227 and @xmath228 . finally , we considered situations for which we can show that udp implies uda . these include : ( i ) the general @xmath27-qubit system ; ( ii ) the @xmath102-qubit system when we consider uniqueness for almost all pure states and the measurements corresponds to @xmath27-rdms ; ( iii ) when only two observables are measured ; and ( iv ) the observables measured correspond to some symmetry of the state space . however we do not know how far we are from enumerating all the possible situations that udp implies uda , when considering uniqueness for all pure states or almost all pure states . in principle one can even consider the relationship between udp and uda for special subsets of pure states . we believe our systematic study of the uniqueness of quantum states compatible with given measurement results shed light on several aspects of quantum information theory and its connection to different topics in mathematics . these include quantum tomography and the space of hermitian operators , unique ground states of local hamiltonians and general solutions to certain linear equations , measurements and numerical ranges of operators , and the geometric meaning of measurements and the symmetry of state space . we thus conclude with several open questions that we believe warrant further investigation . theorem [ th : uda ] can be implied by the following lemma . [ lemma:2 + - ] there exists a set of @xmath94 linearly independent hermitian matrices @xmath229 , such that the hermitian matrix @xmath96 has at least two positive eigenvalues for any nonzero real vector @xmath97 . we prove the statement by giving an explicit construction . our proof is motivated by and similar to the diagonal filling technique used in ref . @xcite , but along the other direction of the diagonals . we will need the lemma 9 from ref . @xcite about totally non - singular matrix , which we restate as lemma [ lem : cmw08 ] in the following . for simplicity , we also assume that the totally non - singular matrix is real . therefore , for any length @xmath230 and @xmath231 , there is @xmath232 linearly independent real vectors such that every nonzero linear combination of them has at least @xmath27 nonzero entries . let @xmath233 be a matrix . we will always fix the diagonal to be zero , namely @xmath234 for @xmath235 . in the upper triangular part of the matrix not including the diagonal , there are @xmath236 lines of entries parallel to the antidiagonal . that is , each line contains entries @xmath237 with @xmath238 and @xmath239 where @xmath23 goes from @xmath5 to @xmath236 . we will call it the @xmath23-th line of the matrix in the following . we also call the set of entries @xmath237 with @xmath239 the @xmath23-th antidiagonal . it is easy to see that the length @xmath240 of the @xmath23-th line is @xmath241 & \text{for } k\le d-1,\\[.5em ] \displaystyle \bigl [ \frac{2d-1-k}{2 } \bigr ] & \text{otherwise . } \end{cases}\ ] ] so the length @xmath242 for @xmath243 , and we can find @xmath244 real vectors for which every nonzero linear combination has at least @xmath27 nonzero entries . for each of the @xmath244 vectors , we can form two hermitian matrices . one of them is the symmetric one whose @xmath23-th line is filled with the vector , and the lower triangular part determined by the hermitian condition . such a matrix is a real symmetric matrix having nonzero entries only on the @xmath23-th antidiagonal . we will call it a real @xmath23-th line matrix . the other is the one with @xmath23-th line filled with the vector multiplied by @xmath245 , and lower part is determined by the hermitian condition . this is a matrix consisting of purely imaginary entries on the @xmath23-th antidiagonal and we call it an imaginary @xmath23-th line matrix . now we prove that the constructed matrices satisfy our requirement . first we prove that the matrices are linearly independent . it suffices to show that the matrices of nonzero @xmath23-th line is linearly independent . let @xmath246 be the set of linearly independent real vectors chosen for the @xmath23-th line . we need to show that @xmath247 is linearly independent over @xmath248 . if the contrary is true , that is , there exists complex numbers @xmath249 not all zero such that @xmath250 this is equivalent to @xmath251 from the above two equations , we get @xmath252 and @xmath253 which is a contradiction . next , we prove that for any nonzero real coefficient @xmath254 , the matrix @xmath255 has at least two positive eigenvalues . let @xmath256 be the largest @xmath23 such that there is a @xmath23-th line matrix @xmath257 whose coefficient @xmath258 is nonzero . then , either the real @xmath256-th line matrices or the imaginary ones have nonzero coefficients . by the construction , this implies that there is at least two nonzero entries on the @xmath256th line of the matrix @xmath259 . let the nonzero entries be @xmath260 . we then have a principle submatrix of @xmath259 that has the form @xmath261 where @xmath262 are two unknown number and @xmath263 represents the complex conjugate of @xmath264 . this matrix has trace @xmath265 and determinant @xmath266 . therefore , it has exactly two positive eigenvalues . as it is a principle submatrix of matrix @xmath259 , follows from theorem [ thm : horn90 ] , @xmath259 has at least two positive eigenvalues . the number of matrices thus constructed is the summation @xmath267 which can be computed to be @xmath268 discussion : we note that our construction will also imply that the matrix has at least two negative eigenvalues , thus at least rank @xmath269 . but our bound is even better than the @xmath270 bound on the dimension of subspaces in which every matrix has rank @xmath271 . this is not a contradiction as we are considering all real combinations . for example , the case of @xmath272 has two matrices for our purpose , namely @xmath273 these two matrices do satisfy our requirements , but their span over @xmath248 contains a rank @xmath27 matrix @xmath274 . generalization : similarly , length @xmath275 for @xmath276 , and we can find @xmath277 real vectors for which every nonzero linear combination has at least @xmath278 nonzero entries . for each of the @xmath277 vectors , we can also form two hermitian matrices . such constructed matrices are linearly independent and any real linear combination has at least @xmath278 positive eigenvalues . we restate our result as lemma [ lemma : rank - q ] which will lead to theorem [ th : uda ] . [ lemma : rank - q ] there exists a set of @xmath279 linearly independent hermitian matrices @xmath229 , such that the hermitian matrix @xmath96 has at least @xmath278 positive eigenvalues for any nonzero real vector @xmath97 . we just follow the lines of the proof of lemma [ lemma:2 + - ] . to complete our argument , we need to show that any @xmath280 by @xmath280 invertible traceless , hermitian , upper left triangler matrix has exactly @xmath278 positive eigenvalues . let s prove this claim by induction . when @xmath281 , it is already known . let s assume this claim holds true for any @xmath282 . then for @xmath283 , we can write such matrix @xmath284 in the following form @xmath285 one may observe that , by deleting the first and the last rows / columns , we will have a @xmath286 by @xmath286 invertible , traceless , hermitian , upper left triangler submatrix . from our assumption , this submatrix has exactly @xmath287 positive eigenvalues which means @xmath284 has at least @xmath287 positive eigenvalues and at least @xmath287 negative eigenvalues . note that its determinant equals to @xmath288 . this follows that @xmath284 has exactly @xmath289 positive eigenvalues which completes our argument . the number of matrices thus constructed is the summation @xmath290-q)+2\sum\limits_{k = d}^{2d-2q-3}(\bigl [ \frac{2d-1-k}{2 } \bigr]-q)\\ & = & d^2-(4q+1)d+(4q^2 + 2q).\end{aligned}\ ] ] [ lem : cmw08 ] let @xmath291 be a @xmath0 by @xmath0 totally non - singular matrix , with @xmath292 . let @xmath293 be any linear combination of @xmath20 of the columns of @xmath291 . then @xmath293 contains at most @xmath36 zero elements . [ thm : horn90 ] let @xmath284 be a @xmath20 by @xmath20 hermitian matrix , let @xmath287 be an integer with @xmath294 , and let @xmath295 denote any @xmath287 by @xmath287 principle submatrix of @xmath284 ( obtained by deleting @xmath296 rows and the corresponding columns from @xmath284 ) . for each integer @xmath23 such that @xmath297 we have @xmath298 almost every tripartite density operator @xmath299 with rank no more than @xmath300 is uda by its 2-rdms of particle pairs @xmath28 and @xmath29 . for any @xmath301 , we can choose @xmath302 to be the pure state whose 3-rdm of particles @xmath124 is exactly @xmath303 . we can further assume @xmath304 . without loss of generality , we can assume @xmath305 if there is another @xmath306 agrees with @xmath303 in subsystems @xmath28 and @xmath29 . then we can find some pure state @xmath307 whose 3-rdm of particle set @xmath124 is @xmath306 . in general , @xmath308 are linearly independent and they will span the support of @xmath309 . hence , any pure state @xmath307 which agrees with @xmath302 in subsystem @xmath28 can be expanded as the following . @xmath310 recall that @xmath307 and @xmath302 agree in subsystem @xmath29 , we will have @xmath311 firstly , let s look into the left hand side . @xmath312 then , for the right hand side , @xmath313 combining the above two equations , we have @xmath314 for any @xmath315 . similarly , follows from the fact that @xmath307 and @xmath302 agree in subsystem @xmath28 , we have the following equation . @xmath316 for any @xmath317 . let s denote @xmath318 for any @xmath319 . if there is only one solution @xmath320 satisfies the above two linear systems , then @xmath321 is completely determined . there are generically @xmath322 linearly independent linear equations and @xmath323 variables . @xmath324 implies that there is at most one solution . thus a generic low - rank density operator @xmath303 is uda by its 2-rdms of particle pairs @xmath28 and @xmath29 . we begin by presenting without proof a result that was proved as theorem 1(i ) in @xcite . before we state the result , recall from sec . iv that @xmath325 is the numerical range of @xmath326 and is thus convex by the hausdorff toeplitz theorem @xcite . it therefore makes sense to talk about extreme points of @xmath325 in this case . [ prop : embry ] let @xmath327 . then @xmath183 is an extreme point of @xmath325 if and only if @xmath328 is a linear subspace of @xmath3 . we are now in a position to prove theorem [ th : num ] . suppose that @xmath8 is udp and define @xmath329 . then @xmath330 , which is linear , so @xmath183 is an extreme point of @xmath325 by proposition [ prop : embry ] . because @xmath331 in this case by convexity , @xmath183 is also an extreme point of @xmath332 . suppose now that there exists a mixed state @xmath333 with @xmath334 . then @xmath335 . since @xmath183 is extreme in @xmath332 , it follows that @xmath336 for all @xmath111 , which contradicts the fact that @xmath8 is udp unless each @xmath337 is the same up to global phase ( i.e. , @xmath338 ) . based on the proof of theorem [ th : num ] , we might expect that udp implies uda for all pure states whenever @xmath339 is convex . however , the example provided in sec . iv showed this not to be the case . we now expand upon the reason for this apparent discrepancy , which lies buried in the proof of proposition [ prop : embry ] . in the case when @xmath339 is convex , the `` only if '' implication of proposition [ prop : embry ] still holds for arbitrary @xmath9 . however , the proof of the `` if '' implication relies on the fact that if @xmath340 and @xmath341 , then for any @xmath342 we can find @xmath343 such that @xmath344 . in other words , the proof of the proposition uses the fact that @xmath339 is not only convex , but that convex combinations are in a sense well - behaved between the input and output of @xmath345 . for convenience , we refer to this property as _ strong convexity _ for the remainder of this section . the standard proofs of the hausdorff toeplitz theorem show that strong convexity , not just convexity itself , always holds when @xmath346 . to see how strong convexity can fail when @xmath347 even when convexity holds , we again return to the example of sec . iv . in this case , we have @xmath348 and @xmath349 . however , even though @xmath350 is convex and thus there exists a pure state @xmath8 with @xmath351 , the only such pure state is @xmath352 , which is not contained in the span of @xmath353 and @xmath354 . we might hope that strong convexity , rather than convexity itself , provides the natural generalization of theorem [ th : num ] . that is , we might hope that if @xmath339 is strongly convex , then udp implies uda for all pure states . it turns out that this is a true but vacuous statement if @xmath339 is strongly convex then it must be the case that @xmath355 , so theorem [ th : num ] itself applies directly . this fact seems to be implicit in many papers on the joint numerical range , but we prove it here for completeness . before stating the result , we briefly note that we can assume without loss of generality that @xmath356 contains @xmath357 and is linearly independent , as adding the identity to @xmath356 has no effect on convexity , uda , or udp , and furthermore these properties only depend on the span of the observables in @xmath356 . [ prop : strong_convex ] let @xmath358 be a linearly independent set . then @xmath359 is strongly convex if and only if @xmath355 . the `` if '' direction , as already mentioned , follows from any of the usual proofs of the hausdorff toeplitz theorem . for the `` only if '' direction , suppose that that @xmath360 is strongly convex and assume ( in order to get a contradiction ) that @xmath190 . by ( * ? ? ? * theorem 4.1 ) , there exists @xmath361 with @xmath362 such that @xmath363 spans all of @xmath364 . by letting @xmath121 and @xmath8 be the column vectors of @xmath204 , we see that strong convexity of @xmath359 immediately implies convexity of @xmath365 . since convexity of @xmath365 depends only on the span of @xmath366 , it follows that @xmath367 is also convex , where @xmath368 is the pauli basis given by equation . however , it is easily verified that @xmath367 is the bloch sphere embedded in @xmath269-dimensional space and hence is not convex , which gives the desired contradiction . it thus seems that numerical range and convexity arguments are not able to tell us anything non - trivial about the udp / uda problem beyond the @xmath369 case of theorem [ th : num ] . we will see that if there is a compact group of symmetries whose fixed point set is the linear span of observables @xmath370 , then udp for these observables implies uda for any pure state . we start with the following result about fixed points of compact groups . [ haar]let @xmath210 be a compact group of unitaries on a real or complex finite dimensional hilbert space @xmath259 , and let @xmath75 be the set of fixed points of @xmath210 . let @xmath371 be haar measure on @xmath210 , and define @xmath372 to be the linear map satisfying @xmath373 for @xmath374 . then @xmath208 is the orthogonal projection onto @xmath75 , @xmath375 for all @xmath376 , and @xmath208 is in the convex hull of @xmath210 . left and right invariance of haar measure imply that @xmath377 for all @xmath376 . the definition of @xmath208 implies that @xmath378 . now @xmath379 for all @xmath376 implies @xmath380 , and hence @xmath381 . next , @xmath381 and the definition of @xmath208 give @xmath382 . to show that @xmath383 we use the fact that the integrals of @xmath384 and @xmath385 are the same for haar measure , together with the assumption that @xmath210 is a group of unitaries : @xmath386 finally , by the alaoglu birkhoff mean ergodic theorem ( * ? ? ? 4.3.4 ) @xmath208 is in the strong closure of the convex hull of @xmath210 . since @xmath259 is finite dimensional , then the space of linear operators on @xmath259 is also finite dimensional , so the convex hull of the compact set @xmath210 is compact and hence closed . symmetries of @xmath196 are given by conjugation by unitaries or by the transpose map or by composition of these two types of symmetries . ( an affine automorphism of @xmath196 preserves transition probabilities , cf . @xcite , so this is a consequence of wigner s theorem @xcite . ) if we view the space of observables in @xmath387 as a real hilbert space ( with the usual inner product @xmath388 ) , then conjugation by unitaries and the transpose map both preserve this inner product , so are given by unitaries on this hilbert space . if @xmath75 is a ( real ) linear subspace of observables containing the identity , then @xmath75 will be the real linear span of @xmath389 . thus any symmetry of @xmath196 will fix @xmath389 if and only if that symmetry when extended to a map on @xmath387 fixes @xmath75 . if @xmath210 is a compact group of symmetries whose fixed point set is @xmath389 , then the corresponding maps on @xmath387 will have fixed point set @xmath75 . [ symmetry ] let @xmath284 be a finite set of observables on @xmath390 with real linear span @xmath75 . assume there exists a compact group @xmath210 of affine automorphisms of @xmath196 whose fixed point set is @xmath389 . then each pure state which is udp for measuring @xmath284 is also uda . as discussed above , we may view @xmath210 as a compact group of unitaries with fixed point set @xmath75 . define @xmath208 as in theorem [ haar ] . fix a pure state @xmath4 . suppose first that @xmath391 . then there is some @xmath392 such that @xmath393 . since @xmath394 , both @xmath395 and @xmath4 are pure states with the same image in @xmath396 under the map @xmath208 . thus udp fails for @xmath4 ( and hence trivially uda fails ) . now suppose @xmath397 . let @xmath398 be a pre image of @xmath4 under @xmath208 . then @xmath399 and equality can hold only if @xmath400 for all @xmath401 , i.e. , if and only if @xmath402 . then @xmath403 , so for such @xmath4 both udp and uda hold . [ cor:2dim ] for @xmath46 , for all pure states and all sets @xmath404 of observables , udp implies uda . let @xmath405 be observables in @xmath364 and let @xmath406 be their real linear span . we will show that there is a finite group of affine automorphisms @xmath210 of the state space @xmath196 of @xmath364 with fixed point set @xmath407 . there are three cases , depending on the dimension of the fixed point set . the fixed point set in the bloch sphere will be the central point , a diameter of the bloch sphere , or the intersection of a plane ( through the center ) with the bloch sphere . in each case reflection of the bloch sphere in the fixed point set generates an order 2 group of affine automorphisms with fixed point set @xmath407 . now the corollary follows from theorem [ symmetry ] . [ algebra ] let @xmath408 be observables in @xmath387 . if the ( complex ) linear span of @xmath409 is a * -subalgebra @xmath410 of @xmath387 , then udp = uda for pure states measured by these observables . udp and uda for a set of observables are nt affected if we include the identity among those observables , so hereafter we assume that @xmath411 . note that @xmath410 is the linear span of the unitaries in @xmath410 . furthermore , @xmath410 is a von neumann algebra containing the identity @xmath412 , so by the bicommutant theorem ( * ? ? ? 2.77 ) @xmath413 , where for @xmath414 , @xmath415 denotes the algebra of matrices that commute with all matrices in @xmath204 . combining these two statements shows that @xmath410 is the set of matrices that commute with all unitaries in @xmath416 , and thus is the set of fixed points of @xmath417 . it follows that @xmath418 ( the hermitian matrices in @xmath410 ) is the set of observables fixed by the compact group @xmath210 . the corollary follows from theorem [ symmetry ] . let @xmath419 . then the complex linear span of @xmath404 consists of the diagonal matrices . from corollary [ algebra ] it follows that for each pure state on @xmath387 , udp for @xmath404 implies uda . we can generalize the last example by taking the * -algebra consisting of diagonal observables with the restriction that certain diagonal entries coincide . for example , if @xmath420 we can look at diagonal matrices of the form @xmath421 whose linear span will be 4 dimensional . the space of hermitian members of this algebra is four dimensional . if we choose 4 observables including the identity spanning this space , then uda = udp for all pure states when measuring these observables . ( we could drop the identity from this list if we wish . ) in this way for any @xmath0 we can find a set of @xmath23 observables for any @xmath422 for which uda = udp . we can also find many larger sets of observables for which uda = udp . for example , for any @xmath0 we can consider the * -algebra of all block diagonal matrices with @xmath23 blocks that are of size @xmath423 for @xmath424 , where @xmath425 . the subspace of hermitian matrices in this algebra has dimension @xmath426 , so any such dimension is realizable as the number of observables in a set of observables for which uda = udp holds for all pure states . jc is supported by nserc and nsf of china ( grant no . 61179030 ) . zj acknowledges support from nserc , aro . nj is supported by the university of guelph brock scholarship and an nserc postdoctoral fellowship . dwk is supported by nserc discovery grant 400160 and nserc discovery accelerator supplement 400233 . bz is supported by nserc and cifar .

let be an edge - coloured graph . the minimum colour degree of is the largest integer such that , for every vertex , there are at least distinct colours on edges incident to . we say that is properly coloured if no two adjacent edges have the same colour . in this paper , we show that every edge - coloured graph with contains a properly coloured-factor . furthermore , we show that for any there exists an integer such that every edge - coloured graph with and contains a properly coloured cycle of length for every . this result is best possible in the sense that the statement is false for .

1212.6735

a classical theorem of dirac @xcite states that every graph @xmath0 on @xmath13 vertices with minimum degree @xmath14 contains a hamiltonian cycle . in this paper , we generalise this result to edge - coloured graphs . an _ edge - coloured graph _ is a graph @xmath0 with an edge colouring @xmath15 of @xmath0 . we say that @xmath0 is _ properly coloured _ if no two adjacent edges of @xmath0 have the same colour . moreover , @xmath0 is said to be _ rainbow _ if all edges have distinct colours . we consider the following analogue of degree for edge - coloured graphs @xmath0 . given a vertex @xmath16 , the _ colour degree @xmath17 _ is the number of distinct colours of edges incident to @xmath3 . the _ minimum colour degree _ @xmath1 of an edge - coloured graph @xmath0 is the minimum @xmath17 over all vertices @xmath3 in @xmath0 . one elementary result of graph theory states that every graph @xmath0 with @xmath18 contains a cycle . however , for all @xmath19 , there exist edge - coloured graphs @xmath0 with @xmath20 that do not contain any properly coloured cycles . grossman and hggkvist @xcite gave a sufficient condition for the existence of properly coloured cycles in edge - coloured graphs with two colours , which does not depend on @xmath1 . later on , yeo @xcite extended the result to edge - coloured graphs with any number of colours . one natural generalisation of dirac s theorem is to determine the minimum colour degree threshold for the existence of a rainbow hamiltonian cycle . however , such thresholds do not exist for all @xmath21 . indeed , for every even integer @xmath22 , there exists a properly coloured complete graph @xmath23 on @xmath21 vertices using exactly @xmath24 colours . note that @xmath25 , but @xmath23 does not contain a rainbow hamiltonian cycle . in fact , for each @xmath26 , there is a properly coloured @xmath27 that does not contain any rainbow hamiltonian path ( see @xcite ) . hence , a better question would be to ask for the minimum colour degree threshold for the existence of a properly coloured hamiltonian cycle . the problem of finding properly coloured spanning subgraphs in edge - coloured complete graphs @xmath23 has been investigated by numerous researchers . bang - jensen , gutin and yeo @xcite proved that if @xmath23 contains a properly coloured @xmath5-factor , then @xmath23 also contains a properly coloured hamiltonian path . a graph @xmath0 is said to be a _ @xmath28-path - cycle _ if @xmath0 is a vertex - disjoint union of at most one path and a number of cycles . feng , giesen , guo , gutin , jensen and rafiey @xcite showed that @xmath23 contains a properly coloured hamiltonian path if and only if it contains a spanning properly coloured @xmath28-path - cycle . we define _ @xmath29 _ to be the maximum number of edges of the same colour incident to the same vertex . in other words , @xmath30 over all monochromatic subgraphs @xmath31 , where an edge - coloured graph @xmath0 is _ monochromatic _ if all edges have the same colour . notice that @xmath32 . bollobs and erds @xcite proved that if @xmath33 , then @xmath23 contains a properly coloured hamiltonian cycle . they further conjectured that @xmath34 suffices . their result was subsequently improved by chen and daykin @xcite , shearer @xcite and alon and gutin @xcite . in @xcite , the author showed that for any @xmath35 , every @xmath23 with @xmath36 contains a properly coloured hamiltonian cycle , provided @xmath21 is large enough . therefore , the conjecture of bollobs and erds is true asymptotically . for a survey regarding properly coloured subgraphs in edge - coloured graphs , we recommend chapter 16 of @xcite . let @xmath0 be an edge - coloured graph ( not necessarily complete ) . li and wang @xcite proved that @xmath0 contains a properly coloured path of length @xmath37 or a properly coloured cycle of length at least @xmath38 . in @xcite , the author improved this result by showing that @xmath0 contains a properly coloured path of length @xmath37 or a properly coloured cycle of length at least @xmath39 . in the same paper , the author showed that every connected edge - coloured graph @xmath0 contains a properly coloured hamiltonian cycle or a properly coloured path of length at least @xmath40 . furthermore , the author also conjectured the following . [ conj : path ] every connected edge - coloured graph @xmath0 contains a properly coloured hamiltonian cycle or a properly coloured path of length @xmath41 . if the conjecture is true , then every edge - coloured graph @xmath0 with @xmath42 contains a properly coloured hamiltonian cycle . in this paper , we prove the following weaker result that @xmath42 implies the existence of a properly coloured @xmath5-factor in @xmath0 . [ thm:2factor ] every edge - coloured graph @xmath0 with @xmath43 contains a properly coloured @xmath5-factor . we also show that if @xmath44 and @xmath45 is large enough , then @xmath0 does indeed contain a properly coloured hamiltonian cycle . furthermore , @xmath0 contains a properly coloured cycle of length @xmath10 for all @xmath46 . [ thm : pchc ] for any @xmath35 , there exists an integer @xmath7 such that every edge - coloured graph @xmath0 with @xmath47 and @xmath48 contains a properly coloured cycle of length @xmath10 for all @xmath46 . given a subgraph @xmath49 in @xmath0 , we write @xmath50 for the ( edge - coloured ) subgraph obtained from @xmath0 by deleting all edges in @xmath49 . for an edge - coloured graph @xmath0 , let _ @xmath51 _ be the minimum @xmath52 over all monochromatic subgraphs @xmath49 in @xmath0 . note that @xmath53 . note that for an edge - coloured complete graph @xmath23 , we have @xmath54 . we prove the following stronger statement , which implies theorem [ thm : pchc ] . [ thm : pchc2 ] for any @xmath35 , there exists an integer @xmath7 such that every edge - coloured graph @xmath0 with @xmath55 and @xmath48 contains a properly coloured cycle of length @xmath10 for all @xmath46 . we now outline the proof of theorem [ thm : pchc2 ] for the case when @xmath56 , i.e. the existence of a properly coloured hamiltonian cycle . the proof adapts the absorption technique introduced by rdl , ruciski and szemerdi @xcite , which was used to tackle hamiltonicity problems in hypergraphs . the proof is divided into two main steps . in the first step , we find ( by lemma [ lma : abscycle ] ) a small ` absorbing cycle ' @xmath57 in @xmath0 . the absorbing cycle @xmath57 has the property that , given any small number of vertex - disjoint properly coloured paths @xmath58 in @xmath0 with @xmath59 for each @xmath60 , there exists a properly coloured cycle @xmath61 in @xmath0 with @xmath62 . thus , we have reduced the problem to covering the vertex set @xmath63 with small number of vertex - disjoint properly coloured paths . we remove the vertices of @xmath57 from @xmath0 and let @xmath64 be the resulting graph . since @xmath57 is small , we may assume that @xmath65 for some small @xmath66 . then we find a properly coloured @xmath5-factor in @xmath64 using lemma [ lma:2factor1 ] such that every cycle has length at least @xmath67 . ( although theorem [ thm:2factor ] also implies that @xmath64 contains a properly coloured @xmath5-factor , there is no bound on the lengths of the cycles . ) hence @xmath68 can be covered by at most @xmath69 vertex - disjoint properly coloured paths @xmath58 . by the ` absorbing ' property of @xmath57 , there is a properly coloured cycle @xmath61 with @xmath70 . therefore , @xmath61 is a properly coloured hamiltonian cycle as required . the paper is organised as follows . in the next section , we set up some basic notation and give the extremal example for conjecture [ conj : path ] . section [ sec:2fact ] is devoted to finding properly coloured @xmath5-factors and contains the proof of theorem [ thm:2factor ] . the absorbing cycle is constructed in section [ sec : abscycle ] . finally , theorem [ thm : pchc2 ] is proved in section [ sec : proof ] . throughout this paper , unless stated otherwise , @xmath0 will be assumed to be an edge - coloured graph with edge - colouring @xmath15 . for an edge @xmath71 , we denote its colour by @xmath72 . for @xmath16 , we denote by @xmath73 the neighbourhood of @xmath3 in @xmath0 . if the graph @xmath0 is clear from the context , we omit the subscript . let @xmath74 . we write @xmath75 $ ] for the ( edge - coloured ) subgraph of @xmath0 induced by @xmath76 . we also write @xmath77 for the subgraph obtained from @xmath0 by deleting all vertices in @xmath76 , i.e. @xmath78 $ ] . for edge - disjoint ( edge - coloured ) graphs @xmath0 and @xmath49 , we denote by @xmath79 the union of @xmath0 and @xmath49 . further , we write @xmath80 to mean @xmath81 . let @xmath82 not necessarily disjoint . whenever we define an auxiliary bipartite graph @xmath49 with vertex classes @xmath76 and @xmath83 , we mean @xmath49 has vertex classes @xmath84 and @xmath85 , where @xmath84 is a copy of @xmath76 and @xmath85 is a copy of @xmath83 . hence , @xmath76 and @xmath83 are considered to be disjoint in @xmath49 . given an edge @xmath86 in @xmath49 , we say @xmath87 and @xmath88 to mean @xmath89 and @xmath90 . in this paper , every path will be assumed to be directed . hence , the paths @xmath91 and @xmath92 are considered to be different for @xmath93 . we also allow a single vertex to be a ( trivial ) path . given vertex - disjoint paths @xmath94 , we define the path @xmath95 to be the concatenation of @xmath96 . for example , if @xmath97 and @xmath98 are vertex - disjoint paths and @xmath99 is a vertex not in @xmath100 , then @xmath101 denotes the path @xmath102 . let @xmath49 be a union of vertex - disjoint directed cycles . for each @xmath103 , let @xmath104 be the cycle in @xmath49 that contains @xmath105 . denote by @xmath106 and @xmath107 the successor and ancestor of @xmath105 in @xmath104 respectively . write @xmath108 for @xmath109 and @xmath110 for @xmath111 . for distinct vertices @xmath112 in a cycle @xmath57 in @xmath49 , define @xmath113 and @xmath114 to be the paths @xmath115 and @xmath116 in @xmath57 respectively . if @xmath117 , then set @xmath118 . in @xcite , the author gave a construction to show that conjecture [ conj : path ] is best possible for infinitely many values of @xmath45 and @xmath1 . the same construction also shows that theorem [ thm:2factor ] is best possible . we include it here for completeness . for @xmath119 with @xmath120 , there exists a connected edge - coloured graph @xmath0 on @xmath21 vertices with @xmath121 which does not contain a union of vertex - disjoint properly coloured cycles spanning more than @xmath122 vertices . in particular , @xmath0 does not contain a properly coloured @xmath5-factor or a properly coloured hamiltonian cycle . moreover , all properly coloured paths in @xmath0 have length at most @xmath122 . let @xmath119 with @xmath123 . let @xmath124 and @xmath125 be vertex sets with @xmath126 and @xmath127 . define @xmath0 to be the edge - coloured graph on the vertex set @xmath128 as follows . let @xmath129 $ ] be a rainbow complete graph and let @xmath130 $ ] be a set of independent vertices . for each @xmath131 , add an edge of new colour @xmath132 between @xmath133 and @xmath105 for each @xmath134 . by our construction , @xmath135 for all @xmath16 and so @xmath121 . let @xmath57 be a properly coloured cycle in @xmath0 . arbitrarily orient @xmath57 into a directed cycle . note that every vertex @xmath136 must be immediately followed by two consecutive vertices in @xmath137 , so @xmath138 . therefore , if @xmath139 is a collection of vertex - disjoint properly coloured cycles in @xmath0 , then @xmath139 spans at most @xmath140 vertices . the ` moreover ' statement is proved by a similar argument . first we prove theorem [ thm:2factor ] ; that is , every edge - coloured graph with @xmath43 contains a properly coloured @xmath5-factor . we now present a sketch of the proof . suppose that @xmath0 is an edge - coloured graph on @xmath21 vertices . let @xmath49 be a properly coloured subgraph in @xmath0 consisting of vertex - disjoint cycles such that @xmath141 is maximal . if @xmath142 , then we are done . hence we may assume that there exists @xmath143 . colour neighbourhood _ @xmath144 of @xmath99 in @xmath0 to be a maximal subset of neighbours of @xmath99 in @xmath0 such that @xmath145 for all distinct @xmath146 . ( the choice of @xmath144 will be specified later . ) assume that we are in the ideal case that @xmath147 . if @xmath148 is an edge and @xmath149 for some distinct @xmath146 , then @xmath150 is a properly coloured triangle and so @xmath151 contradicts the maximality of @xmath141 . hence , we may assume that for every @xmath146 , @xmath148 is not an edge , or @xmath152 , or @xmath153 . by an averaging argument , there exists a vertex @xmath154 such that the number of @xmath155 such that either @xmath156 or @xmath157 is at least @xmath158 . recall that there are at least @xmath159 neighbours @xmath3 of @xmath160 with @xmath161 . this means that @xmath162 and so @xmath163 . let @xmath0 be an edge - coloured graph on @xmath21 vertices with edge - colouring @xmath15 . set @xmath164 . suppose that @xmath0 does not contain a properly coloured @xmath5-factor . we will show that @xmath120 . this is trivial if @xmath165 , so we may assume that @xmath13 . by deleting edges in @xmath0 , we may assume that @xmath0 is edge - minimal , that is , any additional edge deletion would lead to a decrease in @xmath17 for some vertex @xmath3 in @xmath0 . suppose that @xmath64 is a monochromatic subgraph of @xmath0 , which is not isomorphic to a disjoint union of stars . then there exists an edge @xmath166 in @xmath64 with @xmath167 . set @xmath168 . note that @xmath169 for all vertices @xmath170 . this contradicts the edge - minimality of @xmath0 . therefore , every monochromatic subgraph @xmath64 in @xmath0 is a disjoint union of stars . let @xmath49 be a properly coloured subgraph in @xmath0 consisting of vertex - disjoint cycles such that @xmath141 is maximal . note that @xmath171 . arbitrarily orient each cycle in @xmath49 into a directed cycle . . we will choose @xmath144 to be a maximal subset of neighbours of @xmath99 in @xmath0 such that @xmath172 for all distinct @xmath173 . note that @xmath174 . obtain @xmath144 as follows . if there are at least two vertices @xmath175 and @xmath176 such that @xmath177 , then we choose @xmath178 to be in @xmath144 according to the following order of preferences ( if there are still at least two choices for @xmath105 , pick one arbitrarily ) : * @xmath179 ; * @xmath103 and @xmath180 ; * @xmath103 and @xmath181 ; * @xmath103 and @xmath182 ; * @xmath103 and @xmath183 ; * @xmath103 and @xmath184 . note that if @xmath185 satisfies property ( f ) , then @xmath186 for every vertex @xmath187 in @xmath104 . define @xmath188 to be the set of vertices in @xmath144 chosen due to ( a ) , and define @xmath189 similarly . next , set @xmath190 we have @xmath191 . if @xmath192 , then @xmath193 . for @xmath194 , we have @xmath195 since @xmath196 is a properly coloured cycle . also , for @xmath197 , we have @xmath198 since @xmath199 . thus , together with the definitions of @xmath200 and @xmath201 , we deduce that * if @xmath192 , then @xmath202 and @xmath203 , and * if @xmath204 , then @xmath205 and @xmath206 . let @xmath207 be the edge - coloured subgraph of @xmath0 induced by @xmath208 . for @xmath209 , define the vertex colour @xmath210 to be @xmath211 where @xmath212 is a new colour that does not appear in @xmath0 . the following claim concerns the colours of the edges of @xmath207 . [ clm : e(f ) ] if @xmath213 , then @xmath214 or @xmath215 . in particular , @xmath216 is an independent set in @xmath0 . _ proof of claim @xmath217 . _ suppose the claim is false , so there exists an edge @xmath213 with @xmath218 and @xmath219 . we will show that there is a properly coloured subgraph @xmath220 in @xmath0 consisting of vertex - disjoint cycles with @xmath221 , contradicting the maximality of @xmath141 . if @xmath222 , then @xmath223 . since @xmath224 , then @xmath150 is a properly edge - coloured triangle in @xmath0 . hence @xmath225 contradicts the maximality of @xmath141 . if @xmath226 and @xmath227 , then @xmath228 . by ( i ) we have @xmath229 and @xmath230 . moreover , we have @xmath231 since @xmath232 . therefore , @xmath233 is a properly coloured cycle . we obtain a contradiction by setting @xmath234 . similarly , if @xmath226 and @xmath235 , then we obtain a contradiction by considering @xmath236 instead of @xmath61 . if @xmath226 and @xmath237 , then @xmath238 . since @xmath239 is a properly coloured cycle , we have @xmath240 or @xmath241 . recall that @xmath242 , so @xmath230 and @xmath243 by ( i ) and ( ii ) . also , @xmath244 , so @xmath245 . hence , @xmath246 or @xmath247 would imply a contradiction . therefore , we may assume that @xmath248 . in order to prove the claim , it is enough to show that there exist at most two vertex - disjoint properly coloured cycles @xmath61 , @xmath249 spanning @xmath250 ( which would then imply that @xmath251 is a union of properly coloured cycles with @xmath252 , a contradiction ) . suppose that @xmath104 and @xmath239 are distinct . if @xmath253 , then by ( i ) we have @xmath254 since @xmath255 , we have @xmath256 . note that @xmath257 , so @xmath258 hence , @xmath259 ( see figure [ fig : clm : e(f)](a ) ) is a properly coloured cycle with vertex set @xmath260 . notice that @xmath61 is a properly coloured cycle if both and hold . if @xmath261 , then holds by ( i ) . therefore , we may assume without loss of generality that @xmath262 or @xmath263 with @xmath264 . since @xmath265 and the cycle @xmath104 is properly coloured , we have @xmath204 and @xmath266 . by ( ii ) and reversing the orientation of @xmath104 , we have @xmath202 , @xmath267 and @xmath268 . similarly , by reversing the orientation of @xmath239 if necessary , we may assume that both and hold and so we derive a contradiction . finally , suppose that @xmath269 . if @xmath270 , then @xmath271 . in particular , @xmath272 and @xmath273 . by ( i ) , @xmath274 ( see figure [ fig : clm : e(f)](b ) ) is a properly coloured cycle with vertex set @xmath275 , so we are done . if @xmath276 and @xmath235 , then @xmath277 , otherwise @xmath278 is a properly coloured cycle by ( i ) and ( ii ) . however , both @xmath279 and @xmath280 ( see figure [ fig : clm : e(f)](c ) ) are properly coloured cycles spanning @xmath281 . if @xmath282 , then by reversing the orientation of the cycle @xmath57 we conclude that @xmath270 and so we are done . if @xmath263 or @xmath237 , then we apply the following ` blindness ' argument . suppose that @xmath263 . since the cycle @xmath57 is properly coloured , @xmath283 or @xmath284 . we treat @xmath105 to be in @xmath285 if @xmath283 , and @xmath105 to be in @xmath286 otherwise . we apply a similar treatment when @xmath237 . hence , we are back in the case when @xmath287 . this completes the proof of claim [ clm : e(f ) ] . @xmath288 [ clm : y ] for each vertex @xmath209 there exists a vertex @xmath289 such that either @xmath290 or @xmath291 . _ proof of claim @xmath292 . if @xmath226 , then @xmath293 and so we are done by setting @xmath294 . if @xmath192 with @xmath295 , then @xmath71 is not an edge . set @xmath294 for @xmath276 with @xmath296 . if @xmath192 with @xmath297 , then @xmath298 for @xmath299 by the remark after the definition of @xmath144 . hence , @xmath300 . this implies that @xmath301 and so @xmath302 . set @xmath303 for @xmath276 with @xmath304 . suppose that @xmath192 with @xmath305 ( i.e. @xmath306 ) . we may assume that @xmath307 or else we set @xmath294 . if @xmath308 , then @xmath309 by definition . if @xmath310 , then @xmath311 . recall that every monochromatic subgraph of @xmath0 is a disjoint union of stars . hence , @xmath312 . in summary , we have @xmath313 for @xmath192 with @xmath314 . if @xmath315 , then ( i ) implies that @xmath316 is a properly coloured cycle and so we can enlarge @xmath49 , a contradiction . hence , @xmath317 set @xmath294 for all @xmath318 with @xmath314 . suppose that @xmath204 , so ( ii ) implies that @xmath319 . we may assume that @xmath307 or else we set @xmath294 . recall that every monochromatic subgraph of @xmath0 is a disjoint union of stars . hence , @xmath320 . if @xmath321 , then @xmath322 is a properly coloured cycle and so we can enlarge @xmath49 , a contradiction . therefore @xmath323 hence , we set @xmath294 for all @xmath324 . finally , suppose that @xmath325 . we may further assume that @xmath314 and @xmath307 . since @xmath326 and @xmath314 , we have @xmath327 by . on the other hand , since @xmath328 , we have @xmath329 by . hence , @xmath330 contradicting the fact that the cycle @xmath104 is properly coloured . therefore , @xmath331 for all @xmath263 , so we can set @xmath294 for @xmath263 . this completes the proof of claim [ clm : y ] . @xmath288 if @xmath332 , then @xmath333 by . by claim [ clm : e(f ) ] , @xmath216 is an independent set in @xmath0 . claim [ clm : y ] implies that for each @xmath263 there exists a vertex @xmath334 as the colour @xmath335 does not appear in @xmath0 . therefore @xmath336 which gives @xmath337 as required . thus , we may assume @xmath338 . we are going to show that there exists a vertex @xmath339 such that @xmath340 define @xmath341 to be the directed graph on @xmath342 such that there is a directed edge from @xmath105 to @xmath343 if @xmath344 or @xmath215 . thus , in proving , it suffices to show that there exists a vertex @xmath339 with indegree at least @xmath345 in @xmath341 . by claim [ clm : e(f ) ] , the base graph of @xmath341 is complete . moreover , @xmath346 exists for all @xmath263 and @xmath347 as the colour @xmath348 does not appear in @xmath0 . hence , the number of directed edges into @xmath349 is at least @xmath350 . by an averaging argument , there exists @xmath339 with indegree @xmath351 so holds . recall that the colour degree @xmath352 , so @xmath160 meets at least @xmath353 edges of distinct colours . in particular , there are at least @xmath354 neighbours @xmath355 of @xmath160 with @xmath356 . hence , there are at most @xmath357 vertices @xmath343 in @xmath358 such that either @xmath359 or @xmath360 . by claim [ clm : y ] , there exists @xmath361 such that @xmath362 or @xmath363 . together with , we have @xmath364 thus , @xmath120 as required . this completes the proof of theorem [ thm:2factor ] . the properly coloured @xmath5-factor obtained by theorem [ thm:2factor ] may contain @xmath365 cycles . we would like to minimise the number of cycles in a properly coloured @xmath5-factor . in the next lemma , we show that this can be achieved by assuming a slightly larger @xmath366 . recall that @xmath51 is the minimum @xmath52 over all monochromatic subgraphs @xmath49 in @xmath0 . so @xmath53 . [ lma:2factor1 ] for every integer @xmath367 , every edge - coloured graph @xmath0 with @xmath368 and @xmath369 contains a properly coloured @xmath5-factor in which every cycle has length at least @xmath370 . in particular , @xmath0 can be covered by at most @xmath371 vertex - disjoint properly coloured paths . the idea of the proof is rather simple . recall that a @xmath28-path - cycle is a vertex - disjoint union of at most one path @xmath372 and a number of cycles . we consider a properly coloured @xmath28-path - cycle @xmath49 in @xmath0 such that every cycle has length at least @xmath370 and @xmath141 is maximal . if we suppose that @xmath142 , then we may assume that @xmath49 contains a path @xmath372 as a component or else we are done . our aim is to show that there exists a properly coloured @xmath5-factor @xmath220 in @xmath373 $ ] such that each cycle has length at least @xmath370 . therefore @xmath374 is the desired properly coloured @xmath5-factor . suppose the contrary , and let @xmath0 be a counterexample with @xmath368 and @xmath369 . let @xmath49 be a properly coloured @xmath28-path - cycle in @xmath0 such that every cycle has length at least @xmath370 and @xmath141 is maximal . if @xmath142 , then @xmath49 contains a path @xmath372 as a component or else we are done . if @xmath375 , then we may assume @xmath49 contains a path ( maybe consisting of a single vertex ) . let @xmath376 be the path in @xmath49 . we further assume that @xmath49 is chosen such that @xmath10 is maximal ( subject to @xmath141 maximal ) . suppose that @xmath377 and so @xmath378 . then @xmath379 follows by the maximality of @xmath141 . let @xmath380 , and orient @xmath104 into a directed cycle such that @xmath381 . then @xmath382 is a properly coloured @xmath28-path - cycle on vertex set @xmath383 with path @xmath384 , contradicting the maximality of @xmath10 . therefore , @xmath93 . let @xmath385 and @xmath386 . so latexmath:[\[\begin{aligned } by the maximality of @xmath141 , we have @xmath388 . if @xmath389 , then orient @xmath104 into a directed cycle such that @xmath381 . then @xmath390 contradicts the maximality of @xmath10 . therefore , @xmath391 and similarly @xmath392 . moreover , @xmath393 if @xmath394 and @xmath395 , then @xmath396 is a properly coloured cycle of length at least @xmath397 . hence , @xmath398 consists of vertex - disjoint properly coloured cycles each of length at least @xmath370 . note that @xmath399 spans @xmath383 . if @xmath142 , then we are done . if @xmath375 , then together with a vertex in @xmath400 we obtain a properly coloured @xmath28-path - cycle contradicting the maximality of @xmath141 . therefore , we have @xmath401 or @xmath402 . let @xmath403 for @xmath404 . since @xmath405 and @xmath406 , by , we have @xmath407 for a given @xmath408 , we write @xmath409 for @xmath410 if @xmath411 , and @xmath412 for @xmath413 if @xmath414 . set @xmath415 for @xmath416 and @xmath417 for @xmath418 . for distinct @xmath419 , @xmath420 denotes the subpath of @xmath372 from @xmath99 to @xmath105 . if @xmath421 , set @xmath422 . for @xmath404 , define @xmath423 note that @xmath424 . since @xmath372 is properly coloured , for each @xmath425 , we have @xmath426 or @xmath427 . therefore , the three sets @xmath428 , @xmath429 , @xmath430 are pairwise disjoint . moreover , the union of these three sets is precisely @xmath431 . a similar statement also holds for @xmath432 . hence , @xmath433 for @xmath434 . note that if @xmath435 , then @xmath436 . similarly if @xmath437 , then @xmath438 . for @xmath404 , we set @xmath439 note that @xmath440 and @xmath441 . hence , @xmath442 and @xmath443 . moreover , for @xmath434 , @xmath444 now define an auxiliary directed bipartite graph @xmath207 as follows . the vertex classes of @xmath207 are @xmath445 and @xmath446 . ( recall that even though @xmath445 and @xmath446 might not be disjoint , we treat @xmath445 and @xmath446 to be so in @xmath207 . ) for @xmath447 there is a directed edge in @xmath207 from @xmath448 to @xmath449 if and only if @xmath307 and one of the following statements holds : * @xmath450 and @xmath451 ; * @xmath452 and @xmath453 ; * @xmath454 ; * @xmath455 , @xmath456 and @xmath457 ; * @xmath458 , @xmath459 and @xmath460 . we are going to show that there exists a vertex @xmath461 such that @xmath462 is in at least @xmath2 directed 2-cycles in @xmath207 . if @xmath463 , then @xmath464 exists for @xmath465 if and only if @xmath451 . hence , the outdegree of @xmath99 in @xmath207 is @xmath466 a similar statement holds for @xmath467 . in summary , for all @xmath468 , @xmath469 hence , there are at least @xmath470 directed edges from @xmath471 to @xmath446 , and similarly there are at least @xmath472 directed edges from @xmath473 to @xmath474 . note that if @xmath464 is a directed edge in @xmath207 with @xmath468 and @xmath475 , then @xmath307 and so @xmath476 forms a directed @xmath5-cycle in @xmath207 by ( iii ) . hence , if @xmath464 is a directed edge in @xmath207 not contained in a directed @xmath5-cycle with @xmath468 and @xmath465 , then @xmath477 . a similar statement holds for @xmath464 with @xmath478 and @xmath479 . therefore , at most @xmath480 directed edges from @xmath481 are not contained in a directed 2-cycle in @xmath207 . the number of directed edges @xmath464 in @xmath207 such that @xmath482 and @xmath464 is contained in a directed @xmath5-cycle is at least @xmath483 hence , by an averaging argument , there exists a vertex @xmath461 that is in at least @xmath2 directed @xmath5-cycles in @xmath207 . assume that @xmath484 . there are at least @xmath2 vertices @xmath485 such that @xmath86 is a directed 2-cycle in @xmath207 . pick one such @xmath486 such that the subpath @xmath487 in @xmath0 has length at least @xmath370 . similarly , if @xmath488 , then there is a vertex @xmath489 such that @xmath86 is a directed 2-cycle in @xmath207 and the subpath @xmath487 in @xmath0 has length at least @xmath370 . without loss of generality , we may assume that @xmath490 and @xmath491 . if @xmath492 , then @xmath493 with @xmath494 , or @xmath495 with @xmath496 , as @xmath372 is a properly coloured path and @xmath497 . together with ( i ) , ( ii ) and ( iv ) , we may assume that at least one of the following statements holds : * @xmath498 and @xmath494 ; * @xmath499 and @xmath496 ; * @xmath500 and @xmath501 . if @xmath502 , then the definition of @xmath503 implies that @xmath504 and @xmath505 . since @xmath504 , @xmath506 a similar statement also holds for @xmath507 , so at least one of the following statements holds : * @xmath498 , @xmath494 and @xmath508 ; * @xmath499 , @xmath496 and @xmath509 ; * @xmath500 and @xmath510 . by symmetry , we can deduce that at least one of the following statements holds for @xmath511 : * @xmath512 , @xmath513 and @xmath514 ; * @xmath515 , @xmath516 and @xmath517 ; * @xmath518 and @xmath519 . we now claim that * there exists a properly coloured @xmath5-factor @xmath220 in @xmath373 $ ] such that each cycle has length at least @xmath370 . if ( @xmath520 ) holds , then @xmath521 is a disjoint union of properly coloured cycles each of length at least @xmath370 on vertex set @xmath383 . if @xmath522 , then @xmath523 is the desired properly coloured @xmath5-factor . if @xmath524 , then @xmath525 is a properly coloured 1-path - cycle contradicting the maximality of @xmath526 . therefore , in order to complete the proof , it suffices to prove ( @xmath520 ) for each combination of ( a@xmath527 ) and ( b@xmath528 ) for @xmath529 . we say that @xmath530 if @xmath531 and @xmath532 with @xmath533 . we would like to point out that cases 26 are proved by similar arguments used in case 1 . since the explicit structures of @xmath220 ( see figure [ fig : r1sl ] ) are different , we include their proofs for completeness . * case 0 : ( a@xmath534 ) and ( b@xmath534 ) hold . * since @xmath500 and @xmath518 , ( a@xmath534 ) implies that @xmath535 and ( b@xmath534 ) implies that @xmath395 . recall that @xmath401 or @xmath402 . hence , we have a contradiction . * case 1 : ( a@xmath536 ) and ( b@xmath537 ) hold . * we have the following three statements : @xmath538 if @xmath539 , then we set @xmath540 ( see figure [ fig : r1sl](a ) ) . note that @xmath220 is properly coloured and a union of two cycles on vertex set @xmath383 . by ( a@xmath536 ) and , we have @xmath541 . thus , @xmath542 is a cycle of length at least @xmath370 . by a similar argument , we deduce that @xmath543 is also a cycle of length at least @xmath370 as @xmath544 by ( b@xmath537 ) . hence ( @xmath520 ) holds . next suppose that @xmath545 . if @xmath546 and @xmath547 , then @xmath548 where the second equality is due to the fact that @xmath502 . recall that @xmath549 . this contradicts the fact that @xmath550 . therefore , if @xmath546 , then @xmath551 . together with , we conclude that @xmath552 ( see figure [ fig : r1sl](b ) ) is a properly coloured cycle with vertex set @xmath553 , so ( @xmath520 ) holds . if @xmath554 , then @xmath542 , @xmath555 and @xmath543 ( see figure [ fig : r1sl](c ) ) are properly coloured cycles by . we will show that each is a cycle of length at least @xmath370 ( which then implies ( @xmath520 ) ) . by ( a@xmath536 ) and , we have @xmath541 . thus , @xmath542 is a cycle of length at least @xmath370 . similarly , @xmath543 is also a cycle of length at least @xmath370 as @xmath544 by ( b@xmath537 ) and . since @xmath462 and @xmath511 are chosen such that the subpath @xmath487 has length at least @xmath370 , @xmath555 has length at least @xmath370 . therefore , ( @xmath520 ) holds for ( a@xmath536 ) and ( b@xmath537 ) . * case 2 : ( a@xmath536 ) and ( b@xmath536 ) hold . * we have the following three statements : @xmath556 if @xmath530 , then we set @xmath557 ( see figure [ fig : r1sl](d ) ) . note that @xmath220 is properly coloured . by ( a@xmath536 ) and , we deduce that @xmath542 is a cycle of length at least @xmath370 . recall our choices of @xmath462 and @xmath511 that @xmath487 has length at least @xmath370 . this implies that @xmath558 is also a cycle of length at least @xmath370 . hence ( @xmath520 ) holds . next suppose that @xmath559 . by , we deduce that @xmath560 and so @xmath561 . hence , @xmath562 is a path with length at least one . together with , we conclude that @xmath563 ( see figure [ fig : r1sl](e ) ) is a properly coloured cycle with vertex set @xmath553 , so ( @xmath520 ) holds . therefore , ( @xmath520 ) holds for ( a@xmath536 ) and ( b@xmath536 ) . * case 3 : ( a@xmath536 ) and ( b@xmath534 ) hold . * since @xmath518 , we have @xmath564 note that both @xmath542 and @xmath565 ( see figure [ fig : r1sl](f ) ) are properly coloured cycles . by , both cycles have length at least @xmath370 . therefore , ( @xmath520 ) holds for ( a@xmath536 ) and ( b@xmath534 ) by setting @xmath566 . * case 4 : ( a@xmath537 ) and ( b@xmath536 ) hold . * we have the following three statements : @xmath567 first suppose that @xmath568 . if @xmath569 and @xmath570 , then @xmath571 where the second inequality is due to the fact that @xmath572 . this contradicts the fact that @xmath573 as @xmath574 by . therefore , if @xmath569 , then @xmath551 . together with , we conclude that @xmath575 ( see figure [ fig : r1sl](g ) ) is a properly coloured cycle with vertex set @xmath553 , so ( @xmath520 ) holds . if @xmath576 , then @xmath577 ( see figure [ fig : r1sl](h ) ) is a properly coloured cycle with vertex set @xmath553 by and . thus ( @xmath520 ) holds . suppose that @xmath559 . by , we deduce that @xmath560 and so @xmath561 . hence , @xmath578 is a properly coloured cycle . moroever , it has length at least @xmath370 by our choices of @xmath462 and @xmath511 . by and , we conclude that @xmath579 is a properly coloured cycle . it has length at least @xmath370 by . hence ( @xmath520 ) holds by setting @xmath580 ( see figure [ fig : r1sl](i ) ) . therefore , ( @xmath520 ) holds for ( a@xmath537 ) and ( b@xmath536 ) . * case 5 : ( a@xmath537 ) and ( b@xmath534 ) hold . * since @xmath518 , we have @xmath581 note that @xmath582 ( see figure [ fig : r1sl](j ) ) is a properly coloured cycle with vertex set @xmath553 and so ( @xmath520 ) holds . * case 6 : ( a@xmath527 ) and ( b@xmath528 ) hold for @xmath583 . * let @xmath584 be the properly coloured path obtained by reversing @xmath372 . hence , @xmath585 for all @xmath586 . given @xmath587 with @xmath411 , write @xmath588 to mean @xmath589 and @xmath590 . similarly , given @xmath587 with @xmath591 , we write @xmath592 and @xmath593 . ( hence , these primed notations are defined relative to @xmath594 . ) recall that @xmath594 is the reversed @xmath372 , so @xmath595 and @xmath596 . further , set @xmath597 and @xmath598 . now suppose that ( a@xmath534 ) and ( b@xmath537 ) hold . under the primed notation , we have @xmath599 more importantly , if we ignore the primes , then we obtain . therefore , in hindsight , we are in the case when ( a@xmath536 ) and ( b@xmath534 ) hold with respect to @xmath594 , i.e. we are in case 3 . hence , ( @xmath520 ) holds by setting @xmath600 . similarly , we deduce that the case when ( a@xmath537 ) and ( b@xmath537 ) hold with respect to @xmath372 corresponds to the case when ( a@xmath536 ) and ( b@xmath536 ) hold with respect to @xmath594 . also , ( a@xmath534 ) and ( b@xmath536 ) ( with respect to @xmath372 ) corresponds to ( a@xmath537 ) and ( b@xmath534 ) ( with respect to @xmath594 ) . therefore , we are in case 1 and case 5 , respectively . the proof of the lemma is completed . the aim of this section is to prove the following lemma , that is , to show that there exists a small absorbing cycle . [ lma : abscycle ] let @xmath601 . then there exists an integer @xmath7 such that whenever @xmath602 the following holds . suppose that @xmath0 is an edge - coloured graph of order @xmath21 with @xmath603 . then there exists a properly coloured cycle @xmath57 of length at most @xmath604 such that for all @xmath605 and for any collections @xmath606 of vertex - disjoint properly coloured paths in @xmath607 , there exists a properly coloured cycle with vertex set @xmath608 . given a vertex @xmath99 , we say that a path @xmath372 is an _ absorbing path for @xmath99 _ if the following conditions hold : 1 . @xmath609 is a properly coloured path of length @xmath610 ; 2 . @xmath611 ; 3 . @xmath612 is a properly coloured path . next we define an absorbing path for two disjoint edges . given two vertex - disjoint edges @xmath613 , @xmath614 , we say that a path @xmath372 is an _ absorbing path for @xmath615 _ if the following conditions hold : 1 . @xmath609 is a properly coloured path of length @xmath610 ; 2 . @xmath616 ; 3 . both @xmath617 and @xmath618 are properly coloured paths of length 3 . note that the ordering of @xmath615 is important . given a vertex @xmath99 , let @xmath619 be the set of absorbing paths for @xmath99 . similarly , given two vertex - disjoint edges @xmath613 , @xmath614 , let @xmath620 be the set of absorbing paths for @xmath621 . the following simple proposition follows immediately from the definition of an absorbing path for @xmath621 . [ prp : abspath ] let @xmath622 be a properly coloured path with @xmath623 . let @xmath624 be an absorbing path for @xmath625 with @xmath626 . then @xmath627 is a properly coloured path . lemma [ lma : abscycle ] will be proved as follows . suppose that @xmath628 . in the next lemma , we show that every @xmath619 and every @xmath629 are large . by a simple probabilistic argument , lemma [ lma : absorbing ] shows that there exists a small family @xmath630 of vertex - disjoint properly coloured paths ( each of length @xmath610 ) such that @xmath630 contains a linear number of members of @xmath619 for all @xmath170 and a similar statement holds for @xmath631 . finally , we join all paths in @xmath630 into one short properly coloured cycle @xmath57 using lemma [ lma : ifar ] . moreover , @xmath57 satisfies the desired property in lemma [ lma : abscycle ] . [ lma : abspath ] let @xmath632 and let @xmath633 be an integer . suppose that @xmath0 is an edge - coloured graph on @xmath21 vertices with @xmath634 . then @xmath635 for every @xmath170 and @xmath636 for any distinct vertices @xmath637 with @xmath613 , @xmath638 . fix a vertex @xmath170 . choose a vertex @xmath639 . next pick another vertex @xmath640 such that @xmath641 . notice that the number of such @xmath642 is at least @xmath643 . since @xmath644 , @xmath645 for all monochromatic subgraphs @xmath49 in @xmath0 . hence , the number of @xmath646 such that @xmath647 is at least @xmath648 . fix one such @xmath649 . similarly , there are at least @xmath650 choices for @xmath651 such that @xmath652 . notice that @xmath653 is an absorbing path for @xmath99 . recall that every path is assumed to be directed . therefore , there are at least @xmath654 many absorbing paths for @xmath99 . next , fix vertex - disjoint edges @xmath655 and @xmath656 in @xmath0 . choose a vertex @xmath657 such that @xmath658 . pick another vertex @xmath659 such that @xmath660 . the number of such @xmath642 is at least @xmath661 . recall that @xmath645 for all monochromatic subgraphs @xmath49 in @xmath0 . hence , the number of @xmath662 such that @xmath663 is at least @xmath664 . fix one such @xmath649 . similarly , there are at least @xmath665 choices for @xmath666 such that @xmath667 . note that @xmath653 is an absorbing path for @xmath668 . therefore , there are at least @xmath669 many absorbing paths for @xmath670 . lemma [ lma : absorbing ] is proved by a simple probabilistic argument since each of @xmath619 and @xmath671 is large . we will need the following chernoff bound for the binomial distribution ( see e.g. @xcite ) . recall that the binomial random variable with parameters @xmath672 is the sum of @xmath21 independent bernoulli variables , each taking value @xmath28 with probability @xmath673 , or @xmath674 with probability @xmath675 . [ prop : chernoff ] suppose that @xmath137 has the binomial distribution and @xmath676 . then @xmath677 . [ lma : absorbing ] let @xmath678 . then there exists an integer @xmath7 such that whenever @xmath602 the following holds . let @xmath0 be an edge - coloured graph on @xmath21 vertices . suppose that @xmath679 for every @xmath170 and @xmath680 for all distinct vertices @xmath637 with @xmath613 , @xmath638 . then there exists a family @xmath630 of vertex - disjoint properly coloured paths each of length @xmath610 , which satisfies the following properties : latexmath:[\[\begin{aligned } for all @xmath170 and for all distinct vertices @xmath637 with @xmath613 , @xmath638 . choose @xmath682 large so that @xmath683 recall that each path is assumed to be directed . so a path @xmath653 will be considered as a @xmath684-tuple @xmath685 . choose a family @xmath686 of @xmath684-tuples in @xmath687 by selecting each of the @xmath688 possible @xmath684-tuples independently at random with probability @xmath689 notice that @xmath690 for every @xmath170 and for all distinct @xmath637 with @xmath613 , @xmath638 . then by proposition [ prop : chernoff ] , the union bound and with probability at least @xmath691 , the family @xmath686 satisfies the following properties : @xmath692 for every @xmath170 and for all distinct @xmath637 with @xmath613 , @xmath638 . we say that two @xmath684-tuples @xmath693 and @xmath694 are _ intersecting _ if @xmath695 for some @xmath696 . furthermore , we can bound the expected number of pairs of @xmath684-tuples in @xmath686 that are intersecting from above by @xmath697 thus , using markov s inequality , we derive that with probability at least @xmath698 , @xmath699 hence , with positive probability the family @xmath686 has all properties stated in . remove one @xmath684-tuple in each intersecting pair in such a family @xmath686 . further remove those @xmath684-tuples that are not absorbing paths . we get a subfamily @xmath630 consisting of pairwise disjoint @xmath684-tuples , which satisfies latexmath:[\[\begin{aligned } for every @xmath170 and a similar statement holds for @xmath701 . since each @xmath684-tuple in @xmath630 is an absorbing path , @xmath630 is a set of vertex - disjoint properly coloured paths of length @xmath610 . in order to prove lemma [ lma : abscycle ] , it is sufficient to join the paths in @xmath630 given by lemma [ lma : absorbing ] into a short properly coloured cycle @xmath57 . the next lemma shows that we can join any two disjoint edges into a properly coloured path of length @xmath702 . [ lma : ifar ] suppose that @xmath0 is an edge - coloured graph of order @xmath21 with @xmath703 . let @xmath613 , @xmath614 be two edges in @xmath0 with @xmath704 . then there exists an edge @xmath705 with @xmath649 , @xmath706 such that both @xmath707 and @xmath708 are properly coloured paths . in particular , if @xmath709 , @xmath710 , @xmath175 and @xmath176 are distinct , then @xmath711 is a properly coloured path . let @xmath137 be the set of vertices @xmath712 such that @xmath713 . similarly , let @xmath125 be the set of vertices @xmath714 such that @xmath715 . so @xmath716 . define an auxiliary bipartite directed graph @xmath49 with vertex classes @xmath137 and @xmath125 such that for @xmath717 and @xmath134 * @xmath718 if and only if @xmath307 and @xmath719 ; * @xmath720 if and only if @xmath307 and @xmath721 . ( recall that in @xmath49 we treat @xmath137 and @xmath125 to be disjoint . ) the outdegree of each @xmath717 in @xmath49 is @xmath722 and similarly for each @xmath134 , @xmath723 . hence the number of directed @xmath5-cycles in @xmath49 is at least @xmath724 let @xmath705 be a directed @xmath5-cycle in @xmath49 with @xmath725 and @xmath726 . this implies that @xmath727 . hence , @xmath728 and @xmath729 are properly coloured paths . therefore , the lemma follows . we are ready to prove lemma [ lma : abscycle ] . given a graph family @xmath686 , we write @xmath730 . let @xmath731 . choose @xmath682 large so that @xmath732 and lemma [ lma : abspath ] holds . by lemma [ lma : abspath ] , @xmath733 for all @xmath170 and @xmath734 for all distinct vertices @xmath637 with @xmath613 , @xmath638 . let @xmath630 be the set of properly coloured paths obtained by lemma [ lma : absorbing ] . therefore , @xmath735 , latexmath:[\[\begin{aligned } |\mathcal{l}(x ) \cap \mathcal{f}'| & \ge 2^{-9 } \gamma^2 n = \frac{8 \eps^2 n}{81 } \text { and } for all @xmath170 and for all distinct vertices @xmath637 with @xmath613 , @xmath638 . we now show that @xmath57 has the desired property . let @xmath737 be the properly coloured paths in @xmath630 . for each @xmath738 , we are going to find an edge @xmath739 such that @xmath740 , @xmath741 is a properly coloured path , where we take @xmath742 , and @xmath743 for all @xmath744 . assume that we have already found edges @xmath745 for some @xmath738 . let @xmath746 and @xmath747 . set @xmath748 note that @xmath749 $ ] , so @xmath750 . apply lemma [ lma : ifar ] with @xmath751 , @xmath752 , @xmath753 , @xmath754 and @xmath755 , to obtain an edge @xmath756 such that @xmath757 is a properly coloured path in @xmath758 . this implies that @xmath759 is a properly coloured path in @xmath0 . therefore , there exist vertex - disjoint edges @xmath760 as desired . let @xmath57 be the properly coloured cycle obtained by concatenating @xmath761 . note that @xmath762 . suppose that @xmath763 is a family of at most @xmath764 vertex - disjoint properly coloured paths in @xmath63 . let @xmath765 be the family of vertex - disjoint properly coloured paths obtained from @xmath763 by breaking up every path @xmath766 with @xmath767 into isolated vertices . hence , @xmath765 contains at most @xmath768 paths and , for each path @xmath769 , either @xmath770 or @xmath771 . now , we assign each @xmath772 to a path @xmath773 such that @xmath774 if @xmath775 and @xmath776 if @xmath777 with @xmath623 . moreover , no two paths in @xmath765 are assigned to the same @xmath778 . note that such an assignment exists by . apply proposition [ prp : abspath ] to each pair @xmath779 and obtain a family @xmath780 of vertex - disjoint paths such that @xmath781 . note that @xmath782 moreover , there is a one - to - one correspondence between paths @xmath783 and @xmath784 such that the endedges of @xmath594 and @xmath785 are the same . recall that @xmath57 is a properly coloured cycle containing @xmath630 . let @xmath61 be the cycle obtained from @xmath57 by replacing the paths in @xmath630 with paths in @xmath780 . note that @xmath61 is properly coloured and @xmath786 . this completes the proof of lemma [ lma : abscycle ] . first , we prove that @xmath0 contains a properly coloured triangle if @xmath788 . [ prp : triangle ] let @xmath0 be an edge - coloured graph on @xmath21 vertices with @xmath788 . then every vertex in @xmath0 is contained in a properly coloured triangle . let @xmath99 be a vertex in @xmath0 and let @xmath15 be the edge - colouring on @xmath0 . define @xmath49 to be the directed graph on @xmath789 with directed edges @xmath346 if and only if @xmath148 is an edge in @xmath790 $ ] with @xmath791 . for @xmath103 , the outdegree of @xmath105 in @xmath49 is @xmath792 note that if @xmath148 is a directed 2-cycle in @xmath49 , then @xmath793 forms a properly coloured triangle in @xmath0 . hence , we may assume that @xmath49 does not contain any directed 2-cycles . moreover , we may assume that if @xmath794 is in @xmath49 , then @xmath152 . by an averaging argument , there exists a vertex @xmath103 such that @xmath795 . therefore @xmath796 since @xmath797 , the inequality above implies that @xmath798 , a contradiction . finally , we prove theorem [ thm : pchc2 ] . without loss of generality , we may assume that @xmath799 . choose @xmath682 large so that @xmath800 , @xmath801 , and lemma [ lma : abscycle ] holds . let @xmath0 be an edge - coloured graph on @xmath21 vertices with @xmath603 as stated in theorem [ thm : pchc2 ] . by proposition [ prp : triangle ] , @xmath0 contains a properly coloured triangle . suppose that @xmath10 is an integer with @xmath802 . since @xmath803 , we can greedily construct a properly coloured path of length @xmath804 . in particular , @xmath0 contains a properly coloured path @xmath805 of length @xmath806 . let @xmath64 be the subgraph of @xmath0 obtained by removing all the vertices @xmath807 . note that @xmath808 hence , lemma [ lma : ifar ] implies that there exist an edge @xmath705 in @xmath64 such that @xmath809 is a properly coloured path . therefore , @xmath810 is a properly coloured cycle of length @xmath10 . suppose that @xmath811 . let @xmath57 be the properly coloured cycle given by lemma [ lma : abscycle ] , so @xmath812 . let @xmath813 be the subgraph of @xmath0 obtained after removing all the vertices of @xmath57 . note that @xmath814 note that @xmath815 . by lemma [ lma:2factor1 ] , @xmath813 can be covered by at most @xmath816 vertex - disjoint properly coloured paths . that is , we can find vertex - disjoint properly coloured paths @xmath817 in @xmath813 such that @xmath818 , @xmath819 for all @xmath820 and @xmath821 . by removing vertices in the paths , we may assume that the paths @xmath817 span exactly @xmath822 vertices . by the property of @xmath57 guaranteed by lemma [ lma : abscycle ] , there exists a properly coloured cycle @xmath61 with @xmath823 . note that @xmath824 . therefore , @xmath61 is a properly coloured cycle of length @xmath10 as required . the author would like to thank katherine staden for her valuable comments . the author thanks the referees for their careful reading , one of whom in particular read the paper extremely carefully and suggested helpful clarifications .

we study and solve the problem of distilling secret key from quantum states representing correlation between two parties ( alice and bob ) and an eavesdropper ( eve ) via one way public discussion : we prove a coding theorem to achieve the `` wire tapper '' bound , the difference of the mutual information alice bob and that of alice eve , for so called cqq correlations , via one way public communication . this result yields information theoretic formulas for the distillable secret key , giving `` ultimate '' key rate bounds if eve is assumed to possess a purification of alice and bob s joint state . specialising our protocol somewhat and making it coherent leads us to a protocol of entanglement distillation via one way locc ( local operations and classical communication ) which is asymptotically optimal : in fact we prove the so called `` hashing inequality '' which says that the coherent information ( i.e. , the negative conditional von neumann entropy ) is an achievable epr rate . this result is well known to imply a whole set of distillation and capacity formulas which we briefly review .

quant-ph0306078

entanglement and secret correlation share an `` exclusiveness '' in the one case towards the total outside world , in the other towards an entity `` eve '' that has led quantum information scientists to speculate on a systematic relation between their theories : the works in this direction range from building analogies @xcite to using entanglement to prove information theoretic security of quantum key distribution @xcite , to attempts to prove the equivalence of the distillability of secret key and of entanglement @xcite . of course there are also conceptual differences : while the task of distilling secret perfect correlation derives from potential cryptographic applications ( and requires a third , malicious , party to formulate the operational problem ) , entanglement is useful for simple transmission tasks between two perfectly cooperating parties , as exemplified by dense coding @xcite and teleportation @xcite . the present paper falls into the third of the above categories , for we address the two questions , of distilling secret key from many copies of a quantum state ( itself a generalisation of classical information theoretic work begun by maurer @xcite and ahlswede and csiszr @xcite ) by public discussion and of distilling epr pairs by local operations and classical communication ( locc ) , in a unified way . to be more precise , after describing a protocol for secret key distillation from a state by one way public discussion , we show how secrecy codes of a particular structure can be converted into one way locc entanglement distillation protocols achieving the _ coherent information _ , as was conjectured for some time under the name of the `` hashing inequality '' ( after the hashing protocol in @xcite which attains the bound for bell diagonal two qubit states ) . it is well known from @xcite that this inequality yields information theoretic characterisations of distillable entanglement under general locc , as well as the quantum transmission capacity , without , with forward and with bidirectional classical side channel ( the first of these capacity theorems proved recently by shor @xcite , following a heuristic argument of lloyd @xcite , and subseqently in @xcite ) . our approach is very close to that of @xcite , and as far as secret key distillation is concerned the work @xcite : while here our resource is a three party quantum state ( `` static '' model ) , these papers deal with the `` dynamic '' analogue , where the resource is a quantum / wiretap channel . as for the structure of the paper : the main result of the cryptographic part is theorem [ thm : cqq:1way : coding ] in section [ sec : oneway - secret : cqq ] ; the form of the optimal rates is then not hard to obtain , as we shall show in the detailed discussion . it is theorem [ thm : cqq:1way : coding ] which we return to in the entanglement distillation part : a very general modification of the coding procedure will give us theorem [ thm : hashing ] , the hashing inequality ; and as before , the form of the optimal rates is not hard to get from there . a reader only interested in entanglement distillation can thus skip the second part of section [ sec : oneway - secret : cqq ] : there the general form of optimal one way distillable secret key is derived . in section [ sec : oneway - ent ] we turn to one way entanglement distillation , proving the hashing inequality and exhibiting the general form of optimal one way distillation ; then in section [ sec : cap - and - ent ] the consequences of the hashing inequality are detailed . appendices collect the necessary facts about typical subspaces ( [ app : types ] ) , some miscellaneous lemmas ( [ app : facts ] ) and miscellaneous proofs ( [ app : proofs ] ) . we will first study and solve the case of cqq correlations , i.e. , the initial state @xmath0 has the form @xmath1 then @xmath2 copies of that state can be written @xmath3 with @xmath4 and @xmath5 let @xmath6 be a random variable with distribition @xmath7 , and corresponding to the @xmath2 copies of @xmath8 consider independent identically distributed ( i.i.d . ) realisations @xmath9 of @xmath6 . a _ one way key distillation protocol _ consists of : * a channel @xmath10 , with range @xmath11 and @xmath12 . * a povm @xmath13 on @xmath14 for every @xmath15 . the idea is that alice generates @xmath16 ; her version of the key is @xmath17 , while she sends @xmath18 to bob . he obtains his @xmath19 by measuring his system @xmath20 using @xmath21 : @xmath22 for technical reasons we assume that the communication has a rate @xmath23 , for some constant @xmath24 . we call this an _ @xmath25protocol _ if 1 . @xmath26 . 2 . @xmath27 . 3 . there is a state @xmath28 such that for all @xmath29 , @xmath30 we call @xmath31 an _ achievable rate _ if for all @xmath2 there exist @xmath25protocols with @xmath32 and @xmath33 as @xmath34 . ( the convention in this paper is that @xmath35 and @xmath36 are understood to be to basis @xmath37 . ) finally define @xmath38 the _ one way ( or forward ) secret key capacity of @xmath8_. before we can formulate our first main result , we have to introduce some information notation : for a quantum state @xmath8 we denote the von neumann entropy @xmath39 , and the shannon entropy of a probability distribution @xmath7 , @xmath40 . if the state is the reduced state of a multi party state , like the @xmath0 above , we write @xmath41 , etc . in the particular case of eq . ( [ eq : cqq ] ) , obviously @xmath42 . for a general bipartite state @xmath43 define the _ ( quantum ) mutual information _ @xmath44 which for the cq state of eq . ( [ eq : cqq ] ) is easily checked to be equal to @xmath45 a quantity known as the _ holevo bound _ @xcite and which we denote @xmath46 , reflecting in the notation the distribution @xmath7 and the cq channel with channel states @xmath47 . we shall often use the abbreviation @xmath48 for this latter , if the states and distribution of the random variable @xmath6 are implicitly clear : this latter notation has the advantage that for any @xmath49 jointly distributed with @xmath6 , @xmath50 makes sense immediately , without our having to write down a composite state . finally , for a tripartite state @xmath51 , define the _ ( quantum ) conditional mutual information _ @xmath52 which is non negative by strong subadditivity @xcite . usually the state these notations refer too will be clear from the context ; where not we add it in subscript . observe that for a classically correlated system @xmath20 , the conditional mutual information takes the form of a probability average over mutual informations : e.g. , for the state of eq . ( [ eq : cqq ] ) , @xmath53 also for conditional mutual information we make use of the hybrid notation involving random variables : for example , for random variables @xmath54 and @xmath49 , jointly distributed with @xmath6 , @xmath55 is the average over @xmath54 of holevo quantities as above . [ thm : cqq:1way : coding ] for every cqq state @xmath8 , @xmath56 the idea is as follows : the state @xmath57 contains the description of a cq channel with channel states @xmath58 . we will cover `` evenly '' all typical type classes of block length @xmath2 by channel codes @xmath59 to transmit @xmath60 bits , most of which are `` good '' in the sense that they have small error probability . all of them are of the kind that the state of @xmath61 , when taking the average over the last @xmath62 bits of the input , is almost a constant operator , @xmath63 , independent of the leading bits . the key distillation scheme works then as follows : on observing @xmath64 , which is typical with high probability , alice announces the type of it and a random @xmath15 , such that @xmath64 is a codeword of the code @xmath59 , to bob . he is able to decode it with high probability ( because the code will be good with high probability ) , and they take the leading @xmath65 of the message as the key . this is uniformly distributed because the code is entirely within one type class . eve knows almost nothing about the key since she only has a state very close to @xmath63 , independent of the key . [ fig : babushka ] , which are good transmission codes for @xmath20 . a magnified view of one @xmath59 ( to the lower right ) reveals its inner structure : it is composed of @xmath66 , which are good privacy amplification codes against @xmath61.,title="fig:",width=321 ] in precise detail : let @xmath67 be an @xmath2type ( ultimately we will only be interested in typical @xmath67 , i.e. @xmath68 . ) consider random variables @xmath69 , independent identically distributed ( i.i.d . ) according to the uniform distribution on the type class @xmath70 ( see appendix [ app : types ] ) , @xmath71 , @xmath72 , @xmath73 . let @xmath74 we are interested in the probability of various random events ( for @xmath75 ) : * @xmath76evenness : * for all @xmath77 , @xmath78 with the indicator functions @xmath79 on @xmath70 . * @xmath76-secrecy : * for all @xmath80 , the average of @xmath81 over @xmath82 is close to @xmath83 : @xmath84 * codes @xmath59 are @xmath76good : * define the code @xmath59 as the collection of codewords @xmath85 . we call it _ @xmath76good _ if there exists a povm @xmath86 such that @xmath87 using the chernoff bound for the indicator functions @xmath79 evaluated at all points in @xmath70 ( lemma [ lemma : op : chernoff ] and following remarks ) , we obtain @xmath88 proposition [ prop : covering ] gives us ( observing @xmath89 ) , for every @xmath90 and sufficiently large @xmath2 , @xmath91 with @xmath92 . finally , proposition [ prop : hsw : type ] yields for every @xmath90 and if @xmath93 ( @xmath2 sufficiently large ) , @xmath94 since the individual events in this equation are independent , another application of the chernoff bound ( to the indiator function of `` @xmath76goodness '' ) gives , @xmath95 thus , if we pick @xmath96 , \\ m & = \exp\bigl [ n\bigl(i(q;\rho^b)-i(q;\rho^e)-3\delta\bigr ) \bigr ] , \\ l & = \exp\bigl [ n\bigl(h(q)-i(q;\rho^b)+2\delta\bigr ) \bigr ] , \end{aligned}\ ] ] and observing @xmath97 , the right hand sides of eqs . ( [ eq : evenness ] ) , ( [ eq : secrecy ] ) and ( [ eq : most : good - codes ] ) converge to @xmath98 as @xmath34 , and hence by the union bound alse the conjunction of these three events approaches unit probability asymptotically . thus , for sufficiently large @xmath2 , _ there exist _ codewords @xmath99 which together have the property of @xmath76evenness , @xmath76secrecy and that a fraction of at least @xmath100 of the @xmath101 is @xmath76good . clearly , we can construct such code sets for all types @xmath67 , of which there are at most @xmath102 many . now , the protocol works as follows : on observing @xmath64 from the source , alice determines its type @xmath67 and sends it to bob . if @xmath64 is not typical , i.e. if @xmath103 , the protocol aborts here . otherwise she selects a random @xmath15 such that @xmath64 is a codeword of @xmath59 , as well as random @xmath104 such that @xmath105 . ( the latter choice of course is unique most of the time : if @xmath59 is a good code , only a fraction of at most @xmath76 of the codewords have a collision . ) she informs bob also of @xmath15 ; if @xmath106 is not @xmath76good the protocol aborts . note that by the @xmath76evenness of the codewords , the state of @xmath107 conditional on @xmath67 and @xmath15 is @xmath108 ( by way of notation , `` @xmath109 '' stands for any number in the interval @xmath110 $ ] . ) now , since @xmath59 is a good code , bob can apply the decoding povm @xmath21 to his part of the system , and transform the state in eq . ( [ eq : key : almost ] ) into a state @xmath111 with the property @xmath112 both alice and bob measure @xmath29 and end up with a perfectly uniformly distributed key of length @xmath113 with probability @xmath114 , where @xmath115 with the dimensions @xmath116 and @xmath117 of bob s and eve s local system , respectively . ( recall that @xmath67 is typical , and use the fannes inequality , stated in appendix [ app : facts ] as lemma [ lemma : fannes ] . ) by the above property of @xmath111 , alice and bob disagree with probability @xmath118 . finally , thanks to the @xmath76secrecy , for all @xmath15 and @xmath29 , @xmath119 so eve s state after the protocol ( including her knowledge of @xmath67 ond @xmath15 ) is almost constant , whatever the value of @xmath29 . [ rem : secret - cc ] the communication cost of the protocol described in the above proof is asymptotically @xmath120 bits of forward communication ( per copy of the state ) : the information which code @xmath59 to apply from alice to bob . here are the facts we use in the proof : [ lemma : op : chernoff ] let @xmath121 be i.i.d . random variables taking values in the operators @xmath122 on the @xmath123dimensional hilbert space @xmath124 , @xmath125 , with @xmath126 , and let @xmath127 . then @xmath128 \right\ } \\ & \phantom{=============== } \leq 2d \exp\left ( -m\frac{\alpha\eta^2}{2\ln 2 } \right ) , \end{split}\ ] ] where @xmath129=\{x : a\leq x\leq b\}$ ] is an interval in the operator order . @xmath130 note that for the case @xmath131 this reduces to the classical chernoff bound for bounded real random variables @xcite . also the case of finite vectors of bounded real random variables is included by considering the matrices with vector entries on the diagonal and zero elsewhere . it is essential in the proof of the following result . [ prop : covering ] for a cq channel @xmath132 and a type @xmath7 , let @xmath133 be i.i.d . according to the uniform distribution on the type class @xmath134 , @xmath135 . define the state @xmath136 then for every @xmath137 , and sufficiently large @xmath2 , @xmath138 with @xmath139 . the proof is very close to that of the compression theorem for povms @xcite . we reproduce a version of the argument in appendix [ app : proofs ] . [ prop : hsw : type ] consider a cq channel @xmath132 and a type @xmath7 , and let @xmath140 be i.i.d . according to the uniform distribution on the type class @xmath134 , @xmath141 . then for every @xmath137 and sufficiently large @xmath2 , if @xmath142 , @xmath143 here we call a collection of codewords _ @xmath76good _ if there exists a povm @xmath144 on @xmath145 such that @xmath146 this really is only a slight modification of the holevo schumacher westmoreland argument @xcite : we give the proof in appendix [ app : proofs ] . this coding theorem puts us in the position to prove the following formula for the one way secret key distillation capacity of a cqq state : for conditional distributions @xmath147 and @xmath148 define the states [ thm : cqq:1way - key ] for very cqq state @xmath8 , @xmath149 with @xmath150,\ ] ] where the maximisation runs over all random variables @xmath49 depending on @xmath6 and @xmath54 depending on @xmath49 , i.e. there are channels @xmath67 and @xmath31 such that @xmath151 and @xmath152 , and the above formula refers to the state @xmath153 the ranges of @xmath49 and @xmath54 may be taken to have cardinalities @xmath154 and @xmath155 , and furthermore @xmath54 can be taken a ( deterministic ) function of @xmath49 . let us begin with the converse part , i.e. the inequality `` @xmath156 '' : consider an @xmath25protocol with rate @xmath31 ; then by its definition , and using standard information inequalities and the fannes inequality lemma [ lemma : fannes ] @xmath157 letting @xmath158 and @xmath159 we obtain @xmath160 with arbitrarily small @xmath161 as @xmath34 . the proof of the properties of @xmath49 and @xmath54 is given in appendix [ app : proofs ] . now we come to the proof of the direct part , i.e. the inequality `` @xmath162 '' : it is clearly sufficient to show that , for given @xmath49 and @xmath54 , the rate @xmath163 is achievable . to this end , consider a protocol , where alice generates @xmath49 and @xmath54 for each copy of the state i.i.d . , and broadcasts @xmath54 : this leaves alice , bob and eve in @xmath2 copies of @xmath164 observing @xmath165 we can invoke theorem [ thm : cqq:1way : coding ] , and are done . [ rem : single - letter ] comparing this with the classical analogue in @xcite , it is a slight disappointment to see that here we do nt get a single letter formula . the reader may want to verify that the technique used there to single letterise the upper bound does not work here , as it introduces conditioning on quantum registers , while our @xmath54 has to be classical . one can clearly also use a general three party state @xmath0 to generate secret key between alice and bob : a particular strategy certainly is for alice to perform a quantum measurement described by the povm @xmath166 , which leads to the state @xmath167 then , starting from many copies of the original state @xmath8 , we now have many copies of @xmath168 , and theorem [ thm : cqq:1way - key ] can be applied . because we can absorb the channel @xmath169 into the povm , we obtain the direct part ( `` @xmath162 '' ) in the following statement : [ thm : qqq:1way - key ] for every state @xmath0 , @xmath170 with @xmath171 where the maximisation is over all povms @xmath166 and channels @xmath31 such that @xmath172 , and the information quantities refer to the state @xmath173 the range of the measurement @xmath67 and the random variable @xmath54 may assumed to be bounded as follows : @xmath174 and @xmath175 , and furthermore @xmath54 can be taken a ( deterministic ) function of @xmath6 . after our remarks preceding the statement of the theorem , we have only the converse to prove . this will look very similar to the converse of theorem [ thm : cqq:1way - key ] . even though we havent so far defined what a key distillation protocol is in the present context , we can easily do that now ( and check that the procedure above is of this type ) : it consists of a measurement povm @xmath176 for alice and the povms @xmath21 for bob , with the same conditions ( 1)(3 ) as in the first paragraphs of this section , where as before we assume a rate bound on the public discussion : @xmath23 . this obviously generalises the definition we gave for cqq states . consider an @xmath25protocol with rate @xmath31 ; using standard information inequalities and the fannes inequality lemma [ lemma : fannes ] we can estimate as follows : @xmath157 the measurement @xmath67 and @xmath177 are permissible in the definition of @xmath178 , hence we obtain @xmath160 with arbitrarily small @xmath161 as @xmath34 . it remains to prove the bounds on the range of @xmath6 and @xmath54 for which we imitate the proof of the corresponding statement in theorem [ thm : cqq:1way - key ] : the full argument is given in appendix [ app : proofs ] . [ rem : worstcase ] clearly the worst case for alice and bob is when eve holds the system @xmath61 of a purification @xmath179 of @xmath43 , because clearly every other extension @xmath0 of @xmath43 can be obtained from the purification by a quantum operation acting on @xmath61 . our result ( at least in principle ) characterises those bipartite states @xmath43 for which one way key distillation is possible at positive rate . we have to leave open the question of characterising the states for which positive rates can be obtained by general two way public discussion ( compare the classical case @xcite ! ) . note that the classical analogue of the `` worst case '' is total knowledge of eve about both alice s and bob s random variables which makes key distillation totally impossible . for quantum states thus , it must be some `` non classical '' correlation which makes positive rates possible ; it is tempting to speculate that a manifestation of entanglement is behind this effect . we do not fully resolve this issue in the present paper ; nevertheless , in a similar vein , we show in the following section that if @xmath43 allows one - way distillation of epr pairs at positive rates , then our cryptographic techniques give a construction of an entanglement distillation protocol by a modification of key distillation protocols of a particular form . consider an arbitrary state @xmath43 between alice and bob . in @xcite the task of distilling epr pairs at optimal rate from many copies of @xmath8 , via local operations and classical communication ( locc ) , was introduced . a _ one way entanglement distillation protocol _ consists of * a quantum instrument @xmath180 for alice . ( an _ instrument _ @xcite is a quantum operation with both classical and quantum outputs it is modelled in general as a cp map valued measure ; for our purposes it is a finite collection of cp maps which sum to a cptp map . ) * for each @xmath15 a quantum operation @xmath181 for bob . we call it an _ @xmath25protocol _ if it acts on @xmath2 copies of the state @xmath8 and produces a maximally entangled state @xmath182 up to fidelity @xmath183 : @xmath184 note that we may assume without loss of generality that @xmath185 and @xmath181 output states supported on the reduced states of @xmath186 on alice s and bob s system , respectively : otherwise we could improve the fidelity . a number @xmath31 is an _ achievable rate _ if there exist , for every @xmath2 , @xmath25-protocol , with @xmath32 and @xmath33 as @xmath34 . finally , @xmath187 is the _ one way ( or forward ) entanglement capacity of @xmath8_. in @xcite the case of bell diagonal two qubit states , @xmath188 was considered and proved that @xmath189 , by a method called `` hashing protocol '' ( this was generalised recently to higher dimensions in @xcite ) . concerning lower bounds not much more is known , but there are numerous works dealing with upper bounds on the distillable entanglement : the _ entanglement of formation _ @xmath190 @xcite , the _ relative entropy of entanglement _ @xmath191 @xcite , the _ rains bound _ @xmath192 @xcite , and the recently introduced _ squashed entanglement _ @xmath193 @xcite . to connect to the cryptographic setting discussed so far , construct a purification @xmath194 of @xmath8 , of which we are particularly interested in its schmidt form @xmath195 consider the following special strategy for a one way secret key distillation protocol , on @xmath2 copies of the state : alice measures @xmath64 ( i.e. the above schmidt basis ) , and applies the secret key distillation protocol from theorem [ thm : cqq:1way : coding ] : it is easy to evaluate the key rate : @xmath196 by letting alice and bob execute this protocol `` coherently '' , we can prove : [ thm : hashing ] @xmath197 the right hand side here equals the negative conditional von neumann entropy , @xmath198 , a quantity known as _ coherent information _ @xcite , which we denote ( acknowledging is directionality ) @xmath199 . if the state this is referring to is not apparent from the context , we add it in subscript : @xmath200 . recall the structure of the protocol in the proof of theorem [ thm : cqq:1way : coding ] : for each typical type @xmath67 we have a collection of codewords @xmath201 , @xmath71 , @xmath72 and @xmath73 from @xmath70 satisfying @xmath76evenness , @xmath76secrecy , and a fraction of at least @xmath100 of the codes @xmath202 are @xmath76-good . the first step of the protocol is that alice measures the type @xmath67 non destructively and informes bob about the result . the protocol aborts if @xmath67 is not typical , i.e. if @xmath203 . this leaves the post measurement state @xmath204 define now a quantum operation for alice , with kraus elements @xmath205 which we interpret as an instrument with outcomes @xmath15 and @xmath206 @xcite : @xmath207 . this outcome is communicated to bob . that these are really permissible kraus operators we obtain from the @xmath76evenness condition . the outcome @xmath206 ( resulting in abortion of the protocol ) is observed with probability at most @xmath76 ( if @xmath2 is large enough ) . the other outcomes @xmath15 all occur with the probability @xmath208 in which case the output state of the instrument is @xmath209 ( the absence of the @xmath109 factors when compared to the analogous eq . ( [ eq : key : almost ] ) in the proof of theorem [ thm : cqq:1way : coding ] ) is due to our having introduced the error event @xmath206 . ) now , just as in the proof of theorem [ thm : cqq:1way : coding ] , bob decodes @xmath29 and @xmath210 , at least if @xmath59 is @xmath76good ( which fails to happen with probability only @xmath211 ) . but he does it coherently , by introducing an ancilla system @xmath212 in a standard state @xmath213 and applying a unitary to extract @xmath214 into @xmath20 , leaving in @xmath212 whatever is necessary to make the map unitary . this transforms the state in eq . ( [ eq : ent : almost ] ) into a state @xmath215 where @xmath216 is the probability of the code incorrectly identifying @xmath214 , and @xmath217 is orthogonal to @xmath218 . now , because the code is @xmath76good , @xmath219 where we have used the markov inequality : at most a fraction of @xmath220 of the @xmath216 can be larger than @xmath220 . since the decoding only affects bob s registers , but certainly not @xmath61 , we have @xmath221 and hence we can assume that @xmath222 this implies @xmath223 at this point , alice and bob almost have their maximal entanglement of the @xmath29variable . all that remains is to be done is to disentangle eve : to begin , alice measures the @xmath210component of her register in the fourier transformed basis : @xmath224 and tells bob the result @xmath225 , who applies the phase shift @xmath226 to the @xmath210component of his register @xmath20 . this transforms @xmath227 into a state @xmath228 with @xmath229 invoking the non decrease of the fidelity under quantum operations , applied to eq . ( [ eq : ent : fidelity ] ) . absorbing @xmath210 into the register @xmath212 , the right hand state in the last equation can be rewritten as @xmath230 with @xmath231 the reduced states of eve of the @xmath232 is @xmath233 where we made use of eq . ( [ eq : eveisout ] ) , which is , by the @xmath76secrecy , at trace distance at most @xmath76 from a state we denoted @xmath83 in the proof of theorem [ thm : cqq:1way : coding ] . by lemma [ lemma : norm - fidelity ] in appendix [ app : facts ] , @xmath234 . choosing a purification @xmath235 of @xmath83 , this means that there are unitaries @xmath236 on @xmath212 such that @xmath237 because the mixed state fidelity equals the maximum pure state fidelity over all purifications of the states and all purifications are related by unitaries on the purifying system @xcite . hence , if bob applies @xmath238 to his share of the state , then the state in eq . ( [ eq : only : s : remains ] ) is transformed into a state @xmath239 with @xmath240 of course , he actually works on @xmath241 , so they end up with the state @xmath242 , which has fidelity @xmath243 to @xmath244 , hence with eq . ( [ eq : ent : fidelity ] ) we conclude ( by simple geometry ) that it has fidelity @xmath245 to @xmath246 . nontypical @xmath67 , the event @xmath206 or bad code @xmath59 happen with total probability at most @xmath247 . in the `` good '' case , alice and bob distill up to fidelity @xmath248 a maximally entangled state of log schmidt rank @xmath249 with @xmath250 just as at the end of the proof of theorem [ thm : cqq:1way : coding ] . [ rem : ent - cc ] the communication cost of the above protocol is asymptotically @xmath251 bits of forward classical communication per copy of the state : the information which code @xmath59 to use , plus the information from the measurement of the fourier transformed basis @xmath252 . even though at first sight there seems to be little reason to believe that our procedure is optimal for this resource ( consider for example a separable initial state : alice will have mutual information with a purification but clearly the best thing is to do nothing ) , it is amusing to see the quantum mutual information show up here . it is in fact possible to show that subject to another optimisation , the quantum mutual information between alice and eve gives indeed the minimum forward communication cost @xcite . [ expl : bell - states ] it is interesting to compare our method to the original hashing protocol of @xcite , for the case of mixtures of bell states @xmath253 with the numbering of the bell states introduced in @xcite : @xmath254 the purification we use in the proof reads @xmath255 with @xmath256 note that this is indeed a schmidt decomposition . first of all , the communication cost of our protocol evaluates ( using the symmetry between a and b ) to @xmath257 which is the same as in @xcite . but the way of the hashing protocol is to `` hash '' information about the identity of the state in the bell ensemble into approximately @xmath258 of the states , which then are measured locally and the results communicated . our protocol in contrast has two very distinct communication parts : there is the `` code information '' ( which amounts to error correction between alice and bob , with built in privacy amplification for eve s information about the basis state ) , and there is the `` phase information '' from the measurement in the fourier transformed basis . the first amounts to @xmath259 while the second is @xmath260 our result leads to the general formula for one way distillable entanglement : [ thm:1way - ent ] for any bipartite state @xmath43 , @xmath261 with @xmath262 where the maximisation is over quantum instruments @xmath263 on alice s system , @xmath264 and @xmath265 . the range of @xmath15 can be assumed to be bounded , @xmath266 , and moreover each @xmath185 can be assumed to have only one kraus operator : @xmath267 . first , for the direct part , it is sufficient to consider an instrument @xmath268 on one copy of the state : if alice performs the instrument @xmath268 on each copy and communicates the result to bob , they end up with the new state @xmath269 observe that @xmath270 thus application of theorem [ thm : hashing ] to @xmath168 gives achievability . for the converse , consider any one way distillation protocol with rate @xmath31 , and denote alice s instrument by @xmath271 , bob s quantum operations by @xmath181 . write @xmath272 then , using fannes inequality lemma [ lemma : fannes ] , the convexity of the coherent information in the state @xcite and quantum data processing @xcite , @xmath273 where @xmath274 . hence we get @xmath275 with arbitrarily small @xmath250 as @xmath34 , and we are done . as for the bound on @xmath276 and the structure of @xmath268 , observe that if one @xmath185 has more than one kraus element , one can decompose @xmath277 into a sum of terms @xmath278 : for the corresponding probabilities @xmath279 and for the post measurement states @xmath280 . then by the convexity of @xmath281 in the state @xcite , @xmath282 by the polar decomposition and invariance of @xmath281 under local unitaries we may further assume that @xmath283 , i.e. @xmath284 ; in this form the whole instrument is actually described by the povm @xmath285 , and each povm corresponds to an instrument by taking as the @xmath286 the square roots of the povm operators . now , invoking a theorem of davies @xcite ( which actually is another application of caratheodory s theorem , lemma [ lemma : caratheodory ] ) , any povm is a convex combination of extremal povms , which have at most @xmath287 non zero elements each , and this convex decomposition clearly carries over to the instruments : @xmath288 . since then @xmath289 is the same convex combination of similar such terms for the instruments @xmath290 , at least one of these gives a higher yield @xmath291 : note that the cp maps of @xmath290 are scalar multiples of the @xmath185 , hence the output state of @xmath290 with classical result @xmath15 is @xmath292 . horodecki@xmath293 @xcite have observed that the hashing inequality implies information theoretic formulas for a number of quantum capacities and the distillable entanglement : in particular , the quantum capacity of a quantum channel , either unassisted or assisted by forward or two way communication is given by a formula involving coherent information ( where we indicate the assisting resource in the subscript ) : [ thm : q : capacities ] let @xmath294 be any quantum channel . then , @xmath295 with any pure state @xmath296 on @xmath297 and the state @xmath298 furthermore , @xmath299 with any pure state @xmath296 on @xmath297 , two way locc operation @xmath300 and the state @xmath301.\ ] ] see @xcite . that forward communication does not help was proved in @xcite , and that the right hand side is an upper bound to @xmath302 was shown in @xcite . the idea of achievability is to distill the state @xmath303 and then use teleportation this involves forward communication but either it is free or the whole procedure including the distillation and teleportation uses only forward communication , which by @xcite can be removed . in @xcite a similar formula ( involving a coherent information @xmath304 ) was proposed for the quantum capacity with classical feedback . however , the proof as indicated above does not work in this case : indeed we may use the back communication to help distillation , but teleportation needs a _ forward _ communication , so we end up with a quantum channel code utilising two way classical side communication , which is not known to be reducible to just back communication : in fact , the results of bowen @xcite might be taken as indication that for the erasure channel the capacities with feedback and with two way side communication are different . we have already given a formula for the distillable entanglement using one way locc in theorem [ thm:1way - ent ] : [ thm : ent : capacities ] for any state @xmath43 , @xmath305 with any two way locc operation @xmath300 and the coherent information refers to the state @xmath306 . for the direct part ( `` @xmath162 '' ) it is obviously sufficient to consider any two way locc operation @xmath300 on the bipartite system , which applied to @xmath8 gives a state @xmath307 . doing that for @xmath2 copies of @xmath8 , application of theorem [ thm : hashing ] shows that we can distill epr pairs at rate @xmath308 from this . conversely , let @xmath309 be a two way locc producing a state @xmath310 with @xmath311 , @xmath312 . without loss of generality we may assume that @xmath310 is supported within the tensor product of the supports of the reduced states of @xmath186 . thus , @xmath313 and we are done . it was shown furthermore in @xcite that for an ensemble @xmath314 of bipartite pure states the hashing inequality implies @xmath315 this inequality was first exhibited in @xcite for a class of examples , and conjectured to be true in general . note that the inequality is trivially true ( using only concavity of the entropy ) for the _ loss of coherent information _ on the left hand side . * history and relation to other work : * the coherent information made its appearance in @xcite where its relation to quantum channel capacity was conjectured and many of its properties proved . independently @xcite proposed this quantity and a heuristic for a proof which however fell short of a proof . only recently hamada @xcite succeeded in giving a lower bound on quantum channel capacity in terms of coherent information still with a crucial restriction to stabiliser codes . it took until @xcite for a full proof to be found but then quite quickly one of us @xcite discovered a proof based on private information transmission , an idea inspired by the work of schumacher and westmoreland @xcite . regarding entanglement distillation , the hashing inequality appears to have been a folk conjecture from the publication of @xcite on , which however has received much less attention than the quantum channel coding problem . it was codified as an important conjecture in @xcite . while completing the writing of the present paper we learned of the work @xcite , in which it is shown that the proof by random coding of the channel capacity theorem can be used to obtin the hashing inequality . it may be interesting to compare the proofs @xcite for the achievability of the coherent information to ours and @xcite . while we , on the face of it , take a detour via secret key distillation , the final procedure can be argued more direct : in particular , we do nt require the `` double blocking '' which in the other approaches seem necessary to reduce to a situation in which alice s end is in a maximally mixed state . thus , presumably , our codes achieve rates approaching the coherent information more quickly , i.e. for smaller block length . our findings not only transport an existing classical theory of distilling secret key from prior correlation ( maurer @xcite , ahlswede and csiszr @xcite , and follow up work ) to the quantum case , but also link this subject to entanglement distillation in an operational way : a coherent implementation of the basic secret key distillation protocol yielded an entanglement distillation protocol achieving the coherent information this then implies information theoretic formulas for distillable entanglement and quantum capacities . we want to draw the reader s attention to several open question that we have to leave : first of all , are there states for which @xmath316 ? are there maybe even bound entangled states with positive key rate ? a first step might be to find states such that @xmath317 . note that the potential gap between @xmath318 and @xmath319 comes from the possibility to have more general measurements at alice s side than the complete von neumann measurement in the schmidt basis that was our starting point in the proof of theorem [ thm : hashing ] ( actually any complete von neumann measurement would do ) : namely , in key distillation , a viable option for alice is to discard part of her state ( corresponding to using higher rank povm elements ) , but keep that part secret from eve all the same ; while in entanglement distillation , ` eve' is everything except alice and bob , so it is as if she would get the parts alice decided to toss away . a second group of open questions : in general , the optimisations in theorems [ thm : cqq:1way - key ] , [ thm : qqq:1way - key ] and [ thm:1way - ent ] are quite nasty , most so because they involve a limit of many copies of the state . in the classical theory of secret key distillation , a single letter formula for the optimal one way key rate is proved in @xcite , so there might be hope at least for theorem [ thm : cqq:1way - key ] . in contrast , the optimal rate of two way protocols or even a procedure to decide if it is nonzero is still to be found ( see the very well informed reviews @xcite and @xcite ) , which is why we concentrate on one way protocols for now . it is known that distillability of entanglement may be absent for a single copy of a state , but could appear for collective operations on several copies ( see again the review @xcite , sections 6.3 and 7.2 and references therein ) , so there are only limited possibilities for making theorem [ thm:1way - ent ] into a single letter formula . note in particular that the results of @xcite see also the discussion of barnum , nielsen and schumacher @xcite where the failure of subadditivity for the coherent information is observed imply that single letter maximisation of the coherent information will certainly not achieve the optimum distillability . it would therefore be good to have at least an a priori bound on the number @xmath2 of copies of the state which we have to consider to have @xmath320 within , say @xmath76 , of the optimal rate . in general , good single letter lower and upper bounds @xcite are still wanted ! finally , we would like to know what the public / classical communication cost is of distilling secret key and entanglement , respectively , in particular in the one way scenario ( which at any rate seems to be the one open to analysis ) . more generally , if we limit the amount of communication , can we determine the optimal distillation rates ( see @xcite for the communication cost of _ common randomness _ distillation ) ? in the entanglement case this should link up with initial efforts to understand the communication cost of various state transformation tasks @xcite . a study concerning the forward communication cost of entanglement distillation is in perparation @xcite . we wish to thank patrick hayden and debbie leung for a discussion at the conception of this work . after the present research had been done , we learned from micha and pawe horodecki that they had a different approach to proving the hashing inequality , which they were able to carry through after hearing of our result . we want to thank them for showing us an early draft of their paper @xcite . i d is supported in part by the nsa under us army research office ( aro ) grants daag5598c0041 and daad1901106 . aw is supported by the u.k . engineering and physical sciences research council . the following material can be found in most textbooks on information theory , e.g. @xcite , or in the original literature on quantum information theory @xcite . for strings of length @xmath2 from a finite alphabet @xmath321 , which we generically denote @xmath322 , we define the _ type of @xmath64 _ as the empirical distribution of letters in @xmath64 : i.e. , @xmath7 is the type of @xmath64 if @xmath323 it is easy to see that the total number of types is upper bounded by @xmath102 . the _ type class of @xmath7 _ , denoted @xmath134 , is defined as all strings of length @xmath2 of type @xmath7 . obviously , the type class is obtained by taking all permutations of an arbitrary string of type @xmath7 . the following is an elementary property of the type class : @xmath324 with the ( shannon ) entropy @xmath325 . for @xmath90 , and for an arbitary probability distribution @xmath7 , define the set of _ @xmath7typical sequences _ as @xmath326 by the law of large numbers , for every @xmath327 and sufficiently large @xmath2 , @xmath328 furthermore : @xmath329 for a ( classical ) channel @xmath330 ( i.e. a stochastic map , taking @xmath331 to a probability distribution @xmath332 on @xmath333 ) and a string @xmath334 of type @xmath7 we denote the output distribution of @xmath64 in @xmath2 independent uses of the channel by @xmath335 let @xmath90 , and define the set of _ conditonal @xmath336typical sequences _ as @xmath337 where @xmath338 is the conditional entropy . once more by the law of large numbers , for every @xmath76 and sufficiently large @xmath2 , @xmath339 furthermore : @xmath340 all of these concepts and formulas have analogues as `` typical projectors '' @xmath341 for quantum state : by virtue of the spectral decomposition , the eigenvalues of a density operator can be interpreted as a probability distribution over eigenstates . the subspaces spanned by the typical eigenstates are the `` typical subspaces '' . the trace of a density operator with one of its typical projectors is then the probability of the corresponding set of typical sequences . notations like @xmath342 etc . should be clear from this . there is only one such statement for density operators that we shall use , which is not of this form : [ lemma : law : large - numbers ] let @xmath334 be of type @xmath7 , and let @xmath343 be a cq channel . denote the average output state of @xmath336 under @xmath7 as @xmath344 then , for every @xmath327 and sufficiently large @xmath2 , @xmath345 see @xcite , lemma 6 . this appendix collects some standard facts about various functionals we use : entropy , fidelity and trace norm . [ lemma : fannes ] let @xmath8 and @xmath307 be states on a @xmath346dimensional hilbert space , with @xmath347 . then @xmath348 , with @xmath349 note that @xmath350 is a monotone and concave function . [ lemma : norm - fidelity ] let @xmath8 and @xmath307 be any two states on a hilbert space . then @xmath351 @xmath130 [ lemma : gentle ] let @xmath8 be a ( subnormalized ) density operator , i.e. @xmath352 and @xmath353 , and let @xmath354 . then , if @xmath355 , @xmath356 @xmath130 [ lemma : caratheodory ] let @xmath357 be points in a @xmath346dimensional @xmath358vector space , and let @xmath359 be probabilities ( i.e. , non negative and summing to @xmath98 . then the convex combination @xmath360 can be expressed as a convex combination of ( at most ) @xmath361 of the @xmath362 . as a consequence , there exist probability distributions @xmath363 on @xmath364 and probability weights @xmath365 such that for all @xmath366 , @xmath367 @xmath130 the proof follows closely the argument of @xcite and of @xcite : begin by constructing the typical projectors @xmath368 of the @xmath369 , which , for @xmath64 of type @xmath7 , is defined as the sum of the eigenstate projectors of @xmath369 with eigenvalues in the interval @xmath370,\ ] ] with the conditional entropy @xmath338 . for sufficiently large @xmath2 , by the law of large numbers , @xmath371 . now define @xmath372 where @xmath341 is the typical projector of @xmath373 , i.e. the sum of the eigenstate projectors with eigenvalues in the interval @xmath374.\ ] ] these concepts are taken from @xcite and @xcite , but see also appendix [ app : types ] . by the law of large numbers , for sufficiently large @xmath2 , @xmath375 from this and the gentle measurement lemma [ lemma : gentle ] , we get @xmath376 the strategy is now to apply the operator chernoff bound to the @xmath377 : they are supported on a subspace of dimension @xmath378 , and are all upper bounded by @xmath379 . the only remaining obstacle is that we need a lower bound on @xmath380 to this end , let @xmath381 be the projector onto the subspace spanned by eigenvectors of @xmath382 with eigenvalues @xmath383 . in this way , for sufficiently large @xmath2 , @xmath384 defining the operators @xmath385 we can now apply lemma [ lemma : op : chernoff ] to the ( rescaled ) @xmath386 , and get @xmath387 \right\ } \\ & \phantom { : } \leq 2d^n\exp\left ( -m\exp\bigl(-n(i(p;w)+3\delta)\bigr)\frac{\epsilon^2}{2\ln 2 } \right)\!. \end{split}\ ] ] but @xmath388\ ] ] implies @xmath389 , which in turn implies @xmath390 in particular we get , invoking eq . ( [ eq : truncated ] ) , @xmath391 hence , with the gentle measurement lemma [ lemma : gentle ] appendix [ app : facts ] , we obtain @xmath392 combining eqs . ( [ eq : step - one ] ) and ( [ eq : step - two ] ) via the triangle inequality gives @xmath393 and using eq . ( [ eq : truncated ] ) to replace @xmath377 by @xmath394 in both above operators , we get finally @xmath395 the complement of this event has probability smaller than eq . ( [ eq : there : it : is ] ) , and since @xmath161 was arbitrary we obtain our claim . in @xcite , it is proved that selecting @xmath396 codewords @xmath397 at random , i.i.d . according to @xmath398 , one can construct a canonical decoding povm such that for the expectation ( over the code @xmath399 ) of the average error probability , @xmath400 , goes to zero : @xmath401 ( see @xcite , eq . the first thing we note is that ( for sufficiently large @xmath2 ) @xmath402 for a constant @xmath403 depending on @xmath161 : this follows by inspection of section iii of @xcite , where @xmath76 is introduced as the loss of probability mass by removing non typical contributions . as non typicality is defined as large deviation events for a sum of independent random variables , of the form @xmath404 the chernoff bound allows us to put exponential bounds on the non typical mass . we want to show that @xmath407 is a good approximation to a random code from the type class @xmath134 . of course , it is not quite that , if only because it has a variable number of codewords ! there is an easy fix to this problem : define , with @xmath408 , @xmath409 which makes sense because we can put the codewords in the order we select them . if the intersection is too small , define @xmath410 to be empty . now , that @xmath412 , happens extremely rarely : because @xmath413 , the expected cardinality of the intersection is larger than @xmath414 , for sufficiently large @xmath2 . but then , using the chernoff bound , @xmath415 by symmetry , it is clear that conditional on @xmath416 , the code @xmath410 is a uniformly random code of @xmath417 words from @xmath134 , i.e. it can be described by i.i.d . and uniformly picking codewords . hence , denoting by @xmath418 a truly random code of @xmath417 words from @xmath134 , we have @xmath419 observe that the left hand side is the total variational distance of distributions . thus , putting this together with eqs.([eq : participation ] ) and ( [ eq : exp - error ] ) , we obtain @xmath420 for sufficiently large @xmath2 . but this in turn implies that @xmath421 by the markov inequality . observing @xmath423 , \end{split}\ ] ] we aim at writing the four conditional entropies on the right as averages over similar such quantities but with limited range of @xmath49 and @xmath54 . to this end , observe that the channels @xmath67 and @xmath31 induce a probability distribution @xmath424 on the values @xmath425 of @xmath426 ( of which @xmath54 clearly is a deterministic function ) , and that for each @xmath425 there is the conditional distribution @xmath427 on @xmath321 : @xmath428 which has the property @xmath429 . with these notations , @xmath430 and similarly for @xmath431 . for each @xmath225 , let @xmath432 , which allows us to write down the conditional distribution @xmath433 on the points @xmath425 : @xmath434 with this , the conditional distribution @xmath435 on @xmath321 can be written @xmath436 for which clearly @xmath437 . this allows us to write @xmath438 and similarly for @xmath61 . now , invoking caratheodory s theorem , lemma [ lemma : caratheodory ] , we can write , for all @xmath225 , @xmath439 with probabilities @xmath440 and conditional distributions @xmath441 such that for all @xmath366 , @xmath442 and @xmath443 . another application of caratheodory s theorem gives a convex decomposition @xmath444 such that the support of all the @xmath445 has cardinality @xmath446 and for all @xmath447 , @xmath448 eqs . ( [ eq : q - cond : decomp ] ) and ( [ eq : q : decomp ] ) define random variables @xmath449 and @xmath450 , respectively : by eqs . ( [ eq : pt : preserved ] ) and ( [ eq : p : preserved ] ) , for each value @xmath451 the conditional distribution of @xmath54 and @xmath452 define variables @xmath453 , and so @xmath454 . \end{split}\ ] ] in the same way , @xmath455 . \end{split}\ ] ] hence there exist @xmath366 and @xmath447 such that @xmath456 and @xmath457 and @xmath458 satisfy the range constraints . denote the povm elements of the measurement producing @xmath460 and @xmath225 by @xmath461 , and introduce the coarse grained operators @xmath462 . to decompose the povm using convexity arguments , we rewrite the completeness conditions as @xmath463 using caratheodory s theorem , lemma [ lemma : caratheodory ] , we can write @xmath464 with distributions @xmath445 of support @xmath465 and such that for all @xmath447 , @xmath466 using caratheodory s theorem once more , we obtain , for each @xmath225 , a decomposition @xmath467 with conditional distributions @xmath468 of support @xmath465 and such that for all @xmath366 , @xmath469 now , let @xmath470 , which @xmath54 clearly is a function of . then , eqs . ( [ eq : op : q : decomp ] ) and ( [ eq : op : q - cond : decomp ] ) define random variables @xmath449 and @xmath450 , respectively : by eqs . ( [ eq : op : mm : preserved ] ) and ( [ eq : op : pi : preserved ] ) , for each value @xmath451 we have a povm @xmath471 ( whose output variable we denote @xmath472 the function @xmath54 of which we denote @xmath458 ) . then ( compare the previous proof ) , @xmath473 . \\ -&h(b|xt ) + h(e|xt ) = - h(b|\widetilde{x}jk ) + h(e|\widetilde{x}jk ) \\ & \phantom{=== } = \sum_{jk } \pr\{jk = jk\ } \bigl [ -h(b|\widetilde{x}_{jk})+h(e|\widetilde{x}_{jk } ) \bigr ] . \end{aligned}\ ] ] hence there exist @xmath366 and @xmath447 such that @xmath474 and @xmath472 and @xmath458 satisfy the range constraints . c. h. bennett , g. brassard , c. crpeau , r. jozsa , a. peres , w. k. wootters , `` teleporting an unknown quantum state via dual classical and einstein podolsky rosen channels '' , phys . letters , vol . 70 , no . 13 , pp . 18951899 , 1993 . d. bru , m. christandl , a. ekert , b.g . englert , d. kaszlikowski , c. macchiavello , `` tomographic quantum cryptography : equivalence of quantum and classical key distillation '' , e print quant - ph/0303184 , 2003 . d. p. divincenzo , p. w. shor , j. a. smolin , `` quantum channel capacity of very noisy channels '' , phys . rev . a , vol . 2 , pp . 830839 , 1998 . erratum : phys . rev . a , vol . 2 , p. 1717 , 1999 . see also p. w. shor , j. a. smolin , `` quantum error correcting codes need not completely reveal the error syndrome '' , e print quant - ph/9604006 , 1996 . a. s. holevo , `` some estimates for the amount of information transmittable by a quantum communications channel '' ( russian ) , probl . 3 , pp . 311 , 1973 . english translation : probl . , vol . 9 , no . 3 , pp . 177183 , 1973 . e. m. rains , `` rigorous treatment of distillable entanglement '' , phys . . a , vol . 1 , pp . 173178 , 1999 . e. m. rains , `` bound on distillable entanglement '' , phys . . a , vol . 1 , pp . 179179 , 1999 ; erratum : phys . 63 , 019902(e ) , 2001 . b. schumacher , `` quantum coding '' , phys . rev . a , vol . 4 , pp . 27382747 , 1995 . r. jozsa , b. schumacher , `` a new proof of the quantum noiseless coding theorem '' , j. mod . optics , vol . 12 , pp . 23432349 , 1994 . b. schumacher , `` sending entanglement through noisy quantum channels '' , phys . rev . a , vol . 4 , pp . 26142628 , 1996 . b. schumacher , m. a. nielsen , , `` quantum data processing and error correction '' , phys . 4 , pp . 26292635 , 1996 . h. barnum , m. a. nielsen , b. schumacher , `` information transmission through a noisy quantum channel '' , phys . rev . a , vol . 57 , no . 6 , pp . 41534175 , 1998 . p. w. shor , `` the quantum channel capacity and coherent information '' , in preparation . lecture notes and video ( realplayer ) available at http://www.msri.org/publications/ln/msri/2002/ quantumcrypto / shor/1/index.html .

we generalize the unique decoding algorithm for one - point ag codes over the miura - kamiya curves proposed by to general one - point ag codes , without any assumption . we also extend their unique decoding algorithm to list decoding , modify it so that it can be used with the feng - rao improved code construction , prove equality between its error correcting capability and half the minimum distance lower bound by that has not been done in the original proposal except for one - point hermitian codes , remove the unnecessary computational steps so that it can run faster , and analyze its computational complexity in terms of multiplications and divisions in the finite field . as a unique decoding algorithm , the proposed one is as fast as the bms algorithm for one - point hermitian codes , and as a list decoding algorithm it is much faster than the algorithm by . + * keywords : * algebraic geometry code , grbner basis , list decoding + * msc 2010 : * primary : 94b35 ; secondary : 13p10 , 94b27 , 14g50

1203.6127

we consider the list decoding of one - point algebraic geometry ( ag ) codes . @xcite proposed the well - known list decoding algorithm for one - point ag codes , which consists of the interpolation step and the factorization step . the interpolation step has large computational complexity and many researchers have proposed faster interpolation steps , see ( * ? ? ? * figure 1 ) . by modifying the unique decoding algorithm @xcite for primal one - point ag codes , we propose another list decoding algorithm based on voting in grbner bases whose error correcting capability is higher than @xcite and whose computational complexity is smaller than @xcite in many cases . a decoding algorithm for primal one - point ag codes was proposed in @xcite , which was a straightforward adaptation of the original feng - rao majority voting for the dual ag codes @xcite to the primal ones . the feng - rao majority voting in @xcite for one - point primal codes was generalized to multi - point primal codes in ( * ? ? ? * section 2.5 ) . the one - point primal codes can also be decoded as multi - point dual codes with majority voting @xcite , whose faster version was proposed in @xcite for the multi - point hermitian codes . @xcite proposed another unique decoding ( not list decoding ) algorithm for primal codes based on the majority voting inside grbner bases . the module used by them @xcite is a curve theoretic generalization of one used for reed - solomon codes in @xcite that is a special case of the module used in @xcite . an interesting feature in @xcite is that it did not use differentials and residues on curves for its majority voting , while they were used in @xcite . the above studies @xcite dealt with the primal codes . we recently proved in @xcite that the error - correcting capabilities of @xcite are the same . the earlier papers @xcite suggest that central observations in @xcite were known to the dutch group , which is actually the case @xcite . @xcite , @xcite and @xcite studied the error - correcting capability of the feng - rao @xcite or the bms algorithm @xcite with majority voting beyond half the designed distance that are applicable to the dual one - point codes . there was room for improvements in the original result @xcite , namely , ( a ) they have not clarified the relation between its error - correcting capability and existing minimum distance lower bounds except for the one - point hermitian codes , ( b ) they have not analyzed the computational complexity , ( c ) they assumed that the maximum pole order used for code construction is less than the code length , and ( d ) they have not shown how to use the method with the feng - rao improved code construction @xcite . we shall ( 1 ) prove that the error - correcting capability of the original proposal is always equal to half of the bound in @xcite for the minimum distance of one - point primal codes ( proposition [ prop : ag ] ) , ( 2 ) generalize their algorithm to work with any one - point ag codes , ( 3 ) modify their algorithm to a list decoding algorithm , ( 4 ) remove the assumptions ( c ) and ( d ) above , ( 5 ) remove unnecessary computational steps from the original proposal , ( 6 ) analyze the computational complexity in terms of the number of multiplications and divisions in the finite field . the proposed algorithm is implemented on the singular computer algebra system @xcite , and we verified that the proposed algorithm can correct more errors than @xcite with manageable computational complexity . this paper is organized as follows : section [ sec2 ] introduces notations and relevant facts . section [ sec3 ] improves @xcite in various ways , and the differences to the original @xcite are summarized in section [ sec : diff ] . section [ sec4 ] shows that the proposed modification to @xcite works as claimed . section [ sec : experiment ] compares its computational complexity with the conventional methods . section [ sec6 ] concludes the paper . part of this paper was presented at 2012 ieee international symposium on information theory , cambridge , ma , usa , july 2012 @xcite . our study heavily relies on the standard form of algebraic curves introduced independently by @xcite and @xcite , which is an enhancement of earlier results @xcite . let @xmath1 be an algebraic function field of one variable over a finite field @xmath2 with @xmath3 elements . let @xmath4 be the genus of @xmath5 . fix @xmath6 distinct places @xmath7 , @xmath8 , , @xmath9 of degree one in @xmath5 and a nonnegative integer @xmath10 . we consider the following one - point algebraic geometry ( ag ) code @xmath11 where @xmath12 , , @xmath13 . suppose that the weierstrass semigroup @xmath14 at @xmath7 is generated by @xmath15 , , @xmath16 , and choose @xmath17 elements @xmath18 , , @xmath19 in @xmath5 whose pole divisors are @xmath20 for @xmath21 , , @xmath17 . we do _ not _ assume that @xmath15 is the smallest among @xmath15 , , @xmath16 . without loss of generality we may assume the availability of such @xmath18 , , @xmath19 , because otherwise we can not find a basis of @xmath22 for every @xmath10 . then we have that @xmath23 is equal to @xmath24 , , @xmath25 $ ] @xcite . we express @xmath26 as a residue class ring @xmath27 , , @xmath28/i$ ] of the polynomial ring @xmath27 , , @xmath28 $ ] , where @xmath29 , , @xmath30 are transcendental over @xmath2 , and @xmath31 is the kernel of the canonical homomorphism sending @xmath32 to @xmath33 . @xcite identified the following convenient representation of @xmath26 by using grbner basis theory @xcite . the following review is borrowed from @xcite . hereafter , we assume that the reader is familiar with the grbner basis theory in @xcite . let @xmath34 be the set of nonnegative integers . for @xmath35 , , @xmath36 , @xmath37 , , @xmath38 , we define the weighted reverse lexicographic monomial order @xmath39 such that @xmath35 , , @xmath36 @xmath39 @xmath37 , , @xmath40 if @xmath41 , or @xmath42 , and @xmath43 , @xmath44 , , @xmath45 , @xmath46 , for some @xmath47 . note that a grbner basis of @xmath31 with respect to @xmath39 can be computed by ( * ? ? ? * theorem 15 ) , @xcite , ( * ? ? ? * theorem 4.1 ) or ( * ? ? ? * proposition 2.17 ) , starting from any affine defining equations of @xmath1 . [ ex1 ] according to ( * ? ? ? * example 3.7 ) , @xmath48 is an affine defining equation for the klein quartic over @xmath49 . there exists a unique @xmath49-rational place @xmath7 such that @xmath50 , @xmath51 , and @xmath52 . the numbers @xmath53 , @xmath54 and @xmath55 is the minimal generating set of the weierstrass semigroup at @xmath7 . choosing @xmath56 as @xmath18 , @xmath57 as @xmath58 and @xmath59 as @xmath60 , by ( * ? ? ? * theorem 4.1 ) we can see that the standard form of the klein quartic is given by @xmath61 which is the reduced grbner basis with respect to the monomial order @xmath39 . we can see that @xmath62 , @xmath63 , and @xmath64 . [ ex : gs1 ] consider the function field @xmath65 with relations @xmath66 this is the third function field in the asymptotically good tower introduced by @xcite . substituting @xmath67 with @xmath68 and @xmath69 with @xmath70 in eq . ( [ eq : gs1 ] ) we have affine defining equations @xmath71 the function @xmath72 has a unique pole @xmath7 in @xmath73 . the minimal generating set of the weiestrass semigroup @xmath14 at @xmath7 is @xmath74 , @xmath75 , @xmath76 , @xmath77 , @xmath78 and @xmath79 ( * ? ? ? * example 4.11 ) . it has genus @xmath76 and @xmath80 @xmath81-rational points different from @xmath7 @xcite . define six functions @xmath82 , @xmath83 , @xmath84 , @xmath85 , @xmath86 and @xmath87 . we have @xmath88 , @xmath89 , @xmath90 , @xmath91 , @xmath92 and @xmath93 @xcite . from this information and ( * theorem 4.1 ) we can compute the @xmath94 polynomials in the reduced grbner basis of the ideal @xmath95 , , @xmath96 $ ] defining @xmath26 as @xmath97 , @xmath98 , @xmath99 , @xmath100 , @xmath101 , @xmath102 , @xmath103 , @xmath104 , @xmath105 , @xmath106 , @xmath107 , @xmath108 , @xmath109 , @xmath110 , @xmath111 . note that polynomials in the above grbner basis are in the ascending order with respect to the monomial order @xmath112 while terms in each polynomial are in the descending order with respect to @xmath112 . for @xmath113 , , @xmath114 , we define @xmath115 , and @xmath116 to be the minimum element @xmath35 , , @xmath117 with respect to @xmath112 such that @xmath118 . note that @xmath119 s are the well - known apry set ( * ? ? ? * lemmas 2.4 and 2.6 ) of the numerical semigroup @xmath14 . then we have @xmath120 if we write @xmath116 as @xmath121 , , @xmath122 . for each @xmath123 , @xmath124 , , @xmath125 , define @xmath126 . the footprint of @xmath31 , denoted by @xmath127 , is @xmath128 , , @xmath129 is not the leading monomial of any nonzero polynomial in @xmath31 with respect to @xmath130 , and define @xmath131 , , @xmath132 . then @xmath133 is a basis of @xmath26 as an @xmath2-linear space @xcite , two distinct elements in @xmath133 have different pole orders at @xmath7 , and @xmath134 equation ( [ eq : footprintform ] ) shows that @xmath26 is a free @xmath135$]-module with a basis @xmath136 , , note that the above structured shape of @xmath133 reflects the well - known property of every weighted reverse lexicographic monomial order , see the paragraph preceding to ( * ? ? ? * proposition 15.12 ) . [ ex2 ] for the curve in example [ ex1 ] , we have @xmath138 , @xmath139 , @xmath140 . let @xmath141 be the unique valuation in @xmath5 associated with the place @xmath7 . the semigroup @xmath14 is equal to @xmath142 ( * ? ? ? * lemma 2.6 ) . by ( * proposition 3.18 ) , for each nongap @xmath143 there is a unique monomial @xmath144 with @xmath145 such that @xmath146 , and let us denote this monomial by @xmath147 . let @xmath148 , and we may consider the one - point codes @xmath149 where @xmath150 denotes the @xmath2-linear space spanned by @xmath151 . since considering linearly dependent rows in a generator matrix has no merit , we assume @xmath152 where @xmath153 . one motivation for considering these codes is that it was shown in @xcite how to increase the dimension of the one - point codes without decreasing the lower bound @xmath154 for the minimum distance . the bound @xmath155 is defined for @xmath156 as follows @xcite : for @xmath157 , let @xmath158 then @xmath159 . it is proved in @xcite that @xmath154 gives the same estimate for the minimum distance as the feng - rao bound @xcite for one - point dual ag codes when both @xmath154 and the feng - rao bound can be applied , that is , when the dual of a one - point code is isometric to a one - point code . furthermore , it is also proved in @xcite that @xmath155 can be obtained from the bounds in @xcite , hence @xmath154 can be understood as a particular case of these bounds @xcite . suppose that we have a received word @xmath160 . we shall modify the unique decoding algorithm proposed by @xcite so that we can find all the codewords in @xmath156 in eq . ( [ eq : cgamma ] ) within the hamming distance @xmath161 from @xmath162 . the overall structure of the modified algorithm is as follows : 1 . [ l1 ] precomputation before getting a received word @xmath162 , 2 . [ l2 ] initialization after getting a received word @xmath162 , 3 . [ l3 ] termination criteria of the iteration , and 4 . [ l4 ] main part of the iteration . steps [ l2 ] and [ l4 ] are based on @xcite . steps [ l1 ] and [ l3 ] are not given in @xcite . each step is described in the following subsections in section [ sec3 ] . we shall analyze time complexity except the precomputation part of the algorithm . we retain notations from section [ sec2 ] . in this subsection , we modify notations and definitions in @xcite to describe the proposed modification to their algorithm . we also introduce several new notations . define a set @xmath163 , @xmath164 , @xmath165 . our @xmath166 is @xmath167 in @xcite . recall also that @xmath168 . since the @xmath135$]-module @xmath169 has a free basis @xmath170 , we can regard @xmath166 as the set of monomials in the grbner basis theory for modules . we introduce a monomial order on @xmath166 as follows . for given two monomials @xmath171 and @xmath172 , first rewrite @xmath173 and @xmath174 by @xmath58 , , @xmath19 defined in section [ sec2 ] and get @xmath175 and @xmath176 . for a nongap @xmath177 , we define the monomial order @xmath178 parametrized by @xmath179 if @xmath180 or @xmath181 and @xmath182 , @xmath183 , , @xmath184 and @xmath185 for some @xmath186 . observe that the restriction of @xmath187 to @xmath133 is equal to @xmath112 defined in section [ sec2 ] . in what follows , every grbner basis , leading term , and leading coefficient is obtained by considering the grbner basis theory for modules , not for ideals . for @xmath188 , @xmath189 denotes the number of nonzero terms in @xmath190 when @xmath190 is expressed as an @xmath2-linear combination of monomials in @xmath166 . @xmath191 denotes the number of nonzero terms whose coefficients are not @xmath192 . for the code @xmath156 in eq . ( [ eq : cgamma ] ) , define the divisor @xmath193 . define @xmath194 for a positive divisor @xmath195 of @xmath1 . then @xmath196 is an ideal of @xmath197 @xcite . let @xmath198 be any element in @xmath196 such that @xmath199 with @xmath200 being the minimal given @xmath201 . then by ( * ? ? ? * proposition 1 ) , @xmath202 , , @xmath203 is a grbner basis for @xmath196 with respect to @xmath187 as an @xmath135$]-module . for a nonnegative integer @xmath179 , define @xmath204 , @xmath205 , and @xmath206 . we define @xmath207 . before getting @xmath162 , we need to compute the pellikaan - miura standard form of the algebraic curve , @xmath208 , @xmath209 , , @xmath210 , and @xmath147 for @xmath177 as defined in section [ sec2 ] . also compute @xmath211 , , @xmath212 , which can be done by @xcite . for each @xmath213 , express @xmath214 as an @xmath2-linear combination of monomials in @xmath133 . such expressions will be used for computing products and quotients in @xmath197 as explained in section [ sec : product ] . from the above data , we can easily know @xmath215 , which will be used in eqs . ( [ eq : quotientpre ] ) and ( [ eq : mui ] ) . find elements @xmath216 with @xmath217 . there are @xmath218 such elements , which we denote by @xmath219 , @xmath220 such that @xmath221 . compute the @xmath222 matrix @xmath223 in both original unique decoding algorithm @xcite and our modified version , we need to quickly compute the product @xmath224 of two elements @xmath225 in the affine coordinate ring @xmath197 . in our modified version , we also need to compute the quotient @xmath226 depending on the choice of iteration termination criterion described in section [ sec : termination ] . since the authors could not find quick computational procedures for those tasks in @xmath197 , we shall present such ones here . the normal form of @xmath4 , for @xmath227 , is the expression of @xmath4 written as an @xmath2-linear combination of monomials @xmath216 . @xmath225 are assumed to be in the normal form . we propose the following procedure to compute the normal form of @xmath224 . let the normal form of @xmath214 be @xmath228 with @xmath229 $ ] , which is computed in section [ sec : precomput ] . we denote by @xmath29 , @xmath230 , , @xmath231 algebraically independent variables over @xmath2 . 1 . assume that @xmath4 and @xmath232 are in their normal forms . change @xmath233 to @xmath234 and @xmath18 to @xmath29 in @xmath225 for @xmath21 , , @xmath114 . recall that @xmath235 . denote the results by @xmath236 . 2 . compute @xmath237 . this step needs @xmath238 multiplications in @xmath2 . 3 . let @xmath239 . then we have @xmath240 computation of @xmath241 needs at most @xmath242 multiplications in @xmath2 . therefore , the total number of multiplications in @xmath2 in this step is at most @xmath243 therefore , the total number of multiplications in @xmath2 is at most @xmath244 define eq . ( [ eq : nfmulti3 ] ) as @xmath245 . we emphasize that when the characteristic of @xmath2 is 2 and all the coefficients of defining equations belong to @xmath246 , which is almost always the case for those cases of interest for applications in coding theory , then @xmath247 in eq . ( [ eq : nfmulti3 ] ) is zero . this means that @xmath26 has little additional overhead over @xmath248 $ ] for computing products of their elements in terms of the number of @xmath2-multiplications and divisions . define @xmath213 to be equivalent to @xmath249 if @xmath250 . denote by @xmath251 $ ] the equivalence class represented by @xmath213 . for @xmath252 $ ] we have @xmath253 , which is denoted by @xmath254,k}(x_1)$ ] . the right hand side of eq . ( [ eq : gh ] ) can be written as @xmath255}\left(\sum_{(i',j')\in[i , j ] } f_{g , h , i',j'}(x_1)\right ) \sum_{k=0}^{a_1 - 1 } y_k f_{[i , j],k}(x_1 ) . \label{eq : gh2}\ ] ] by using eq . ( [ eq : gh2 ] ) instead of eq . ( [ eq : gh ] ) , we have another upper bound on the number of multiplications as @xmath256 } \gamma\left ( \sum_{(i',j')\in[i , j ] } f_{g , h , i',j'}(x_1)\right ) \gamma_{\neq 1}(\sum_{k=0}^{a_1 - 1 } y_k f_{[i , j],k}(x_1 ) ) . \label{eq : nfmulti4}\ ] ] since @xmath257 } f_{g , h , i',j'}(x_1)\right ) \leq \sum_{(i',j')\in[i , j ] } \gamma ( f_{g , h , i',j'}(x_1)),\ ] ] we have eq . ( [ eq : nfmulti4 ] ) @xmath258 eq . ( [ eq : nfmulti3 ] ) . however , eq . ( [ eq : nfmulti4 ] ) is almost always the same as eq . ( [ eq : nfmulti3 ] ) over the curve in example [ ex : gs1 ] , and eq . ( [ eq : nfmulti4 ] ) will not be used in our computer experiments in section [ sec : experiment ] . assume @xmath259 . the following procedure computes the quotient @xmath260 or declares that @xmath4 does not belong to the principal ideal of @xmath197 generated by @xmath232 . 1 . initialize @xmath261 . also initialize @xmath262 . 2 . [ lab ] check if @xmath263 . if not , declare that @xmath4 does not belong to the principal ideal of @xmath197 generated by @xmath232 , and finish the procedure . 3 . let @xmath216 such that @xmath264 . observe that @xmath265 and that @xmath266 is precomputed as section [ sec : precomput ] . let @xmath267 computation of @xmath268 needs one multiplication and one division in @xmath2 . observe that @xmath269 . 4 . compute the normal form of @xmath270 , which requires at most @xmath271 multiplications in @xmath2 . increment @xmath272 by @xmath273 . 5 . update @xmath274 and @xmath275 . if the updated @xmath4 is zero , then output the updated @xmath276 as the quotient and finish the procedure . otherwise go to step [ lab ] . this step has no multiplication nor division . define @xmath277 as @xmath272 after finishing the above procedure . @xmath277 is an upper bound on the number of multiplications and divisions in @xmath2 in the above procedure . the program variable @xmath272 is just to define @xmath277 , and the decoding algorithm does not need to update @xmath272 . observe also that the above procedure is a straightforward generalization of the standard long division of two univariate polynomials @xcite . let @xmath278 , , @xmath279 , where @xmath280 is defined in eq . ( [ eq : invgenmatrix ] ) . . then we have @xmath282 . the computation of @xmath283 from @xmath162 needs at most @xmath284 multiplications in @xmath2 . let @xmath285 . for @xmath113 , , @xmath114 , compute @xmath286 and @xmath287 . the computation of @xmath288 needs at most @xmath289 multiplications in @xmath2 . therefore , the total number of multiplications in the initialization is at most @xmath290 let @xmath291 and execute the following steps . after finishing the initialization step in section [ sec : initialization ] , we iteratively compute @xmath292 and @xmath293 with @xmath294 and @xmath295 with @xmath296 from larger @xmath179 to smaller @xmath179 . the single iteration consists of two parts : the first part is to check if an iteration termination criterion is satisfied . the second part is computation of @xmath292 and @xmath293 for @xmath297 . we describe the first part in section [ sec : termination ] . let @xmath298 having the smallest @xmath299 among @xmath300 , , @xmath301 . in the following subsections , we shall propose three different procedures to judge whether or not iterations in the proposed algorithm can be terminated . in an actual implementation of the proposed algorithm , one criterion is chosen and the chosen one is consistently used throughout the iterations . the first one and the second one are different generalizations of ( * ? ? ? * theorem 12 ) for the case @xmath302 to @xmath303 . * theorem 12 ) proved that if the number @xmath304 of errors satisfies @xmath305 , where @xmath306 is the minimum distance @xmath307 of the @xmath308 $ ] reed - solomon code @xmath309 , then the transmitted information word is obtained by ali - kuijper s algorithm as @xmath310 . to one - point primal ag codes , @xmath306 can be generalized as either @xmath311 or @xmath312 . the former generalization @xmath311 corresponds to the first criterion in section [ sec : first ] and the latter @xmath312 corresponds to the second in section [ sec : second ] . the third one is almost the same as the original procedure in @xcite . the first one was proposed in @xcite while the second and the third ones are new in this paper . we shall compare the three criteria in section [ sec : comparethree ] . throughout this paper , @xmath313 denotes the hamming weight of a vector @xmath314 . if * @xmath157 , * @xmath315 , and * @xmath316 then do the following : 1 . [ first1 ] compute @xmath317 . this needs at most @xmath318 multiplications and divisions in @xmath2 . 2 . if @xmath319 and @xmath320 can be written as a linear combination of monomials in @xmath321 , then do the following : 1 . [ first11 ] if @xmath322 or @xmath323 then include the coefficients of @xmath324 into the list of transmitted information vectors , and avoid proceeding with the rest of the decoding procedure . [ first12 ] otherwise compute @xmath325 . this needs at most @xmath326 multiplications and divisions in @xmath2 . 3 . [ first13 ] if @xmath327 then include the coefficients of @xmath310 @xmath328 @xmath329 into the list of transmitted information vectors , and avoid proceeding with @xmath179 . otherwise , continue the iterations unless @xmath330 . if @xmath331 , then do the following : 1 . [ second1 ] if @xmath332 then stop proceeding with iteration . 2 . otherwise compute @xmath317 . this needs at most @xmath333 multiplications and divisions in @xmath2 . if @xmath334 , @xmath319 , and @xmath320 can be written as a linear combination of monomials in @xmath321 then declare the coefficients of @xmath324 as the only transmitted information and finish . otherwise declare `` decoding failure '' and finish . 4 . if @xmath335 , @xmath319 and @xmath320 can be written as a linear combination of monomials in @xmath321 , then do the following : 1 . [ second11 ] if @xmath323 then include the coefficients of @xmath324 into the list of transmitted information vectors , and avoid proceeding with @xmath179 . [ second12 ] otherwise compute @xmath325 . this needs at most @xmath336 multiplications and divisions in @xmath2 . 3 . if @xmath327 then include the coefficients of @xmath310 @xmath328 @xmath329 into the list of transmitted information vectors . 5 . finish the iteration no matter what happened in the above steps . just repeat the iteration until finding @xmath292 at @xmath337 for @xmath113 , , @xmath114 . if @xmath334 then declare the vector @xmath338 as the only transmitted information and finish . if @xmath335 then do the following : 1 . [ third1 ] if @xmath339 and @xmath323 then include the vector @xmath338 into the list of transmitted information vectors . finish the iteration . [ third2 ] if @xmath340 then finish the iteration . [ third3 ] otherwise compute @xmath341 . this needs at most @xmath342 multiplications and divisions in @xmath2 . [ third4 ] if @xmath343 then include the vector @xmath338 into the list of transmitted information vectors . finish the iteration . the iteration of the original algorithm @xcite consists of three steps , called pairing , voting , and rebasing . we will make a small change to the original . our modified version is described below . let @xmath344,\ ] ] @xmath345,\ ] ] and let @xmath346 . we assume that @xmath347 and @xmath348 . for @xmath349 , as in @xcite there are unique integers @xmath350 and @xmath351 satisfying @xmath352 note that by the definition above @xmath353 and the integer @xmath354 is a nongap if and only if @xmath355 . now let @xmath356 . note that the map @xmath357 is a permutation of @xmath358 and that the integer @xmath359 is defined such that @xmath360 . for each @xmath361 , we set @xmath362}{\mu_i } , ~ \bar{c}_i=\max\{c_i,0\ } , \label{eq : mui}\ ] ] where @xmath363 $ ] denotes the coefficient of @xmath364 of the univariate polynomial @xmath365 $ ] . we remark that the leading coefficient @xmath366 must be considered after expressing @xmath367 by monomials in @xmath133 . observe that @xmath368 and that @xmath369 is already precomputed as section [ sec : precomput ] . by using that precomputed table , computation of @xmath366 needs one multiplication . the total number of multiplications and divisions in eq . ( [ eq : mui ] ) is @xmath370 excluding negation from the number of multiplication . let @xmath371 we consider two different candidates depending on whether @xmath157 or not : * if @xmath372 , set @xmath373 * if @xmath157 , let @xmath374 be one of the element(s ) in @xmath2 with @xmath375 let @xmath376 . if several @xmath374 s satisfy the condition above , repeat the rest of the algorithm for each of them . in all of the following cases , we need to compute the normal form of the product @xmath377 , and the product @xmath378 . for each @xmath201 , the number of multiplications is @xmath379 where @xmath380 is defined in section [ sec : product ] . * if @xmath381 , then let @xmath382where the parentheses denote substitution of the variable @xmath383 and let @xmath384 . the number of multiplications in this case is bounded by eq . ( [ eq : multirebasing ] ) . * if @xmath385 and @xmath386 , then let @xmath387 and let @xmath388 . + computation of @xmath389 needs one multiplication and one division . the product of @xmath389 and @xmath390 needs @xmath391 multiplications , where @xmath392 is defined in section [ sec : product ] . thus , the number of multiplications and divisions is @xmath393 * if @xmath385 and @xmath394 , then let @xmath395 and let @xmath396 . + computation of @xmath389 needs one multiplication and one division . the product of @xmath389 and @xmath390 needs @xmath391 multiplications , where @xmath392 is defined in section [ sec : product ] . thus , the number of multiplications and divisions is @xmath397 eq . ( [ eq : compl2 ] ) . after computing @xmath398 and @xmath399 as above , update the program variable @xmath179 to @xmath400 and go to the beginning of section [ sec : termination ] . in this subsection , we review advantages of our modified algorithm over the original @xcite . * our version can handle any one - point primal ag codes , while the original can handle codes only coming from the @xmath0 curves @xcite . this generalization is enabled only by replacing @xmath401 in @xcite by @xmath173 defined in section [ sec2 ] . * our version can find all the codeword within hamming distance @xmath161 from the received word @xmath162 , while the original is a unique decoding algorithm . * our version does not compute @xmath402 , @xmath403 for a weiestrass gap @xmath404 , while the original computes them for @xmath405 . * the original algorithm assumed @xmath406 , where @xmath10 is as defined in eq . ( [ eq : cu ] ) . this assumption is replaced by another less restrictive assumption ( [ eq : gammaassume ] ) in our version . * our version supports the feng - rao improved code construction @xcite , while the original does not . this extension is made possible by the change at eq . ( [ eq : w0 ] ) . * the first and the second termination criteria come from ( * ? ? ? * theorem 12 ) and do not exist in the original @xcite . * the third termination criterion is essentially the same as the original @xcite , but examination of the hamming distance between the decoded codeword and @xmath162 is added when @xmath407 . * the original @xcite is suitable for parallel implementation on electric circuit similar to the ktter architecture @xcite . our modified version retains this advantage . in this section we prove that our modified algorithm can find all the codewords within hamming distance @xmath161 from the received word @xmath162 . we also give upper bounds on the number of iterations in section [ sec : nriteration ] . in section [ sec : sl ] we shall introduce several lemmas necessary in sections [ sec : nrvotes][sec : firstwhy ] . recall that the execution of our modified algorithm can branch when there are multiple candidates satisfying the condition ( [ eq : acceptedvote ] ) . for a fixed sequence of determined @xmath295 , define @xmath408 and recursively define @xmath409 . by definition @xmath410 . the following lemma explains why the authors include `` grbner bases '' in the paper title . the module @xmath411 was used in @xcite but the use of @xmath412 with @xmath413 was new in @xcite . [ lem : ir ] fix @xmath414 . let @xmath415 correspond to @xmath295 ( @xmath157 ) chosen by the decoding algorithm . define the @xmath135$]-submodule @xmath412 of @xmath169 by @xmath416 where @xmath417 , , @xmath418 . then @xmath419 , @xmath420 , , @xmath421 is a grbner basis of @xmath412 with respect to @xmath187 as an @xmath135$]-module . * proof . * this lemma is a generalization of ( * ? ? ? * proposition 11 ) . we can prove this lemma in exactly the same way as the proof of ( * ? ? ? * proposition 11 ) with replacing @xmath401 in @xcite with @xmath173 and @xmath422 in @xcite by @xmath400 . the following proposition shows that the original decoding algorithm @xcite can correct errors up to half the bound @xmath155 , which was not claimed in @xcite . [ prop : ag ] fix @xmath157 . let @xmath423 as defined in eq . ( [ eq : lambda ] ) and @xmath424 as defined in eq . ( [ eq : nu ] ) . then @xmath425 . * * let @xmath426 , then we have @xmath427 . moreover , observe that @xmath428 therefore , for @xmath157 we have @xmath429 where the third equality holds by eq . ( [ eq : iprime ] ) . by the equalities above , we see @xmath430 which proves the equality @xmath431 . @xcite showed that their original decoding algorithm can correct up to @xmath432 errors , where @xmath433 , @xmath434 . proposition [ prop : ag ] implies that @xmath435 is equivalent to @xmath436 for every one - point primal code @xmath22 , and therefore ( * ? ? ? * theorem 8) implies ( * ? ? ? * proposition 12 ) . in another recent paper @xcite we proved that @xmath154 and @xmath437 are equal to the feng - rao bound as defined in @xcite for @xmath22 . in section [ sec : nrvotes ] we discuss the number of votes ( [ eq : acceptedvote ] ) which a candidate @xmath438 receives . since we study list decoding , we can not assume the original transmitted codeword nor the error vector as in @xcite . nevertheless , the original theorems in @xcite allow natural generalizations to the list decoding context . [ lem : nrvotes ] fix @xmath157 . for @xmath439 , fix a sequence of @xmath440 chosen by the decoding algorithm , and define @xmath415 corresponding to the chosen sequence of @xmath440 . fix @xmath441 . let @xmath442 , , @xmath443 be a nonzero vector with the minimum hamming weight in the coset @xmath444 , where @xmath445 is as defined in eq . ( [ eq : cu ] ) . define @xmath446 let @xmath447 , , @xmath448 be a grbner basis for @xmath449 as an @xmath135$]-module with respect to @xmath187 ( for any integer @xmath179 ) , such that @xmath450 . under the above notations , we have @xmath451 for @xmath201 with @xmath452 , and @xmath453 for @xmath201 with @xmath454 . * proof . * the proof is the same as those of ( * ? ? ? * propositions 7 and 8) , with replacing @xmath401 in @xcite by @xmath173 , @xmath455 in @xcite by @xmath456 . the following lemma is a modification to ( * ? ? ? * proposition 9 ) for the list decoding . we retain notations from lemma [ lem : nrvotes ] . we have @xmath457 * proof . * lemma [ lem : nrvotes ] implies @xmath458 and @xmath459 now we have a chain of inequalities @xmath460 where at eq . ( [ eq : this ] ) we used the equality @xmath461 shown in ( * ? ? ? * lemma 2 and eq . finally note that @xmath462 by ( * ? ? ? * eq . ( 3 ) ) . the following lemma is a modification to ( * ? ? ? * proposition 10 ) for list decoding , and provides a lower bound for the number of votes ( [ eq : acceptedvote ] ) received by any candidate @xmath441 , as indicated in the section title . [ prop : nrvotes ] we retain notations from lemma [ lem : nrvotes ] . let @xmath424 be as defined in eq . ( [ eq : nu ] ) . we have @xmath463 * proof . * we have @xmath464 as @xmath465 for @xmath466 . in this subsection and the following sections , we shall prove that the proposed list decoding algorithm will find all the codewords within the hamming distance @xmath161 from the received word @xmath162 . since the third iteration termination criterion is the easiest to analyze , we start with the third one . fix a sequence @xmath295 for @xmath467 . if @xmath468 then the sequence @xmath295 is found by the algorithm because of proposition [ prop : nrvotes ] . when @xmath334 , by proposition [ prop : ag ] the decoding is not list decoding , and the algorithm just declares the sequence @xmath295 as the transmitted information . on the other hand , if @xmath335 , then the found sequence could correspond to a codeword more distant than hamming distance @xmath161 , and the algorithm examines the hamming distance between the found codeword and the received word @xmath162 . since computing @xmath469 for @xmath470 needs many multiplications in @xmath2 , the algorithm checks some sufficient conditions to decide the hamming distance between the found codeword and the received word @xmath162 . let @xmath471 , , @xmath472 . when @xmath339 in section [ sec:3rd ] , by lemma [ lem : ir ] , we have @xmath473 because eq . ( [ eq : ir ] ) and @xmath339 implies that @xmath474 for @xmath475 . by the above equation , @xmath323 implies that the found codeword is within hamming distance @xmath161 from @xmath162 . this explains why the algorithm can avoid computation of the evaluation map @xmath476 in step [ third1 ] in section [ sec:3rd ] . in order to explain step [ third2 ] in section [ sec:3rd ] , we shall show that the condition of step [ third2 ] in section [ sec:3rd ] implies that @xmath477 . suppose that @xmath478 . then there exists @xmath479 such that @xmath480 for @xmath475 , @xmath481 , and @xmath482 . because the leading term of @xmath483 must be divisible by @xmath484 for some @xmath201 by the property of grbner bases , we must have @xmath485 . this explains why the algorithm can avoid computation of the evaluation map @xmath476 in step [ third2 ] in section [ sec:3rd ] . otherwise , the algorithm computes the hamming distance between the found codeword and @xmath162 in steps [ third3 ] and [ third4 ] in section [ sec:3rd ] . we shall explain why the second criterion in section [ sec : second ] correctly finds the required codewords . for explanation , we present slightly rephrased version of facts in @xcite . [ lem:23 ] ( * ? ? ? * lemma 2.3 ) let @xmath486 with @xmath487 with respect to @xmath187 and @xmath488 . if there exists @xmath489 such that @xmath490 , then we have @xmath491 . * proof . * observe that @xmath487 implies that @xmath492 . the claim of lemma [ lem:23 ] is equivalent to ( * ? ? ? * lemma 2.3 ) with @xmath493 and @xmath494 . note that the assumption @xmath495 was not used in ( * ? ? ? * lemma 2.3 ) but only in ( * ? ? ? * lemma 2.4 ) . note that the following proposition was essentially proved in ( * ? ? ? * proposition 2.10 ) , ( * ? ? ? * section 14.2 ) , and ( * ? ? ? * theorem 2.1 ) with @xmath496 . [ prop : candivide ] let @xmath497 and @xmath498 be as in section [ sec : second ] . if @xmath499 and there exists @xmath489 such that @xmath490 , then we have @xmath500 . * proof . * let @xmath227 such that @xmath501 if @xmath502 , and assume that @xmath4 has the minimum pole order at @xmath7 among such elements in @xmath197 . then @xmath503 . one has that @xmath504 and @xmath505 with respect to @xmath187 . by the property of grbner bases , @xmath506 is divisible by @xmath507 for some @xmath201 , which implies @xmath508 . by lemma [ lem:23 ] we have @xmath500 . we explain how the procedure in section [ sec : second ] works as desired . when the condition in step [ second1 ] in section [ sec : second ] is true , then there can not be a codeword within hamming distance @xmath161 from @xmath415 by the same reason as section [ sec : thirdwhy ] . so the algorithm stops processing with @xmath415 . when @xmath334 , then the algorithm declares @xmath324 as the unique codeword . when @xmath335 , then the algorithm examines the found codeword close enough to @xmath162 in steps [ second11 ] and [ second12 ] in section [ sec : second ] . when @xmath323 we can avoid computation of the evaluation map @xmath476 by the same reason as section [ sec : thirdwhy ] , which is checked at step [ second11 ] . otherwise we compute the codeword vector at step [ second12 ] and examine its hamming distance to @xmath415 . by proposition [ prop : candivide ] , the codeword must be found at @xmath509 . therefore , we do not execute the iteration at @xmath510 . we shall explain why the first criterion in section [ sec : first ] correctly finds the required codewords . the idea behind the first criterion is that there can not be another codeword within hamming distance @xmath161 from @xmath415 when the algorithm already found one . so the algorithm can stop iteration with smaller @xmath179 once a codeword is found as @xmath325 . the algorithm does not examine conditions when @xmath332 by the same reason as sections [ sec : thirdwhy ] and [ sec : secondwhy ] . when @xmath334 , then the algorithm declares @xmath324 as the unique codeword . when @xmath335 , then the algorithm examines the found codeword close enough to @xmath162 in steps [ first11][first13 ] in section [ sec : first ] . when @xmath323 we can avoid computation of the evaluation map @xmath476 by the same reason as section [ sec : thirdwhy ] , which is checked at step [ first11 ] . by proposition [ prop : candivide ] , the codeword must be found at some @xmath511 . therefore , we do not execute the iteration at @xmath512 . observe that for @xmath513 we set always @xmath295 to @xmath514 . for each @xmath157 satisfying @xmath515 , the number of accepted candidates satisfying eq . ( [ eq : acceptedvote ] ) can be at most @xmath3 . on the other hand , for @xmath179 with @xmath516 , the number of candidates is either zero or one , because at most one @xmath517 can satisfy eq . ( [ eq : acceptedvote ] ) . therefore , we have upper bounds for the number of iterations , counting executions of rebasing in section [ sec : rebasing ] , as @xmath518 for the third criterion for judging termination , where @xmath519 , and @xmath520 for the first and the second criteria for judging termination . we will use @xmath521 in place of @xmath522 in eqs . ( [ iteration - ub1 ] ) and ( [ iteration - ub2 ] ) for computation in tables [ tab1][tab3 ] , because @xmath522 depents on @xmath162 and @xmath523 . observe that the list decoding can be implemented as @xmath524 parallel execution of the unique decoding . therefore , when one can afford @xmath524 parallel implementation , which increases the circuit size , the decoding time of list decoding is the same as that of the unique decoding . we have provided an upper bound on the number of multiplications and divisions at each step of the proposed algorithm . we simulated @xmath525 transmissions of codewords with the one - point primal codes on klein quartic over @xmath49 with @xmath526 by using examples [ ex1 ] and [ ex2 ] , the one - point hermitian codes over @xmath527 with @xmath528 , and the one - point primal codes on the curve in example [ ex : gs1 ] over @xmath81 with @xmath529 . the program is implemented on the singular computer algebra system @xcite . the program used for this simulation is available from http://arxiv.org/src/1203.6127/anc . in the execution , we counted the number of iterations , ( executions of rebasing in section [ sec : rebasing ] ) , the sum of upper bounds on the number of multiplications and divisions given in eqs . ( [ eq : compl : ini ] ) , ( [ eq : compl3 ] ) , ( [ eq : compl4 ] ) , ( [ eq : compl5 ] ) , ( [ eq : compl6 ] ) , ( [ eq : compl7 ] ) , ( [ eq : compl1 ] ) , ( [ eq : multirebasing ] ) and ( [ eq : compl2 ] ) , and the number of codewords found . note also that eq . ( [ eq : nfmulti3 ] ) instead of eq . ( [ eq : nfmulti4 ] ) is used . the parameter @xmath161 is set to the same as the number of generated errors in each simulation condition . @xmath522 or @xmath530 in the number of errors in tables [ tab2 ] and [ tab22 ] indicates that the error vector is generated toward another codeword nearest from the transmitted codeword or completely randomly , respectively . the distribution of codewords is uniform on @xmath156 . that of error vectors is uniform on the vectors of hamming weight @xmath161 . in the code construction , we always try to use the feng - rao improved construction . specifically , for a given designed distance @xmath304 , we choose @xmath531 , and construct @xmath156 of eq . ( [ eq : cgamma ] ) . in the following , the designed distance is denoted by @xmath155 . it can be seen from tables [ tab1][tab3 ] and the following subsections that the computational complexity of the proposed algorithm tends to explode when the number of errors exceeds the error - correcting capability of the guruswami - sudan algorithm @xcite . in section [ sec : termination ] we proposed three criteria for terminating iteration of the proposed algorithm . from tables [ tab1][tab3 ] , one can see the following . the first criterion has the smallest number of iterations , and the second is the second smallest . on the other hand , the first criterion has the largest number of multiplications and divisions . the second and the third have the similar numbers . only the first criterion was proposed in @xcite and we see that the new criteria are better than the old one . the reason is as follows : the computation of quotient @xmath320 at step [ first1 ] in section [ sec : first ] is costlier than updating @xmath292 and @xmath293 in section [ sec : rebasing ] and the first criterion computes @xmath320 many times , which cancels the effect of decrease in the number of iterations . on the other hand , the second criterion computes @xmath320 only once , so it has the smaller number of multiplications and divisions than the first . the second criterion is faster when @xmath532 , while the third tends to be faster when @xmath407 . in addition to this , the ratio of the number of iterations in the second criterion to that of the third is smaller with @xmath532 than with @xmath407 . we speculate the reason behind them as follows : when @xmath407 and a wrong candidate is chosen at eq . ( [ eq : acceptedvote ] ) , after several iterations of sections [ sec : termination ] and [ sec : mainiteration ] , we often observe in our simulation that no candidate satisfies eq . ( [ eq : acceptedvote ] ) and the iteration stops automatically . under such situation , the second criterion does not help much to decrease the number of iterations nor the computational complexity when a wrong candidate is chosen at eq . ( [ eq : acceptedvote ] ) , and there are many occasions at which a wrong candidate is chosen at eq . ( [ eq : acceptedvote ] ) when @xmath407 . on the other hand , when @xmath532 , the second criterion helps to determine the transmitted information earlier than the third . in table [ tab3 ] , we observe that the upper bounds ( [ iteration - ub1 ] ) and ( [ iteration - ub2 ] ) are much larger than the actual number of iterations for @xmath533 . the disappearance of candidates satisfying eq . ( [ eq : acceptedvote ] ) in the last paragraph may also explain the reason behind the large differences for @xmath533 . on the other hand , we observe that the upper bound ( [ iteration - ub2 ] ) is quite tight for @xmath533 in table [ tab1 ] and @xmath534n in table [ tab22 ] . this suggests that improvement of eq . ( [ iteration - ub2 ] ) may need some additional assumption . we can use @xcite to decode this set of parameters . it is essentially the forward elimination in the gaussian elimination , and it takes roughly @xmath537 multiplications . in this case @xmath538 . the proposed algorithm has the lower complexity than @xcite . the code is @xmath22 with @xmath541 , @xmath542 . there is no previously known algorithm that can handle this case . the code is @xmath22 with @xmath544 , @xmath545 . according to ( * ? ? ? * figure 1 ) , we can use the original guruswami - sudan @xcite but it seems that its faster variants can not be used . we need the multiplicity @xmath55 to correct @xmath54 errors . we have to solve a system of @xmath546 linear equations . it takes @xmath547 multiplications in @xmath49 . the proposed algorithm is much faster . the code is @xmath22 with @xmath544 , @xmath545 . there is no previously known algorithm that can handle this case . we can use the bms algorithm @xcite for this case . the complexity of @xcite is estimated as @xmath551 and @xmath552 . the complexity of the proposed algorithm seems comparable to @xcite . however , we are not sure which one is faster . the code becomes the feng - rao improved code with designed distance @xmath555 . its dimension is @xmath556 . in order to have the same dimension by @xmath22 we have to set @xmath557 , whose ag bound @xcite is @xmath558 and the guruwsami - sudan can correct up to 2 errors . the proposed algorithm finds all codewords in the improved code with @xmath53 and @xmath558 errors . the code is @xmath22 with @xmath560 . the required multiplicity is @xmath561 , and the required designed list size is @xmath562 . the fastest algorithm for the interpolation step seems @xcite . * example 4 ) estimates the complexity of their algorithm as @xmath563 , where @xmath564 is the designed list size . ignoring the log factor and assuming the scaling factor one in the big-@xmath565 notation , the number of multiplications and divisions is @xmath566 . the proposed algorithm needs much fewer number of multiplications and divisions in @xmath527 . the guruwsami - sudan algorithm @xcite can correct up to @xmath568 errors and there seems no previously known algorithm that can handle this case . we can use @xcite to decode this set of parameters . it is essentially the forward elimination in the gaussian elimination , and it takes roughly @xmath537 multiplications . in this case @xmath570 . the proposed algorithm has the lower complexity than @xcite . this is a feng - rao improved code with dimension @xmath572 . in order to realize a code with the same dimension , we have to set @xmath573 in @xmath22 . the guruswami - sudan algorithm @xcite can correct no error in this set of parameters . there seems no previously known algorithm that can handle this case . this is a feng - rao improved code with dimension @xmath575 . in order to realize a code with the same dimension , we have to set @xmath576 in @xmath22 . the guruswami - sudan algorithm @xcite can correct @xmath577 errors in this set of parameters . there seems no previously known algorithm that can handle this case . this is an ordinary one - point ag code @xmath22 with @xmath579 and dimension @xmath580 . the guruswami - sudan algorithm @xcite can correct @xmath568 errors with the multiplicity @xmath581 and the designed list size @xmath582 . we have to solve a system of @xmath583 linear equations . it takes @xmath584 multiplications in @xmath81 . the proposed algorithm is much faster . the guruswami - sudan algorithm @xcite can correct up to @xmath568 errors and there seems no previously known algorithm that can handle this case . in this paper , we modified the unique decoding algorithm for plane ag codes in @xcite so that it can support one - point ag codes on _ any _ curve , and so that it can do the list decoding . the error correction capability of the original @xcite and our modified algorithms are also expressed in terms of the minimum distance lower bound in @xcite . we also proposed procedures to compute products and quotients in coordinate ring of affine algebraic curves , and by using those procedures we demonstrated that the modified decoding algorithm can be executed quickly . specifically , its computational complexity is comparable to the bms algorithm @xcite for one - point hermitian codes , and much faster than the standard list decoding algorithms @xcite for several cases . the original decoding algorithm @xcite allows parallel implementation on circuits like the ktter architecture @xcite . our modified algorithm retains this advantage . moreover , if one can afford large circuit size , the proposed list decoding algorithm can be executed as quickly as the unique decoding algorithm by parallel implementation on a circuit . the authors thank an anonymous reviewer for his / her careful reading that improved the presentation . this research was partially supported by the mext grant - in - aid for scientific research ( a ) no . 23246071 , the villum foundation through their velux visiting professor programme 20112012 , the danish national research foundation and the national science foundation of china ( grant no . 11061130539 ) for the danish - chinese center for applications of algebraic geometry in coding theory and cryptography , the spanish grant mtm2007 - 64704 , and the spanish mineco grant no . mtm2012 - 36917-c03 - 03 . the computer experiments in this research was conducted on singular 3.1.3 @xcite . m. ali and m. kuijper . a parametric approach to list decoding of reed - solomon codes using interpolation . _ ieee trans . inform . theory _ , 57(10):67186728 , oct . doi : 10.1109/tit.2011.2165803 . arxiv:1011.1040 . p. beelen and t. hholdt . the decoding of algebraic geometry codes . in e. martnez - moro , c. munuera , and d. ruano , editors , _ advances in algebraic geometry codes _ , volume 5 of _ coding theory and cryptology _ , pages 4998 . world scientific , 2008 . doi : 10.1142/9789812794017_0002 . m. bras - amors and m. e. osullivan . the correction capability of the berlekamp - massey - sakata algorithm with majority voting . _ applicable algebra in engineering , communication and computing _ , 17(5):315335 , oct doi : 10.1007/s00200 - 006 - 0015 - 8 . h. elbrnd jensen , r. r. nielsen , and t. hholdt . performance analysis of a decoding algorithm for algebraic - geometry codes . _ ieee trans . inform . theory _ , 45(5):17121717 , july 1999 . doi : 10.1109/18.771253 . m. fujisawa , h. matsui , m. kurihara , and s. sakata . with a higher probability one can correct errors up to half the designed distance for primal codes from curves . in _ proc . sita2006 _ , pages 101104 , hakodate , hokkaido , japan , nov . 2006 . o. geil , r. matsumoto , and d. ruano . list decoding algorithms based on grbner bases for general one - point ag codes . in _ proc . isit 2012 _ , pages 8690 , cambridge , ma , usa , july 2012 . doi : 10.1109/isit.2012.6284685 . arxiv:1201.6248 . k. lee , m. bras - amors , and m. e. osullivan . unique decoding of plane ag codes via interpolation . ieee trans . inform . theory _ , 58(6):39413950 , june 2012 . doi : 10.1109/tit.2012.2182757 . arxiv:1110.6251 . r. matsumoto and s. miura . on the feng - rao bound for the @xmath586-construction of algebraic geometry codes . _ ieice trans . fundamentals _ , e83-a(5):926930 , may 2000 . url http://www.rmatsumoto.org/repository/e83-a_5_923.pdf . r. matsumoto and s. miura . finding a basis of a linear system with pairwise distinct discrete valuations on an algebraic curve . _ j. symbolic comput . _ , 30(3):309323 , sept . 2000 . doi : 10.1006/jsco.2000.0372 . r. matsumoto and s. miura . on construction and generalization of algebraic geometry codes . in t. katsura et al . , editors , _ proc . algebraic geometry , number theory , coding theory , and cryptography _ , pages 315 , univ . tokyo , japan , jan . url http://www.rmatsumoto.org/repository/weight-construct.pdf . r. matsumoto , d. ruano , and o. geil . generalization of the lee - osullivan list decoding for one - point ag code . _ j. symbolic comput . _ , 55:19 , aug . 2013 . doi : 10.1016/j.jsc.2013.03.001 . arxiv:1203.6129 . s. miura . algebraic geometric codes on certain plane curves . _ electronics and communications in japan ( part iii : fundamental electronic science ) _ , 76(12):113 , dec . 1993 . doi : 10.1002/ecjc.4430761201 . ( original japanese version published as trans . ieice , vol . j75-a , no . 17351745 , nov . 1992 ) . r. pellikaan . on the efficient decoding of algebraic - geometric codes . in p. camion , p. charpin , and s. harari , editors , _ eurocode 92 international symposium on coding theory and applications _ , number 339 in cism courses and lectures , pages 231253 . cism international centre for mechanical sciences , springer , 1993 . url http://www.win.tue.nl/~ruudp/paper/17.pdf . k. saints and c. heegard . algebraic - geometric codes and multidimensional cyclic codes : a unified theory and algorithms for decoding using grbner bases . _ ieee trans . inform . theory _ , 41(6):17331751 , nov . doi : 10.1109/18.476246 . s. sakata . on fast interpolation method for guruswami - sudan list decoding of one - point algebraic - geometry codes . in s. bozta and i. e. shparlinski , editors , _ proc . aaecc-14 _ , volume 2227 of _ lecture notes in computer science _ , pages 172181 , melbourne , australia , nov . 2001 . springer - verlag . doi : 10.1007/3 - 540 - 45624 - 4_18 . s. sakata . multivariate interpolation and list decoding . in k. kobayashi and h. morita , editors , _ proc . 3rd asian - european workshop on information theory _ , pages 2831 , kamogawa , chiba , japan , june 2003 . society of information theory and its applications . isbn 4 - 902087 - 04 - 9 . s. sakata , h. elbrnd jensen , and t. hholdt . generalized berlekamp - massey decoding of algebraic - geometric codes up to half the feng - rao bound . _ ieee trans . inform . theory _ , 41(5):176211768 , sept . doi : 10.1109/18.476248 . s. sakata , j. justesen , y. madelung , h. elbrnd jensen , and t. hholdt . fast decoding of algebraic geometric codes up to the designed minimum distance . _ ieee trans . inform . theory _ , 41(5):16721677 , sept . doi : 10.1109/18.476240 . j. schicho . inversion of birational maps with grbner bases . in b. buchberger and f. winkler , editors , _ grbner bases and applications _ , volume 251 of _ london mathematical society lecture note series _ , pages 495503 . cambridge university press , 1998 . isbn 9780521632980 . doi : 10.1017/cbo9780511565847.031 .

we present multi - epoch high - resolution optical spectroscopy , uv / radio / x - ray imaging , and archival _ hubble _ and _ spitzer _ observations of an intermediate luminosity optical transient recently discovered in the nearby galaxy ngc300 . we find that the transient ( ngc300ot2008 - 1 ) has a peak absolute magnitude of mag , intermediate between novae and supernovae , and similar to the recent events m85ot2006 - 1 and sn2008s . our high - resolution spectra , the first for this event , are dominated by intermediate velocity ( km s ) hydrogen balmer lines and emission and absorption lines that point to a complex circumstellar environment , reminiscent of the yellow hypergiant irc+10420 . in particular , we detect broad h&k absorption with an asymmetric _ red _ wing extending to km s , indicative of gas infall onto a massive and relatively compact star ( blue supergiant or wolf - rayet star ) ; an extended red supergiant progenitor is unlikely . the origin of the inflowing gas may be a previous ejection from the progenitor or the wind of a massive binary companion . the low luminosity , intermediate velocities , and overall similarity to a known eruptive star indicate that the event did not result in a complete disruption of the progenitor . we identify the progenitor in archival _ spitzer _ observations , with deep upper limits from _ hubble _ data . the spectral energy distribution points to a dust - enshrouded star with a luminosity of about l , indicative of a m progenitor ( or binary system ) . this conclusion is in good agreement with our interpretation of the outburst and circumstellar properties . the lack of significant extinction in the transient spectrum indicates that the dust surrounding the progenitor was cleared by the outburst . we thus predict that the progenitor should be eventually visible with _ hubble _ if the transient event marks an evolutionary transition to a dust - free state , or with _ spitzer _ if the event marks a cyclical process of dust formation .

0901.0710

in recent years dedicated searches and serendipitous discoveries have uncovered several optical transients with luminosities intermediate between the well - studied classes of nova eruptions ( @xmath7 mag ) and supernova ( sn ) explosions ( @xmath8 mag ) . as can be expected from the wide gap in luminosity , these intermediate luminosity optical transients ( hereafter , ilots ) appear to be diverse in their properties and origins . some have been classified as low luminosity core - collapse sne , with inferred energies and @xmath9ni masses that are at least an order of magnitude below typical sn events ( e.g. , @xcite ) . others , sometimes initially classified as type iin sne , have been subsequently tagged with the catch - all designation of `` sn impostors '' ( e.g. , @xcite ) , that includes events resembling luminous blue variable ( lbv ) eruptions . finally , a small number of somewhat dimmer events have been recently grouped under the proposed name `` luminous red novae '' ( lrne ; @xcite ) , based on their red optical colors ; the origin of lrne and their relation to each other remain unclear . overall , the various groupings have been subject to debate , and there is no clear agreement about the nature of individual objects or the degree of overlap between the various designations . equally important , the connection between the different types of events and different classes of progenitors remains unclear . for example , some lbv eruptions have been initially classified as sne ( e.g. , sne 1961v and 1954j ; @xcite ) . similarly , the progenitor of the recent event sn2008s , which was detected in archival _ spitzer space telescope _ images , has been argued to be an extreme agb star @xcite or an lbv @xcite . the nature of the event itself is also unclear , with claims of a low mass electron - capture sn @xcite and an lbv - like outburst @xcite . the recent event 85 , and the possibly related eruptive object v838 mon , have been speculated to possibly result from stellar mergers @xcite , but other possibilities have been proposed such as a low - luminosity sn for 85@xcite , and an agb pulse , nova eruption from an embedded common envelope white dwarf , or planet capture for v838 mon ( e.g. , @xcite and references therein ) . regardless of the exact interpretation , it is clear that the various proposed eruption / explosion scenarios mark important phases in the evolution of massive stars , possibly on the path to diverse types of sne . therefore , a mapping between ilots and their progenitors will provide important constraints on the initial conditions ( i.e. , progenitors and circumstellar properties ) of sn explosions . the present uncertainty in this mapping highlights the need for detailed observations that can uncover the conditions both prior to and during the ilot events . these include the identification of progenitors , a detailed study of the circumstellar environment , and measurements of the event properties across the electromagnetic spectrum . such observations will also allow us to determine the connection between the various proposed categories of ilots , and to map the overall diversity of this seemingly heterogeneous class . in this paper we present such a combination of observations for an ilot discovered in the nearby galaxy ngc300 on 2008 april 24 ut ( hereafter , ) . for the first time for such an event , we combine multi - wavelength observations ( uv , optical , radio , and x - rays ) , high - resolution echelle spectroscopy , and archival _ hubble space telescope _ and _ spitzer space telescope _ data to show that was the result of an eruption from a @xmath10 m@xmath5 dust - obscured and relatively compact star , possibly in a binary system . our high - resolution spectra , the first for this event , resemble the yellow hypergiant star irc+10420 ( see also @xcite ) , suggesting that may mark a transition to a similar evolutionary stage . a comparison to the recent events sn2008s and 85 shows that all three events may have originated from the same phenomenon , but with a scatter in the energy release and circumstellar properties , that may point to a range of progenitor masses . was discovered by @xcite on 2008 apr 24.16 ut in the outskirts of the nearby galaxy ngc300 ( @xmath11 mpc ; @xcite ) with an apparent brightness of 16.5 mag ( unfiltered ) . subsequent observations on may 14.14 ut revealed that the object had brightened to about 14.2 mag @xcite . was not detected in observations from 2008 apr 17.1 ut , but only to a limit of 15.5 mag . it was also not detected in deeper observations taken on 2007 dec 30.8 and 2008 feb 8.75 ut to limiting magnitudes of 18.5 and 18.0 mag , respectively @xcite . since the transient was brightening at the discovery epoch , we adopt an event start time that is consistent with the most recent non - detection , @xmath12 apr 17 ut but note that the actual start time may have occurred earlier . a low resolution spectrum of obtained by @xcite on may 15.4 ut revealed emission lines corresponding to h@xmath13 , h@xmath14 , the ir triplet , and the [ ] @xmath15 doublet ( see also @xcite ) . the marginally - resolved lines indicated velocities of @xmath16 km s@xmath2 , significantly lower than those of both classical novae and supernovae . following the announcement of the discovery and early spectroscopy , and given the unique opportunity to study a nearby ilot in detail , we initiated observations across the electromagnetic spectrum , particularly high - resolution spectroscopy . in addition , we obtained archival _ hubble space telescope _ and _ spitzer space telescope _ observations at the position of the transient to search for a pre - discovery counterpart . the original reported position of was @xmath17 , @xmath18 ( j2000 ) , measured relative to the core of ngc300 @xcite . as part of our initial spectroscopic observations we obtained an @xmath19-band image of the field with the low dispersion survey spectrograph ( ldss-3 ) mounted on the magellan / clay 6.5-m telescope . using 13 objects in common with the naval observatory merged astrometric dataset ( nomad ) , we measure the actual position of the transient to be @xmath17 , @xmath18 ( j2000 ) with an uncertainty of about @xmath20 in each coordinate . this position is @xmath21 west and @xmath22 south of the iauc position . the revised coordinates allowed us to search for a progenitor in archival hst observations , and led @xcite to identify a counterpart in archival _ observations ( see [ sec : prog ] ) . we obtained multiple low- and high - resolution optical spectra of using ldss-3 and the magellan inamori kyocera echelle ( mike ) spectrograph mounted on the magellan / clay 6.5-m telescope . a log of the observations , including the spectral coverage and resolution is provided in table [ tab : spec ] . the ldss-3 spectra were reduced using standard iraf routines , while rectification and sky subtraction were performed using the method and software described in @xcite . wavelength calibration was performed using henear arc lamps , and air - to - vacuum and heliocentric corrections were applied . flux calibration was performed using several spectrophotometric standard stars . the mike spectra were reduced using a custom reduction pipeline written in python . wavelength calibration was performed using thar arc lamps , and air - to - vacuum and heliocentric corrections were applied . we obtained optical photometric observations with ldss-3 in the @xmath23 filters on may 31.44 ut ; see table [ tab : opt ] . the data were reduced using standard routines in iraf . photometry was performed relative to the standard star field e8-a . the resulting spectral energy distribution is shown in figure [ fig : optsed ] . we obtained uv and optical observations with the _ swift _ uv / optical telescope spanning from 2008 may 20 to october 15 ; see table [ tab : uvot ] . the data were processed using standard routines within the heasoft package . photometry of the transient in the uvw1 and @xmath24 images was performed using the standard @xmath25 aperture ; the source was not clearly detected in the uvw2 images . the light curves are shown in figure [ fig : lcs ] , and the spectral energy distribution at the epoch of the ldss-3 observations is shown in figure [ fig : optsed ] . simultaneous with the _ swift_/uvot observations , data were also collected with the co - aligned x - ray telescope for a total exposure time of 20.01 ks . no source is detected at the position of to a @xmath26 limit of @xmath27 erg s@xmath2 @xmath28 , where we have assumed a @xmath29 kev thermal bremsstrahlung model and galactic absorption , @xmath30 @xmath28 . a power law model with a photon index of @xmath31 leads to a similar limit of @xmath32 erg s@xmath2 @xmath28 . the corresponding limit on the luminosity is @xmath33 erg s@xmath2 . this is lower by at least an order of magnitude than any x - ray detected supernova to date . we observed with the very large array ) beginning on 2008 may 21.65 ut . no coincident radio source was detected to a limit of @xmath34 @xmath35jy ( @xmath26 ) at 8.46 ghz . continued monitoring at frequencies between 4.86 and 22.5 ghz over the subsequent seven months revealed no radio emission , with the latest non - detection obtained on 2008 november 24 ut ( @xmath36 @xmath35jy at 8.46 ghz ) . the upper limits imply a radio luminosity @xmath37 erg s@xmath2 hz@xmath2 . this limit is at least two orders of magnitude below the least luminous radio sne ever observed ( type iip sne : @xcite ; type ibc sn2002ap @xcite ) . it is also two orders of magnitude below the detected emission from the `` sn impostor '' sn1961v @xcite . the uv / optical spectral energy distribution ( sed ) on 2008 may 31 is shown in figure [ fig : optsed ] . the data have been corrected for the low galactic extinction , @xmath38 mag @xcite . we find that the sed is well fit by a blackbody spectrum ( @xmath39 for 5 degrees of freedom ) with a temperature of @xmath40 k and a radius of @xmath41 cm . the best - fit value of the host galaxy extinction is @xmath42 mag , with a @xmath43 upper limit of @xmath44 mag . as we show below , the low intrinsic dust extinction is supported by our detection of weak interstellar d absorption lines at the redshift of ngc300 , from which we infer negligible line of sight extinction , @xmath45 mag . using an event start date of april 17 and assuming a constant expansion rate , the inferred velocity is about 600 km s@xmath2 , in good agreement with our measured velocity width of the h@xmath13 emission line , @xmath46 km s@xmath2 ( see [ sec : lowspec ] ) . the optical / uv `` bolometric '' luminosity of on may 31 is @xmath47 erg s@xmath2 , or @xmath48 mag . this is about 4 magnitudes brighter than typical novae , and about @xmath49 mag fainter than sne ( e.g. , @xcite ) . the optical / uv emission fades slowly to @xmath50 d with a decline rate of about 0.02 mag d@xmath2 ( figure [ fig : lcs ] ) . however , observations at @xmath51 d reveal significant steepening relative to the early decay rate , indicating a break in the light curves at @xmath52 d ( detailed optical and near - ir light curves are presented in @xcite ) . we return to this point in [ sec : comp ] . using a roughly constant luminosity out to @xmath53 d , we find that the total radiated energy is about @xmath54 erg . therefore , the outflow kinetic energy is at least as large , indicating an ejected mass of @xmath55 m@xmath5 . the combination of low ejecta velocity and an intermediate luminosity suggest that is not a true sn explosion . indeed , its luminosity is nearly an order of magnitude lower than even the claimed low - luminosity core - collapse sne 1994n , 1997d , 1999br , 1999eu , and 2001dc , with @xmath56 erg s@xmath2 at @xmath57 d @xcite . the low - resolution spectra of are shown in figure [ fig : ldss ] . convolution of the spectrum from may 31 with the transmission functions of the @xmath23 filters indicates that the flux calibration across the full spectral range is good to better than @xmath58 . the most pronounced spectral features are the hydrogen balmer emission lines , the [ ] doublet and the ir triplet in emission , and h&k in absorption . we additionally detect weak and emission lines , as well as and in absorption . a full list of the lines detected at each epoch is provided in table [ tab : ldss ] . the h@xmath13 line is best fit by a lorentzian profile with a velocity width ( fwhm ) of @xmath59 km s@xmath2 ; see figure [ fig : redlines ] . the full width at continuum intensity ( fwci ) is @xmath60 km s@xmath2 . the width of the line is indicates an expansion velocity of about 600 km s@xmath2 , in good agreement with our size and eruption date estimates ( [ sec : prop ] ) . the h@xmath13 flux remains relatively unchanged for the first 2 months , with @xmath61 erg s@xmath2 @xmath28 . the h@xmath14 line flux is @xmath62 erg s@xmath2 @xmath28 , leading to @xmath63 . this value is well in excess of @xmath64 expected for case b recombination ( at @xmath65 k appropriate for the sed temperature of ; [ sec : prop ] ) . a large balmer decrement may be due to dust extinction , but as we already showed above ( and see also [ sec : ext ] ) the line of sight extinction to is negligible . instead , we conclude that the large balmer decrement is due to interaction with a high density circumstellar environment , through a combination of balmer self absorption and collisional excitation @xcite . the observed ratio is indeed in good agreement with circumstellar interaction models ( e.g. , @xcite ) . using the calculations of @xcite for hydrogen emission at high electron densities , we find that the observed @xmath66 ratio requires a density of @xmath67 @xmath68 . at this range of densities we expect the h@xmath69 line to be suppressed to a minimum ratio of @xmath70 ( compared to 0.45 for case b recombination ) . our non - detection of h@xmath69 on may 31 , with @xmath71 , and the detection on june 22 with @xmath72 support this conclusion . we note that the observed ratio @xmath73 ( june 22 ) is somewhat higher than the expected value of about 0.1 . we thus conclude that the balmer lines arise from circumstellar interaction between the event ejecta and the pre - existing environment . the h@xmath13 luminosity resulting from circumstellar interaction is given by @xmath74 erg s@xmath2 , where we have assumed a shell with @xmath75 and selected fiducial parameters for as inferred above ( @xmath76 @xmath68 and @xmath77 cm ) . comparing with the observed h@xmath13 luminosity of @xmath78 erg s@xmath2 , we find that @xmath79 @xmath68 and @xmath80 , in reasonable agreement with the values inferred from the balmer line ratios and the blackbody fit to the sed . the shocked circumstellar material has an inferred mass of @xmath81 m@xmath5 . for a uniform wind with a constant velocity , @xmath82 , this corresponds to a mass loss rate of @xmath83 m@xmath5 yr@xmath2 , where @xmath84 km s@xmath2 . as a sanity check , we find that the inferred thomson optical depth for these parameters is low , @xmath85 . to summarize , the hydrogen balmer emission line properties , coupled with the optical / uv spectral energy distribution , are indicative of an interaction between a low mass shell traveling at about 600 km s@xmath2 with a high density circumstellar medium , resulting from a progenitor mass loss rate of @xmath86 m@xmath5 yr@xmath2 . we next turn to an investigation of the unusual permitted and forbidden lines , which are not generally detected in sn spectra ( particularly at early time ) . as shown in figure [ fig : redlines ] , the [ ] lines are significantly narrower than the h@xmath13 line , with @xmath87 km s@xmath2 . moreover , the lines are clearly asymmetric , ranging at continuum intensity from @xmath88 to @xmath89 km s@xmath2 . we do not find any clear evolution in the line brightness and profile between may 23 and june 22 ut . the ir triplet lines also exhibit a complex structure , with an overall symmetric profile , marked by strong absorption near the line center . the observed brighter blue wing indicates that the center of the absorption feature is redshifted relative to the emission center . we measure this shift to be about 80 km s@xmath2 . the overall width of the lines is @xmath90 km s@xmath2 , while the ( unresolved ) absorption component has @xmath91 km s@xmath2 . finally , the h&k absorption lines exhibit an asymmetric profile with an extended red wing ; see figure [ fig : bluelines ] . the overall velocity range at continuum intensity in the may 31 spectrum is @xmath92 to @xmath93 km s@xmath2 . the lines appear to be somewhat narrower in the later spectrum from june 22 . the large velocity width of the lines indicates that they are not interstellar in origin , but instead arise in the circumstellar environment . however , the fact that we observe in absorption gas that is both inflowing ( red wing ) and outflowing ( blue wing ) , indicates that the lines do not arise exclusively from an outgoing wind associated with the eruption ; in that case we would expect only the outflowing ( blue ) component to be observed in absorption . the existence of material inflowing along the line of sight to at velocities of up to @xmath94 km s@xmath2 can be explained as the result of : ( i ) gas infall from a previous eruption ; or ( ii ) as the signature of wind outflow from a companion star impinging on the progenitor of . in the former scenario , the observed infall velocity , which we expect to be close to the progenitor s escape velocity , points to a massive and compact star such as a wolf - rayet star or a blue supergiant ; the velocity is much larger than would be expected for a red giant , a red supergiant , or an asymptotic giant branch ( agb ) progenitor as proposed by @xcite and @xcite . in the latter scenario , the companion would have to satisfy the same constraints ( massive and compact ) in order to launch a wind with @xmath3 km s@xmath2 . however , as we show in [ sec : prog ] , the total progenitor system mass is in the range of @xmath10 m@xmath5 , making it difficult to accommodate an early - b or o star companion as required from the @xmath3 km s@xmath2 wind velocity @xcite . the general spectral properties outlined in the previous section are borne out in much greater detail and complexity in our high resolution spectra ; see figures [ fig : mike1][fig : mike6 ] . these are the first such observations available for . we further stress that neither 85 nor sn2008s , which appear to share similar properties , have been observed at high spectral resolution so this is the first chance to investigate the properties of an ilot in detail . before we proceed with an analysis of the broad lines associated with the event itself , we use the d absorption lines to determine the line of sight extinction within ngc300 . in the may 20 spectrum we detect unresolved lines with the expected @xmath95 ratio of equivalent widths ( 0.17 and 0.08 ) , indicating that they are optically thin ; see figure [ fig : hei ] . we find no evidence for variation in the line profiles between the various epochs . from the narrowness and lack of variability we conclude that the lines are interstellar in origin . the measured equivalent widths correspond to a negligible line of sight extinction of @xmath45 mag @xcite . this inference is in good agreement with the lack of obvious extinction evidences in the optical / uv spectral energy distribution . in the spectra between june 6 and july 2 we further detect broad d absorption , with @xmath96 km s@xmath2 ( figure [ fig : hei ] ) . the lines are observed in emission in the spectrum from august 23 with a similar velocity width . the maximum absorption equivalent width is about 0.5 in the july 2 spectrum , close to the saturation limit of the relation between extinction and d equivalent width . the allowed range of @xmath97 for this value is @xmath98 mag , in rough agreement with the upper limit of @xmath44 mag inferred from the sed of ( [ sec : prop ] ) . we stress , however , that the relation between extinction and d equivalent width is calibrated in the interstellar medium , and there is therefore no reason that it should equally hold in the complex circumstellar environment of , particularly in the presence of bright uv emission from the eruption itself . indeed , as we show below , we find definitive evidence for dust destruction , which would undoubtedly alter the nature of the relation . the broad and asymmetric h&k absorption lines are perhaps the most striking aspect of the spectrum ( figure [ fig : cahk ] ) . from the high resolution spectrum obtained on june 6 we find that the red wing of the lines is best modeled with a lorentzian profile centered at @xmath99 km s@xmath2 relative to the narrow component ( the systemic velocity ) , and with a width of @xmath100 km s@xmath2 ; a gaussian profile with @xmath101 km s@xmath2 provides a reasonable fit to the red wing as well . an extension of these profiles to the blue wing of the h&k lines significantly over - estimates the line absorption ( figure [ fig : cahk ] ) . the blue wing itself is well fit by a lorentzian with a width of @xmath102 km s@xmath2 . it is unclear whether the two h&k absorption components ( red and blue wings ) arise in physically distinct regions ( with the red side that corresponds to the blue component perhaps contributing to the red wing ) , or whether they arise from a kinematically asymmetric distribution of a single absorbing component . in the former case , we interpret the line profiles to arise from the presence of both outflowing and inflowing gas in the circumstellar environment due to distinct mass loss episodes , fragmentation of a previously ejected shell , or the prsenence of a binary companion wind . this scenario is similar to the model proposed by @xcite for the yellow supergiant star irc+10420 , in which recombined gas no longer responds to radiation pressure , leading to infall at roughly the escape velocity . the measured infall velocity of @xmath3 km s@xmath2 is significantly higher than @xmath103 km s@xmath2 for absorption features in irc+10420 @xcite , indicating that the progenitor has a higher escape velocity ( i.e. , it is significantly more compact ) . subsequent echelle spectra reveal significant evolution in the h$k line profiles . first , the red absorption wing becomes progressively narrower with time , from about @xmath104 km s@xmath2 on june 6 to about @xmath105 km s@xmath2 on july 14 . at the same time , the blue absorption wing becomes narrower and eventually starts to be overtaken by emission with a fwci width of about 500 km s@xmath2 so that the overall line profile eventually resembles a reversed p cygni profile . both aspects of the line evolution can be explained as the result of recombination of ca@xmath106 gas produced through shocking of the initially absorbing gas by the eruption s outflowing material . in this scenario , the blue - shifted ( outflowing ) gas and the highest velocity inflowing gas are located physically closer to the progenitor , while the lower velocity infalling material will be eventually overtaken by the shock at a later time . a hint of this effect is seen in the low signal - to - noise spectrum from august 23 , in which very little absorption in the red wing of the h&k lines is observed . the same process can also explain the shift from absorption to emission in the d lines . the low - resolution spectra of showed that the triplet emission lines are marked by broad absorption features offset to the red from the emission line center . at high resolution , the line profile is more complex , and the absorption is separated into multiple components ; see figure [ fig : cair ] . in particular , in the june 6 spectrum there appear to be three distinct absorption components in addition to a narrow ( interstellar ) component , with velocity widths of 200 , 100 , and 40 km s@xmath2 . all three components are centered redward of the emission line peak , by 30 , 200 , and 80 km s@xmath2 , respectively . however , in our final spectrum from august 23 , the line profiles become simpler , with just a single absorption component redshifted relative to the emission line center . this may again point to the presence of an initially complex circumstellar environment , which is partially overtaken and shocked by the outflowing ejecta . the emission lines are symmetric and have a relatively constant width of @xmath107 km s@xmath2 in the various epochs . since the ionca2 ir lines are produced through radiative de - excitation from an excited state that is populated through absorption in the h&k lines , we expect the overall kinematic profiles of the h&k lines and the ir lines to match . however , the situation appears to be more complex here since the broad red absorption wing of the h$k lines does not appear to have a clear emission counterpart in the ir lines . another striking aspect of the echelle spectra is the strongly asymmetric profile of the [ ] lines , in which the blue wing is completely missing ; see figures [ fig : caii_1 ] and [ fig : caii_2 ] . the red wing is best fit by a lorentzian profile with a width of @xmath108 km s@xmath2 in the may 20 and june 6 spectra , decreasing to @xmath109 km s@xmath2 in the june 14 and july 2 spectra , and finally @xmath110 km s@xmath2 in the august 23 spectra . these lines are therefore significantly narrower than the ir lines and exhibit a more pronounced decrease in width with time . this , along with the missing blue wing , indicate that the forbidden lines are produced in a physically distinct region from the ir lines . indeed , to avoid collisional excitation and de - excitation out of the meta - stable excited state that is responsible for the [ ] emission , the lines need to be produced in a lower density environment compared to the ir lines and the hydrogen balmer lines . while the blue portion of the line is missing , we find no evidence for a p cygni profile . thus , the flux deficit is not due to absorption of the line emission by an optically thick outflow as generally observed for sne . instead , the line asymmetry may reflect an asymmetric distribution of the low density gas . in the context of the complex circumstellar environment inferred from the h&k and ir lines , it is likely that the red wing is produced in the low density outer layers of the inflowing gas , with the blue wing emission being absorbed by intervening material . this scenario also explains the lower velocity width compared to the h&k lines . the spectra from july 2 and august 23 reveal the emergence of low level emission on the blue side of the forbidden lines . this increase coincides with the emergence of h&k emission on the blue side of the lines ( [ sec : hk ] ) , and it is likely due to the same process of ca@xmath106 recombination . finally , as discussed in @xcite , we note that the energy required to ionize ca@xmath111 to ca@xmath106 from the meta - stable excited state ( which if left undisturbed would produce the [ ] doublet through radiative de - excitation ) corresponds to 1218.9 , or about 3.2 above the ly@xmath13 line . thus , when the ly@xmath13 line is broader than about 800 km / s , the [ ] doublet emission can be strongly suppressed by ionization . for narrow ly@xmath13 emission this effect is inefficient and we expect stronger [ ] emission . this is indeed the case for , for which the typical lines widths are @xmath112 km s@xmath2 . as a result , we expect that events like will generally exhibit strong [ ] emission lines . the h@xmath13 line , which appears to have a symmetric profile at low resolution , exhibits a more complex combination of emission and absorption when viewed at high spectral resolution ; see figure [ fig : ha ] . at all epochs we find a narrow absorption feature close to the line center , with a width of @xmath113 km s@xmath2 . the absorption feature is redshifted by about 30 km s@xmath2 relative to the interstellar absorption of the lines . a second , but weaker , absorption component with a similar velocity width is detected at a velocity of @xmath114 km s@xmath2 . the overall emission profile can be fit with either a lorentzian profile or a combination of narrow and broad gaussian profiles , which are slightly offset in velocity . the motivation behind the latter model is that the broad wings of the line appear to be somewhat asymmetric , with a slight deficit in the blue portion of the line . in the lorentzian case , the line width remains relatively unchanged between may 20 and june 14 with @xmath115 km s@xmath2 , and decreases mildly to @xmath116 km s@xmath2 on july 2 and august 23 . in the gaussian model the narrow component appears to increase in width from about 295 km s@xmath2 on may 20 to about 400 km s@xmath2 in the spectra from june 6 to august 23 the broad component remains relatively unchanged with @xmath117 km s@xmath2 . with only a mild change in width , and an increase in brightness relative to the continuum level , the h@xmath13 equivalent width increases significantly with time , ranging from about 10 to 300 between may 20 and august 23 . the h@xmath14 line exhibits the same profile as h@xmath13 ; see figure [ fig : hb ] . in the second epoch ( which has a higher signal to noise ratio ) , the line is marked by a deep absorption feature with @xmath118 km s@xmath2 , which is centered about 80 km s@xmath2 redward of the emission line center . this component is thus wider than its counterpart h@xmath13 absorber , and it also exhibits a more pronounced shift relative to the line center . the h@xmath14 line is again well fit by either a lorentzian profile or a combination of narrow and wide gaussians with offset line centers . in the former case , the line width remains unchanged with @xmath119 km s@xmath2 until july 14 , and decreases to about 330 km s@xmath2 on august 23 . in the gaussian model , the line widths remain unchanged at @xmath120 and @xmath121 km s@xmath2 . finally , the h@xmath69 and h@xmath122 lines exhibit similar profiles to h@xmath13 and h@xmath14 . in particular , the h@xmath69 line exhibits blue - shifted weak absorption similar to that seen in h@xmath13 ; see figures [ fig : balmer3 ] and [ fig : balmer4 ] . in addition to the strong hydrogen balmer lines and lines , we also detect the @xmath123 emission line ( figure [ fig : hei ] ) and the triplets at 7774 and 8446 in absorption ( figure [ fig : oi ] ) . in all cases the line widths are about 300 km s@xmath2 . moreover , the lines are clearly asymmetric with broad blue - shifted absorption extending to about 300 km s@xmath2 . this is the opposite effect compared to the h&k lines , indicating that the absorption in these two species arises in distinct kinematic components . in addition , we detect narrow absorption features of h&k , the ir triplet , and , with a typical width of @xmath124 km s@xmath2 . it is unclear whether these narrow features are interstellar in origin , or are due to previous mass loss from the progenitor of , with a much lower wind velocity ( possibly during a preceding red supergiant phase ) . our precise astrometry of ( [ sec : astrom ] ) enabled a search for the progenitor in archival _ hubble space telescope _ and _ spitzer space telescope _ observations ( see also @xcite ) . the location of was observed with the advanced camera for surveys ( acs ) on 2006 november 8 ut in the f475w ( 1488 s ) , f606w ( 1515 s ) , and f814w ( 1542 s ) filters as part of program 10915 ( pi : delcanton ) . we retrieved the drizzled images from the hst archive and performed an astrometric tie relative to our initial ldss-3 image ( [ sec : astrom ] ) . the resulting astrometry has an rms of 30 mas in each coordinate . we do not detect any objects at the position of to the following @xmath125 limits : @xmath126 mag ( f475w ) , @xmath127 mag ( f606w ) , @xmath128 mag ( f814w ) . the nearest object to the position of is located about @xmath129 away , or about 1.75 pc at the distance of ngc300 . images of the field at the location of are shown in figure [ fig : hst ] . the location of was observed with the infra - red array camera ( irac ) on 2007 december 28 ut , and with the multiband imaging photometer for spitzer ( mips ) on 2007 july 6 and 16 ut . we retrieved the post - bcd data from the _ spitzer _ archive and performed an astrometric tie relative to our initial ldss-3 image ( [ sec : astrom ] ) . the resulting astrometry has an rms of @xmath130 in each coordinate . images of the field at the location of are shown in figures [ fig : spitzer1 ] and [ fig : spitzer2 ] . as noted by @xcite based on our reported astrometry of the transient @xcite , a coincident object is detected in the _ spitzer _ data ( see also @xcite ) . we performed photometry of this object using a 2-pixel aperture on the 3.6 and 4.5 @xmath35 m images , and a 3-pixel aperture on the 5.8 , 8.0 , and 24 @xmath35 m images . the aperture size was selected to reduce contamination from nearby objects . we further used the standard zeropoints and aperture corrections provided in the _ spitzer _ manual . the resulting spectral energy distribution of the coincident source , along with the hst upper limits , is shown in figure [ fig : spitzer_hst ] . a detailed discussion of the progenitor is presented by @xcite who conclude that it is an extreme agb star during a short - lived phase ( @xmath131 yr ) of its evolution . below we provide an independent analysis of the _ spitzer _ and hst observations . the spectral energy distribution is shown in figure [ fig : spitzer_hst ] . we find that the sed is well fit by a blackbody profile ( @xmath132 for 3 degrees of freedom ) , indicating that the progenitor was enshrouded by dust prior to the eruption . the best - fit blackbody parameters are @xmath133 k and @xmath134 au , with a resulting luminosity of @xmath135 l@xmath5 . the total dust mass can be inferred by setting @xmath136 , which for @xmath137 @xmath138 g@xmath2 @xcite , corresponds to @xmath139 m@xmath5 . the fact that the progenitor was completely obscured by dust , while exhibits negligible dust extinction ( @xmath44 mag ; [ sec : prop ] ) demonstrates that the eruption destroyed the obscuring dust . since the size of the dust - obscured region exceeds the shock radius of by about a factor of 20 ( [ sec : prop ] ) , we conclude that the dust was destroyed through sublimation rather than shock heating . this process has been investigated by several authors , and we use here the formulation of @xcite developed in the context of gamma - ray bursts . the dust destruction radius is given by @xmath140 cm , for a typical grain size of 0.1 @xmath35 m and a dust absorption efficiency factor of order unity @xcite . for the radius of the dust - obscured region inferred above , we find that the luminosity required for sublimation is @xmath141 erg s@xmath2 . this is about an order of magnitude larger than the bolometric luminosity measured for on may 31 ( @xmath142 d ) . the measured decay rate of about 0.02 mag d@xmath2 extrapolated back to the estimated eruption date is not sufficient to overcome this discrepancy . we therefore conclude that either the early emission from was characterized by a higher photospheric temperature , or the initial event produced a relatively bright uv flash with a luminosity of about @xmath143 erg s@xmath2 . in either case , the destruction of the obscuring dust may be responsible for the strong emission , since the calcium is initially strongly depleted on dust grains . our detailed spectroscopic observations of , and the detection of its progenitor system in archival data , allow us to construct a basic model of the event and its environment . the key observational results presented in the preceding sections are summarized as follows : * the line of sight extinction is negligible as inferred from the uv / optical sed and the weak d lines . * prior to the event , the progenitor system was enshrouded by dust , and had a luminosity of about @xmath4 l@xmath5 , corresponding to a @xmath10 m@xmath5 star . * the wide and asymmetric h&k absorption lines require gas infall at velocities of about @xmath144 km s@xmath2 , and hence a massive and relatively compact progenitor or companion . * the ir lines and the hydrogen balmer lines are symmetric and broad ( @xmath145 km s@xmath2 ) , and marked by narrow absorption redward of the emission line peak . * the [ ] lines are extremely asymmetric , and significantly narrower than the other lines and hydrogen balmer lines . * none of the lines exhibit velocities in excess of @xmath3 km s@xmath2 , or p cygni profiles . these properties bear a striking resemblance to those of the massive yellow hypergiant irc+10420 ( e.g. , @xcite ) , which is thought to be rapidly transitioning from a red supergiant phase to a luminous blue variable or wolf - rayet phase . like the progenitor of , irc+10420 is marked by a bright infrared excess resulting from a dusty nebula , but unlike the central object is not completely obscured . the luminosity of irc+10420 , @xmath146 l@xmath5 , is about an order of magnitude larger than our inferred luminosity of the progenitor of ( [ sec : prop ] ) . this points either to a lower mass or a somewhat earlier evolutionary stage for the progenitor of . equally important , irc+10420 exhibits essentially identical hydrogen balmer and ir line profiles to those seen in our high resolution spectra , including the redshifted narrow absorption feature ( see figures 4 and 5 of @xcite and figures 3 and 4 of @xcite ) . at the same time , there are some distinct difference between the spectroscopic properties of and irc+10420 . first , the ionca2 h&k lines in irc+10420 are symmetric and significantly narrower than those observed in @xcite . second , the [ ] emission lines in irc+10420 appear to be symmetric @xcite , although like in the case of they are significantly narrower than the hydrogen balmer lines and the ir lines . third , the ir lines , which extend to about @xmath147 km s@xmath2 @xcite , are significantly narrower than in . finally , while the h@xmath13 profiles in both objects are generally similar , the red and blue emission peaks of irc+10420 have a similar width to the central absorption @xcite , whereas in the case of the emission peaks are significantly wider . in both cases the h@xmath13 line exhibit a broad wing , which is interpreted as the effect of electron scattering in the case of irc+10420 @xcite . in the spectrum of , however , the core of the line is broader and the wings are less pronounced , and we interpret the profile as due to circumstellar interaction of an outflow with a velocity of about 600 km s@xmath2 ( as opposed to a outflowing wind velocity of only 50 km s@xmath2 inferred for irc+10420 ) . the combination of these various properties suggests that in broad terms the progenitor of may be similar to irc+10420 , but it may have been in a somewhat different evolutionary stage prior to the eruption , with still complete dust obscuration . in addition , the eruption itself appears to have produced a significantly faster , more energetic , and more luminous outflow compared to the `` quiescent '' state of irc+10420 . thus , it is possible that marks the initial transition to an irc+10420-like object , and that as the eruption fades , its properties will converge to those of irc+10420 . we finally turn to a comparison of with two recent events that exhibit several similar properties : 85 and sn2008s . sn2008s was discovered on february 1.8 ut in ngc6946 ( @xmath148 mpc ) and had an absolute magnitude of @xmath149 mag ( taking into account significant galactic and host obscuration of @xmath150 mag ; @xcite ) . the @xmath151-band light curve of sn2008s from @xcite is shown in figure [ fig : lcs ] . the temporal evolution is remarkably similar to that of , namely a relatively stable decay rate of about 0.035 mag d@xmath2 from about 30 to 100 days . however , unlike , the optical emission from sn2008s does not exhibit a clear steepening at later times , and it may indeed flatten to a decay rate of about 0.01 mag d@xmath2 from 130 to 270 days . in figure [ fig:2008s ] we present a low - resolution spectrum of sn2008s that we obtained on 2008 mar 15.5 ut ( @xmath152 d ) using the arc 3.5-m telescope at apache point observatory . the resolution of the spectrum is 6.5 , or about 300 km s@xmath2 . neglecting extinction , we find that the overall continuum shape and brightness are characterized by a blackbody profile with @xmath153 k and @xmath154 cm , similar to the properties inferred for at a similar epoch of about 44 d ( figure [ fig : spitzer_hst ] ) . the inferred bolometric absolute magnitude is @xmath155 mag . however , if we include the estimated extinction of @xmath150 mag , we find that the spectrum requires a higher temperature , @xmath156 k , and a smaller radius , @xmath157 cm , leading to an overall absolute magnitude of @xmath158 mag , similar to the values inferred by @xcite from similar spectroscopic observations . thus , sn2008s has a bolometric luminosity that is nearly an order of magnitude larger than . combined with the similar temporal evolution , we conclude that sn2008s released about an order of magnitude more energy than , @xmath159 erg ( see also @xcite ) . the overall similarity in also evident in the dominant spectral features of both events . sn2008s exhibits strong and narrow hydrogen balmer lines , [ ] lines , and ir lines . the h@xmath13 line is resolved with a width of @xmath160 km s@xmath2 , about a factor of 2 times broader than for . as in the case of , the ir triplet lines exhibit a similar width to that of h@xmath13 , @xmath161 km s@xmath2 . the [ ] lines are only marginally resolved , with @xmath162 km s@xmath2 . this is again a factor of 2 times broader than in , and comparably narrower than the h@xmath13 line . due to the low resolution of the spectrum it is unclear whether the [ ] lines are asymmetric . unlike in the case of , we do not detect h&k absorption or emission . finally , perhaps the most important similarity between sn2008s and is the detection of a coincident _ spitzer _ source in both events @xcite . as shown in figure [ fig : spitzer_hst ] , the seds of both coincident sources are similar , with @xmath163 k , @xmath164 au , and @xmath165 l@xmath5 for sn2008s . thus , both events appear to arise from objects of a similar luminosity and in a similar evolutionary stage . @xcite recently argued that the progenitors of both events are extreme agb stars with masses of @xmath166 m@xmath5 , and that the events themselves are likely electron - capture sne . @xcite , on the other hand , advocated a scenario for sn2008s of a super - eddington eruption that results in significant mass loss . they further note an overall similarity to irc+10420 , and speculate that sn2008s resembles an eruption of a @xmath167 m@xmath5 blue supergiant , rather than a lower mass red supergiant as advocated by @xcite . our high resolution spectroscopic observations of , and their close resemblance to the spectrum of irc+10420 , support the interpretation of @xcite . still , while both events appear to arise from similar objects , these progenitors are clearly capable of producing eruptions with an energy release that spans at least an order of magnitude . we next turn to a comparison with 85 . as shown in figure [ fig : optsed ] , the spectral energy distribution of this event closely resembles , with a nearly identical bolometric luminosity of about @xmath168 mag @xcite . the light curve evolution is also similar ( figure [ fig : lcs ] ) , with a comparable decay rate during the first @xmath169 days , followed by a significant steepening at later time . thus , 85 appears to be a clear analog of both and sn2008s . unfortunately , the low - resolution spectra of 85 , obtained about 1 and 48 d after discovery @xcite , had a low signal - to - noise ratio that precludes a detailed analysis . still , the second spectrum does exhibit a narrow h@xmath13 emission line with @xmath170 km s@xmath2 , about two times narrower than in . additional weak features possibly corresponding to h&k and @xmath171 may also be discerned in the spectrum ( see figure 3 of @xcite ) . these possible features provide additional support for a common origin with . on the other hand , the environments of and 85 differ significantly . in the case of 85 , the host is an early type galaxy ( with a mean age of @xmath172 gyr ) , and the environment of the transient itself contains no stars brighter than @xmath173 mag , corresponding to a mass of @xmath174 m@xmath5 @xcite . the local environment of , on the other hand , contains stars with inferred masses of @xmath175 m@xmath5 @xcite , in line with the inferred mass of the progenitor system . thus , it appears that different channels / progenitors may lead to the formation of similar transient outbursts . given the overall similarity between the three events it is reasonable to conclude that they share a common origin . thanks to our detailed high - resolution spectroscopic observations of , an intimate connection with objects like irc+10420 appears likely ( in agreement with previous speculation @xcite ) but may not be required ( e.g. , 85 ) . from the overall similarity we further conclude that none of the three transients are likely to have resulted from the explosion of the progenitor star . instead , the eruptions may mark a shift in the evolutionary stage from complete obscuration to a yellow supergiant phase , and eventually a wolf - rayet star . clearly , such a transition apparently leads to overall common features , but in detail the energy and velocity of the outflow appear to vary by at least an order of magnitude . we presented the first high - resolution spectroscopic observations for the intermediate luminosity optical transient , as well as multi - wavelength observations ( uv / optical / radio / x - ray ) and an independent analysis of archival _ spitzer _ and hst observations of its progenitor ( see also @xcite ) . the preponderance of photometric and spectroscopic evidence indicate that this event is not a supernova , but is instead the result of an eruption on a @xmath6 m@xmath5 blue supergiant or pre - wolf - rayet star , possibly in a binary system . the high - resolution spectra reveal the presence of a complex circumstellar environment , marked by outflows and possible inflow from previous ejections or from a companion wind . the observed inflow velocities are indicative of a massive and relatively compact progenitor , not a highly extended red supergiant , in contrast to the interpretation of @xcite and @xcite . alternatively , the progenitor may be part of a binary system in which the companion wind ( with a velocity of @xmath3 km s@xmath2 ) is observed in absorption along the line of sight . in this scenario , however , it may be difficult to accommodate the relatively low total system mass of @xmath176 m@xmath5 . the outburst itself gave rise to an outflow with a velocity of @xmath177 km s@xmath2 and a kinetic energy of at least @xmath54 erg . no radio or x - ray emission are detected to limits that are significantly dimmer than the faintest sne detected to date , indicating a lack of fast ejecta ( with @xmath178 ) . closely resembles two previous events , 85 and sn2008s , in terms of its intermediate luminosity and spectral properties . it also shares with sn2008s a dust - obscured progenitor with a luminosity of a @xmath179 l@xmath5 . our detailed comparison reveals that the outburst and progenitor properties point to a common origin , but with a range of outburst energy and velocity of nearly an order of magnitude . we also note a resemblance between the spectroscopic properties of and those of the luminous yellow supergiant irc+10420 . however , the latter is about an order of magnitude more luminous than the progenitor of , and its observed `` quiescent '' outflow has a velocity of only @xmath180 km s@xmath2 . these differences raise the possibility that marks the transition to an irc+10420-like state , or that its progenitor is simply a lower mass analogue . while our optical / uv photometry and low - resolution spectroscopy are of similar quality to those of sn2008s and 85 , our extensive echelle spectroscopy provides a much deeper view of the circumstellar environment than previously available for this class of events . indeed , we stress that the overall photometric behavior and general spectroscopic features observed in may be generic to a broad class of stellar eruptions with an intermediate velocity in the presence of a dense circumstellar environment . for example , the 2002 eruption of v838 mon was similar in broad terms to that of , 85 , and sn2008s , based on a comparable light curve behavior and spectroscopic evidence for narrow h@xmath13 and lines ( e.g. , @xcite ) . however , the eruption was about an order of magnitude less luminous than and 85 , and the progenitor appears to have been a lower mass star . we conclude that a clear understanding of the broad range of ilots requires high - resolution spectroscopy to assess the detailed properties of the circumstellar environment , as well as the identification of progenitors ; optical / ir photometry and low - resolution spectroscopy alone are not likely to be sufficient . finally , the main prediction of our model is that the progenitor of should be visible following the decay of the event . if the current eruption is similar to previous events that produced the now - destroyed dust shell , than the progenitor would again be visible as an obscured ir source . however , if only a minor amount of dust reforms , we predict that the spectrum will more closely resemble that of irc+10420 . in either case it appears that events like represent an important stage in the evolution of massive stars toward their eventual demise in sn explosions . the advent of facilities such as pan - starrs and lsst will lead to an increased rate of detections of events similar to . from the observations presented here it is clear that multi - wavelength follow - up , in particular high - resolution spectroscopy , which is not typically attempted for sne ( cf . sn2008s ) , is critical for a detailed understanding of the eruption process and the properties of the progenitor(s ) . this paper includes data gathered with the 6.5 meter magellan telescopes located at las campanas observatory , chile . we thank david sand for obtaining the ldss-3 spectrum on 2008 june 5 , bruce draine for insightful discussions about the spectral properties of , and ranga chary for assistance with the _ spitzer _ data . ams acknowledges support by nasa through a hubble fellowship grant . rac acknowledges support from an nsf grant ast-0807727 . llccccc 2008 may 20.42 & mike & @xmath181 & blue & @xmath182 & @xmath183 & 0.10 + & mike & @xmath181 & red & @xmath182 & @xmath184 & 0.21 + 2008 may 23.43 & ldss-3 & @xmath185 & vph - red & @xmath186 & @xmath187 & 6.2 + 2008 may 31.40 & ldss-3 & @xmath181 & vph - blue & @xmath186 & @xmath188 & 3.4 + & ldss-3 & @xmath189 & vph - red & @xmath186 & @xmath187 & 6.2 + 2008 june 5.41 & ldss-3 & @xmath185 & vph - red & @xmath186 & @xmath187 & 6.2 + 2008 june 6.40 & mike & @xmath190 & blue & @xmath186 & @xmath183 & 0.13 + & mike & @xmath190 & red & @xmath186 & @xmath184 & 0.28 + 2008 june 14.40 & mike & @xmath191 & blue & @xmath182 & @xmath183 & 0.10 + & mike & @xmath191 & red & @xmath182 & @xmath184 & 0.21 + 2008 june 22.42 & ldss-3 & @xmath185 & vph - blue & @xmath186 & @xmath192 & 3.4 + & ldss-3 & @xmath185 & vph - red & @xmath186 & @xmath193 & 6.2 + 2008 july 2.35 & mike & @xmath194 & blue & @xmath186 & @xmath183 & 0.13 + & mike & @xmath194 & red & @xmath186 & @xmath184 & 0.28 + 2008 july 14.36 & mike & @xmath194 & blue & @xmath186 & @xmath183 & 0.13 + & mike & @xmath194 & red & @xmath186 & @xmath184 & 0.28 + 2008 august 23.32 & mike & @xmath194 & blue & @xmath182 & @xmath183 & 0.10 + & mike & @xmath194 & red & @xmath182 & @xmath184 & 0.21 + lcccc 2008 may 20.01 & uvw2 & 5723 & @xmath195 & @xmath196 + 2008 may 25.03 & uvw1 & 3877 & @xmath197 & @xmath198 + 2008 may 25.04 & u & 1551 & @xmath199 & @xmath200 + 2008 may 25.04 & b & 773 & @xmath201 & @xmath202 + 2008 may 25.04 & v & 691 & @xmath203 & @xmath204 + 2008 june 4.68 & uvw1 & 1295 & @xmath205 & @xmath196 + 2008 june 4.67 & u & 1599 & @xmath206 & @xmath207 + 2008 june 4.68 & b & 1856 & @xmath208 & @xmath209 + 2008 june 4.67 & v & 2006 & @xmath210 & @xmath211 + 2008 june 19.39 & uvw1 & 1765 & @xmath212 & @xmath213 + 2008 june 19.39 & u & 879 & @xmath214 & @xmath215 + 2008 june 19.39 & b & 879 & @xmath216 & @xmath217 + 2008 june 19.39 & v & 879 & @xmath218 & @xmath219 + 2008 august 14.17 & uvw1 & 746 & @xmath220 & @xmath221 + 2008 august 14.17 & u & 428 & @xmath222 & @xmath223 + 2008 august 14.17 & b & 428 & @xmath224 & @xmath225 + 2008 august 14.17 & v & 428 & @xmath226 & @xmath227 + 2008 august 15.04 & uvw1 & 727 & @xmath228 & @xmath221 + 2008 august 15.04 & u & 467 & @xmath229 & @xmath230 + 2008 august 15.04 & b & 467 & @xmath231 & @xmath232 + 2008 august 15.04 & v & 467 & @xmath233 & @xmath234 + 2008 august 26.60 & uvw1 & 1522 & @xmath235 & @xmath236 + 2008 august 26.60 & u & 1024 & @xmath237 & @xmath238 + 2008 august 26.60 & b & 1024 & @xmath239 & @xmath240 + 2008 august 26.59 & v & 1024 & @xmath241 & @xmath240 + 2008 october 15.02 & u & 716 & @xmath242 & @xmath236 + 2008 october 15.02 & b & 646 & @xmath242 & @xmath221 + 2008 october 15.01 & v & 716 & @xmath243 & @xmath244 llccccc h & may 31 & 3938.84 & 310 & @xmath249 & 12.1 & + & june 22 & 3938.30 & 270 & @xmath250 & 8.7 & + k & may 31 & 3973.78 & 315 & @xmath251 & 9.3 & + & june 22 & 3973.47 & 295 & @xmath252 & 7.0 & + h@xmath122 & may 31 & 4102.51 & @xmath253 & @xmath254 & 14.1 & @xmath255 + & june 22 & 4104.32 & 105 & @xmath256 & 6.7 & @xmath257 + h@xmath69 & june 22 & 4343.25 & 110 & @xmath258 & 7.2 & @xmath259 + h@xmath14 & may 31 & 4864.35 & 100 & @xmath260 & 9.0 & @xmath261 + & june 22 & 4864.28 & 100 & @xmath262 & 6.7 & @xmath263 + ? & may 31 & 5857.10 & & @xmath264 & 7.2 & + & june 22 & 5856.00 & & @xmath265 & 7.9 & + 5877.25 & may 31 & 5878.96 & 90 & @xmath266 & 9.8 & @xmath267 + & june 22 & 5880.13 & 145 & @xmath268 & 7.6 & @xmath269 + d & may 31 & 5895.65 & 55 & @xmath270 & 10.0 & + & june 22 & 5894.75 & 10 & @xmath271 & 10.2 & + 5993.03 & may 31 & 5999.51 & 325 & @xmath272 & 9.9 & @xmath273 + & june 22 & 5998.22 & 260 & @xmath274 & 7.1 & @xmath275 + 6434.46 & may 23 & 6437.31 & 130 & @xmath274 & 21.3 & @xmath276 + & may 31 & 6434.78 & 15 & @xmath277 & 20.7 & @xmath278 + & june 5 & 6436.56 & 100 & @xmath279 & 18.5 & @xmath280 + & june 22 & 6438.13 & 170 & @xmath266 & 9.0 & @xmath278 + 6517.85 & may 23 & 6523.81 & 270 & @xmath281 & 7.3 & @xmath282 + & may 31 & 6522.95 & 235 & @xmath283 & 9.6 & @xmath273 + & june 5 & 6522.66 & 220 & @xmath272 & 7.5 & @xmath284 + & june 22 & 6521.58 & 170 & @xmath285 & 8.0 & @xmath286 + h@xmath13 & may 23 & 6569.94 & 240 & @xmath287 & 16.2 & @xmath288 + & may 31 & 6568.86 & 190 & @xmath289 & 15.3 & @xmath290 + & june 5 & 6568.74 & 190 & @xmath291 & 14.9 & @xmath292 + & june 22 & 6468.02 & 155 & @xmath293 & 13.3 & @xmath294 + 6679.99 & may 23 & 6684.60 & 210 & @xmath274 & 17.5 & @xmath280 + & may 31 & 6684.99 & 225 & @xmath277 & 13.0 & @xmath295 + & june 5 & 6684.64 & 210 & @xmath274 & 8.9 & @xmath296 + & june 22 & 6684.46 & 200 & @xmath297 & 11.5 & @xmath298 + 7067.14 & may 23 & 7068.33 & 50 & @xmath299 & 6.4 & @xmath300 + & may 31 & 7068.51 & 60 & @xmath281 & 5.4 & @xmath301 + & june 5 & 7069.89 & 120 & @xmath302 & 6.6 & @xmath282 + & june 22 & 7071.10 & 170 & @xmath274 & 10.7 & @xmath273 + @xmath303 $ ] 7293.48 & may 23 & 7299.60 & 250 & @xmath254 & 4.3 & @xmath304 + & may 31 & 7299.20 & 235 & @xmath305 & 5.5 & @xmath306 + & june 5 & 7299.45 & 245 & @xmath307 & 4.0 & @xmath308 + & june 22 & 7298.22 & 195 & @xmath309 & 5.2 & @xmath310 + @xmath303 $ ] 7325.91 & may 23 & 7332.04 & 250 & @xmath311 & 5.4 & @xmath62 + & may 31 & 7331.62 & 235 & @xmath312 & 6.5 & @xmath310 + & june 5 & 7331.80 & 240 & @xmath313 & 5.2 & @xmath308 + & june 22 & 7330.61 & 195 & @xmath314 & 6.1 & @xmath315 + 7464.44 & may 23 & 7468.98 & 180 & @xmath277 & 24.5 & @xmath316 + & may 31 & 7468.14 & 150 & @xmath274 & 16.0 & @xmath317 + & june 5 & 7467.85 & 135 & @xmath266 & 20.1 & @xmath295 + & june 22 & 7467.54 & 125 & @xmath285 & 19.2 & @xmath318 + 7713.83 & may 23 & 7723.98 & 395 & @xmath266 & 24.2 & @xmath318 + & may 31 & 7720.87 & 275 & @xmath274 & 13.8 & @xmath296 + & june 5 & 7721.19 & 285 & @xmath274 & 11.6 & @xmath273 + & june 22 & 7720.93 & 275 & @xmath319 & 11.1 & @xmath278 + 7774.08@xmath320 & may 23 & 7780.29 & 240 & @xmath321 & 8.9 & + & may 31 & 7778.85 & 185 & @xmath322 & 10.6 & + & june 5 & 7778.87 & 185 & @xmath322 & 10.3 & + & june 22 & 7776.96 & 110 & @xmath323 & 9.4 & + 8448.57@xmath324 & may 23 & 8454.78 & 220 & @xmath325 & 4.9 & + & may 31 & 8452.95 & 155 & @xmath326 & 5.6 & + & june 5 & 8452.76 & 150 & @xmath326 & 5.8 & + & june 22 & 7776.96 & 110 & @xmath323 & 9.4 & ) . the h&k lines are somewhat narrower in the later spectrum , while the hydrogen balmer lines and the @xmath327 line exhibit a significant increase relative to the continuum brightness . similarly , the d broad components exhibit increased absorption . [ fig : bluelines],width=672 ]

one of the most promising and versatile approaches to creating new quantum algorithms is based on the quantum hidden subgroup ( qhs ) paradigm , originally suggested by alexei kitaev . this class of quantum algorithms encompasses the deutsch - jozsa , simon , shor algorithms , and many more . in this paper , our strategy for finding new quantum algorithms is to decompose shor s quantum factoring algorithm into its basic primitives , then to generalize these primitives , and finally to show how to reassemble them into new qhs algorithms . taking an `` alphabetic building blocks approach , '' we use these primitives to form an `` algorithmic toolkit '' for the creation of new quantum algorithms , such as wandering shor algorithms , continuous shor algorithms , the quantum circle algorithm , the dual shor algorithm , a qhs algorithm for feynman integrals , free qhs algorithms , and more . toward the end of this paper , we show how grover s algorithm is most surprisingly almost a qhs algorithm , and how this result suggests the possibility of an even more complete `` algorithmic tookit '' beyond the qhs algorithms .

quant-ph0607046

one major obstacle to the fulfillment of the promise of quantum computing is the current scarcity of quantum algorithms . quantum computing researchers simply have not yet found enough quantum algorithms to determine whether or not future quantum computers will be general purpose or special purpose computing devices . as a result , much more research is crucially needed to determine the algorithmic limits of quantum computing . one of the most promising and versatile approaches to creating new quantum algorithms is based on the quantum hidden subgroup ( qhs ) paradigm , originally suggested by alexei kitaev @xcite . this class of quantum algorithms encompasses the deutsch - jozsa , simon , shor algorithms , and many more . in this paper , our strategy for finding new quantum algorithms is to decompose shor s quantum factoring algorithm into its basic primitives , then to generalize these primitives , and finally to show how to reassemble them into new qhs algorithms . taking an `` alphabetic building blocks approach , '' we will use these primitives to form an `` algorithmic toolkit '' for the creation of new quantum algorithms , such as wandering shor algorithms , continuous shor algorithms , the quantum circle algorithm , the dual shor algorithm , a qhs algorithm for feynman integrals , free qhs algorithms , and more . toward the end of this paper , we show how grover s algorithm is most surprisingly almost a qhs algorithm , and how this suggests the possibility of an even more complete `` algorithmic tookit '' beyond the qhs algorithms . before discussing how shor s algorithm can be decomposed into its primitive components , let s take a quick look at an example of the execution of shor s factoring algorithm . as we discuss this example , we suggest that the reader , as an exercise , try to find the basic qhs primitives that make up this algorithm . can you see them ? shor s quantum factoring algorithm reduces the task of factoring a positive integer @xmath0 to first finding a random integer @xmath1 relatively prime to @xmath0 , and then next to determining the period @xmath2 of the following function @xmath3{ccl}\mathbb{z } & \overset{\varphi}{\longrightarrow } & \mathbb{z}\operatorname{mod}n\\ x & \longmapsto & a^{x}\operatorname{mod}n\text { , } \end{array}\ ] ] where @xmath4 denotes the additive group of integers , and where @xmath5 denotes the integers @xmath6 under multiplication with @xmath7 is found by selecting a random integer , and then applying the euclidean algorithm to determine whether or not it is relatively prime to @xmath0 . if not , then the @xmath8 is a non - trivial factor of @xmath0 , and there is no need to proceed futher . however , this possibility is highly unlikely if @xmath0 is large . ] . since @xmath4 is an infinite group , shor chooses to work instead with the finite additive cyclic group @xmath9 of order @xmath10 , where @xmath11 and with the `` approximating '' map @xmath3{ccll}\mathbb{z}_{q } & \overset{\widetilde{\varphi}}{\longrightarrow } & \mathbb{z}\operatorname{mod}n & \\ x & \longmapsto & a^{x}\operatorname{mod}n\text { , } & 0\leq x < q \end{array}\ ] ] shor begins by constructing a quantum system with two quantum registers @xmath12 the left intended for holding the arguments @xmath13 of @xmath14 , the right for holding the corresponding values of @xmath14 . this quantum system has been constructed with a unitary transformation @xmath15 implementing the approximating map @xmath14 . as an example , let us use shor s algorithm to factor the integer @xmath16 , assuming that @xmath17 has been randomly chosen . thus , @xmath18 . unknown to us , the period is @xmath19 , and hence , @xmath20 . we proceed by executing the following steps : * initialize@xmath21 * apply the inverse fourier transform @xmath22 to the left register , where @xmath23 is a primitive @xmath24-th root of unity , to obtain @xmath25 * apply the unitary transformation @xmath26 to obtain @xmath27 * apply the fourier transform @xmath28 to the left register to obtain @xmath29 where @xmath30 * measure the left register . then with probability @xmath31 the state will `` collapse '' to @xmath32 with the value measured being the integer @xmath33 , where @xmath34 . a plot of @xmath35 is shown in figure 1 . ( see @xcite and @xcite for details . ) fig1.ps + * figure 1 . a plot of * @xmath36**. * * the peaks in the above plot of @xmath35 occur at the integers @xmath37 the probability that at least one of these six integers will occur is quite high . it is actually @xmath38 . indeed , the probability distribution has been intentionally engineered to make the probability of these particular integers as high as possible . and there is a good reason for doing so . the above six integers are those for which the corresponding rational @xmath39 is closest to a rational of the form @xmath40 . by closest we mean that @xmath41 in particular , @xmath42 are rationals respectively closest to the rationals @xmath43 the six rational numbers @xmath44 are `` closest '' in the sense that they are convergents of the continued fraction expansions of @xmath45 , respectively . hence , each of the six rationals @xmath44 can be found using the standard continued fraction recursion formulas . but ... , we are not searching for rationals of the form @xmath40 . instead , we seek only the denominator @xmath19 . unfortunately , the denominator @xmath19 can only be obtained from the continued fraction recursion when the numerator and denominator of @xmath40 are relatively prime . given that the algorithm has selected one of the random integers @xmath46 , the probability that the corresponding rational @xmath40 has relatively prime numerator and denominator is @xmath47 , where @xmath48 denotes the euler phi ( totient ) function . so the probability of finding @xmath19 is actually not @xmath38 , but is instead @xmath49 . as it turns out , if he repeats the algorithm @xmath50 times , we will obtain the desired period @xmath2 with probability bounded below by approximately @xmath51 . however , this is not the end of the story . once we have in our possession a candidate @xmath52 for the actual period @xmath19 , the only way we can be sure we have the correct period @xmath2 is to test @xmath52 by computing @xmath53 . if the result is @xmath54 , we are certain we have found the correct period @xmath2 . this last part of the computation is done by the repeated squaring algorithm via the expression @xmath55 where @xmath56 is the radix 2 expansion of @xmath52 . ] . now that we have taken a quick look at shor s algorithm , let s see how it can be decomposed into its primitive algorithmic components . we will first need to answer the following question : _ what is a quantum hidden subgroup algorithm ? _ but before we can answer the this question , we need to provide an answer to an even more fundamental question : _ what is a hidden subgroup problem ? _ a map @xmath57 from a group @xmath58 into a set @xmath59 is said to have * hidden subgroup structure * if there exists a subgroup @xmath60 of @xmath58 , called a * hidden subgroup * , and an injection @xmath61 , called a * hidden injection * , such that the diagram@xmath3{ccc}g & \overset{\varphi}{\longrightarrow } & s\\ \nu\searrow & & \nearrow\iota_{\mathbf{\varphi}}\\ & g / k_{\mathbf{\varphi } } & \end{array}\ ] ] is commutative . the notion generalizes in an obvious way to more complicated diagrams . ] , where @xmath62 denotes the collection of right cosets of @xmath63 in @xmath58 , and where @xmath64 is the natural surjection of @xmath58 onto @xmath65 . we refer to the group @xmath58 as the * ambient group * and to the set @xmath59 as the * target set*. if @xmath60 is a normal subgroup of @xmath58 , then @xmath66 is a group , called the * hidden quotient group * , and @xmath67 is an epimorphism , called the * hidden epimorphism*. we will call the above diagram the * hidden subgroup structure * of the map @xmath68 . ( see @xcite,@xcite . ) the underlying intuition motivating this formal definition is as follows : given a natural surjection ( or epimorphism ) @xmath64 , an `` archvillain with malice of forethought '' hides the algebraic structure of @xmath69 by intentionally renaming all the elements of @xmath65 , and `` maliciously tossing in for good measure '' some extra elements to form a set @xmath59 and a map @xmath57 . the hidden subgroup problem can be stated as follows : * hidden subgroup problem ( hsp ) . * @xmath68 _ be a map with hidden subgroup structure . the problem of determining a hidden subgroup _ @xmath63 _ of _ @xmath58 _ is called a * hidden subgroup problem ( hsp)*. an algorithm solving this problem is called a * hidden subgroup algorithm*. _ the corresponding quantum form of this hsp is stated as follows : * hidden subgroup problem ( quantum version ) . * @xmath57 _ be a map with hidden subgroup structure . construct a quantum implementation of the map _ @xmath70 _ as follows : let _ @xmath71 _ and _ @xmath72 _ be hilbert spaces defined respectively by the orthonormal bases _ @xmath73__and _ _ @xmath74 _ and let _ @xmath75 _ _ , where _ _ @xmath54 _ denotes the identity . ] of the ambient group _ @xmath76__. finally , let _ _ @xmath77 _ _ be a unitary transformation such that__@xmath3{ccc}\mathcal{h}_{g}\otimes\mathcal{h}_{s } & \longrightarrow & \mathcal{h}_{g}\otimes\mathcal{h}_{s}\\ \left\vert g\right\rangle \left\vert s_{0}\right\rangle & \mapsto & \left\vert g\right\rangle \left\vert \varphi(g)\right\rangle \end{array}\ ] ] _ determine the hidden subgroup _ @xmath63 _ with bounded probability of error by making as few queries as possible to the blackbox _ @xmath78__. a quantum algorithm solving this problem is called a * quantum hidden subgroup ( qhs ) algorithm*. _ _ we are now in a position to construct one of the fundamental algorithmic primitives found in shor s algorithm . let @xmath57 be a map from a group @xmath58 to a set @xmath59 with hidden subgroup structure . we assume that all representations of @xmath58 are equivalent to unitary representations . let @xmath79 denote a * complete set of distinct irreducible unitary representations * of @xmath58 . using multiplicative notation for @xmath58 , we let @xmath54 denote the * identity * of @xmath58 , and let @xmath80 denote its image in @xmath59 . finally , let @xmath81 denote the * trivial representation * of @xmath58 . if @xmath58 is abelian , then @xmath82 becomes the * dual group * of characters . the generic qhs algorithm is given below : quantum subroutine qrand@xmath83 * initialization@xmath84 * application of the inverse fourier transform @xmath85of @xmath58 to the left register@xmath86 where @xmath87 denotes the cardinality of the group @xmath58 . * application of the unitary transformation @xmath88@xmath89 * application of the fourier transform @xmath90 of @xmath58 to the left register@xmath91 where @xmath92 denotes the degree of the representation @xmath93 , where @xmath94 denotes the contragradient representation ( i.e. , @xmath95 ) , where @xmath96 , and where @xmath97 . * measurement of the left quantum register with respect to the orthonormal basis@xmath98 thus , with probability@xmath99 the resulting measured value is the entry @xmath100 , and the quantum system `` collapses '' to the state@xmath101 * step 5 . output @xmath102 , and stop . but shor s algorithm consists of more than the primitive qrand . for many ( but not all ) hidden subgroup problems ( hsps ) @xmath68 , the corresponding generic qhs algorithm qrand either is not physically implementable or is too expensive to implement physically . for example , the hsp @xmath70 is usually not physically implementable if the ambient group is infinite ( e.g. , @xmath58 is the infinite cyclic group @xmath4 ) , and is too expensive to implement if the ambient group is too large ( e.g. , @xmath58 is the symmetric group @xmath103 ) . in this case , there is a standard generic way of `` tweaking '' the hsp to get around this problem , which we will call * pushing*. let @xmath57 be a map from a group @xmath58 to a set @xmath59 . a map @xmath104 from a group @xmath105 to the set @xmath59 is said to be a * push * of @xmath70 , written @xmath106 provided there exists an epimorphism @xmath107 from @xmath58 onto @xmath105 , and a transversal be an epimorphism from a group @xmath76 to a group @xmath108 . then a transversal @xmath109 of @xmath69 is is a map @xmath110 such that @xmath111 is the identity map @xmath112 . ( it immediately follows that @xmath109 is an injection . ) in othere words , a transversal @xmath109 of an epimorphism @xmath69 is a map which maps eacl element @xmath113 of @xmath108 to an element of @xmath76 contained in the coset @xmath113 , i.e. , to a coset representative of @xmath113 . ] @xmath114 of @xmath69 such that @xmath115 , i.e. , such that the following diagram is commutative@xmath3{ccc}g & \overset{\varphi}{\longrightarrow}\quad & s\\ \quad\uparrow\tau & \quad\nearrow\widetilde{\varphi}\quad & \\ \widetilde{g } & & \end{array}\ ] ] if the epimorphism @xmath116 and the transversal @xmath109 are chosen in an appropriate way , then execution of the generic qhs subroutine with input @xmath117 , i.e. , execution of @xmath118 will with high probability produce an irreducible representation @xmath119 of the group @xmath105 which is sufficiently close to an irreducible representation @xmath93 of the group @xmath58 . if this is the case , then there is a polynomial time classical algorithm which upon input @xmath119 produces the representation @xmath93 . obviously , much more can be said about pushing . but unfortunately that would take us far afield from the objectives of this paper . for more information on pushing , we refer the reader to @xcite . it would be amiss not to mention that the above algorithmic primitive of pushing suggests the definition of a second primitive which we will call * lifting*. let @xmath57 be a map from a group @xmath58 to a set @xmath59 . a map @xmath120 from a group @xmath121 to the set @xmath59 is said to be a * lift * of @xmath70 , written @xmath122 provided there exists an morphism @xmath123 from to @xmath58 such that @xmath124 , i.e. , such that the following diagram is commutative@xmath3{cl}\underline{g } & \\ \ \ \eta\downarrow\quad & \quad\searrow\underline{\varphi}\quad\\ g & \quad\overset{\varphi}{\longrightarrow}\quad s \end{array}\ ] ] fig2.eps + * figure 2 . pushing and lifting hsps . * we are now in position to describe shor s algorithm in terms of its primitive components . in particular , we are now in a position to see that shor s factoring algorithm is a classic example of a qhs algorithm created from the push of an hsp . let @xmath0 be the integer to be factored . let @xmath4 denote the additive group of integers , and @xmath125 denote the integers @xmath6 under multiplication . shor s algorithm is a qhs algorithm that solves the following hsp @xmath3{rrc}\varphi:\mathbb{z } & \longrightarrow & \mathbb{z}_{n}^{\times}\\ m & \longmapsto & a^{m}\operatorname{mod}n \end{array}\ ] ] with unknown hidden subgroup structure given by the following commutative diagram @xmath3{ccc}\mathbb{z } & \overset{\varphi}{\longrightarrow } & \mathbb{z}_{n}^{\times}\\ \nu\searrow & & \nearrow\iota\\ & \mathbb{z}/p\mathbb{z } & \end{array } \text { \ , } \ ] ] where @xmath1 is an integer relatively prime to @xmath0 , where @xmath2 is the hidden integer period of the map @xmath126 , where @xmath127 is the additive subgroup of all integer multiples of @xmath2 ( i.e. , the hidden subgroup ) , where @xmath128 is the natural epimorpism of the integers onto the quotient group @xmath129 ( i.e. , the hidden epimorphism ) , and where @xmath130 is the hidden monomorphism . an obstacle to creating a physically implementable algorithm for this hsp is that the domain @xmath4 of @xmath70 is infinite . as observed by shor , a way to work around this difficulty is to push the hsp . in particular , as illustrated by the following commutative diagram@xmath3{ccl}\mathbb{z\qquad } & \overset{\varphi}{\longrightarrow } & \qquad\mathbb{z}_{n}^{\times}\\ \mu\searrow\nwarrow\tau & & \nearrow\varphi = push\left ( \varphi\right ) = \varphi\circ\tau\\ & \mathbb{z}_{q } & \end{array } \text { \ \ , } \ ] ] a push @xmath117 is constructed by selecting the epimorphism @xmath131 of @xmath4 onto the finite cyclic group @xmath9 of order @xmath132 , where the integer @xmath132 is the unique power of @xmath133 such that @xmath134 , and then choosing the transversal is an injection @xmath135 such that @xmath136 is the identity map on @xmath9 , i.e. , a map that takes each element of @xmath9 onto a coset representative of the element in @xmath4 . ] @xmath3{rrc}\tau:\mathbb{z}_{q } & \longrightarrow & \mathbb{z}\\ m\operatorname{mod}q & \longmapsto & m \end{array } \text { \ , } \ ] ] where @xmath137 . _ this push _ @xmath138 _ is called _ * shor s oracle*. shor s algorithm consists in first executing the quantum subroutine qrand@xmath139 , thereby producing a random character @xmath140 of the finite cyclic group @xmath9 . the transversal @xmath109 used in pushing has been engineered in such a way as to assure that the character @xmath141 is sufficiently close to a character @xmath142 of the hidden quotient group @xmath143 . in this case `` sufficiently close '' means that @xmath144 which means that @xmath40 is a continued fraction convergent of @xmath39 , and thus can be found found by the classical polynomial time continued fraction algorithm and @xmath145 can in the obvious way be identified with points of in the unit circle in the complex plane . with this identification , we can see that this inequalty is equivalent to saying the the chordal distance bewteen these two rational points on the unit circle is less than or equan to @xmath146 . hence , shor s algorithm is using the topology of the unit circle . ] . now let s use the primitives described in sections 3 , 4 , and 5 to create other new qhs algorithms , called wandering shor algorithms . wandering shor algorithms are essentially qhs algorithms on free abelian finite rank @xmath147 groups @xmath76 which , with each iteration , first select a random cyclic direct summand @xmath4 of the group @xmath76 , and then apply one iteration of the standard shor algorithm to produce a random character of the approximating finite group @xmath148 , called a * * group probe * * , we mean an epimorphic image of the ambient group @xmath76 . ] . three different wandering shor algorithms are created in @xcite . the first two wandering shor algorithms given in @xcite are quantum algorithms which find the order @xmath2 of a maximal cyclic subgroup of the hidden quotient group @xmath149 . the third computes the entire hidden quotient group @xmath149 . the first step in creating a wandering shor algorithm is to find the right generalization one of the primitives found in shor s algorithm , namely , the transversal @xmath150 of shor s factoring algorithm . in other words , we need to construct the `` correct '' generalization of the transversal from @xmath9 to a free abelian group @xmath76 of rank @xmath147 . for this reason , we have created the following definition : let @xmath76 be the free abelian group of rank @xmath147 , let @xmath151 onto the cyclic group @xmath9 of order @xmath132 with selected generator @xmath152 . a transversal be an epimorphism from a group @xmath76 to a group @xmath108 . then a transversal @xmath109 of @xmath69 is is a map @xmath110 such that @xmath153 is the identity map @xmath154 . ( it immediately follows that @xmath109 is an injection . ) in othere words , a transversal @xmath109 of an epimorphism @xmath69 is a map which maps eacl element @xmath113 of @xmath108 to an element of @xmath76 contained in the coset @xmath113 , i.e. , to a coset representative of @xmath113 . ] @xmath155 of @xmath69 is said to be a * shor transversal * provided that : * @xmath156 for all @xmath157 * for each ( free abelian ) basis @xmath158 of @xmath76 , the coefficients @xmath159 of @xmath160 satisfy @xmath161 . later , when we construct a generalization of shor transversals to free groups of finite rank @xmath147 , we will see that the first condition simply states that a shor transversal is nothing more than a 2-sided schreier transversal . the second condition of the above definition simply says that @xmath162 maps the generator @xmath152 of @xmath9 onto a generator of a free direct summand @xmath4 of @xmath76 . ( for more details , please refer to section 12 of this paper . ) in @xcite , we show how to use the extended euclidean algorithm to construct the epimorphism @xmath163 and the transversal @xmath155 . fig3.eps + * figure 3 . flowchart for the first wandering shor algorithm ( a.k.a . , a vintage shor algorithm ) . this algorithm finds the order * @xmath2 * * of a maximal cyclic subgroup of the hidden quotient group * * @xmath149**. * * flow charts for the three wandering shor algorithms created in @xcite are given in figures 3 through 5 . in @xcite , these were also called * vintage shor algorithms*. fig4.eps + * figure 4 . flowchart for the second wandering shor algorithm ( a.k.a . , a vintage shor algorithm ) . this algorithm finds the order * @xmath2 * * of a maximal cyclic subgroup of the hidden quotient group * * @xmath149**. * * fig5.eps + * * figure 5 . flowchart for the third wandering shor algorithm , a.k.a . , a vintage shor algorithm . this algorithm finds the entire hidden quotient group * * @xmath149*. * the algorithmic complexities of the above wandering shor algorithms is given in @xcite . for example , the first wandering shor algorithm is of time complexity@xmath164 where @xmath147 is the rank of the free abelian group @xmath76 . this can be readily deduced from the abbreviated flowchart given in figure 6 . fig6.ps + * figure 6 . abbreviated flowchart for the first wandering shor algorithm . * in in @xcite and in @xcite , the algorithmic primitives found in above sections of this paper were used to create a class of algorithms called continuous shor algorithms . by a * continuous variable shor algorithm * , we mean a quantum hidden subgroup algorithm that finds the hidden period @xmath2 of an admissible function @xmath165 from the reals @xmath166 to itself . by an admissible function , we mean a function belonging to any sufficiently well behaved class of functions . for example , the class of functions which are lebesgue integrable on every closed interval of @xmath166 . there , are many other classes of functions that work equally as well . actually , the papers @xcite , @xcite give in succession three such continuous shor algorithms , each successively more general than the previous . for the first algorithm , we assume that the unknown hidden period @xmath2 is an integer . the algorithm is then constructed by using rigged hilbert spaces@xcite , @xcite , linear combinations of dirac delta functions , and a subtle extension of the fourier transform found in the generic qhs subroutine qrand@xmath167 , which has been described previously in section 4 of this paper . in step 5 of qrand@xmath167 , the observable@xmath168 is measured , where @xmath132 is an integer chosen so that @xmath169 . it then follows that the output of this algorithm is a rational @xmath170 which is a convergent of the continued fraction expansion of a rational of the form @xmath171 . the above quantum algorithm is then extended to a second quantum algorithm that finds the hidden period @xmath2 of functions @xmath165 , where the unknown period @xmath2 is a rational . finally , the second algorithm is extended to a third algorithm which finds the hidden period @xmath2 of functions @xmath165 , when @xmath2 is an _ arbitrary real number_. we point out that for the third and last algorithm to work , we must impose a very restrictive condition on the map @xmath165 , i.e. , the condition that the map @xmath70 is continuous . we have shown in previous sections how the mathematical primitives of pushing and lifting can be used to create new quantum algorithms . in particular , we have described how pushing and lifting can be used to derive new hsps from an hsp @xmath57 on an arbitrary group @xmath58 . we now see how group duality can be exploited by these two primitives to create even more quantum algorithms . fig7.eps + * figure 7 . using duality to create new qhs algorithms . * to this end , we assume that @xmath58 is an _ abelian _ group . hence , its dual group of characters @xmath79 exists is non - abelian , then its dual is not a group , but instead the representation algebra @xmath172 over the group ring @xmath173 . the methods described in this section can also be used to create new quantum algorithms for hsps @xmath174 on on the representation algebra @xmath172 . ] . it now follows that pushing and lifting can also be used to derive new hsps from an arbitrary hsp @xmath175 on the dual group @xmath79 . in @xcite , this method is used to create a number of new quantum algorithms derived from shor - like hsps @xmath176 . a roadmap is shown in figure 8 of the developmental steps taken to find and to create a new qhs algorithm on @xmath9 , which is ( in the sense described below ) dual to shor s original algorithm . we call the algorithm developed in the final step of figure 8 the * dual shor algorithm*. fig8.eps + * figure 8 . roadmap for creating the dual shor algorithm . * as indicated in figure 5 , our first step is to create an intermediate qhs algorithm based on a shor - like hsp @xmath176 from the additive group of integers @xmath4 to a target set @xmath59 . the resulting algorithm `` lives '' in the infinite dimensional space @xmath177 defined by the orthonormal basis @xmath178 . this is a physically unemplementable quantum algorithm created as a first steping stone in our algorithmic development sequence . intuitively , this algorithm can be viewed as a `` distillation '' or a `` purification '' of shor s original algorithm . as a next step , * duality * is used to create the * quantum circle algorithm*. this is accomplished by devising a qhs algorithm for an hsp @xmath179 on the dual group @xmath180 of the additive group of integers @xmath4 . ( by @xmath180 , we mean the * additive group of reals * @xmath181 , which is isomorphic to the multiplicative group @xmath182 , i.e. , the * unit circle * in the complex plane . ) once again , this is probably a physcally unemplementable quantum algorithm . but its utility lies in the fact that it leads to the physically implementable quantum algorithm created in the last and final developmental step , as indicated in figure 8 . for in the final step , a physically implementable qhs algorithm is created by * lifting * the hsp @xmath179 to an hsp @xmath183 . for the obvious reason , we call the resulting algorithm a * dual shor algorithm*. for detailed descriptions of each of these quantum algorithms , i.e. , the `` distilled '' shor , the quantum circle , and the dual shor algorithms , the reader is referred to @xcite and @xcite . we give below brief descriptions of the quantum circle and the dual shor algorithms . for the * quantum circle algorithm * , we make use of the following spaces ( each of which is used in quantum optics ) : * the rigged hilbert space @xmath184 with orthonormal basis @xmath185 . by orthonormal we mean that @xmath186 , where @xmath187 denotes the dirac delta function . the elements of @xmath184 are * formal integrals * of the form @xmath188 . ( the physicist dirac in his classic book@xcite on quantum mechanics refers to these integrals as infinite sums . see also @xcite and @xcite . ) * the complex vector space @xmath189 of formal sums@xmath190 with orthonormal basis @xmath191 . by orthonormal we mean that @xmath192 , where @xmath193 denotes the kronecker delta . we can now design an algorithm which solves the following hidden subgroup problem : * hidden subgroup problem for the circle . * _ let _ @xmath194 _ be an admissible function from the circle group _ @xmath180 _ to the complex numbers _ @xmath195 _ with hidden rational period _ @xmath196 _ _ , where _ _ @xmath197 _ denotes the rational circle , i.e. , the rationals _ @xmath198__. _ _ by an admissible function , we mean a function belonging to any sufficiently well behaved class of functions . for example , the class of functions which are lebesgue integrable on @xmath180 . there , are many other classes of functions that work equally as well . if @xmath199 ( with @xmath200 ) is a rational period of a function @xmath201 , then @xmath202is also a period of @xmath203 . hence , the minimal rational period of @xmath203 is always a reciprocal integer @xmath198 . the following quantum algorithm finds the reciprocal integer period of the function @xmath203 . circle - algorithm@xmath204 * initialization@xmath205 * application of the inverse fourier transform @xmath206@xmath207 * step 2 . application of the unitary transformation @xmath208@xmath209 * application of the fourier transform @xmath210 remark . letting @xmath211 , we have@xmath212 where @xmath213 is the unknown reciprocal period . but@xmath214{ll}a & \text{if \ } n=0\operatorname*{mod}a\\ 0 & \text{otherwise}\end{array } \right.\ ] ] hence,@xmath215 * measurement of@xmath216 with respect to the observable@xmath217 to produce a random eigenvalue @xmath218 . the above quantum circle algorithm can be extended to a quantum algorithm which finds the hidden period @xmath219 of a function @xmath194 , when @xmath219 is an arbitrary real number @xmath198 . but in creating this extended quantum algorithm , a very restrictive condition must be imposed on the map @xmath194 , namely , the condition that @xmath203 be continuous . we now give a brief description of the * dual shor algorithm*. the dual shor algorithm is a qhs algorithm created by making a discrete approximation of the quantum circle algorithm . more specifically , it is created by lifting the qhs circle algorithm for @xmath220 to the finite cyclic group @xmath9 , as illustrated in the commutative diagram given below:@xmath3{ll}\;\,\mathbb{z}_{q } & \\ \mu\downarrow\;\ ; & \;\;\searrow\widetilde{\varphi}=push\left ( \varphi \right ) = \varphi\circ\mu\\ \mathbb{r}\mathbf{/}\mathbb{z } & \longrightarrow\;\;s \end{array}\ ] ] intuitively , just as in shor s algorithm , the circle group @xmath180 is `` approximated '' with the finite cyclic group @xmath9 , where the group @xmath9 is identified with the additive group@xmath221 and where the hidden subgroup @xmath222 is identified with the additive group@xmath223 with @xmath224 . this is a physically implementable quantum algorithm . in a certain sense , it is actually faster than shor s algorithm . for the last step of shor s algorithm uses the standard continued fraction algorithm to determine the unknown period . on the other hand , the last step of the dual shor algorithm uses the much faster euclidean algorithm to compute the greatest common divisor of the integers @xmath225 , thereby determining the desired reciprocal integer period @xmath213 . for more details , please refer to @xcite and @xcite . we now discuss a qhs algorithm based on feynman path integrals . this quantum algorithm was developed at the mathematical sciences research institute ( msri ) in berkeley , california when one of the authors of this paper was challenged with an invitation to give a talk on the relation between feynmann path integrals and quantum computing at an msri conference on feynman path integrals . until recently , both authors of this paper thought that the quantum algorithm to be described below was a highly speculative quantum algorithm . for the existence of feynman path integrals is very difficult ( if not impossible ) to determine in a mathematically rigorous fashion . but surprisingly , jeremy becnel in his doctoral dissertation @xcite actually succeeded in creating a firm mathematical foundation for this algorithm . we should mention , however , that the physical implementability of this algorithm is still yet to be determined . definition . let paths be the real vector space of all continuous paths @xmath226 \longrightarrow\mathbb{r}^{n}$ ] which are @xmath227 with respect to the inner product@xmath228 with scalar multiplication and vector sum defined as * @xmath229 * @xmath230 we wish to create a qhs algorithm for the following hidden subgroup problem : * hidden subgroup problem for * paths . @xmath231paths@xmath232 _ be a functional with a hidden subspace _ @xmath233 _ of _ paths _ _ such that__@xmath234 our objective is to create a qhs algorithm which solves the above problem , i.e. , which finds the hidden subspace @xmath233 . let @xmath235 be the rigged hilbert space with orthonormal basis @xmath236 , and with bracket product @xmath237 . we will use the following observation to create the qhs algorithm : * observation . * paths@xmath238 _ , where _ @xmath239 _ denotes the orthogonal complement of the hidden vector subspace _ @xmath233__. _ _ the qhs algorithm for feynman path integral is given below : feynman@xmath83 * initialize@xmath240 * apply @xmath206@xmath241 * apply @xmath242@xmath243 * apply @xmath210@xmath3{lll}\left\vert \psi_{3}\right\rangle & = & { \displaystyle\int\limits_{\text{\textsc{paths } } } } \mathcal{d}y{\displaystyle\int\limits_{\text{\textsc{paths } } } } \mathcal{d}x\ e^{-2\pi ix\cdot y}\left\vert y\right\rangle \left\vert \varphi\left ( x\right ) \right\rangle \\ & = & { \displaystyle\int\limits_{\text{\textsc{paths } } } } \mathcal{d}y\ \left\vert y\right\rangle { \displaystyle\int\limits_{\text{\textsc{paths } } } } \mathcal{d}x\ e^{-2\pi ix\cdot y}\left\vert \varphi\left ( x\right ) \right\rangle \end{array}\ ] ] but @xmath244 however,@xmath245 so , @xmath246 * measure@xmath247 with respect to the observable@xmath248 to produce a random element of @xmath239 the above algorithm suggests an intriguing question . can the above qhs feynman integral algorithm be modified in such a way as to create a quantum algorithm for the jones polynomial ? in other words , can it be modified by replacing paths with the space of gauge connections , and making suitable modifications ? this question is motivated by the fact that the integral over gauge transformations@xmath249 looks very much like a fourier transform , where@xmath250 denotes the * wilson loop * over the knot @xmath251 . in this and the following section of this paper , our objective is to show that a free group is the the most natural domain for qhs algorithms . in retrospect , this is not so surprising if one takes a discerning look at shor s factoring algorithm . for in section 6 , we have seen that shor s algorithm is essentially a qhs algorithm on the free group @xmath4 which has been pushed onto the finite group @xmath9 . in particular , let @xmath57 be a map with hidden subgroup structure from a finitely generated ( f.g . ) group @xmath58 to a set @xmath59 . we assume that the hidden subgroup @xmath251 is a normal subgroup of @xmath58 of finite index . then the objectives of this section are to demonstrate the following : * every hidden subgroup problem ( hsp ) @xmath57 on an arbitray f.g . group @xmath58 can be lifted to an hsp @xmath252 on a free group @xmath253 of finite rank . * moreover , a solution for the lifted hsp @xmath252 is for all practical purposes the same as the solution for the original hsp @xmath57 . _ thus , one need only investigate qhs algorithms for free groups of finite rank ! _ before we can describe the above results , we need to review a number of definitions . we begin with the definition of a free group : [ universal definition]a group @xmath253 is said to be * free * of finite rank @xmath147 if there exists a finite set of @xmath147 generators @xmath254 such that , for every group @xmath58 and for every map @xmath255 of the set @xmath256 into the group @xmath58 , the map @xmath257 extends to a morphism @xmath258 . we call the set @xmath256 a * free basis * of the group @xmath253 , and frequently denote the group @xmath253 by @xmath259 , . it follows from this definition that the morphism @xmath260 is unique . the intuitive idea encapsulated by this definition is that a free group is an unconstrained group ( very much analogous to a physical system without boundary conditions . ) in other words , a group is free provided it has a set of generators such that the only relations among those generators are those required for @xmath253 to be a group . for example , * @xmath261 is an allowed relation * @xmath262 is not an allowed relation for @xmath263 * @xmath264 is not an allowed relation as an immediate consequence of the above definition , we have the following proposition : let @xmath58 be an arbitrary f.g . group with finite set of @xmath147 generators @xmath265 , and let @xmath266 be the free group of rank @xmath147 with free basis @xmath267 . then by the above definition , the map @xmath268 @xmath269 induces a uniques epimorphism @xmath270 from @xmath253 onto @xmath58 . with this epimorphism , every hsp @xmath68 on the group @xmath58 uniquely lifts to the hsp @xmath271 on the free group @xmath253 . moreover , if @xmath251 and @xmath272 are the hidden subgroups of the hsps @xmath70 and @xmath273 , respectively , the corresponding hidden quotient groups @xmath274 and @xmath275 of these two hsps are isomorphic . hence , every solution of the hsp @xmath252 immediately produces a solution of the original hsp @xmath57 . we close this section with the defintion of a group resentation , a concept that will be needed in the next section for generalizing shor s algorithm to free groups . let @xmath58 be a group . a * * group presentation**@xmath276 for @xmath58 is a set of free generators @xmath277 of a free group @xmath253 and a set of words @xmath278 in @xmath259 , called * relators * , such that the group @xmath58 is isomorphic to the quotient group @xmath279 , where @xmath280 , called the * consequence * of @xmath278 , is the smallest normal subgroup of @xmath259 containing the relators @xmath278 . the intuition captured by the above definition is that @xmath277 are the generators of @xmath58 , and @xmath281 is a complete set of relations among these generators , i.e. , every relation among the generators of @xmath58 is a * consequence * of ( derivable from ) the relations @xmath281 . for example , * @xmath282 and @xmath283 are both presentations of the free group @xmath259 * @xmath284 and @xmath285 are both presentations of the cyclic group @xmath9of order @xmath132 , where @xmath1 and @xmath113 are integers such that @xmath286 . * @xmath287 is a presentation of the symmetric group @xmath288 on three symbols . the objective of this section is to generalize shor s algorithm to free groups of finite rank of rank 1 constructed by a push onto the cyclic group @xmath9 . in light of this and of the results outlined in the previous section , it is a natural objective to generalize shor s algorithm to free groups of finite rank . ] . the chief obstacle to accomplishing this goal is finding a correct generalization of the shor transversal @xmath3{cccc}\mathbb{z}_{q } & \overset{\tau}{\longrightarrow } & \mathbb{z } & \\ n\operatorname{mod}q & \longmapsto & n & \left ( \text{\ } 0\leq n < q\right ) \end{array}\ ] ] unfortunately , there appear to be few mathematical clues indicating how to go about making such a generalization . however , as we shall see , the generalization of the shor transversal to the transversal found in the wandering shor algorithm does provide a crucial clue , suggesting that a generalized shor transversal must be a 2-sided schreier transversal . ( see section 7 . ) we begin by formulating a constructive approach to free groups : let @xmath259 be a free group with free basis @xmath277 . then a * word * is a finite string of the symbols @xmath289 . a * reduced word * is a word in which there is no substring of the form @xmath290 or @xmath291 . two words are said to be * equivalent * if one can be transformed into the other by applying a finite number of substring insertions or deletions of the form @xmath290 or @xmath291 . we denote an * arbitrary word * @xmath292 by @xmath293 , where each @xmath294 . the * length * @xmath295 of a word @xmath296 is number of symbols @xmath297 that appear in @xmath292 , i.e. , @xmath298 . for example , @xmath299 is a word of length @xmath300 which is equivalent to the reduced word @xmath301 of length @xmath302 . it easily follows that : a free group @xmath259 is simply the set of reduced words together with the obvious definition of product , i.e. , concatenation followed by full reduction . we can now use this constructive approach to create a special kind of transversal @xmath303 of an epimorphism @xmath270 , called a 2-sided scheier transversal@xcite : a set @xmath304 of reduced words in a free group @xmath266 is said to be a * 2-sided schreier system * provided * the empty word @xmath54 lies in @xmath304 . * @xmath305 , and * @xmath306 given an epimorphism @xmath307 of the free group @xmath253 onto a group @xmath58 , a * 2-sided schreier transversal * @xmath308 for @xmath69 is a transversal of @xmath69 for which there exists a 2-sided schreier system such that @xmath309 . a 2-sided schreier transversal is said to be * minimal * provided the length of each word @xmath292 is less than or equal to the length of each reduced word in the coset @xmath310 , where @xmath311 denotes the kernel of the epimorpism @xmath69 . the wandering shor algorithm found in section 7 suggests that a correct generalization of the shor transversal @xmath312 @xmath313 must at least have the property that it is a minimal 2-sided schreier transversal . whatetever other additional properties this generalization must have is simply not clear . in @xcite , we construct and investigate a number of different qhs algorithms on free groups that arise from the application of various additional conditions imposed upon the minimal 2-sided schreier transversal requirement . in this section , we only give a descriptive sketch of the simplest of these algorithms , i.e. , a qhs algorithm on free groups with only the minimal 2-sided schreier transversal requirement imposed . let @xmath266 be the free group of finite rank @xmath147 with free basis @xmath254 , and let @xmath314 be an hsp on the free group @xmath253 . we assume that the hidden subgroup @xmath251 is normal and of finite index in @xmath253 . ( please note that @xmath315 . ) * choose a finite group probe @xmath58 with presentation @xmath316 , where the subscript @xmath69 denotes the epimorphism @xmath307 induced by the map @xmath317 . * choose a minimal 2-sided schreier transversal @xmath303 of the epimorphism @xmath307 . * finally , construct the push @xmath318 our generalized shor algorithm for the free group @xmath253 consists of the following steps : * call qrand@xmath319 to produce a word @xmath320 in @xmath253 close to a word @xmath321 lying in @xmath322 . * with input @xmath320 , use a polytime classical algorithm to determine @xmath321 . ( see @xcite . ) * repeat steps 1 and 2 until enough relators @xmath321 s are found to produce a presentation @xmath323 of the hidden subgroup @xmath324 , then output the presentation @xmath325 , and stop . obviously , much more needs to be said . for , example , we have not explained how one chooses the relators @xmath326 so that @xmath327 is a good group probe . moreover , we have not explained what classical algorithm is used to transform the words @xmath320 into the relators @xmath321 . for more details , we refer the reader to @xcite . in this section , our objective is to factor grover s algorithm into the qhs primitives developed in the previous sections of this paper . as a result , we will show that grover s algorithm is more closely related to shor s algorithm than one might at first expect . in particular , we will show that grover s algorithm is a qhs algorithm in the sense that it solves an hsp @xmath328 , which we will refer to as the * grover hsp . * however , we will then show that the standard qhs algorithm for this hsp can not possibly find a solution . we begin with a question : _ does grover s algorithm have symmetries that we can exploit ? _ the problem solved by grover s algorithm @xcite , @xcite , @xcite , @xcite is that of finding an unknown integer label @xmath329 in an unstructured database with items labeled by the integers : @xmath330 given the oracle@xmath331{ll}1 & \text{if \ } j = j_{0}\\ 0 & \text{otherwise}\end{array } \right.\ ] ] let @xmath332 be the hilbert space with orthonormal basis @xmath333 . grover s oracle is essentially given by the unitary transformation@xmath334 where @xmath335 is inversion in the hyperplane orthogonal to @xmath336 . let @xmath337 denote the hadamard transformation on the hilbert space @xmath338 . then grover s algorithm is as follows : * ( initialization)@xmath339 * loop until @xmath340@xmath341 * measure @xmath342 with respect to the standard basis@xmath343 to obtain the unknown state @xmath344 with@xmath345 but where is the hidden symmetry in grover s algorithm ? let @xmath346 be the symmetric group on the symbols @xmath347 . then grover s algorithm is invariant under the * hidden subgroup * @xmath348,called the * stabilizer subgroup * for @xmath329 , i.e. , grover s algorithm is invariant under the group action@xmath349 moreover , if we know the hidden subgroup @xmath350 , then we know @xmath329 , and vice versa . in other words , the problem of finding the unknown label @xmath329 is informationally the same as the problem of finding the hidden subgroup @xmath350 . let @xmath351 denote the permutation that interchanges integers @xmath352 and @xmath353 , and leaves all other integers fixed . thus , @xmath354is a transposition if @xmath263 , and the identity permutation @xmath54 if @xmath355 . the set @xmath356 is a complete set of distinct coset representatives for the hidden subgroup @xmath350 of @xmath346 , i.e. , the coset space @xmath357is given by the following complete set of distinct cosets:@xmath358 we can now see that grover s algorithm is a hidden subgroup algorithm in the sense that it is a quantum algorithm which solves the following hidden subgroup problem : * grover s hidden subgroup problem . @xmath359 _ be a map from the symmetric group _ @xmath346 _ to a set _ @xmath360 _ _ with hidden subgroup structure given by the commutative diagram__@xmath3{ccc}s_{n } & \longrightarrow & \quad\;\;s\\ \nu_{j_{0}}\searrow & & \nearrow\iota\\ & s_{n}/stab_{j_{0 } } & \end{array } \text { \ , } \ ] ] _ where _ @xmath361 _ is the natural surjection of _ _ on to the coset space _ @xmath357 _ _ , and where _ _ @xmath3{ccc}\iota:\;s_{n}/stab_{j_{0 } } & \longrightarrow & s\\ \quad\left ( jj_{0}\right ) stab_{j_{0 } } & \longmapsto & j \end{array}\ ] ] _ is the unknown relabeling ( bijection ) of the coset space _ @xmath357__onto the set _ _ @xmath59__. find the hidden subgroup _ _ @xmath350__with bounded probability of error . _ _ now let us compare shor s algorithm with grover s . from section 6 , we know that shor s algorithm @xcite , @xcite , @xcite , @xcite solves the hidden subgroup problem @xmath362 with hidden subgroup structure@xmath3{ccc}\mathbb{z } & \longrightarrow & \quad\;\;\mathbb{z}_{n}\\ \;\;\nu\searrow & & \nearrow\iota\\ & \mathbb{z}/p\mathbb{z } & \end{array}\ ] ] moreover , as stated in section 6 , shor has created his algorithm by pushing the above hidden subgroup problem @xmath362 to the hidden subgroup problem @xmath363 ( called shor s oracle ) , where the hidden subgroup structure of @xmath14 is given by the commutative diagram@xmath3{ccc}z & \longrightarrow & \quad\;\;z_{n}\\ \qquad\;\;\alpha\searrow\nwarrow\tau & & \nearrow\widetilde{\varphi}=\varphi\circ\tau\\ & z_{q } & \end{array } \text { \ , } \ ] ] where @xmath219 is the natural epimorphism of @xmath4 onto @xmath9 , and where @xmath109 is shor s chosen transversal for the epimorphism @xmath219 . surprisingly , grover s algorithm , viewed as an algorithm that solves the grover hidden subgroup problem , is very similar to shor s algorithm . like shor s algorithm , grover s algorithm solves a hidden subgroup problem , i.e. , the grover hidden subgroup problem @xmath359 with hidden subgroup structure@xmath3{ccc}s_{n } & \longrightarrow & \quad\;\;s\\ \;\;\nu\searrow & & \nearrow\iota\\ & s_{n}/stab_{j_{0 } } & \end{array } \text { \ , } \ ] ] where @xmath360 denotes the set resulting from an unknown relabeling ( bijection ) @xmath364 of the coset space@xmath365 also , like shor s algorithm , we can think of grover s algorithm as one created by pushing the grover hidden subgroup problem @xmath359 to the hidden subgroup problem @xmath366 , where the pushing is defined by the following commutative diagram@xmath3{ccc}s_{n } & \longrightarrow & \qquad\qquad\;\;s = s_{n}/stab_{j_{0}}\\ \qquad\;\;\alpha\searrow\nwarrow\tau & & \nearrow\widetilde{\varphi}=\varphi\circ\tau\\ & s_{n}/stab_{0 } & \end{array } \text { \ \ , } \ ] ] where @xmath367 denotes the natural surjection of @xmath346 onto the coset space @xmath368 , and where @xmath369denotes the transversal of @xmath219 given by@xmath3{ccc}s_{n}/stab_{0 } & \longrightarrow & s_{n}\\ \left ( j0\right ) stab_{0 } & \longmapsto & \left ( j0\right ) \end{array } \text { \ \ .}\ ] ] again also like shor s algorithm , the map @xmath14 given by @xmath370 is ( if @xmath371 ) actually a disguised grover s oracle . for the map @xmath14 can easily be shown to simply to@xmath372{ll}(j0)stab_{j_{0 } } & \text{if \ } j = j_{0}\\ stab_{j_{0 } } & \text{otherwise\ \ , } \end{array } \right.\ ] ] which is informationally the same as grover s oracle@xmath331{ll}j & \text{if \ } j = j_{0}\\ 1 & \text{otherwise}\end{array } \right.\ ] ] hence , we can conclude that grover s algorithm is a quantum algorithm very much like shor s algorithm , in that it is a quantum algorithm that solves the grover hidden subgroup problem . however , , this appears to be where the similarity between grover s and shor s algorithms ends . for the standard non - abelian qhs algorithm for @xmath346 can not find the hidden subgroup @xmath350 for each of following two reasons : * since the subgroups @xmath373 are not normal subgroups of @xmath346 , it follows from the work of hallgren et al @xcite , @xcite that the standard non - abelian hidden subgroup algorithm will find the largest normal subgroup of @xmath346 lying in @xmath350 . but unfortunately , the largest normal subgroup of @xmath346 lying in @xmath373 is the trivial subgroup of @xmath346 . * the subgroups @xmath374 are mutually conjugate subgroups of @xmath346 . moreover , one can not hope to use this qhs approach to grover s algorithm to find a faster quantum algorithm . for zalka @xcite has shown that grover s algorithm is optimal . the arguments given above suggest that grover s and shor s algorithms are more closely related that one might at first expect . although the standard non - abelian qhs algorithm on @xmath346 can not solve the grover hidden subgroup problem , there does remain an intriguing question : * question . * _ is there some modification of ( or extension of ) the standard qhs algorithm on the symmetric group _ @xmath346 _ that actually solves grover s hidden subgroup problem ? _ for a more in - depth discussion of the results found in this section , we refer the reader to @xcite . in this paper , we have decomposed shor s quantum factoring algorithm into primitives , generalized these primitives , and then reassembled them into a wealth of new qhs algorithms . but as the results found in the previous section suggest , this list of quantum algorithmic primitives is far from complete . this is expressed by the following question : _ where can we find more algorithmic primitives to create a more well rounded toolkit for quantum algorithmic development ? _ the previous section suggests that indeed all quantum algorithms may well be hidden subgroup algorithms in the sense that they all find hidden symmetries , i.e. , hidden subgroups . this is suggestive of the following meta - procedure for quantum algorithm development : * explicitly state the problem to be solved . * rephrase the problem as a hidden symmetry problem . * create a quantum algorithm to find the hidden symmetry . _ can this meta - procedure be made more explicit ? _ perhaps some reader to this paper will be able to answer this question . this work is partially supported by the defense advanced research projects agency ( darpa ) and air forche research laboratory , air force materiel command , usaf , under agreement number f30602 - 01 - 2 - 0522 . government is authorized to reproduce and distribute reprints for governmental purposes notwithstanding any copyright annotation thereon . this work also partially supported by the institute for scientific interchange ( isi ) , torino , the national institute of standards and technology ( nist ) , the mathematical sciences research institute ( msri ) , the isaac newton institute for mathematical sciences , and the l - o - o - p fund . hallgren , sean , alexander russell , amnon ta - shma , * the hidden subgroup problem and quantum computation using group representations , * proceedings of the thirty - second annual acm symposium on theory of computing , portland , oregon , may 2000 , 627 - 635 . lomonaco , samuel j. , jr . , ( ed . ) , * `` quantum computation : a grand mathematical challenge for thetwenty - first century and the millennium , '' * proceedings of the symposia of applied mathematics , vol . 58 , american mathematical society , providence , rhode island , ( 2002 ) . ( 358 pages)(http://www.ams.org / bookstore?fn=20&arg1=whatsnew&item = psapm-58)(http://www.csee.umbc.edu / lomonaco / ams / lecture_notes.html ) lomonaco , samuel j. , jr . , and howard e. brandt , ( eds . ) , * `` quantum computation and information , '' * ams contemporary mathematics , vol . 305 , american mathematical society , providence , ri , ( 2002 ) . ( 310 pages)(http://www.csee.umbc.edu/@xmath375lomonaco / ams / special.html ) lomonaco , samuel j. , jr . , and louis h. kauffman , * continuous quantum hidden subgroup algroithms * , spie proceedings on quantum information and computation , vol . 5105 , 11 , ( 2003 ) , 80 - 89 . ( http://arxiv.org/abs/quant-ph/0304084 ) lomonaco , samuel j. , jr . , and , louis h. kauffman , * quantum hidden subgroup algorithms : the devil is in the details , * 2004 proceedings of spie proceedings on quantum information and computation , ( 2004 ) , 137 - 141 . http://arxiv.org/abs/quant-ph/0403229 mosca , michelle , and artur ekert , * the hidden subgroup problem and eigenvalue estimation on a quantum computer , * proceedings of the 1st nasa international conference on quantum computing and quantum communication , springer - verlag , ( 2001 ) . ( http://xxx.lanl.gov/abs/quant-ph/9903071 ) shor , peter w. , * polynomial time algorithms for prime factorization and discrete logarithms on a quantum computer , * siam j. on computing , 26(5 ) ( 1997 ) , pp 1484 - 1509 . ( http://xxx.lanl.gov/abs/quant-ph/9508027 )

i present a generalized power - law diagnostic that allows to identify the presence of active galactic nuclei ( agn ) in infrared ( ir ) galaxies at , down to flux densities at which the extragalactic ir background is mostly resolved . i derive this diagnostic from the analysis of 174 galaxies with and spectroscopic redshifts in the chandra deep field south , for which i study the rest - frame uv / optical / near - ir spectral energy distributions ( seds ) , after subtracting a hot - dust , power - law component with three possible spectral indices , 2.0 and 3.0 . i obtain that 35% of these sources are power - law composite galaxies ( plcgs ) , which i define as those galaxies for which the sed fitting with stellar templates , without any previous power - law subtraction , can be rejected with confidence . subtracting the power - law component from the plcg seds produces stellar - mass correction factors in of cases . the plcg incidence is especially high ( 47% ) at . to unveil which plcgs host agn , i conduct a combined analysis of 4ms x - ray data , galaxy morphologies , and a greybody modelling of the hot dust . i find that : 1 ) 77% of all the x - ray agn in my sample at are recognised by the plcg criterion ; 2 ) plcgs with or 2.0 have regular morphologies and $ \sim$}}~}1000 \,\rm k$ ] , indicating nuclear activity . instead , plcgs with are characterised by disturbed galaxy dynamics , and a hot interstellar medium can explain their dust temperatures . overall , my results indicate that the fraction of agn among sources is between% and 52% at .

1303.1906

the study of the individual dust - obscured sources that make the extragalactic infrared ( ir ) background @xcite has made enormous progress over the last decade , since the launch of the _ spitzer space telescope _ @xcite , and successively the _ akari telescope _ @xcite and the _ herschel space observatory _ @xcite . in particular , the scientific output of these missions has revealed the importance of powerful , dust - obscured star - formation and nuclear activity in shaping galaxy evolution at high redshifts ( @xmath0 ) . this activity was dominated by luminous and ultra - luminous infrared galaxies ( lirgs and ulirgs ; * ? ? ? * ; * ? ? ? * ) , which had a main role in the global star formation history of the universe ( e.g. * ? ? ? * ) , and the process of massive galaxy buildup at @xmath0 @xcite . the availability of multi - wavelength ancillary data from deep galaxy surveys has been crucial to investigate the presence and properties of ir galaxies at high @xmath13 . most redshift estimates and the derivation of other parameters , such as the galaxy stellar mass , rely on the fitting of spectral energy distribution ( sed ) templates to broad - band photometry that traces the galaxy rest uv / optical and near - ir light ( @xmath14 $ \sim$}}~}3 \ , \rm \mu m$ ] ) . the derivation of reliable values for these galaxy parameters requires a proper wavelength coverage of the photometric data , and also considering the possible variations that the galaxy seds may have , through the choice of sufficiently representative galaxy templates , and the analysis of departures of these seds from models of pure stellar evolution . a long - standing problem regarding the composition of the extragalactic ir background is understanding to which extent dust - obscured active galactic nuclei ( agn ) are part of the ir galaxy population at high redshifts . a direct way to reveal nuclear activity is searching for x - ray detections ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but dust - obscured agn can remain undetected even in typically deep x - ray maps . spectral line diagnostics at optical and ir wavelengths are also useful to recognise agn , but they are usually limited to relatively small samples of bright ir sources ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? an alternative method that has proven to be quite efficient to reveal agn in large samples of dust - obscured galaxies is the use of mid - ir colour - colour diagrams @xcite , based on photometry at @xmath15 , taken with the _ spitzer _ infrared array camera ( irac ; * ? ? ? the segregated locus that some agn occupy in these diagrams is related to the fact that pure agn seds are characterised by a power - law shape at rest - frame optical / near - ir wavelengths . in fact , the presence of a pure irac power - law sed has been proposed by some authors as a criterion to select agn - dominated galaxies among _ spitzer _ @xmath4 and irac - selected sources @xcite . but agn selection through irac colour - colour diagrams has well - known limitations : it is complete only for bright ir sources ( e.g. * ? ? ? * ) , and the resulting samples are typically contaminated by star - forming galaxies @xcite . this is especially the case at high redshifts , as the galaxy stellar emission is shifted into the irac bands , producing similar irac colours to those of agn at lower redshifts . in addition to the identification , there is a second problem , which consists in understanding what fraction of the galaxy light is due to the agn component at different wavelengths . at mid- and far - ir wavelengths ( @xmath16 $ \sim$}}~}\lambda_{\rm rest } { \raisebox{-0.13cm}{~\shortstack{$<$ \\[-0.07 cm ] $ \sim$}}~}1000 \ , \rm \mu m$ ] ) , disentangling the star - formation / agn contributions is required to properly derive the on - going , obscured star formation rates in ir galaxies . this issue has been tackled in different studies of lirgs and ulirgs at high @xmath13 , by analysing mid - ir spectroscopic data , or mid-/far - ir broad - band data ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in the rest - frame optical / near - ir ( @xmath17 $ \sim$}}~}\lambda_{\rm rest } { \raisebox{-0.13cm}{~\shortstack{$<$ \\[-0.07 cm ] $ \sim$}}~}3 \ , \rm \mu m$ ] ) , a power - law component associated with an agn distorts the light of the underlying stellar populations in the host galaxy , and may affect the derived galaxy stellar mass . for galaxies that are identified with x - ray luminous agn , the sed fitting with stellar templates , and derivation of stellar masses are usually avoided ( e.g. * ? ? ? * ) , or attempted only after subtracting empirical agn templates ( e.g. * ? ? ? * ; * ? ? ? * ) . in all other cases , however , the derivation of stellar masses is of common practice , even when there is the suspicion that the rest - frame near - ir light could be partly contaminated by an agn component ( for example , in the case of an irac - band excess in the galaxy sed ) . quantifying the importance of this effect in ir - selected galaxies is then necessary to understand the reliability of the derived host galaxy properties . in this work i present the analysis of the rest - frame optical / near - ir seds of 174 @xmath4-selected galaxies with @xmath18 and secure spectroscopic redshifts @xmath2 . my aim is to investigate whether , and when , a power - law component makes a relevant contribution to the galaxy sed , i.e. the subtraction of a power - law component produces a significant improvement of the galaxy sed fitting with stellar templates . hereafter , i will refer to these galaxies as _ power - law - component galaxies _ ( plcgs ) . note that this criterion is more relaxed than the irac power - law shape imposed by other authors to select potential ir agn candidates @xcite , and therefore , the sample analysed here should also include galaxies for which the agn emission is less dominant over the underlying host galaxy light . this paper is organised as follows . in section [ sec - sample ] , i give details on the sample selection and compilation of multi - wavelength photometry . in section [ sec - seds ] , i present the results of the sed analysis , and quantify the effect on the derived stellar masses . later , in section [ sec - nature ] , i investigate the nature of the plcgs at @xmath0 . i make use of ultra - deep x - ray data to obtain an independent diagnostic of nuclear activity among the @xmath4 sources , and understand in which cases the significant sed power - law component is related to the presence of an agn . i also analyse the plcg morphologies , and derive the greybody temperatures that are necessary to explain the sed power law . in section [ sec - iraccol ] , i investigate how the latest proposed irac colour - colour criteria deal with the selection of plcgs and other x - ray - detected agn . finally , in section [ sec - summ ] , i summarise my findings and present some concluding remarks . all magnitudes and colours quoted in this paper are total and refer to the ab system @xcite . i adopt a cosmology with @xmath19 , @xmath20 and @xmath21 . all stellar masses refer to a salpeter ( 1955 ) initial mass function ( imf ) over star masses of @xmath22 . the great observatories origins deep survey ( goods ) programme @xcite comprises a wide range of deep galaxy surveys conducted with main astronomical facilities , including the _ spitzer _ and _ hubble space telescopes _ , the _ chandra x - ray observatory _ , and the largest optical / near - ir ground - based telescopes . in particular , the _ spitzer _ images for the goods fields have been collected as part of the goods _ spitzer _ legacy programme ( pi : m. dickinson ) , and include deep data from both irac and the multiband imaging photometer for _ spitzer _ ( mips ; * ? ? ? the 174 @xmath4-selected galaxies analysed here have been extracted from the publicly released @xmath4 catalogue of the goods - south ( goods - s ) field , which has a flux density limit of @xmath23 , down to which this catalogue is highly reliable and complete . sources down to this flux density limit make @xmath24 of the extragalactic @xmath4 background @xcite . to decide which @xmath4 sources would be part of the analysis sample , i applied the following two criteria : 1 ) the source should have a secure ( good - quality flag ) spectroscopic redshift ( @xmath25 ) from any of the multiple spectroscopic galaxy surveys of the goods - s field ( le fvre et al . 2004 ; szokoly et al . 2004 ; teplitz et al . 2007 ; vanzella et al . 2008 ; popesso et al . 2009 ; fadda et al . 2010 ; kurk et al . 2013 ) ; 2 ) the redshift should be @xmath2 . i searched for spectroscopic counterparts of the @xmath4 sources within a 1.5 arcsec matching radius , and considered only one - to - one identifications . the final sample contains 174 @xmath4 sources with @xmath2 , with similar numbers of galaxies at @xmath8 and @xmath26 ( 89 and 85 sources , respectively ) . restricting the sed study only to galaxies with secure spectroscopic redshifts ensures that all the results and conclusions in this paper are free from the uncertainties that are usually introduced by photometric redshift determinations at high redshifts . on the other hand , the multiple goods - s spectroscopic datasets correspond to a wide variety of selection criteria , and even include some ir spectra @xcite , so the sub - sample of @xmath4 galaxies with spectroscopic redshifts is reasonably representative of the entire sample of @xmath27 galaxies at @xmath0 . this has been verified by comparing the @xmath28 colour distributions of the current spectroscopic sample and the overall goods - s @xmath4 sample with @xmath0 , with spectroscopic and photometric redshifts ( caputi et al . 2006a , b ; 2007 ) . the @xmath29 flux densities versus spectroscopic redshifts @xmath25 of the 174 galaxies are shown in fig . [ fig_stfmvsz ] . the dashed curve in this plot separates the lirg and ulirg regimes at bolometric ir luminosities @xmath30 . to compute total ir luminosities , i have considered the bavouzet et al . ( 2008 ) monochromatic - to - total ir luminosity conversion , re - calibrated for galaxies with @xmath31 ( see * ? ? ? * ) . as in this work , the k - correction factors are based on a mixture of ir star - forming galaxy templates @xcite and empirical ir spectra . although , in a strict sense , this lirg / ulirg separation curve corresponds to star - forming galaxies , it is suitable for the purpose of illustrating the appoximate regions of the lirg and ulirg regimes in the @xmath29-@xmath13 diagram . independently , i made use of the publicly released goods - s irac maps and the sextractor software @xcite to extract a catalogue of irac sources at 3.6 and @xmath32 . i measured the photometry at 5.8 and @xmath33 using sextractor in dual - map mode . in order to construct a multi - wavelength optical / near - ir catalogue for the irac - selected sources , i searched for counterparts of these sources in the goods - s _ very large telescope ( vlt ) _ isaac near - ir images , and the _ hubble space telescope _ advanced camera for surveys ( acs ) maps , within a matching radius of 0.5 arcsec the finally compiled catalogue includes photometry in 11 bands : @xmath34 , @xmath35 , @xmath36 , @xmath13 , @xmath37 , @xmath38 , @xmath39 , [ 3.6 ] , [ 4.5 ] , [ 5.8 ] and [ 8.0 ] . all the magnitudes in this catalogue are total and have been corrected for galactic extinction . to obtain total magnitudes , i considered aperture magnitudes in circles of 4-arcsec ( irac ) and 2-arcsec ( isaac and acs ) diameter , and applied the corresponding aperture corrections . within the catalogue of irac - selected sources with multi - wavelength photometry , i identified all the 174 @xmath4 galaxies with spectroscopic redshifts @xmath2 . the sed fitting analysis explained in next section is based on this photometry . a total of 22 galaxies ( 12.5% of the sample ) are out of the field of view in all the acs bands , or at least one of the isaac bands . for this minority of sources , the sed fitting is based on 7,8,9 or 10 bands , depending on the case . for all the remaining galaxies , the sed fitting is based on the 11-band photometry . in case of a non - detection , a 2@xmath40 upper limit has been considered for the flux density in the corresponding band . i considered that the flux density @xmath41 of each galaxy at observed wavelengths @xmath42 can be decomposed as : @xmath43 where the two terms on the right - hand side correspond to a stellar component and a power - law component with spectral index @xmath44 , respectively . the power - law term is normalised to the @xmath45 flux density , so the constant @xmath46 can vary between 0 and 1 . this decomposition is similar to that proposed by hainline et al . ( 2011 ) to analyse the rest optical / near - ir seds of sub - millimetre galaxies at @xmath47 . for each of the 174 galaxies in my @xmath4 sample , i produced a suite of 31 photometric sets @xmath48 in the @xmath49[3.6][4.5][5.8][8.0 ] bands , corresponding to different power - law subtractions . i considered that @xmath44 could take three , and @xmath46 ten possible values : @xmath50 and @xmath51 , and @xmath52 . the case with @xmath53 and any @xmath44 value corresponds to the original photometry @xmath54 . the three adopted @xmath44 values are representative of the ir power - law indices that characterise different types of agn , with different hot - dust components ( e.g. alonso - herrero et al . 2003 , 2006 ; polletta et al . 2006 ; hnig et al . 2010 ) . i performed the sed modelling on the 31 photometric sets @xmath48 of each galaxy , using a customised @xmath55-minimisation sed - fitting code that incorporates the 2007 version of the bruzual & charlot synthetic stellar template library , with solar metallicity @xcite . these templates correspond to different star formation histories : a single stellar population , and different exponentially - declining star - formation histories with @xmath56 through 5 gyr . in all cases , the redshift of the galaxy has been fixed to the known @xmath25 value . to account for internal extinction , i convolved the stellar templates with the calzetti et al . ( 2000 ) reddening law , allowing for @xmath57 with a step of 0.1 . stellar masses are obtained in the output of the same sed - fitting code . the technique is illustrated in figure [ fig_exam ] : once a power - law component is subtracted from the original photometry ( open circles ) , a new photometric dataset is obtained ( filled circles ) . for each galaxy , i have tested 31 possible power - law subtractions ( including the original photometry ) , as explained above . i have then fitted stellar templates to all these photometric sets , and compared the resulting minimum @xmath55 values . figure [ fig_chi ] shows the minimum reduced @xmath55 value after power - law subtraction ( i.e. the absolute minimum obtained considering all the different ( @xmath44 , @xmath46 ) combinations ) , versus the minimum reduced @xmath55 value obtained with the original photometry , for each galaxy . the points lying on the identity line correspond to galaxies for which the original photometry , without a power - law subtraction , produces the absolutely best fitting . the points below the identity line indicate the galaxies for which the sed fitting with stellar templates improves after the subtraction of a power - law component . this is the case for the majority of the @xmath4 galaxies . in particular , in 60 out of 174 cases ( 35% ) , the best sed - fitting solution on the original photometry , without a power - law subtraction , can be rejected with @xmath58 confidence . these 60 galaxies are , according to my definition , the plcgs in the @xmath4 sample with @xmath2 . interestingly , around two - thirds of the plcgs ( 42 out of 60 ) lie at @xmath59 ( fig . [ fig_plcgstfmz ] ) . this implies that @xmath60 of all the lirgs and ulirgs at @xmath8 are plcgs , i.e. they have a significant power - law component in their rest - frame near - ir seds . the percentage of plcgs at @xmath61 is much higher than the known percentage of agn among lirgs and ulirgs at these redshifts , and raises the question of whether the power - law component is actually related to an agn presence in all these cases . i investigate this issue further in section [ sec - nature ] . at @xmath62 , instead , the fraction of plcgs among @xmath4 galaxies ( the vast majority of which are ulirgs ) is of only @xmath63 . i repeated the same sed modelling using the 2003 version of the bruzual & charlot stellar template library , and found very similar results : @xmath64 of the @xmath4 sources with @xmath2 are classified as plcg . the resulting minimum @xmath55 values are very similar to those obtained with the 2007 templates ( the median of the minimum @xmath55 differences is 0.009 ) . all the following analysis will be based on the 2007 template library run , but no conclusion in this paper would change if i adopted the results obtained with the 2003 templates . the host galaxy properties , as determined from the best - fitting stellar templates , are similar for plcgs and non - plcgs . for both sub - samples , the best - fitting star formation histories correspond to single stellar populations or @xmath65 models with @xmath66 in @xmath7 of cases . after the power - law subtraction , these models provide the best fitting for 93% of the plcgs . the best - fitting extinction values for plcgs ( before power - law subtraction ) and non - plcgs are also very similar : the median and r.m.s . of the distributions are @xmath67 and @xmath68 , respectively . after power - law subtraction , the plcgs have @xmath69 . figure [ fig_zdisalpha ] shows the extinction @xmath70 versus spectroscopic redshifts for the 60 plcgs , classified according to their power - law indices @xmath44 . the non - plcgs are also shown as a reference . a bit more than a half of the plcgs ( 55% ) are characterised by a power - law component with @xmath71 , while the remaining plcgs have lower best - fitting power - law indices , i.e. @xmath3 or 2 . these indices are related to the temperature of the hot - dust component in the galaxy , which is higher for lower @xmath44 values ( cf . section [ sec - gbtemp ] ) . the effect of a decrease in the best - fitting @xmath70 value is especially evident for the @xmath3 plcgs : a shallow power - law component can mimic the effect of additional reddening . after power - law subtraction , the majority of the @xmath3 and @xmath72 plcgs have @xmath73 , while most of the @xmath71 plcgs have @xmath74 . the normalisation factor of the power - law component is @xmath75 for virtually all the plcgs , which means that at least 50% of the @xmath45 flux density is in the power - law component . for around a half of the plcgs , this contribution is @xmath76 . this confirms that the plcg definition adopted here , based on the @xmath58 improvement of the sed fitting , truly selects sources for which the power - law component makes an important contribution to the rest - frame near - ir light of the galaxy . for galaxies at @xmath77 , the emission from polycyclic aromatic hydrocarbons ( pahs ) at rest @xmath78 will enter the observed @xmath45 band . although this is a relatively faint pah emission feature , it can potentially mimic the effect of a steep power - law component in the galaxy sed . to assess the importance of this effect on the plcg selection , i have performed the sed fitting on additional photometric sets for each galaxy , with power - law subtractions corresponding to larger spectral indices , namely , @xmath79 and @xmath80 . within the total sample , only 13 galaxies appear to be plcgs with such large best - fitting spectral index . however , their redshifts are not concentrated around @xmath77 or any other specific redshift , so the steeper @xmath44 values are unlikely the effect of emission features . in fact , 11 out of these 13 galaxies have been recognised as plcgs with @xmath71 in my original analysis . only two galaxies appear as new plcgs : one at @xmath81 ( corresponding to an x - ray luminous agn ) , and another one at @xmath82 ( for which the @xmath83 pah emission could be partly within the @xmath45-filter wavelength coverage ) . these results indicate that : a ) the @xmath83 pah emission plays a minor role in the overall ir sed for most ir galaxies ; b ) considering @xmath84 values for the power - law subtraction produces a virtually negligible effect on the identified plcg sample . to investigate whether the high fraction of plcgs is truly a characteristic of the @xmath4 galaxy sample , i performed a similar sed fitting analysis on a control sample of 247 galaxies selected from the irac catalogue described in section [ sec - sample ] , also with secure @xmath2 , but which are not @xmath4-detected ( i.e. @xmath85 ) . these galaxies have been selected to have a similar stellar mass distribution as the @xmath4 galaxies ( @xmath86 $ \sim$}}~}10^{10 } \ , \rm m_\odot$ ] , based on the sed modelling of the original photometry ) . for each of these galaxies , i have produced 31 sets of photometric variations in the same manner explained above , and fitted the resulting seds with stellar templates in all cases . the comparison of minimum @xmath55 values with and without power - law subtraction is shown in fig . [ fig_chicon ] . for the control sample , the percentage of galaxies that significantly improve their sed fitting with stellar templates after a power - law subtraction is a factor of two smaller than for the @xmath4 galaxies : a pure stellar sed fitting can be discarded with @xmath58 confidence in only 18% of cases ( 44 out of 247 galaxies ) . these 44 plcgs in the control sample have best - fitting @xmath44 values shared in similar proportions as for the @xmath4-detected plcgs : 50% have best - fitting @xmath71 , while the other 50% have @xmath3 or 2 . these results obtained on the control sample indicate that the incidence of plcgs among mid - ir - selected galaxies is twice more important than among other similarly massive galaxies at @xmath0 . figure [ fig_stmratio ] shows the distribution of corrected - to - original stellar mass ratios for the 60 @xmath4-selected plcgs , classified according to their best - fitting power - law index . the corrected stellar masses are those derived from the sed modelling with stellar templates after the best - fitting power - law subtraction . the inspection of these distributions shows that the correction on the stellar masses is non - trivial : the exact correction needs to be derived on a case - by - case basis . however , for @xmath8780% ( 90% ) of the plcgs the correction is within a factor of @xmath88 ( 2.0 ) . the widest correction - factor distributions are those corresponding to power - law spectral indices @xmath3 and 2 , for which the stellar mass ratios range between @xmath89 $ \sim$}}~}0.1 $ ] and @xmath90 . the median values of these stellar mass ratios are 0.67 and 0.85 , respectively . the corrections on the stellar masses are the smallest for the case @xmath71 ( the median of the ratio distribution is 0.92 ) . this behaviour is expected , as the subtraction of such a steep power - law has a rapidly declining effect on the photometry at wavelengths shorter than @xmath45 , i.e the overall sed shape is not substantially modified except for a single photometric point . it may a priori seem surprising that the sed fitting for some galaxies yields _ larger _ stellar masses after the pl subtraction . this effect is produced because the best - fitting stellar template properties ( star formation history , age and extinction ) are usually different for the pl - subtracted photometry to those for the original photometry , and , in some cases , the resulting overall normalisation factors ( i.e. mass - to - light ratios ) are larger . for 5 out of the 60 plcgs , all with best - fitting @xmath3 , the corrected stellar mass is less than @xmath91 of the stellar mass derived from the original photometry . these cases correspond to `` pure power - law sources '' , and all except one of them are identified as x - ray luminous agn with @xmath92 in the chandra 4 ms catalogue for the goods - s field @xcite . the only source that is not x - ray detected is the highest - redshift of these pure power - law sources , with @xmath93 and @xmath94 , so it is likely an intrinsically luminous agn as well . for these sources , a stellar mass calculation based on the original photometry should clearly be avoided . but , as said above , in all other cases the correction to the stellar mass is within a factor of two . this result validates the approach of obtaining approximate stellar mass estimates for e.g. low - luminosity agn , using the original un - corrected photometry , that is followed by some authors ( e.g. * ? ? ? * ; * ? ? ? in order to investigate whether the plcgs actually host agn , i searched for counterparts of these sources in the publicly available , ultra - deep chandra 4ms x - ray catalogue for the goods - s field @xcite . this diagnostic is very useful to assess to which extent the plcg criterion can be used to identify agn in the general case that such deep x - ray catalogues are not available ( to date , x - ray maps of such a depth have only been obtained for the goods - s field ) . [ fig_xrayid ] shows the @xmath29-@xmath13 diagram for the @xmath4 galaxies in the @xmath2 sample that are identified with x - ray sources in the chandra 4ms catalogue . the plcgs are highlighted with couloured circles . there are a total of 59 @xmath4 galaxies identified with an x - ray source , 48 out of which correspond to x - ray - classified agn , and 11 are x - ray ` normal galaxies ' . the x - ray classification is based on the intrinsic x - ray luminosities , x - ray colours ( hardness ratios ) , and derived neutral gas column densities ( e.g. * ? ? ? * ; * ? ? ? although the x - ray - based agn classification is among the most secure , the criteria typically adopted to classify x - ray sources could still result in some low - luminosity or highly - obscured agn being classified as normal galaxies ( see xue et al . 2011 and references therein ) . the analysis of fig . [ fig_xrayid ] shows the following . @xmath95 at @xmath96 , 77% of the x - ray agn are recognised as plcgs , which indicates that the plcg criterion produces a high completeness in the identification of agn at these redshifts . and , interestingly , among the x - ray - classified ` normal galaxies ' , the ratio of plcgs is very similar ( @xmath97 ) . so , either these galaxies also host an agn , or there is another mechanism that gives rise to the x - ray emission , and is also responsible for the hot - dust component manifested in the plcg classification . in starburst galaxies , supernova explosions can produce hot - gas bubbles with significant x - ray emission . the resulting shock waves can heat the surrounding dust grains up to sublimation temperatures @xmath98 $ \sim$}}~}1000 \ , \rm k$ ] . however , if the x - ray luminosities of x - ray galaxies were only a product of supernova explosions , they would correlate with the ir luminosities , as both trace the on - going star formation rates @xcite . _ this is not the case for the plcgs at @xmath96 , for which two sources can differ in more than a factor of three in ir luminosity , but both be classified as x - ray normal galaxies with similar x - ray luminosities . _ this becomes clear by inspection of fig . [ fig_plcgstfmz ] , where some bright @xmath4 sources at @xmath8 are neither plcgs nor x - ray galaxies at all . thus , the sed power - law excess of these plcgs is not simply tracing high star formation rates . at @xmath96 , it is interesting to note the different distributions of best - fitting spectral indices @xmath44 among the x - ray agn and normal galaxies that are recognised as plcgs : 9 out of 13 ( 69% ) x - ray agn / plcgs have @xmath3 or 2 , while only 2 out of 7 ( 29% ) of the x - ray galaxy / plcgs have @xmath3 or 2 . or , equivalently , only 31% of the x - ray agn classified as plcgs have @xmath71 , while 71% of the x - ray galaxies classified as plcgs have @xmath71 . i will carry on this discussion in section [ sec - morph ] , where i will argue that the spectral index @xmath44 is a good discriminator of plcgs of different nature . @xmath95 only 42% of the x - ray - detected agn at @xmath62 are classified as plcgs . this effect could be due to a real difference in the properties of x - ray agn at different redshifts : e.g. at @xmath62 the light of the host galaxies may dominate the rest near - ir sed , making a power - law component to appear less significant . alternatively , the lower percentage of agn incidence could be the consequence of a subtle k - correction effect , as the observed @xmath45 photometry only traces rest wavelengths @xmath14 $ \sim$}}~}3 \ , \rm \mu m$ ] at @xmath62 , while it traces up to rest @xmath99 at @xmath100 . to investigate these two possible explanations , i consider those @xmath4 galaxies at @xmath47 that form part of the fadda et al . ( 2010 ) ir galaxy sample . pozzi et al . ( 2012 ) have studied the mid-/far - ir sed decomposition of these galaxies , and found that nine out of 24 of them have a significant agn component . only three out of these nine sources are recognised as plcgs within my sample . this result suggests that data beyond observed @xmath45 are necessary to reveal more effectively an agn presence in galaxies at @xmath47 . in section [ sec - gbtemp ] , i will show that very hot dust temperatures ( @xmath9 $ \sim$}}~}1300 \ , \rm k$ ] ) are necessary to generate a power - law sed component detectable in the irac bands at @xmath62 . such hot dust is typically found in agn inner dusty tori @xcite , but some agn are characterised by lower dust temperatures , or the inner torus may be less exposed ( e.g. * ? ? ? this will restrict the fraction of agn that can be identified with my plcg criterion , which is based on sed fitting up to observed @xmath45 , at @xmath47 . @xmath95 around a half of the plcgs at @xmath8 are not x - ray detected . the mean optical / near - ir sed properties of these plcgs are not significantly different to those of x - ray - detected plcgs classified as x - ray galaxies : e.g. the median extinction is @xmath101 for the x - ray undetected , against @xmath102 for the x - ray detected . their irac colours are also similar ( see section [ sec - iraccol ] ) . the plcg classification , and the x - ray detection , do not follow a simple correlation with the host galaxy ir luminosities : there are x - ray detected plcgs both in the lirg and ulirg regimes . but all the x - ray undetected plcgs are lirgs , which suggests that the x - ray non - detection is due to the sensitivity limit of the x - ray surveys . to probe whether the x - ray non - detection of half of the plcgs at @xmath8 is simply due to the sensitivity limit of the chandra 4ms x - ray survey , i performed a stacking analysis of these sources in x rays . fig . [ fig_xraystack ] shows the stacked x - ray images of the plcgs that are not individually detected in x rays at @xmath8 . in the hard x - ray band ( 2.0 - 8.0 kev ) , the stacking does not produce any detection , while in the soft band ( 0.5 - 2.0 kev ) a detection with a @xmath103 significance is obtained . this stacked signal confirms that these sources are x - ray emitters , but their fluxes are below the sensitivity limits of the 4ms chandra survey . nuclear activity can not be discarded as an agent partly responsible for the soft x - ray emission in these sources , but it is probably not the main cause : on average they have @xmath104 , too low a value for an agn classification @xcite . around two thirds of the stacked plcgs have spectral index @xmath71 which , as i discuss below , should mainly correspond to star - forming galaxies with hot interstellar media ( ism ) at @xmath8 . the stacking of the @xmath71 plcgs alone yields a @xmath105 detection in the ( 0.5 - 2.0 kev ) band , which corresponds to an average flux density @xmath106 ( after correction for galactic extinction ; see stark et al . 1992 ) . at the median redshift of these sources @xmath107 , this corresponds to a luminosity @xmath108 , assuming a photon index @xmath109 , which is appropriate for star - forming galaxies ( e.g. * ? ? ? * ) . using the relation calibrated by mineo et al . ( 2013 ) , i find that this corresponds to an instantaneous star formation rate @xmath110 . this is significantly lower than the sfr derived from the average total ir luminosities of these sources , which is @xmath111 , as obtained using the sfr-@xmath112 relation given by kennicutt ( 1998 ) . this difference indicates the presence of internal absorption . to reconcile both sfr values , i find that these galaxies should have an average column density @xmath113 . in any case , as commented above , one has to be careful by considering the plcg x - ray luminosity as a tracer of the instantaneous star formation rate , as many of these galaxies are likely composite star - forming / agn systems . also , the x - ray emission may be the result of energetic processes in the galaxy ism that are not simply related to the on - going star formation rate . some of these effects probably account for the dispersion observed in the @xmath114-sfr relation . further clues about the nature of plcgs , particularly at @xmath8 , can be obtained by comparing their morphologies . figure [ fig_stamps ] shows _ hst_/acs @xmath13-band postage stamps of 15 plcgs at @xmath8 , grouped in three different categories : x - ray detected sources , classified as agn and galaxies , and plcgs that are not individually x - ray detected . the three classes display objects with different morphologies , including galaxies with compact and regular matter concentrations , and galaxies with very disturbed morphologies . the latter are seemingly the product of galaxy mergers , or are characterised by other violent processes , such as the radial ejection of material ( see the x - ray galaxy at the right end of the middle row ) . the striking conclusion that can be extracted from fig.[fig_stamps ] is the trend between the plcg morphology and the value of the best - fitting spectral index @xmath44 : those plcgs with @xmath3 or 2.0 show quite a regular morphology , while the plcgs with @xmath10 are irregular and show signs of different kinds of galaxy interactions . this trend is independent of the x - ray detection and classification . this fact suggests that the spectral index @xmath44 contains information about the nature of the source : in the plcgs with @xmath3 or 2.0 , the sed power - law component is very likely the signature of nuclear activity , while in the @xmath71 plcgs the power law appears to be the consequence of a hot ism produced by the violent galaxy dynamics . shocks of hot gas can compress and heat up the ism dust to several hundred kelvin , and even up to sublimation temperatures ( @xmath98 $ \sim$}}~}1000 \ , \rm k$ ] ; * ? ? ? this hot dust , not related to an agn origin , will also manifest itself as a power - law component in the rest near - ir galaxy sed . of course , in some cases , an agn and a host galaxy with a disturbed ism may co - exist , producing x - ray confirmed agn that are plcgs with @xmath71 ( upper right stamps in fig . [ fig_stamps ] ; see also kocevski et al . 2012 for a discussion on agn morphology ) . i estimated the dust temperatures necessary to produce a power - law component in the galaxy seds , assuming a greybody function for the dust emission : @xmath115 where the second factor on the right - hand side of equation ( [ eq - grey ] ) corresponds to the black - body emission at temperature @xmath116 , and @xmath117 and @xmath118 are the planck s and boltzmann s constants , respectively . i assumed @xmath119 , as it is common in the literature . [ fig_coltemp ] shows the observed flux density ratios @xmath120 produced by greybody emitters at different temperatures , as a function of redshift . the horizontal solid lines indicate the fixed , observed flux density ratios produced by power - laws with the three spectral indices considered in this paper , i.e. @xmath3 , 2.0 and 3.0 . two main results can be derived from this simple diagram . firstly , to generate a power - law component with @xmath3 or 2 at @xmath8 , temperatures as high as @xmath121 are needed . instead , for a power - law with spectral index @xmath10 , temperatures @xmath122 are sufficient . this may explain why most of the x - ray agn , for which dust temperatures close to the sublimation point are expected , are associated with plcgs with spectral indices @xmath3 or 2 , at @xmath8 . the plcgs with spectral index @xmath10 can be explained with a shock - heated ism , with no need of dust getting close to the sublimation point ( the exact dust sublimation temperature depends on the dust composition , but it is typically @xmath98 $ \sim$}}~}1000 \ , \rm k$ ] ) . the existence of hot interstellar dust in ir galaxies has been previously discussed in the literature . particularly , lu et al . ( 2003 ) found that a colour dust temperature of @xmath123 can account for the near - ir excess observed in some local ir normal galaxies , similarly to what i find here for lirgs at @xmath8 . secondly , at redshifts @xmath124 , producing a power - law component in the observed irac sed requires hot dust with temperatures @xmath98 $ \sim$}}~}1300 \ , \rm k$ ] , for any of the spectral indices . this mostly explains why the incidence of plcgs is lower at @xmath62 , as only sources with such an extremely hot dust component will be classified as plcg with the criterion adopted in this paper . the use of _ spitzer_/irac colour - colour diagrams has proven to be quite successful for identifying agn among ir - selected galaxies @xcite . over the years , these colour selections have been refined to be more reliable for agn identification among fainter ir sources ( e.g. * ? ? ? here i analyse how the plcgs , and other @xmath4 galaxies at @xmath2 , are classified in an irac colour - colour diagram . [ fig_iraccol ] shows the irac colour - colour diagram for all the @xmath4 galaxies at @xmath2 , with the plcgs and x - ray sources highlighted . the colours shown here are overall colours , i.e. there is no separation of stellar and power - law sed components . one can see that the range of overall irac colours displayed is quite wide , with some @xmath4 galaxies having much redder irac colours than others . plcgs , in particular , display a wider range of colours than the non - plcgs , as there are very few non - plcgs with very red irac colours @xmath125 $ \sim$}}~}0.2 $ ] and @xmath126 $ \sim$}}~}0.1 $ ] . for plcgs , i find no correlation between the irac colours and the spectral index of the sed power - law component . the solid lines in fig . [ fig_iraccol ] delimit the revised colour region for agn selection proposed by donley et al . more than 80% of the @xmath4 galaxies that lie within that region are plcgs or x - ray agn , indicating that indeed the proposed colour selection is highly reliable . however , these colour criteria are too strict and miss a substantial fraction of plcgs , many of which are x - ray confirmed agn . the incompleteness of this colour selection was discussed by donley et al . ( 2012 ) , who argued that their criteria quickly lose completeness for agn with x - ray luminosities @xmath127 . within the current sample , all the x - ray - detected sources in the donley et al . ( 2012 ) region have @xmath128 $ \sim$}}~}5 \times 10^{43 } \ , \rm erg \ , s^{-1}$ ] . there are three galaxies that lie within the donley et al . agn colour - colour region , but which are neither identified as plcgs , nor x - ray detected . one of them is a @xmath129 galaxy for which the sed fitting with a power - law subtraction just lies below the 2@xmath40 criterion for plcg identification . so , this galaxy could still have a buried agn . the other two galaxies are at redshifts @xmath130 and 2.577 . for these , the stellar bump centred at rest - frame @xmath131 is within the observed @xmath132 filter , producing a red @xmath133 colour . these two galaxies are close to the @xmath134 lower boundary of the donley et al . colour wedge , so their presence within the wedge may simply be due to photometric scattering ( note the size of the error bars ) . galaxy colour selections are attractive for their simplicity , but no criterion has been found to safely identify the bulk of the agn population at any redshift . the plcg criterion proposed here offers a suitable alternative , which is particularly successful at identifying agn of different luminosities at @xmath8 . note that it would not be enough to extend the donley et al . ( 2012 ) to bluer colours , because most of the plcgs display similar irac colours as other ir sources which do not have a significant power - law component in their seds , so an extended colour selection would produce a highly contaminated agn sample . in this paper i have studied the rest - frame uv / optical / near - ir ( @xmath14 $ \sim$}}~}3 \ , \rm \mu m$ ] ) seds of 174 @xmath4-selected galaxies with secure spectroscopic redshifts @xmath2 , analysing explicitly the presence and significance of a hot - dust , power - law component . i modelled the residual light produced after subtracting different possible power - law components , using stellar templates , on 11 broad - bands from the @xmath34-band through @xmath45 . the subtracted power - law components are characterised by three different spectral indices @xmath3 , 2.0 and 3.0 , and different normalisation weights . i found that in 60 out of 174 ( 35% ) cases , the sed fitting with stellar templates without any previous power - law subtraction can be rejected with @xmath5 confidence , i.e. 35% of the @xmath4 galaxies at @xmath0 are characterised by a significant power - law component in their rest near - ir sed . i referred to these galaxies as plcgs . this high percentage of plcgs at @xmath0 appears to be inherent to the ir bright galaxy population . a similar analysis performed on a control sample of 247 irac - selected galaxies with @xmath2 and similar stellar masses as the @xmath4 galaxies , but which are not @xmath4-detected , shows that the incidence of plcgs is of only 18% . the incidence of plcgs among @xmath4 galaxies is much higher at @xmath8 ( 47% ) than at @xmath62 ( 21% ) . the lower incidence at @xmath62 is explained by the fact that the presence of very hot dust ( @xmath9 $ \sim$}}~}1300 \ , \rm k$ ] ) is necessary to have a significant power - law component manifested at observed wavelengths @xmath42 . therefore , only sources with the most extreme dust conditions would be recognised with the plcg criterion at these higher redshifts . the impact of a significant power - law component in the rest near - ir sed on the derived galaxy stellar masses needs to be analysed on a case - by - case basis , but it is small in general . the correction for stellar masses is within a factor of 1.5 in @xmath7 of cases . only in a small percentage ( @xmath135 ) of the plcgs , basically all corresponding to powerful agn , the correction to the stellar mass is so large that a stellar mass calculation based on the original photometry should clearly be avoided . a key issue is understanding whether the plcgs are truly related to the presence of an agn , particularly at @xmath8 , where the plcg incidence is very high . to investigate this , i studied different properties of these sources . firstly , i looked for identifications of the plcgs in the ultra - deep 4ms x - ray catalogue for the cdfs . i found that 77% of the x - ray agn within my @xmath4 sample with @xmath61 are classified as plcgs . this indicates that the plcg criterion provides a high completeness in selecting agn at these redshifts . interestingly , 78% of the x - ray normal galaxies within my @xmath4 sample with @xmath61 are also plcgs . but their plcg classification or x - ray luminosities do not correlate with the @xmath4 luminosities , so the x ray emission or the power - law sed excess is not simply a tracer of high star formation rates . a stacking analysis of the plcgs that are not individually x - ray detected at @xmath8 reveals that they have on average similar properties as x - ray - detected normal galaxies . the stacked x - ray images yield a significant detection in the soft x - ray band ( @xmath136 ) , and no signal in the hard x - ray band ( @xmath137 ) . around two thirds of the stacked plcgs have a best - fitting power law with @xmath71 . the visual inspection of optical _ hst_/acs images has proven very useful to shed light on the nature of the plcgs at @xmath8 . i found a clear trend relating the plcg morphology and the spectral - index @xmath44 of the best - fitting sed power - law component : plcgs with @xmath3 or 2.0 look regular and typically have a nuclear matter concentration . instead , the plcgs with @xmath10 typically show irregular morphologies , indicating a disturbed galaxy dynamics , which in some cases suggest the presence of galaxy mergers . this morphology trend is independent of the plcg x - ray detection and classification . the @xmath10 power - law component appears then to be a signature of quite extreme ism conditions , where the interstellar dust is heated to temperatures @xmath98 $ \sim$}}~}700 \ , \rm k$ ] by the gas shocks produced in the disturbed ism . of course , this does not preclude the @xmath10 plcgs to also host agn . actually , the presence of an agn is confirmed in some of them by the x - ray data . but an agn presence does not appear to be necessary to produce a significant @xmath10 power - law component . for the plcgs with @xmath3 or 2.0 , instead , higher dust temperatures are necessary @xmath98 $ \sim$}}~}900 \ , \rm k$ ] . this property , along with the regular morphologies , take to the direct conclusion that these sources must host an agn . one could speculate that the @xmath10 plcgs may all be on the way to form an agn as well , but this hypothesis can not be tested with the current data . considering that all the plcgs with @xmath3 or 2.0 in my sample contain agn , and adding all the other x - ray classified agn within my @xmath4 sample , i obtain that a total of 30% of the @xmath4 sources at @xmath8 host agn . this should be considered as a lower limit , as some of the @xmath10 plcgs that are not individually x - ray detected could also contain agn . if all of them had agn , that would give an upper limit of @xmath138 on the agn fraction among lirgs and ulirgs at @xmath8 . very recently , different works have concluded on a higher fraction of agn among ir galaxies than previously known at @xmath139 , obtaining percentages of 37 - 50% ( e.g. * ? ? ? * ; * ? ? ? the results of my study are in line with such conclusions . at @xmath62 , a similar criterion as that used at @xmath96 would indicate that @xmath140 of the @xmath4 galaxies contain an agn , although this figure is largely based on the x - ray detections , rather than the plcg incidence , which is relatively low ( for a comparison , pozzi et al . 2012 derived a fraction of 35% based on the mid-/far - ir sed analysis of 24 ulirgs at @xmath47 ) . in this work , i have calibrated the plcg analysis technique making use of galaxies with secure spectroscopic redshifts . a complete analysis of the applicability of this technique with the simultaneous determination of photometric redshifts ( @xmath141 ) will be presented in a future paper . a preliminary run of my sed fitting code leaving the redshift as a free parameter indicates that , in fact , the consideration of a power - law subtraction produces some improvement in the overall @xmath141-@xmath25 comparison , reducing the number of catastrophic outliers . this suggests that the identification of plcgs can be done , and the power - law subtraction is actually recommendable , when attempting a @xmath141 determination . although less straightforward than a simple irac colour - colour selection , the plcg identification criterion introduced here is much more complete to select agn of different luminosities , especially in the redshift range @xmath8 . keeping only those plcgs with spectral index @xmath3 and 2 should result in the best compromise between completeness and reliability , as most x - ray normal galaxies and x - ray non - detections are among the plcgs with @xmath71 . in the absence of very deep x - ray data , the plcg selection with spectral index segregation offers the most efficient method to identify the agn presence in any large galaxy sample . based on observations made with the _ spitzer space telescope _ , which is operated by the jet propulsion laboratory , california institute of technology under a contract with nasa . also based on observations undertaken at the european southern observatory ( eso ) _ very large telescope ( vlt ) _ under different programmes ; the nasa / esa _ hubble space telescope _ , obtained at the space telescope science institute , operated by the association of universities for research in astronomy , inc . ( aura ) , under nasa contract nas 5 - 26555 ; and the _ chandra x - ray observatory _ , which is operated by the smithsonian astrophysical observatory for and on behalf of nasa under contract nas8 - 03060 . i thank almudena alonso - herrero , maurilio pannella , brigitte rocca - volmerange , and john silverman for very useful discussions , and an anonymous referee for a very constructive report . i am also grateful to the goods team for the public release of their superb data products . alonso - herrero , a. , quillen , a. c. , rieke , g. h. , ivanov , v. d. , efstathiou , a. , 2003 , , 126 , 81 alonso - herrero , a. , prez - gonzlez , p. g. , alexander , d. m. et al . , 2006 , , 640 , 167 alonso - herrero , a. , pereira - santaella , m. , rieke , g. h. , rigopulou , d. , 2012 , , 744 , 2 armus , l. , charmandaris , v. , bernard - salas , j. et al . , 2007 , , 656 , 148 barthel , p. , haas , m. , leipski , c. , wilkes , b. , 2012 , , 757 , l26 barvainis , r. , 1987 , , 320 , 537 bauer , f. e. , alexander , d. m. , brandt , w. n. et al . , 2004 , , 128 , 2048 bavouzet , n. , dole , h. , le floch , e. , caputi , k. i. , lagache , g. , kochanek , c. s. , 2008 , , 479 , 83 bertin , e. , arnouts , s. , 1996 , , 117 , 393 bruzual , g. , charlot , s. , 2003 , , 344 , 1000 bruzual , g. , 2007 , aspc , 374 , 303 calzetti , d. , armus , l. , bohlin , r. c. et al . , 2000 , , 533 , 682 caputi , k. i. , dole , h. , lagache , g. et al . , 2006a , , 454 , 143 caputi , k. i. , dole , h. , lagache , g. et al . , 2006b , , 637 , 727 caputi , k. i. , lagache , g. , yan , l. et al . , 2007 , , 660 , 97 caputi , k. i. , lilly , s. j. , aussel , h. et al . , 2008 , , 680 , 939 desai , v. , soifer , b. t , dey , a. et al . , 2008 , , 679 , 1204 dole , h. , lagache , g. , puget , j .- l . et al . , 2006 , , 451 , 417 donley , j. l. , rieke , g. h. , prez - gonzlez , p. g. , rigby , j. r. , alonso - herrero , a. , 2007 , , 660 , 167 donley , j. l. , koekemoer , a. m. , brusa , m. et al . , 2012 , , 748 , 142 fadda , d. , yan , l. , lagache , g. et al . , 2010 , , 719 , 425 fazio , g. g. , hora , j. l. , allen , l. e. et al . , 2004 , , 154 , 10 fiore , f. , grazian , a. santini , p. et al . , 2008 , , 672 , 94 giavalisco , m. , ferguson , h. c. koekemoer , a. m. et al . , 2004 , , 600 , l93 gruppioni , c. , pozzi , f. , andreani , p. et al . , 2010 , , 518 , l27 hainline , l. j. , blain , a. w. , smail , i. et al . , 2011 , , 740 , 96 hernn - caballero , a. , prez - fournon , i. , hatziminaoglou , e. et al . , 2009 , , 395 , 1695 hanami , h. , ishigaki , t. , fujishiro , n. et al . , 2012 , , 64 , 70 hnig , s. f. , kishimoto , m. , gandhi , p. et al . , 2010 , , 515a , 23 hwang , h. s. , geller , m. j. , kurtz , m. j. , dellantonio , i. p. , fabricant , d. g. , 2012 , , 758 , 25 hopkins , a. m. & beacom , j. f. , 2006 , , 651 , 142 juneau , s. , dickinson , m. , bournaud , f. et al . , 2013 , , 764 , 176 kennicutt , r. c. jr . , 1998 , , 36 , 189 kocevski , d. d , faber , s. m. , mozena , m. et al . , 2012 , , 744 , 148 kurk , j. , cimatti , a. , daddi , e. et al . , 2013 , , 549a , 63 lacy , m. , storrie - lombardi , l. j. , sajina , a. et al . , , 154 , 166 lacy , m. , petric , a. , sajina , a. , et al . , 2007 , , 133 , 186 lagache , g. , dole , h. , puget , j .- l . , 2004 , , 154 , 112 le fvre , o. , vettolani , g. , paltani , s. et al . , 2004 , , 428 , 1043 le floch , e. , papovich , c. , dole , h. et al . , 2005 , , 632 , 169 lehmer , b. d. , brandt , w. n. , alexander , d. m. et al . , 2008 , , 681 , 1163 lu , n. , helou , g. , werner , m. w. et al . , 2003 , , 588 , 199 mainieri , v. , bongiorno , a. , merloni , a. et al . , 2011 , , 535a , 80 mas - hesse , j. m. , ot - floranes , h. , cervio , m. , 2008 , , 483 , 71 matsuta , k. , gandhi , p. , dotani , t. et al . , 2012 , , 753 , 104 merloni , a. , bongiorno , a. , bolzonella , m. et al . , 2010 , , 708 , 137 mineo , s. , gilfanov , m. , sunyaev , r. , 2013 , , submitted , arxiv : 1207.2157 mor , r. , netzer , h. , 2012 , , 420 , 526 mullaney , j. r. , pannella , m. , daddi , e. et al . , 2012 , , 419 , 95 murakami h. , baba , h. , barthel , p. et al . , 2007 , , 59s , 369 nardini , e. , risaliti , g. , salvati , m. et al . , 2008 , , 385 , l130 oke , j. b. , gunn , j. e. , 1983 , , 266 , 713 osterbrock , d. e. & ferland , g. j. , 2006 , _ astrophysics of gaseous nebulae and active galactic nuclei _ , 2nd edition , university science books papovich , c. , dole , h. , egami , e.et al . , 2004 , , 154 , 70 petric , a. o. , armus , l. , howell , j. et al . , 2011 , , 730 , 28 pilbratt , g. l. , 2003 , spie , 4850 , 586 polletta , m. , wilkes , b. c. , siana , b. et al . , 2006 , , 642 , 673 popesso , p. , dickinson , m. , nonino , m. et al . , , 494 , 443 pozzi , f. , vignali , c. , gruppioni , c. et al . , 2012 , , 423 , 1909 puget , j .- l . , abergel , a. , bernard , j .- p . , boulanger , f. , burton , w. b. , dsert , f .- x . , hartmann , d. , 1996 , , 308 , l5 rieke , g. h. , young , e. t. , engelbracht , c. w. et al . , 2004 , , 154 , 25 rigby , j. r. et al . , 2004 , , 154 , 160 rosario , d. j. , mozena , m. , wuyts , s. et al . , 2013 , , 763 , 59 sajina , a. et al . , 2007 , , 664 , 713 salpeter , e. e. , 1955 , , 121 , 161 schartmann , m. , meisenheimer , k. , camenzind , m. , wolf , s. , henning , t. , 2005 , , 437 , 861 silverman , j. d. , lamareille , f. , maier , c. et al . , 2009 , , 696 , 396 stark , a. a. , gammie , c. , wilson r. w. et al . , 1992 , , 79 , 77 stern , d. , eisenhardt , p. , gorjian , v. et al . , 2005 , , 631 , 163 symeonidis , m. , georgakakis , a. , seymour , n. et al . , 2011 , , 417 , 2239 szokoly , g. p. , bergeron , j. , hasinger , g. et al . , 2004 , , 155 , 271 teplitz , h. i. , desai , v. , armus , l. et al . , 2007 , , 659 , 941 vanzella , e. , cristiani , s. , dickinson , m. et al . , 2008 , , 478 , 83 werner , m. w. et al . , 2004 , , 154 , 1 xue , y. q. , luo , b. , brandt , w. n. et al . , 2011 , apjs , 195 , 10

motivated by the recent cms excess in a flavor violating higgs decay as well as the anomaly of muon anomalous magnetic moment ( muon g-2 ) , we consider a scenario where both the excess in and the anomaly of muon g-2 are explained by the flavor violation in a general two higgs doublet model . we study various processes involving and , and then discuss the typical predictions and constraints in this scenario . especially , we find that the prediction of can be within the reach of the belle ii experiment . we also show that the lepton non - universality between and can be sizable , and hence the analysis of the current belle data and the future experimental improvement would have an impact on this model . besides , processes such as , , , , and muon edm can be accessible , depending on the unknown yukawa couplings . on the other hand , the processes like and could not be sizable to observe because of the current strong constraints on the and flavor violations . then we also conclude that contrary to decay mode , the lepton flavor violating higgs boson decay modes and are strongly suppressed , and hence it will be difficult to observe these modes at the lhc experiment .

1511.08880

the standard model ( sm ) of the elementary particles describes particle physics phenomena remarkably well up to the electroweak scale . in addition , the recent discovery of a higgs boson at the lhc @xcite strengthens the success of the sm . on the other hand , the detailed measurements of the higgs boson properties have just started , and the whole structure of the higgs sector may have not been unveiled . therefore , theoretical and experimental studies of the extended higgs sector would be important to understand the nature of the higgs sector . one of simple extensions of the higgs sector in the sm is a two higgs doublet model ( 2hdm ) where additional higgs doublet is introduced and both higgs doublets can couple to all fermions . as a result , flavor violating phenomena mediated by the higgs bosons are predicted @xcite . in most cases , such a flavor violation has been considered to be avoided because of lack of the experimental supports for the anomalous flavor - violating phenomena @xcite . however , the cms collaboration has recently reported an event excess in a flavor - violating higgs decay @xmath0 @xcite , and it suggests that the best fit value of the branching ratio is @xmath18 where the final state is a sum of @xmath19 and @xmath20 , and the deviation from the sm prediction is @xmath21 . in addition , the result of the atlas experiment has also appeared recently @xcite , and it is shown as @xmath22 which is consistent with the cms result within @xmath23 . although these results have not been conclusive yet , these become strong motivations to study the flavor violating phenomena predicted by the beyond standard models @xcite . in ref . @xcite , we pointed out a possibility that the @xmath2 flavor violation in general 2hdm can explain not only the cms excess in the higgs decay @xmath0 , but also the anomaly of muon anomalous magnetic moment ( muon g-2 ) @xcite . this possibility is interesting because two unexplained phenomena can be accommodated in the 2hdm , and hence it is worth further investigating this possibility . in this paper , we study phenomena related to @xmath3 and @xmath4 lepton physics in the scenario to see whether there are any interesting predictions and constraints caused by the @xmath2 flavor violation . the paper is organized as follows . in section ii , we present a general 2hdm where both higgs doublets couple to all fermions . we discuss the yukawa interactions and higgs mass spectrum in the model . in section iii , we consider a solution where the cms excess in @xmath1 decay as well as muon g-2 anomaly can be explained by the @xmath2 flavor violating yukawa interactions in the model . we show the typical parameter space where both anomalies are explained . in section iv , we discuss @xmath4- and @xmath3-physics in the interesting region studied in the previous section . especially , we study @xmath5 , @xmath24 , muon electric dipole moment ( muon edm ) , @xmath6 , @xmath25 ( @xmath26 ) , @xmath27 , and @xmath28 . the prediction of @xmath5 can be within the reach of the belle ii experiment , which will start near future . the extra higgs boson correction to @xmath6 can be as large as @xmath29 , but it is not large in @xmath7 mode . the future improvement of measurement on lepton flavor universality in @xmath30 decay will be important in the scenario . we also find that unlike the @xmath2 flavor violation suggested by the cms result , the @xmath15 and @xmath14 flavor violations are severely constrained since the constraint from the @xmath24 is strong . therefore , the processes related to @xmath15 and @xmath14 flavor violations are suppressed . in section v , we also discuss an implication to higgs physics . since @xmath15 and @xmath14 flavor violating yukawa couplings are strongly suppressed , @xmath17 and @xmath31 higgs decay modes will not be observed in this scenario , contrary to @xmath0 mode . in section vi , we summarize our results . in a general two higgs doublet model , there are no symmetry to distinguish the two different higgs doublets . thus , both the higgs doublets can couple to all fermions , and hence there are flavor violating interactions in higgs sector . in general , when the higgs potential is minimized in the sm - like vacuum , both neutral components of higgs doublets develop nonzero vacuum expectation values ( vevs ) . taking a certain linear combination , we can define the basis where only one of the higgs doublets obtains the nonzero vev as follows : @xmath32 where @xmath33 and @xmath34 are nambu - goldstone bosons , and @xmath35 and @xmath36 are a charged higgs boson and a cp - odd higgs boson , respectively . then @xmath35 and @xmath36 are in the mass eigenstates . the cp - even neutral higgs bosons @xmath37 and @xmath38 can mix and form mass eigenstates , @xmath39 and @xmath40 ( @xmath41 ) , in general : @xmath42 here @xmath43 is the mixing angle and fixed by the higgs potential analysis . note that when @xmath44 ( @xmath45 ) , the interactions of @xmath37 approach to those of the sm higgs boson . in mass eigenbasis for the fermions , the yukawa interactions are expressed by @xmath46 where @xmath47 and @xmath48 represent flavor indices , and @xmath49 , @xmath50 are defined . @xmath51 is the cabbibo - kobayashi - maskawa ( ckm ) matrix and @xmath52 is the maki - nakagawa - sakata ( mns ) matrix . fermions @xmath53 @xmath54 are mass eigenstates . here we have assumed that the tiny neutrino masses are achieved by the seesaw mechanism introducing super - heavy right - handed neutrinos , so that in the low - energy effective theory , the left - handed neutrinos have a @xmath55 majorana mass matrix . the yukawa coupling constants @xmath56 are general @xmath57 complex matrices and can be sources of the higgs - mediated flavor changing processes . in mass eigenstates of higgs bosons , the yukawa interactions are given by @xmath58 where @xmath59 are defined with @xmath60 and @xmath61 . note that when @xmath62 is small , the yukawa interactions of @xmath39 are almost equal to those of the sm higgs boson , however , there are small flavor violating interactions @xmath56 which are suppressed by @xmath62 . on the other hand , the yukawa interactions of heavy higgs bosons ( @xmath40 , @xmath36 , and @xmath35 ) mainly come from the @xmath56 couplings . let us comment on the relation between the higgs masses and the parameters in the higgs potential . the renormalizable higgs potential in the general 2hdm is given by @xmath63 in the basis shown in eq . ( [ higgsbasis ] ) , the higgs boson masses can be described by the dimensionless parameters and @xmath64 using the stationary conditions for the higgs doublets : @xmath65 especially , when @xmath66 is close to zero ( that is , @xmath67 ) , we obtain the following simple expressions for the higgs boson masses : @xmath68 note that fixing the couplings , @xmath69 , @xmath70 and @xmath71 , the heavy higgs boson masses are expressed by the cp - odd higgs boson mass @xmath72 , which we treat as a free parameter of the model . the contribution to the peskin - takeuchi t - parameter @xcite should be taken into account , so that we assume that it is suppressed by the degeneracy between @xmath72 and @xmath73 as well as the small higgs mixing parameter @xmath66 . therefore , we set @xmath74 in our analysis , which corresponds to @xmath75 . the cms collaboration has reported an excess in a higgs boson decay mode @xmath0 . furthermore , it is known that there is a discrepancy between the measured value and the sm prediction of the muon anomalous magnetic moment ( muon g-2 ) . the both anomalies can not be explained by the sm , and hence they might be an indication of physics beyond the sm . in this section , we discuss whether the general 2hdm can accommodate both anomalies simultaneously , and investigate the parameter space where both anomalies can be explained . an excess in @xmath0 decay mode has been reported by the cms collaboration : the best fit value of the branching ratio is @xmath76 @xcite . as discussed in the introduction , the atlas collaboration has also shown the result , @xmath77 @xcite , which is consistent with the cms one . it would be an indication of new physics because the sm can not accommodate the excess . since in the general 2hdm the sm - like higgs boson has flavor violating yukawa interactions as discussed in the previous section , the excess can be easily explained . the expression of the branching ratio of the @xmath0 process is given by @xmath78 where @xmath79 is a total decay width of higgs boson @xmath39 and we adopt @xmath80 mev in this paper . in order to accommodate the cms excess , the @xmath2 flavor violating yukawa couplings need to satisfy the following condition : @xmath81 it is interesting to note that even in the small higgs mixing @xmath82 , the @xmath2 flavor violating yukawa couplings with the order of 0.1 can achieve the cms excess . we have shown that the @xmath2 flavor violating yukawa couplings in the general 2hdm explain the cms excess in the @xmath0 decay mode . here we consider extra contributions to the muon anomalous magnetic moment ( muon g-2 ) generated by the @xmath2 flavor violating yukawa couplings . a discrepancy between the measured value ( @xmath83 ) and the standard model prediction ( @xmath84 ) of the muon g-2 has been reported @xcite . for example , @xcite shows the following result : @xmath85 here we consider whether the extra contributions induced by the @xmath2 flavor violating interactions can accommodate the muon g-2 anomaly . the effective operator for the muon g-2 is expressed by @xmath86 we note that the chirality of muon is flipped in the operator . therefore , if there is a large chirality flip induced by the new physics , it can enhance the extra contributions to the muon g-2 @xcite . a feynman diagram for the extra contributions of the neutral higgs bosons to the muon g-2 , induced by the @xmath2 flavor violating yukawa couplings , is shown in fig . [ muong2 ] . as shown in fig . [ muong2 ] , the chirality flip occurs in the internal line of @xmath4 lepton in the diagram . therefore , it induces the @xmath87 enhancement in the extra contributions to the muon g-2 , compared with the one generated by the flavor - diagonal yukawa coupling . we stress that the @xmath2 flavor violating interaction is essential to obtain such an enhancement . note that both couplings @xmath88 and @xmath89 should be nonzero to get the chirality flip in the internal @xmath4 lepton line . the expression of the enhanced extra contribution @xmath90 is given by @xmath91 , \label{a_mu}\end{aligned}\ ] ] where we have assumed that @xmath92 is real , for simplicity . we will discuss the effect of an imaginary part of these yukawa couplings later . we note that a degeneracy of all neutral higgs bosons suppresses the extra contribution to the muon g-2 , as seen in eq . ( [ a_mu ] ) . as a function of @xmath62 and @xmath93 for @xmath94 gev ( upper figure ) and 350 gev ( lower figure ) . regions where the muon g-2 anomaly in eq . ( [ muong2_anomaly ] ) is explained within @xmath95 , @xmath96 and @xmath97 are shown . here we determine the mass spectrum of heavy higgs bosons assuming @xmath98 in eq . ( [ higgs_spectrum2 ] ) . we have assumed @xmath99 with @xmath100 to obtain the positive contribution to @xmath90 . , title="fig:",scaledwidth=70.0% ] as a function of @xmath62 and @xmath93 for @xmath94 gev ( upper figure ) and 350 gev ( lower figure ) . regions where the muon g-2 anomaly in eq . ( [ muong2_anomaly ] ) is explained within @xmath95 , @xmath96 and @xmath97 are shown . here we determine the mass spectrum of heavy higgs bosons assuming @xmath98 in eq . ( [ higgs_spectrum2 ] ) . we have assumed @xmath99 with @xmath100 to obtain the positive contribution to @xmath90 . , title="fig:",scaledwidth=70.0% ] in fig . [ muong2_vs_htm ] , we show numerical results for the extra contribution to muon g-2 ( @xmath90 ) as a function of @xmath62 and @xmath93 for @xmath94 gev ( upper figure ) and 350 gev ( lower figure ) . regions where the muon g-2 anomaly in eq . ( [ muong2_anomaly ] ) is explained within @xmath95 , @xmath96 and @xmath97 are shown . here we have fixed the mass spectrum of heavy higgs bosons assuming @xmath98 in eq . ( [ higgs_spectrum2 ] ) . we have assumed that @xmath99 with @xmath100 to obtain the positive contribution to @xmath90 . we only discuss the case with @xmath101 , however , the predictions of @xmath90 and @xmath102 do not change even if the sign of @xmath62 is flipped ( @xmath103 ) . one can see that there are regions where both anomalies of the muon g-2 and @xmath104 can be explained in the 2hdm . we note that although larger @xmath105 is preferred by the muon g-2 anomaly , the regions where @xmath105 is smaller than the one suggested by the cms result are also allowed by the muon g-2 anomaly if @xmath106 is small . in fig . [ para_muong2 ] , the numerical result for the @xmath90 is shown as a function of @xmath72 and @xmath62 fixing that @xmath107 . regions that explain the muon g-2 anomaly in eq . ( [ muong2_anomaly ] ) within @xmath108 , @xmath109 and @xmath110 are shown . here we take @xmath98 to determine the mass spectrum of heavy higgs bosons ( shown in eq . ( [ higgs_spectrum2 ] ) ) as a function of @xmath72 . we assume that the yukawa couplings @xmath89 and @xmath88 are determined to realize @xmath107 with @xmath100 . when @xmath106 gets smaller , @xmath90 increases because the yukawa couplings @xmath111 become larger with the fixed @xmath93 . it is interesting to see that the 2hdm can explain both anomalies of the muon g-2 and @xmath0 when @xmath106 is small ( @xmath112 ) for @xmath113 gev . we note that the small mixing @xmath106 is consistent with the current results of the higgs coupling measurements as well as the constraints from the electroweak observables . as a function of @xmath72 and @xmath62 assuming @xmath107 . regions which explain the muon g-2 anomaly in eq . ( [ muong2_anomaly ] ) within @xmath95 , @xmath96 and @xmath97 are shown . here we determine the mass spectrum of heavy higgs bosons as a function of @xmath72 assuming @xmath98 in eq . ( [ higgs_spectrum2 ] ) . we have assumed @xmath100 to obtain the positive contribution to @xmath90 and the yukawa couplings @xmath111 are fixed to realize @xmath107 . , scaledwidth=70.0% ] as a function of @xmath72 and @xmath71 assuming @xmath107 with the fixed @xmath114 . here we have assumed @xmath74 and @xmath100.,scaledwidth=70.0% ] in fig . [ para2_muong2 ] , similar to fig . [ para_muong2 ] , the numerical result for the @xmath90 is shown as a function of @xmath72 and @xmath71 fixing that @xmath115 and @xmath116 . we have assumed that @xmath117 and @xmath74 . as @xmath71 gets larger , the @xmath90 becomes larger because the non - degeneracy between @xmath40 and @xmath36 increases and it enhances the @xmath90 . [ muong2_vs_htm ] , [ para_muong2 ] and [ para2_muong2 ] show the typical interesting regions which explain both anomalies of the muon g-2 and @xmath0 . so far , we have seen that the general 2hdm with the @xmath2 flavor violation accommodates the muon g-2 anomaly and the cms excess in @xmath0 decay , simultaneously . the parameter regions with @xmath112 and @xmath118 gev are typically interesting . in this section , we investigate what kinds of predictions and/or constraints in @xmath4- and @xmath3-physics we have in this scenario . the first process we would like to discuss is @xmath119 . the @xmath2 flavor violating yukawa couplings induce the flavor violating phenomena @xmath119 , as shown , for example , in fig . [ tmg_diagrams ] . processes at one - loop level ( left figure ) which are induced by @xmath120 yukawa couplings and at two - loop level ( right figure ) which are barr - zee type contributions and induced by the third generation fermions via @xmath121 yukawa couplings . diagrams where the fermion chiralities are flipped also contribute . we also have a barr - zee type two - loop contribution induced by w - loop , which is not shown here . , title="fig:",scaledwidth=35.0% ] processes at one - loop level ( left figure ) which are induced by @xmath120 yukawa couplings and at two - loop level ( right figure ) which are barr - zee type contributions and induced by the third generation fermions via @xmath121 yukawa couplings . diagrams where the fermion chiralities are flipped also contribute . we also have a barr - zee type two - loop contribution induced by w - loop , which is not shown here . , title="fig:",scaledwidth=49.0% ] we parametrize the decay amplitude ( @xmath122 ) as follows : @xmath123 where @xmath124 are chirality projection operators , and @xmath125 and @xmath126 are the electric charge , a photon polarization vector , a photon momentum , a @xmath4 momentum , and a spinor of the fermion @xmath127 , respectively . the branching ratio is given by @xmath128 where @xmath129 and @xmath130 are the fine structure constant and the fermi constant , respectively . the lepton flavor violating higgs contributions to @xmath131 and @xmath132 via @xmath120 yukawa interactions at one - loop level ( left figure in fig . [ tmg_diagrams ] ) are given by also contribute to @xmath133 . however , the sm part of @xmath134 is smaller than the one of @xmath135 , and @xmath136 is strongly constrained by @xmath137 process as discussed later . therefore we have neglected the contributions from @xmath138 . ] @xmath139,~~(\phi = h,~h,~a)\nonumber \\ a_r^{\phi}&=a_l^{\phi}|_{y^{e*}_{\phi\ ; \tau\mu}\rightarrow y^e_{\phi\ ; \mu\tau } , ~~y^{e}_{\phi \;\tau \tau } \leftrightarrow y^{e*}_{\phi\ ; \tau\tau}},~~ ( \phi = h,~h,~a),\nonumber \\ a_l^{h^-}&= -\frac{(\rho_e^\dagger \rho_e)^{\mu \tau}}{192\pi^2 m_{h^-}^2},~~ a_r^{h^-}=0,\end{aligned}\ ] ] where @xmath140 are the @xmath141 contributions at the one loop level . the yukawa couplings @xmath120 are given in eq . ( [ yukawa ] ) . here we have neglected the @xmath142 contributions . we also find that the barr - zee type contributions ( @xmath143 ) at the two loop level are important and dominant in the most of cases . the third generation fermion contributions via @xmath144 yukawa couplings . ] ( shown in the right figure in fig . [ tmg_diagrams ] ) and the w - boson contribution ( not shown in fig . [ tmg_diagrams ] ) are given by have been studied in ref . the application to @xmath119 is apparent and we adopt their results for @xmath119 . ] @xmath145 \nonumber \\ & + \sum_{\phi = h , h}\frac{\alpha}{16\pi^3}\frac{g_{\phi ww } y^{e*}_{\phi\;\tau\mu}}{m_\tau v } \left [ 3f_h\left(x_{w\phi}\right ) + \frac{23}{4 } f_a\left(x_{w\phi}\right ) \nonumber \right.\\ & \hspace{4 cm } + \frac{3}{4 } g\left(x_{w\phi}\right ) + \frac{m_\phi^2}{2 m_w^2}\left\ { f_h\left(x_{w\phi}\right)-f_a\left(x_{w\phi}\right ) \right\}\nonumber \\ & + \frac{1 - 4s_w^2}{8s_w^2}\left\{\left ( 5-t_w^2+\frac{1-t_w^2}{2 x_{w\phi}}\right ) \tilde{f}_h ( x_{w\phi } , x_{wz } ) \right.\nonumber \\ & \left.\left.+\left(7 - 3t_w^2 -\frac{1-t_w^2}{2x_{w\phi}}\right ) \tilde{f}_a(x_{w\phi } , x_{wz } ) + \frac{3}{2}\left\{f_a(x_{w\phi})+g(x_{w\phi})\right\ } \right\ } \right ] , \label{barr - zee } \\ a_r^{\rm bz } & = a_l^{\rm bz}(y^{e*}_{\phi\;\tau\mu}\rightarrow y^{e}_{\phi\;\mu\tau},~i\rightarrow -i),\end{aligned}\ ] ] where @xmath146 , @xmath147 , @xmath148 and @xmath149 , and @xmath150 and @xmath151 . @xmath152 denotes the isospin of the fermion . here the couplings @xmath153 for @xmath154 ( @xmath155 ) . functions @xmath156 , @xmath34 and @xmath157 are defined by @xmath158,\\ \tilde{f}_h(x , y)&=\frac{xf_h(y)-yf_h(x)}{x - y},\\ \tilde{f}_a(x , y)&=\frac{xf_a(y)-yf_a(x)}{x - y}.\end{aligned}\ ] ] note if the yukawa couplings @xmath56 are real , @xmath159 for @xmath160 , and @xmath161 for @xmath162 are satisfied , as shown in eq . ( [ yukawa ] ) . for simplicity , we assume that all @xmath56 are real in the calculation of @xmath5 . the contribution in the first line ( the second line ) of eq . ( [ barr - zee ] ) comes from the effective @xmath163 vertex ( @xmath164 vertex ) induced by the third generation fermion loop , and the one in the third and the forth lines ( the fifth and sixth lines ) originates from the effective @xmath163 vertex ( @xmath164 vertex ) generated by the w - boson loop . in the analysis of @xmath5 in ref . @xcite , we have not included the barr - zee type contributions induced by the effective @xmath165 vertex since they are sub - dominant contributions . here we include them and find they change the results by about @xmath166 . the total amplitude @xmath167 is a sum of all contributions , @xmath168 in fig . [ tmg_vs_htm ] , numerical results for @xmath169 as a function of @xmath62 and @xmath102 are shown . here the same parameter set as the one in fig . [ muong2_vs_htm ] is taken . we have assumed that the extra yukawa couplings @xmath56 other than @xmath111 are negligible . lines for @xmath170 and @xmath171 ( upper figure ) and for @xmath172 and @xmath173 ( lower figure ) are shown for @xmath94 gev and @xmath174 gev , respectively . as one can see , if @xmath107 as suggested by the cms experiment , the branching ratio for @xmath133 can be larger than @xmath175 , which might be within the reach of the future b - factory experiment , the belle ii . note that the branching ratio @xmath176 almost does not depend on the higgs mixing parameter @xmath66 when @xmath93 is fixed and @xmath66 is small . we also note that the cancellation between the one - loop and two - loop barr - zee type contributions happens , and hence the branching ratio @xmath177 is not simply suppressed by the heavy higgs boson masses . if the extra yukawa couplings other than @xmath111 are not negligible , the branching ratio @xmath169 could be further enhanced . for example , the extra yukawa coupling @xmath178 can contribute at the one - loop level , and on the other hand , @xmath179 can affect the branching ratio via the barr - zee type two - loop contribution . in fig . [ tmg_extrayukawa ] , numerical results for @xmath176 are shown as a function of @xmath180 and @xmath179 . lines for @xmath181 and @xmath182 ( current experimental limit ) are shown . here we have assumed @xmath183 gev with @xmath98 , @xmath184 and @xmath115 with @xmath100 . this parameter set can enhance the muon g-2 as @xmath185 which is within the @xmath23 . at present , the extra yukawa couplings @xmath178 and @xmath179 can be still larger than , for example , @xmath186 with some correlation , however , the future experimental constraint would be significant for this scenario . therefore , the @xmath133 process would be important to probe the scenario . as a function of @xmath62 and @xmath93 in the same parameter set of fig . [ muong2_vs_htm ] . lines for @xmath170 and @xmath171 ( @xmath187 and @xmath173 ) are shown for @xmath94 gev ( @xmath183 gev ) . here we have assumed that the extra yukawa couplings @xmath188 other than @xmath189 are negligible . , title="fig:",scaledwidth=65.0% ] as a function of @xmath62 and @xmath93 in the same parameter set of fig . [ muong2_vs_htm ] . lines for @xmath170 and @xmath171 ( @xmath187 and @xmath173 ) are shown for @xmath94 gev ( @xmath183 gev ) . here we have assumed that the extra yukawa couplings @xmath188 other than @xmath189 are negligible . , title="fig:",scaledwidth=65.0% ] as a function of @xmath180 and @xmath179 . lines for @xmath181 and @xmath182 ( current experimental limit ) are shown . here we have assumed @xmath183 gev with @xmath98 , @xmath184 and @xmath115 with @xmath100 . we note that for this parameter set , @xmath185 which explains the muon g-2 anomaly within the @xmath23 . , scaledwidth=70.0% ] the @xmath2 flavor violating yukawa couplings themselves do not generate @xmath190 . however , together with @xmath14 or @xmath15 flavor - violation , @xmath24 is induced . since there is a strong constraint from this process , @xmath14 and @xmath15 flavor violating couplings are strongly constrained . similar to @xmath119 , we parametrize the decay amplitude @xmath191 as @xmath192 and the branching ratio is given by @xmath193 the neutral higgs contributions @xmath194 ( @xmath195 ) to @xmath167 at the one - loop are given by @xmath196,\\ a_r^{\phi}&=\frac{1}{16\pi^2}\sum_{i=\mu,\tau}\frac{y^{e}_{\phi\ ; e i } } { m_\phi^2 } \left[\frac{m_i}{m_\mu}y^{e } _ { \phi\ ; i \mu}\left ( \log\frac{m_\phi^2}{m_i^2}-\frac{3}{2}\right)+\frac{y^{e * } _ { \phi\ ; \mu i}}{6}\right ] , \label{one - loop_meg}\end{aligned}\ ] ] where the yukawa couplings @xmath197 are defined in eq . ( [ yukawa ] ) . here we neglect an electron mass and we assume that the yukawa coupling @xmath198 is negligible . is strongly constrained by @xmath199 process , as studied later . therefore , our assumption will be justified . ] the charged higgs contribution to @xmath167 is @xmath200 for nonzero @xmath201 , the barr - zee type contributions ( @xmath143 ) at two - loop level are significant . the expression of @xmath202 is the same as one for @xmath5 case shown in eq . ( [ barr - zee ] ) except that the flavor violating yukawa couplings @xmath203 should be replaced by @xmath204 , and the @xmath4 mass ( @xmath205 ) should be replaced by the @xmath3 mass ( @xmath206 ) . the total @xmath167 is a sum of all contributions ; @xmath207 similar to the muon g-2 , the contributions from the @xmath2 flavor violating yukawa interactions together with the @xmath15 flavor violation have @xmath87 enhancement , and induce significant contributions to @xmath190 . as a function of @xmath208 and @xmath209 . lines for @xmath210 ( current limit ) , @xmath211 , @xmath212 and @xmath213 are shown . here we have assumed that @xmath183 gev with @xmath98 , @xmath184 and @xmath107 with @xmath100 , and extra yukawa couplings @xmath188 other than @xmath111 and @xmath214 are negligible . we note that for this parameter set , @xmath185.,scaledwidth=60.0% ] in fig . [ meg_tau - e ] , we show numerical results for @xmath215 as a function of @xmath208 and @xmath209 . here we have taken @xmath183 gev with @xmath98 , @xmath216 and @xmath115 with @xmath100 . we have assumed that extra yukawa couplings @xmath56 other than @xmath111 and @xmath214 are negligible . this parameter set corresponds to @xmath185 . one can see that the current limit on @xmath217 strongly constrains the @xmath15 flavor violating couplings @xmath214 if the cms excess of @xmath105 is true . if we change the value of @xmath102 in fig . [ meg_tau - e ] , the experimental bound of @xmath218 is relaxed by the factor , @xmath219 when @xmath220 is assumed . if yukawa couplings @xmath218 are negligible but @xmath221 are not , the barr - zee type two loop contributions are dominant . is also nonzero , there are also one - loop contributions as shown in eq . ( [ one - loop_meg ] ) . however , the coupling @xmath136 is strongly constrained by the @xmath222 bound , as discussed later . therefore , the effect from @xmath223 is negligible and we neglect it in our numerical analysis . ] we show numerical results for @xmath217 as a function of @xmath224 and @xmath225 . as a function of @xmath224 and @xmath225 . lines for @xmath210 ( current limit ) , @xmath211 , @xmath212 and @xmath213 are shown . here we have assumed that @xmath183 gev with @xmath98 , @xmath184 , and extra yukawa couplings @xmath188 other than @xmath111 and @xmath226 are negligible.,scaledwidth=60.0% ] here we have assumed that @xmath183 gev with @xmath98 , @xmath227 and extra yukawa couplings @xmath56 other than @xmath111 and @xmath226 are negligible . as one can see from the figure , @xmath226 couplings are also severely constrained by the @xmath10 bound . note that the prediction of @xmath24 for this case does not depend on the value of @xmath93 . the future improvement of @xmath228 at the level of @xmath229 as proposed by the meg ii experiment @xcite would significantly probe the flavor structure of this scenario . the effective operator for @xmath230 also generate @xmath231 conversion process in nuclei . besides , the extra yukawa couplings , @xmath232 and @xmath233 , may enhance the @xmath231 conversion through the tree - level higgs exchanging . the contribution depends on the extra yukawa couplings in the quark sector as well , and then our model may be also tested by the experiments @xcite , although our prediction is vague because of the ambiguity of the yukawa couplings . we comment on the consequence of the strong constraints on the @xmath15 and @xmath14 flavor - violations . unlike the @xmath2 flavor violation , the @xmath15 flavor violating yukawa couplings in this scenario is strongly constrained as we have seen above . therefore , the prediction of @xmath234 is expected to be small . similarly , because of the smallness of the @xmath15 and @xmath14 flavor violation , we also expect that the new physics contributions to the anomalous magnetic moment of electron ( electron g-2 ) should be small . when we discussed the muon g-2 , we have assumed that the @xmath2 flavor violating yukawa couplings are real . if the @xmath2 yukawa couplings are complex , the couplings @xmath235 in eq . ( [ a_mu ] ) should be replaced by @xmath236 . in addition , the imaginary parts of the yukawa couplings generate an electric dipole moment ( edm ) of muon . since the muon g-2 and the muon edm are induced by the same feynman diagram shown in fig . [ muong2 ] , these quantities are correlated via the unknown cp - violating phase . the effective operators for the muon g-2 @xmath237 and the muon edm @xmath238 are expressed by @xmath239 if we parametrize the complex yukawa couplings as follows : @xmath240 the relation between the muon g-2 ( @xmath90 ) and the muon edm ( @xmath241 ) induced by the @xmath2 flavor - violating yukawa couplings is given by @xmath242 therefore , the predicted muon edm is @xmath243 the current limit @xcite is @xmath244 @xcite would be significant to probe the scenario . the yukawa couplings @xmath111 induce a correction to @xmath245 via a charged higgs mediation , where the flavor of final neutrino and anti - neutrino states is summed up since it is not detected . and @xmath246 generate the extra corrections to @xmath247 . however , the yukawa couplings @xmath248 and @xmath136 are strongly constrained by @xmath10 and @xmath249 , respectively . therefore , the contributions from these couplings are negligible . the unknown yukawa coupling @xmath178 can be sizable , and hence it can increase the prediction of the @xmath247 . thus our result of @xmath247 induced from @xmath189 is viewed as a conservative estimate . ] the correction @xmath247 is given as follows ; @xmath250 , @xmath247 given in eq . ( [ taudecay_eq ] ) as a function of @xmath62 and @xmath93 in the same parameter set of fig . [ muong2_vs_htm ] . black solid lines for @xmath251 ( from left to right ) are shown for @xmath94 gev ( upper figure ) and @xmath183 gev ( lower figure ) . here we also show the region where the muon g-2 anomaly is explained within @xmath23 ( light green ) and @xmath252 ( light blue ) . here we have assumed that the extra yukawa couplings @xmath188 other than @xmath111 are negligible.,title="fig:",scaledwidth=70.0% ] , @xmath247 given in eq . ( [ taudecay_eq ] ) as a function of @xmath62 and @xmath93 in the same parameter set of fig . [ muong2_vs_htm ] . black solid lines for @xmath251 ( from left to right ) are shown for @xmath94 gev ( upper figure ) and @xmath183 gev ( lower figure ) . here we also show the region where the muon g-2 anomaly is explained within @xmath23 ( light green ) and @xmath252 ( light blue ) . here we have assumed that the extra yukawa couplings @xmath188 other than @xmath111 are negligible.,title="fig:",scaledwidth=70.0% ] in fig . [ taudecay ] , numerical results for the correction @xmath247 given above are shown as a function of @xmath62 and @xmath102 in the same parameter set of fig . [ muong2_vs_htm ] . one can see that as the correction to the muon g-2 ( @xmath90 ) gets larger , the size of @xmath247 also becomes larger , and they are correlated each other , independent of @xmath93 . the interesting regions which explain the muon g-2 anomaly within @xmath23 predict @xmath253-@xmath254 . the current precision of the measurement of the decay rate @xmath255 is at the level of @xmath254 @xcite . therefore , the further improvement of the precision would be important for this scenario . in addition , from the @xmath4 decay , the babar collaboration has reported a measurement of the charged current lepton universality @xcite , given by @xmath256 where @xmath257 , which is a phase space factor . the universality of the gauge interaction in the sm predicts @xmath258 and the current experimental results are @xmath259 in our scenario , we expect the correction to @xmath7 would be small because of the strong constraint on @xmath15 flavor violation from @xmath24 process . therefore , the charged higgs contribution to @xmath6 with @xmath2 flavor violating yukawa couplings induces the significant correction to the violation of the lepton universality above , @xmath260 the result from the belle collaboration and the further improvement of the precision of the lepton universality would have an important impact on our scenario . the nonzero yukawa couplings @xmath111 also generate processes @xmath264 and @xmath265 ( @xmath266 and @xmath267 for short , respectively ) at the tree level . they are induced without unknown @xmath223 and @xmath268 yukawa couplings . the branching ratios , however , are too small to be observed . therefore , nonzero @xmath136 and @xmath268 are important for these processes and @xmath226 couplings also induce the @xmath267 process . however , these yukawa couplings are strongly constrained by @xmath24 process as discussed in previous sections . therefore , we neglect these effects . ] . the branching ratios for @xmath222 and @xmath269 are given by @xcite @xmath270 fig . [ tau_to_3mu ] shows @xmath271 and @xmath272 as a function of @xmath273 ( @xmath274 for @xmath222 and @xmath275 for @xmath267 ) . it is assumed that @xmath184 , @xmath183 gev with @xmath98 and @xmath107 with @xmath100 in fig . [ tau_to_3mu ] . one can see that the current experimental bounds , @xmath276 set the strong constraints on the @xmath273 yukawa couplings . for example , the parameter set shown in fig . [ tau_to_3mu ] , requires @xmath277 ( @xmath278 ) . we note that the constraint on the @xmath136 is still larger than the value of the muon yukawa coupling in the sm ( @xmath279 ) . ( above ) and @xmath272 ( below ) as a function of @xmath273 ( @xmath274 for @xmath222 and @xmath275 for @xmath267 ) . here we have assumed that @xmath184 , @xmath183 gev with @xmath98 and @xmath107 with @xmath100.,title="fig:",scaledwidth=70.0% ] ( above ) and @xmath272 ( below ) as a function of @xmath273 ( @xmath274 for @xmath222 and @xmath275 for @xmath267 ) . here we have assumed that @xmath184 , @xmath183 gev with @xmath98 and @xmath107 with @xmath100.,title="fig:",scaledwidth=70.0% ] contrary to @xmath25 , the @xmath280 @xmath263 process is suppressed in this scenario because the @xmath281 flavor violation is strongly constrained by @xmath24 process . furthermore , since the constraints on @xmath282 are stronger than those on @xmath283 , @xmath284 is expected to be smaller than @xmath25 . ( needless to say , @xmath285 is much suppressed . ) therefore , we expect that the @xmath286 and @xmath287 processes will be small . we also study @xmath288 ( @xmath11 in short ) which depends on the @xmath231 flavor violating yukawa couplings @xmath282 and the flavor diagonal element @xmath268 . as we have seen , the @xmath231 flavor violating yukawa couplings @xmath289 are constrained by the @xmath24 process and the @xmath268 coupling is restricted by the @xmath290 process . from fig . [ meg_mu - e ] and fig . [ tau_to_3mu ] , the current limits on @xmath291 and @xmath268 are @xmath292 for @xmath293 and @xmath294 , respectively , assuming @xmath183 gev with @xmath98 and @xmath184 . under these constraints , it will be interesting to see how large branching ratio of @xmath11 is expected . as a function of @xmath268 and @xmath224 . here we have assumed that @xmath293 , @xmath184 and @xmath183 gev with @xmath98.,scaledwidth=70.0% ] in fig . [ mu_to_3e ] , we show the @xmath295 as a function of @xmath268 and @xmath224 . in the parameter region where the constraints from @xmath10 and @xmath290 are satisfied , the branching ratio can be as large as about @xmath296 . this is consistent with the current limit @xcite @xmath297 the improvement of the branching ratio at the level of @xmath298 @xcite which has been proposed by the mu3e experiment would have a significant impact on this scenario together with the improvement of @xmath24 @xcite and @xmath299 conversion in nuclei @xcite . the @xmath28 is also generated by the extra @xmath301 yukawa coupling via the mediation of the cp - odd higgs boson at the tree level . the expression for the branching ratio of @xmath9 is given by @xcite @xmath302 where @xmath303 and @xmath304 are the mass and the decay constant of @xmath305 . for @xmath306 mev and @xmath307 mev , we obtain a constraint on @xmath301 ; @xmath308 we have a strong constraint although it is still larger than the sm value of the strange quark yukawa coupling ( @xmath309 ) . the other hadronic @xmath4-lepton decays have been studied in ref . they potentially provide constraints on the other extra yukawa couplings @xmath188 in quark sector . for details , see ref . we have seen that the cms excess in @xmath310 is consistent with the anomaly of muon g-2 as well as the other experimental constraints . it will be interesting to note whether other lepton flavor violating higgs boson decays would be possible . as we have already seen , the @xmath14 and @xmath15 flavor violating yukawa couplings are strongly constrained mainly by the @xmath24 constraint . as a consequence , the lepton flavor violating higgs boson decays @xmath16 and @xmath17 are strongly suppressed so that the near future experiments such as the ones at the lhc could not observe these decay modes , contrary to the @xmath1 mode . therefore , the non - observation of these decays is one of interesting predictions of this scenario . the anomalous event in @xmath1 has been observed by the cms collaboration . the discrepancy of the muon g-2 is also one of the longstanding issues in the particle physics . these anomalous phenomena may be a hint of physics beyond the standard model . at glance , these anomalies are not related each others . however , we have found that the both anomalies are related and accommodated by the @xmath2 flavor violating yukawa interactions in a general two higgs doublet model , and hence this motivates further studies to see whether there are any interesting predictions and indications in the scenario . we have identified the parameter space where the cms excess in @xmath0 and the muon g-2 anomaly are both explained , and especially we have studied @xmath4- and @xmath3- physics in this interesting parameter space . one of the interesting processes in the presence of the @xmath2 flavor violation is @xmath5 . the @xmath2 flavor violation suggested by the cms excess in @xmath0 and the muon g-2 anomaly induces the large branching ratio , and it can be as large as @xmath175 which is within the reach of the future experiment at the superkekb . the imaginary parts of the @xmath2 flavor violating yukawa couplings also induce the extra contributions to the muon edm , which may be also within the planning future experiments . the necessary @xmath2 flavor violation also generates the correction to @xmath245 decay and also induces a violation of lepton universality between @xmath311 and @xmath312 . the improvement of their precisions would be interesting . the tree - level @xmath4 decays such as @xmath313 and @xmath28 are also interesting because the extra yukawa couplings @xmath314 and @xmath301 could also induce the observable effects . on the other hand , we have found that the @xmath14 and @xmath15 flavor violating yukawa couplings are severely constrained by mainly @xmath10 process . because of these constraints , phenomena such as @xmath12 , @xmath315 and extra contributions to the electron g-2 would not be accessible in the near future experiments . although there are many unknown yukawa couplings in a general 2hdm , there are many interesting indications to @xmath4- and @xmath3-physics . we have also commented on an implication to higgs physics . contrary to the @xmath2 flavor violation suggested by the cms result , the @xmath14 and @xmath15 flavor violations in the higgs coupling are strongly limited . therefore , the observation of @xmath310 and non - observation of @xmath16 and @xmath17 would be the important implication of the scenario . we summarize our findings in table [ summary_table ] . if the cms excess in @xmath310 is justified in coming lhc run , these phenomena in @xmath4- and @xmath3- physics would be key to reveal the physics beyond the standard model . this work was supported in part by grants - in - aid for scientific research from the ministry of education , culture , sports , science , and technology ( mext ) , japan ( no.22224003 for k.t . and no . 23104011 for y.o . ) and japan society for promotion of science ( jsps ) ( no.26104705 for k.t . ) . is supported in part by the ministry of science and technology , taiwan under grant no . most 104 - 2811-m-008 - 011 . .observalilities in various processes are summarized . if there is an observability in the planning experiments without introducing unknown yukawa couplings @xmath188 other than @xmath189 , the circle mark `` @xmath316 '' is shown . if there is an observability , but it depends on unknown yukawa couplings ( other than @xmath189 ) , `` @xmath317 ) '' is indicated . if there is an observability when the ( currently unknown ) experimental improvement is achieved , the triangle mark `` @xmath318 '' is shown . if the event rate is expected to be too small to be observed , `` @xmath319 '' is shown . [ cols="^,^,^",options="header " , ] xx g. aad _ et al . _ [ atlas collaboration ] , `` observation of a new particle in the search for the standard model higgs boson with the atlas detector at the lhc , '' phys . b * 716 * , 1 ( 2013 ) [ arxiv:1207.7214 [ hep - ex ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , `` observation of a new boson at a mass of 125 gev with the cms experiment at the lhc , '' phys . lett . b * 716 * , 30 ( 2013 ) [ arxiv:1207.7235 [ hep - ex ] ] . j. d. bjorken and s. weinberg , `` a mechanism for nonconservation of muon number , '' phys . lett . * 38 * , 622 ( 1977 ) . s. l. glashow and s. weinberg , `` natural conservation laws for neutral currents , '' phys . d * 15 * , 1958 ( 1977 ) . b. mcwilliams and l. f. li , `` virtual effects of higgs particles , '' nucl . b * 179 * , 62 ( 1981 ) . o. u. shanker , `` flavor violation , scalar particles and leptoquarks , '' nucl . b * 206 * , 253 ( 1982 ) . t. p. cheng and m. sher , `` mass matrix ansatz and flavor nonconservation in models with multiple higgs doublets , '' phys . d * 35 * , 3484 ( 1987 ) . v. khachatryan _ et al . _ [ cms collaboration ] , `` search for lepton - flavour - violating decays of the higgs boson , '' arxiv:1502.07400 [ hep - ex ] . talk by pierre savard in european physical society conference on high energy physics 2015 , 27 july 2015 , vienna , austria ; the atlas collaboration , `` search for lepton - flavour - violating @xmath320 decays of the higgs boson with the atlas detector '' , arxiv:1508.03372 [ hep - ex ] . d. aristizabal sierra and a. vicente , `` explaining the cms higgs flavor violating decay excess , '' phys . d * 90 * , no . 11 , 115004 ( 2014 ) [ arxiv:1409.7690 [ hep - ph ] ] . j. heeck , m. holthausen , w. rodejohann and y. shimizu , `` @xmath320 in abelian and non - abelian flavor symmetry models , '' nucl . b * 896 * , 281 ( 2015 ) [ arxiv:1412.3671 [ hep - ph ] ] . a. crivellin , g. dambrosio and j. heeck , `` explaining @xmath321 , @xmath322 and @xmath323 in a two - higgs - doublet model with gauged @xmath324 , '' phys . lett . * 114 * , 151801 ( 2015 ) [ arxiv:1501.00993 [ hep - ph ] ] . y. omura , e. senaha and k. tobe , `` lepton - flavor - violating higgs decay @xmath325 and muon anomalous magnetic moment in a general two higgs doublet model , '' jhep * 1505 * , 028 ( 2015 ) [ arxiv:1502.07824 [ hep - ph ] ] . s. p. das , j. hernndez - snchez , a. rosado and r. xoxocotzi , `` flavor violating signatures of lighter and heavier higgs bosons within two higgs doublet model type iii at the lhec , '' arxiv:1503.01464 [ hep - ph ] . a. crivellin , g. dambrosio and j. heeck , `` addressing the lhc flavor anomalies with horizontal gauge symmetries , '' phys . d * 91 * , no . 7 , 075006 ( 2015 ) [ arxiv:1503.03477 [ hep - ph ] ] . d. das and a. kundu , `` two hidden scalars around 125 gev and h , '' phys . rev . d * 92 * , no . 1 , 015009 ( 2015 ) [ arxiv:1504.01125 [ hep - ph ] ] . c. x. yue , c. pang and y. c. guo , `` lepton flavor violating higgs couplings and single production of the higgs boson via e@xmath326 collision , '' j. phys . g * 42 * , 075003 ( 2015 ) [ arxiv:1505.02209 [ hep - ph ] ] . w. huang and y. l. tang , `` flavor anomalies at the lhc and the r - parity violating supersymmetric model extended with vectorlike particles , '' phys . d * 92 * , no . 9 , 094015 ( 2015 ) [ arxiv:1509.08599 [ hep - ph ] ] . s. baek and z. f. kang , `` naturally large radiative lepton flavor violating higgs decay mediated by lepton - flavored dark matter , '' arxiv:1510.00100 [ hep - ph ] . e. arganda , m. j. herrero , r. morales and a. szynkman , `` analysis of the @xmath335 decays induced from susy loops within the mass insertion approximation , '' arxiv:1510.04685 [ hep - ph ] . d. aloni , y. nir and e. stamou , `` large br(@xmath336 ) in the mssm , '' arxiv:1511.00979 [ hep - ph ] . k. a. assamagan , a. deandrea and p. a. delsart , `` search for the lepton flavor violating decay @xmath337 at hadron colliders , '' phys . d * 67 * , 035001 ( 2003 ) [ hep - ph/0207302 ] . a. brignole and a. rossi , `` lepton flavor violating decays of supersymmetric higgs bosons , '' phys . b * 566 * , 217 ( 2003 ) [ hep - ph/0304081 ] . s. kanemura , k. matsuda , t. ota , t. shindou , e. takasugi and k. tsumura , `` search for lepton flavor violation in the higgs boson decay at a linear collider , '' phys . b * 599 * , 83 ( 2004 ) [ hep - ph/0406316 ] . e. arganda , a. m. curiel , m. j. herrero and d. temes , `` lepton flavor violating higgs boson decays from massive seesaw neutrinos , '' phys . d * 71 * , 035011 ( 2005 ) [ hep - ph/0407302 ] . s. kanemura , t. ota and k. tsumura , `` lepton flavor violation in higgs boson decays under the rare tau decay results , '' phys . d * 73 * , 016006 ( 2006 ) [ hep - ph/0505191 ] . g. blankenburg , j. ellis and g. isidori , `` flavour - changing decays of a 125 gev higgs - like particle , '' phys . lett . b * 712 * , 386 ( 2014 ) [ arxiv:1202.5704 [ hep - ph ] ] . r. harnik , j. kopp and j. zupan , `` flavor violating higgs decays , '' jhep * 1303 * , 026 ( 2013 ) [ arxiv:1209.1397 [ hep - ph ] ] . k. hagiwara , r. liao , a. d. martin , d. nomura and t. teubner , `` @xmath338 and @xmath339 re - evaluated using new precise data , '' j. phys . g * 38 * , 085003 ( 2011 ) [ arxiv:1105.3149 [ hep - ph ] ] . s. kanemitsu and k. tobe , `` new physics for muon anomalous magnetic moment and its electroweak precision analysis , '' phys . d * 86 * , 095025 ( 2012 ) [ arxiv:1207.1313 [ hep - ph ] ] . d. chang , w. s. hou and w. y. keung , `` two loop contributions of flavor changing neutral higgs bosons to @xmath24 , '' phys . d * 48 * , 217 ( 1993 ) [ hep - ph/9302267 ] . a. m. baldini _ et al . _ , `` meg upgrade proposal , '' arxiv:1301.7225 [ physics.ins-det ] . g. w. bennett _ et al . _ [ muon ( g-2 ) collaboration ] , `` an improved limit on the muon electric dipole moment , '' phys . d * 80 * , 052008 ( 2009 ) [ arxiv:0811.1207 [ hep - ex ] ] . y. k. semertzidis _ et al . _ , `` sensitive search for a permanent muon electric dipole moment , '' aip conf . proc . * 564 * , 263 ( 2001 ) [ hep - ph/0012087 ] ; r.m . carey _ et al . _ `` ags letter of intent : search for a permanent muon electric dipole moment '' , feb . 2000 ; a. silenko _ et al . _ [ muon edm collaboration ] , `` j - parc letter of intent : search for a permanent muon electric moment at the @xmath340 level '' , jan . b. aubert _ et al . _ [ babar collaboration ] , `` measurements of charged current lepton universality and @xmath341 using tau lepton decays to @xmath342 , @xmath343 , @xmath344 and @xmath345 , '' phys . lett . * 105 * , 051602 ( 2010 ) [ arxiv:0912.0242 [ hep - ex ] ] . a. crivellin , a. kokulu and c. greub , `` flavor - phenomenology of two - higgs - doublet models with generic yukawa structure , '' phys . rev . d * 87 * , no . 9 , 094031 ( 2013 ) [ arxiv:1303.5877 [ hep - ph ] ] . a. blondel _ et al . _ , `` research proposal for an experiment to search for the decay @xmath346 '' , dec . 2012 . m. sher , `` tau @xmath347 mu eta in supersymmetric models , '' phys . d * 66 * , 057301 ( 2002 ) [ hep - ph/0207136 ] . d. black , t. han , h. j. he and m. sher , `` tau - mu flavor violation as a probe of the scale of new physics , '' phys . d * 66 * , 053002 ( 2002 ) [ hep - ph/0206056 ] . a. celis , v. cirigliano and e. passemar , `` lepton flavor violation in the higgs sector and the role of hadronic @xmath4-lepton decays , '' phys . d * 89 * , 013008 ( 2014 ) [ arxiv:1309.3564 [ hep - ph ] ] .

the concept of symbolic sequences play important role in study of complex systems . in the work we are interested in ultrametric structure of the set of cyclic sequences naturally arising in theory of dynamical systems . aimed at construction of analytic and numerical methods for investigation of clusters we introduce operator language on the space of symbolic sequences and propose an approach based on wavelet analysis for study of the cluster hierarchy . the analytic power of the approach is demonstrated by derivation of a formula for counting of _ two - fold de bruijn sequences _ , the extension of the notion of de bruijn sequences . possible advantages of the developed description is also discussed in context of applied problem of construction of efficient dna sequence assembly algorithms .

1601.02097

nowadays symbolic sequences is a fundamental concept widely used in various fields of natural sciences @xcite . in bioinformatics , theory of information , and theory of discreet markov chains the intrinsic structure of objects under consideration provides direct mapping on symbolic sequences @xcite . less obvious extension of the symbolic approach one can find in study of complex behavior in dynamical systems @xcite . in essence , the underlying idea is in organization of a stroboscopic sampling of the multidimensional trajectory . in case of hamiltonian systems one constructs a poincar section surface in phase - space @xcite , such that it is oriented orthogonal to the dynamical flow at each point . linearization of the dynamics allows to separate stable and unstable directions of motion and thus define the set of feasible positions at the next crossing of the poincar section surface . all intersections falling within the same sub - region of the surface are designated by a certain symbol . for the system possessing a chaotic dynamics one can aver existence of the markov partitioning of the surface , meaning that each next symbol in the symbolic dynamics is defined only by the previous one . for infinite or cyclic sequences the real trajectory can be uniquely restored from the symbolic dynamics . often , the obtained markov alphabet includes an infinite number of symbols , as , for instance , in chaotic billiards , out of the flow discontinuity on the billiard walls . it is not forbidden , however , to define some other partition prompted by physical ( geometrical ) properties of the system , which would generate a finite alphabet . the fee for such convenience is formation of constraints on combinations of more then two subsequent symbols in the symbolic dynamics . in this article we consider a special case of symbolic dynamics generated by baker s map @xcite . it is a chaotic map from the unit square into itself , its symbolic dynamics do not assume any constraints , i.e. all possible symbolic sequences correspond to some trajectory generated by the baker s map . study of periodic orbits and consequently _ cycled _ symbolic sequences plays a special role in theory of quantum chaotic systems @xcite . unstable periodic orbits form `` skeleton '' of a dynamical system and knowledge of their hierarchy can give one s access to many of the system s dynamical averages , such as , natural measure , lyapunov exponents , fractal dimensions , entropy . they can efficiently expressed in terms of a sum over the unstable periodic orbits @xcite . for instance , by means of semiclassical gutzwiller trace formula @xcite the eigenenergy density of chaotic quantum system can be expressed as a sum over long ( of order heisenberg time @xmath0 , where @xmath1 dimension of the system ) classical periodic orbits and , in turn , the density - density correlations as the double sum . into the correlator formula the orbits come in pairs . the non - trivial oscillatory sub - leading terms in the correlator @xcite result from interference between special type of periodic orbits , such that the partner orbits are close to each other everywhere in configuration space , but the same configuration points are visited in different time - order ( see fig . [ fig4 ] a ) . the time - order switching happens in the regions called encounters , where each of the orbits comes onto the shortest distance to itself , being , at the same time , mostly distanced from the partner . symbolic representation of periodic orbits is not sensitive enough to distinguish points within the encounters and the free loops on the level of a single symbol ( fig . [ fig4 ] b ) . the closeness of the partner orbits is reflected in the fact that the corresponding symbolic sequences being globally different are locally identical , i.e. they both consist of the same set of sub - strings of some given length @xmath2 ( fig . [ fig4 ] c ) . the above picture was recently formalized in @xcite under the notion of @xmath2-closeness : two cyclic symbolic sequences @xmath3 and @xmath4 of the same length @xmath5 are @xmath2-close to each other ( @xmath6 ) if any sub - string of the length @xmath2 appears both in @xmath3 and in @xmath4 the same number of times ( might be zero ) . the properties of the equivalence relation @xmath7 allows to introduce an ultrametric distance @xmath8 on the set of periodic sequences . such distance , in addition to the standard properties of distances , also satisfies the strong triangle inequality : @xmath9 $ ] ( see for instance @xcite ) . in other words , all cyclic sequences of a given length @xmath5 are distributed over hierarchically nested clusters with respect to their @xmath2-closeness and thus they can be classified with respect to their local content , see fig . the @xmath2-closeness can be seen as generalization of de bruijn sequences @xcite , which first appearance can be traced back to the mathematical work of flye sainte - marie @xcite published in 1894 . de bruijn sequence is a cyclic sequence characterized by the property that each possible sub - string of a given length @xmath2 enters the sequence and only once . a ) schematic representation of a periodic orbit with two encounters crossing the poincar section surface in points @xmath10 and @xmath1 ( the colored loops ) . b ) possible choice of the finite partition on the poincar section surface . the 8 letters of the emerging alphabet are encoded by binary numbers . c ) symbolic representation for the pair of periodic orbits . the sequences are different but all triples of symbols containing in the upper sequence enter the same number of times into the lower one . d ) a part of dna sequence and the frequencies of the three - letters sub - strings . e ) graphical representation of the dna sequences as a path on de bruijn graph.,title="fig : " ] + a ) schematic representation of a periodic orbit with two encounters crossing the poincar section surface in points @xmath10 and @xmath1 ( the colored loops ) . b ) possible choice of the finite partition on the poincar section surface . the 8 letters of the emerging alphabet are encoded by binary numbers . c ) symbolic representation for the pair of periodic orbits . the sequences are different but all triples of symbols containing in the upper sequence enter the same number of times into the lower one . d ) a part of dna sequence and the frequencies of the three - letters sub - strings . e ) graphical representation of the dna sequences as a path on de bruijn graph.,title="fig : " ] clustering of cyclic binary sequences of the length @xmath11 ( only half of the tree , 94 sequences out of 188 , is shown ) . the end - points of the tree ( black circles ) represent the sequences , the numbers in the circles give the total number of sequences in the corresponding cluster . the tree is constructed by the procedure developed in section 3 . cluster of sequences with @xmath12 ones and 9 zeros contains 5 sequences , with @xmath13 ones 15 , with @xmath14 30 , with @xmath15 42 , the sequences with @xmath16 do not form non - trivial clusters , see eq . ( [ cluster ] ) . parameter @xmath17 determines the number of branches on the level @xmath18 of the tree ( first from the top ) , the tree brunches up to the maximal level @xmath19 , see section 3.4 . ] interesting , that the similar need in searching of symbolic sequences from the set of their sub - strings emerges in area absolutely different from dynamical systems , the problem of contiguous dna sequence assembly from the shorter dna fragments @xcite . in more detailed exposition , the procedure of genome reading represents a set of multiple readings of small dna pieces called reads , while reconstruction of the whole original nucleotide sequence becomes an algorithmic problem of assembling a number of reads into a single sequence . in the modern _ de novo _ assembly approach the sequence assembly is considered as a search of an eulerian cycle ( path on a graph which traverses all edges and each only once ) on a directed graph @xcite ( see fig . [ fig4 ] d , e ) , a sub - graph of de bruijn graph ( more precisely de bruijn - good graph @xcite ) . the vertices of the sub - graph are associated with the strings of a fixed length @xmath2 extracted from the set of all reads . two vertices are connected if the corresponding strings share an exact @xmath20 overlap ( exact definition of de bruijn graph and its properties will be discussed in section 2.3 ) . taking into account that a certain sub - string can appear in the native dna several times one have to deal with sequences possessing a certain encounter structure . there is a number of principal questions arising in the field , which has been reported in the literature : how the number of possible realizations of sequences behaves with the length of the sequences segments , @xmath2 , and the related question what is the optimal length of the reads @xcite ? the computational bottleneck in realization of de novo algorithms is the amount of memory required for saving the de bruijn sub - graph data , which , obviously , becomes central for assembling of large genomes @xcite . generally speaking , both the study of topological structures of periodic orbits in dynamical systems and the problem of dna sequence assembly are based on the same mathematical apparatus . our recent finding of the ultrametric structure of the set of symbolic sequences @xcite allows to extend our understanding of the problems and develop novel analytic and algorithmic approaches . some of these ideas we demonstrate in the present article . moreover , we believe that the introduced below operator language and wavelet analysis on cyclic symbolic sequences will be useful as well for formulation of advanced theoretical approaches in theories describing multi - scale stochastic processes on the set of symbolic sequences ( see for instance @xcite ) . in section 2 we introduce the operator formalism for the work with symbolic sequences in their frequency representation and reproduce the main known facts about the ultrametric structure of sequences ( section 2.2 ) , their geometric representation by means of de bruijn graphs , as well as formula for the number of de bruijn sequences obtained by counting euler and hamiltonian cycles on the de bruijn graphs ( section 2.3 ) . here ( section 2.4 ) we also repeat the known results regarding the distribution of clusters of symbolic sequences and derive some related basic formulae . the main analytic structures discussed in the present research are rising ( section 2.2 ) and lowering ( section 3.1 ) operators on the cluster tree . analysis of them in basis of wavelets , constructed by analogy with @xmath2-adic wavelets @xcite ( see also @xcite ) , allows to define rigorous procedure for reconstruction of sequences belonging to certain clusters ( see example on fig . [ fig5 ] ) . according to the procedure , on each hierarchical level one reduces the problem to the solution of a number of independent linear systems of diophantine equations ( section 3.2 ) . the main advantage of the approach is that after a certain hierarchy level @xmath2 , such that @xmath21 ( @xmath22 is the number of letters in alphabet , @xmath5 is the length of the sequence ) there is only a finite number of relevant solutions and this number decreases as @xmath2 increases . as we show in section 3.3 , the `` center of masses '' of our approach lies in analysis of connectivity of the obtained sequences . such `` factorisation '' of the problem made possible derivation of a formula , eq . ( [ mainformula ] ) , for counting of two - fold binary de bruijn sequences . the notion of two - fold de bruijn sequences is introduced in section 3.5 . finally , in section 4 we discuss obtained results and consider possible applications of the revealed classification of sequences in various fields . let @xmath23 denotes a set of all strings of @xmath5 characters chosen from @xmath22-symbols alphabet @xmath24 . the total number of such sequences is known to be @xmath25 , we write @xmath26 . it is convenient to enumerate the letters of the alphabet @xmath24 by integer numbers from zero to @xmath27 . let @xmath28 be an @xmath29-dimensional vector space with the scalar product defined in a standard way . let the set @xmath30 be the elementary basis in the linear space @xmath28 . elementary basis consists of column - vectors , such that all entries of the @xmath31th vector , @xmath32 , except the @xmath31th one , are zeros . below , we also use notation @xmath33 for the elementary basis vectors of the space @xmath34 . let us define the bijective transformation @xmath35 between the symbolic sequences from @xmath23 and the basis vectors of the space @xmath28 by the rule @xmath36\in x_n^ { \mathcal{a}_\ell}\to \bigotimes_{k=1}^n e_{1+\alpha_k}\equiv e^{(n)}_{1+\sum_{k=1}^n \alpha_k \ell^{n - k}}\in v_{\ell^n}.\ ] ] here the symbol @xmath37 stands for the kronecker product , such that @xmath38 representation of matrices in terms of kronecker product is useful out of its basic properties : @xmath39 and @xmath40 . here and below the symbol @xmath41 stands for transponation and , if applicable , simultaneous complex conjugation . to proceed with the formalization we introduce the reduction operator @xmath42 for @xmath43 . contrary to the transformation @xmath35 the aim of the latter construction is to project only the first @xmath2 symbols of the sequence @xmath44\in x_n^ { \mathcal{a}_\ell}$ ] onto the vector space of reduced dimensionality , ignoring information of other symbols : @xmath45= \bigotimes_{k=1}^p e_{1+\alpha_k}\equiv e^{(p)}_{1+\sum_{k=1}^p \alpha_k \ell^{p - k}}\in v_{\ell^p}.\ ] ] in terms of direct product the operator takes the following form ( here and below @xmath46 is @xmath47 identity matrix and for matrices @xmath48 we omit the index @xmath22 ) @xmath49 where @xmath50 means the left multiplication on the vector of one s , i.e. @xmath51 . it is assumed that at @xmath52 the matrix @xmath53 coincides with the @xmath54 identity matrix . let us finally introduce operator @xmath55 of a cyclic shift of symbols in symbolic sequence of the length @xmath5 , such that : @xmath56\to [ \alpha_2,\dots \alpha_n,\alpha_1]$ ] . its matrix representation in the basis @xmath57 , ... @xmath58 is given by the formula @xmath59 with the help of the shift operator one can define the set of cyclic sequences @xmath60 as the factor set @xmath23 over @xmath55 , @xmath61 . the volume of the space can be estimated by the combinatorial formula known as the number of necklaces @xcite @xmath62 where the sum runs over @xmath1 , the divisors of @xmath5 ( including @xmath63 and @xmath64 ) , and @xmath65 is the euler s totient function , @xmath66 means that the product runs over all prime divisors of @xmath1 . for @xmath5 prime the formula becomes particularly simple @xmath67 note also that from ( [ numbersec ] ) one can see that the number of non - prime cyclic sequences is negligible in comparison with the number of prime sequences in the limit @xmath68 . in the article we are interested in the properties of the combined projecting operator @xmath69 the operator @xmath70 acting on the vector @xmath71\in v_{\ell^n}$ ] has the image in @xmath28 . its action on the basis vector @xmath72 produces a vector @xmath73 with non - negative integer entries , we write @xmath74 . this set we denote as @xmath75 . the @xmath76th entry of the vector @xmath77 can be seen as frequency , i.e. the number of entrances ( might be zero ) of the string @xmath78 $ ] of the length @xmath2 , such that @xmath79 , into the original sequence @xmath80 $ ] . consider some properties of @xmath70 . each vector @xmath77 has the property @xmath81_j = n.\ ] ] this follows from the fact that @xmath70 is a sum of @xmath5 terms ( see eq . ( [ ppn ] ) ) , each of them produces an elementary basis vector in @xmath28 . in particular , action of @xmath70 on the vector corresponding to the sequence with identical symbols produces a vector proportional to @xmath5 : @xmath82 this is reflection of the generic property , that if some sequence ( @xmath72 ) has the minimal period @xmath1 , then all entries of @xmath83 are proportional to @xmath84 . all sequences obtained as a shift of a given one , i.e. vectors @xmath85 , correspond to one and the same image @xmath83 . it follows directly from the structure of @xmath70 . below , to avoid this trivial degeneracy we prefer to work with the operator ( [ ppn ] ) acting on the factor space @xmath86 . let us now introduce a raising operator @xmath87 which acts by the rule : for all @xmath88 @xmath89 its matrix form can be read off from the relation @xmath90 , it has the form @xmath91 note , that the form of operator @xmath92 does not depend on @xmath5 . let @xmath93 , @xmath94 . the number of sequences is @xmath95 , the size of the factor set @xmath96 . the elementary basis has the form @xmath97 the operators @xmath98 and @xmath99 in this basis have the forms @xmath100 operator @xmath101 has the following matrix representation if it acts in the space @xmath102 ( on the left ) and in the space @xmath103 ( the right column ) @xmath104 as one can see vectors from the sets @xmath105 and @xmath106 have integer entries and support the general results ( [ xpnnorm ] ) and ( [ ppnnorm ] ) . one can reveal ( generically not uniquely ) the corresponding symbolic sequence form @xmath107 . for instance , @xmath108_j\\ 1&&0&0&&1\\ 2&&0&1&&1\\ 3&&1&0&&1\\ 4&&1&1&&0 \end{array}\ ] ] correspond to the cyclic sequence @xmath109 $ ] . now we are at position to discuss the hierarchical structure of the set of cyclic sequences . first we introduce the notion of @xmath2-closeness ( see @xcite ) . [ pclose ] two sequences @xmath110 are @xmath2-close , @xmath111 , if their images @xmath112 and @xmath113 are equal @xmath114 . there are three important properties of the equivalence relation @xmath7 : * the relations @xmath115 and @xmath116 also imply that @xmath117 ; * the relation @xmath115 implies that @xmath118 . * the relation @xmath118 _ does not _ imply that @xmath111 . the first is the direct sequence of the definition [ pclose ] , while the second follows from the linear properties of the raising operator ( [ raising ] ) , ( [ raising2 ] ) . the third sentence , in particular , means that the lowering operator translating @xmath119 into @xmath120 can not be constructed as a linear operator . the latter claim will be clarified in the part 3 . from the properties i , ii , iii it follows that all cyclic sequences can be distributed over hierarchically nested clusters with respect to their closeness . using the notion of @xmath2-closeness one can naturally introduce the ultrametric distance on the set @xmath121 in the following way : the ultrametric distance @xmath122 between @xmath123 is defined as @xmath124 the distance @xmath122 is positive , symmetric and satisfies the strong triangle inequality : @xmath125 the latter can be easily proved . assume that @xmath126 , @xmath127 but @xmath128 and @xmath129 . from the definition of @xmath2-closeness it follows that @xmath130 and therefore @xmath131 but @xmath132 . the function @xmath133 can be calculated with the help of the operator @xmath92 . if @xmath134 are respectively represent the sequences @xmath135 and @xmath136 then @xmath137 before we start our study of the clusters it is instructive to discuss first the graphical representation of sequences and some basic results . the simplest way to represent the set of sequences @xmath23 graphically is to use regularly branching tree . setting the `` left '' end of the sequence onto the root ( the top point ) of the tree one goes down choosing the branches according to the current letter within the sequence ( see appendix [ apptree ] ) . thus the `` leaves '' are enumerated by integer numbers from @xmath138 to @xmath25 , i.e. by the indexes of the elementary basis vectors of the space @xmath139 . the obtained tree obviously generates ultrametrics on the set of symbolic sequences , _ but _ does not reflect the idea of equivalence of symbols with respect to their positions in the sequence and dissimilarity of symbols with respect to their local surrounding . to this end a more adequate graphical structure , de bruijn graph , has been proposed @xcite . the notion of de bruijn graph is used here for a class of directed graphs @xmath140 labeled by the integer numbers @xmath2 and @xmath22 . graph @xmath140 has @xmath29 vertices connected by @xmath141 directed edges . each vertex is designated with the symbolic string from @xmath142 , figure [ fig6 ] . for our purposes , it is convenient also to mark the vertices by the corresponding basis vectors @xmath143 ( exceptions are the graphs @xmath144 and @xmath145 , otherwise one has to work with zero- and one - dimensional spaces ) , where the vector index @xmath31 is calculated according to the rule ( [ trivultrametricity ] ) . two vertices @xmath32 and @xmath146 are connected by the edge with the label @xmath147 ( @xmath148 ) iff @xmath149 , where @xmath150 is the incidence matrix : @xmath151 consider now some properties of the graph @xmath140 . first , vertices have exactly @xmath22 outgoing and @xmath22 incoming edges . therefore , the graph is eulerian , i.e. one can find a path through all vertices , such that it traverses each edge only once . second , the @xmath152 adjacency matrix @xmath153 of the graph @xmath140 can be restored from matrices @xmath92 and @xmath150 . indeed , starting from some vertex @xmath32 we obtain all possible outgoing edges by the formula @xmath154 . the backward projection onto the set of vertices can be done with the help of @xmath150 . finally we obtain @xmath155 by analogy one can derive the expression for the connectivity @xmath156 matrix @xmath157 ( the adjacency matrix for edges ) . it is @xmath158 , which is , on the other hand , the vertex adjacency matrix @xmath159 of the graph @xmath160 . due to equivalence @xmath161 via the line graph construction @xcite each graph @xmath140 is eulerian and hamiltonian . the number of hamiltonian paths on the graph @xmath160 is equal to the number of eulerian paths on the graph @xmath140 . de bruijn graphs , from left to right , @xmath162 , @xmath163 , @xmath164 . the edges and vertices are encoded by binary sequences , see text for explanation . ] the sequences corresponding to the eulerian paths on de bruijn graph have a joint name of de bruijn sequences @xcite . the de bruijn sequence is characterized by the fact that each possible sub - string of the length @xmath2 enters the sequence and only once , therefore its length is @xmath165 . below we give definition of de bruijn sequences in terms introduced in this article and as a special case of @xmath166-fold de bruijn sequences . for given positive integer @xmath2 , and @xmath166 , the @xmath2-ary @xmath166-fold de bruijn sequences from the set @xmath167 is the set of @xmath2-close sequences corresponding to the frequency vector @xmath168 the set of @xmath2-ary single - fold de bruijn sequences coincide with the set of @xmath2-ary de bruijn sequences . the number of @xmath2-ary single - fold de bruijn sequences can be calculated with the help of best theorem @xcite . it states that the number of eulerian cycles for some directed graph @xmath169 with @xmath5 vertices @xmath170 can be calculated by the formula : @xmath171\prod_{i=1}^n(d_{out}(v_i)-1)!\ , , \ ] ] where @xmath172 is the adjacency matrix , @xmath173_{i , j}\right\}}}_{i=1,n}$ ] , the factor @xmath174 is the number of outgoing edges at vertex @xmath170 . the abbreviation @xmath175 stands for cofactor taken for some arbitrary chosen raw @xmath176 . in case of @xmath177 the second multiplier gives @xmath178 . calculation of the cofactor means that one should remove from @xmath179 the elements in the @xmath31-th raw and @xmath31-th column . let @xmath180 , then one obtains @xmath181=\det\left[\ell \bigotimes_{k=1}^{p-1 } { \mathds{1}}-q_{\ell^{p-1}}^*\right],\ ] ] where the matrix @xmath182 is obtained from @xmath183 by dropping the first raw and the first line . the structure of the graph manifests itself in the special relations for traces , @xmath184 , @xmath185 , and in a special form of the secular equation @xcite for @xmath182 : @xmath186 substitution of @xmath187 into the left - hand - side of the above equation allows us to calculate the sought cofactor and to get the famous formula for the number of the @xmath2-ary de bruijn sequences , it is @xmath188 . first the formula was obtained for @xmath94 in @xcite and later extended for generic @xmath22 @xcite . here we reproduce some results describing the distribution of clusters , see particular realization of the cluster tree on the fig . [ fig5 ] . the zero level ( @xmath189 ) of the cluster tree contains only one cluster @xmath190 of all cyclic sequences of the length @xmath5 . the number of such sequences is given by the formula ( [ numbersec ] ) . the first level ( @xmath18 ) consists of the clusters , which number is given by the simple combinatorial formula @xmath191 . the number of sequences in each cluster of the level @xmath18 can be calculated by burnside s lemma @xcite , it extends the result ( [ numbersec ] ) . let the frequency vector @xmath192 is equal to @xmath193 $ ] , then the number of sequences can be calculated from the expression @xmath194\big]\right\vert}=\frac{1}{n}\sum_{d|{\left\{a_1,a_2,\dots , a_\ell\right\}}}\phi ( d ) \frac{(n / d)!}{(a_1/d)!\dots(a_\ell / d)!},\ ] ] where @xmath1 are common divisors of @xmath195 . since @xmath196 we come to the obvious conclusion , that the cluster corresponding to the sequence of identical elements contains only one sequence . the same result one obtains for the sequence with only one non - identical element ( @xmath197 ) . there are some known asymptotic results obtained for the distribution of clusters in case of two - letters alphabet @xcite . in particular , by means of heuristic arguments it was shown that the total number of clusters in the regime of finite @xmath2 and large @xmath5 grows as @xmath198 , where the coefficient @xmath199 can be calculated explicitly for small values of @xmath2 ( @xmath200 ) . the asymptotics for the average sizes of clusters , @xmath201 , at different ratios of @xmath2 to @xmath5 were derived with the help of the random matrix theory methods . the asymptotic of @xmath201 solely depends on the average number of encounters . for @xmath202 , when the encounters are rare , i.e. the majority of frequency vectors from @xmath75 are @xmath203-vectors , the average cluster size is one , @xmath204 , and the number of clusters is almost equal to the total number of cyclic sequences . decreasing of the parameter @xmath2 leads to formation of encounters . as the average number of encounters becomes of the order one , an exponential growth of cluster sizes starts . for @xmath205 one can find that @xmath206 . in the regime @xmath207 the average size is @xmath208 . in this regime almost every point of a generic cyclic sequence belongs to some encounter and the further growth of @xmath5 does not lead to any essential increase of the number of encounters . this , in turn , results in a slower growth of cluster sizes , @xmath209 . the size of the largest cluster , @xmath210 in the limit of long sequences @xmath211 and finite @xmath2 asymptotically behaves like @xmath212 . note , that to the leading order of @xmath5 the periodic orbits belonging to the largest cluster cover the graph @xmath213 uniformly , i.e. the elements of the corresponding frequency vector @xmath107 are almost identical . the latter observation stimulated our interest to the calculation of the number of two - fold de bruijn sequences ( see section 3.4 ) . generically the distribution of the cluster sizes in the limit of long sequences and finite @xmath2 can be described by the probability density that a randomly chosen cyclic sequence belongs to a cluster of the size less then @xmath214 , @xmath215 $ ] , it is @xmath216 one can also derive the expression for the probability of finding @xmath31 randomly chosen sequences of the same length to be belonging to the same cluster , @xmath217 the above two results were derived from estimation of the moments of cluster sizes in @xcite . the second moment was previously considered in @xcite . classification of long periodic orbits with respect to the number of their encounters and approaches for counting of the corresponding cluster sizes one can find in @xcite . for instance , the clusters of sequences having @xmath218 encounters well separated by non - intersecting free parts ( see fig . [ fig4 ] ) contain @xmath219 sequences . the formula for the number of short length binary cyclic sequences ( @xmath220 ) has been derived in @xcite . it is clear that the lowering operator can not be organized by the same simple way as @xmath222 . it is a non - linear multi - valued operator , generating a set of vectors @xmath223 out of the single vector @xmath77 . thus the problem of finding the lowering operator is , in fact , the problem of restoring information from the data given in integrated form . to this end , we propose an algorithmic solution . to simplify notations it is assumed along this section that @xmath5 is fixed and we settle the following notations * let @xmath224 ; * let @xmath225 , such that @xmath226 ; * let @xmath227 , such that for all @xmath228 , @xmath229 ; * let @xmath230\equiv{\left\{\bar z\right\}}\cap \mathcal{f}_{p+1,n}$ ] . the lowering operator acts in two steps * restoring of the set @xmath231 from the information integrated in @xmath4 ; * taking the operation of intersection in @xmath232 to obtain @xmath230 $ ] . consider both steps separately . here we formulate a set of equations which generate all vectors from @xmath231 and demonstrate the most moderate method to resolve them based on the usage of wavelet basis . elements of the set @xmath231 satisfy the following set of diophantine equations @xmath233 to make sure in correctness of the above proposition it is instructive to use the analogy of cyclic sequences with the periodic trajectories on the de bruijn graph . the @xmath76th entry ( @xmath234 ) of the vector @xmath228 contains information of the number of times the trajectory passes through the vertex @xmath235 . among all trajectories we choose only the `` smooth '' one s , i.e. those for which the balance conditions hold : the incoming and outgoing `` currents '' in each vertex must be equal . this gives @xmath236 the system of linear equations ( [ closeness ] ) has to be supplemented by the initializing conditions , @xmath237 . it and ( [ closeness ] ) generate the set of equations ( [ eq1 ] ) and ( [ initialization ] ) . the obtained equations form a system of linear diophantine equations subject to the condition that solutions @xmath228 has to have non - negative entries , i.e. @xmath238 , eq . ( [ condition ] ) . to resolve the obtained system ( [ eq1])-([condition ] ) we use the basis of wavelets , constructed by analogy with the basis of @xmath2-adic wavelets @xcite . to build the basis vectors we utilize the ( non - normalized ) basis @xmath239 of discreet fourier transform . it is orthogonal on the space @xmath34 , the vectors elements are @xmath240_i = e^{\frac{2\pi { \textrm{i}}}{\ell}(i-1)j},\qquad i=1,\dots,\ell,\qquad j=0,\dots,\ell-1.\ ] ] its orthogonality follows from the nullification of the sum @xmath241 . the vectors of the wavelet basis in the space @xmath242 for @xmath243 have the form @xmath244 @xmath245 the total number of vectors is @xmath29 , their orthogonality follows from orthogonality of vectors @xmath246 and the mutual orthogonality of @xmath247 and @xmath248 . consider now the action of operators @xmath150 and @xmath92 on the above basis vectors : @xmath249 @xmath250 @xmath251 it follows from ( [ initialization ] ) and ( [ rp1 ] ) that the first @xmath29 coefficients ( corresponding to @xmath252 ) in the wavelet expansion of @xmath228 , @xmath253 coincide , up to the factor @xmath254 , with the corresponding coefficients @xmath255 in the expansion of @xmath4 , @xmath256 on the other hand since @xmath257 we derive @xmath258\psi^{(p)}_{\gamma-1,j,{\left\{\alpha_k\right\}}}\\+\sum_{j=1}^{\ell-1}\sum_{{\left\{\alpha_k\right\}}}\left[\sum_\beta c_{p , j,{\left\{\beta,\alpha_k\right\}}}\right]\psi^{(p)}_{p-1,j,{\left\{\alpha_k\right\}}}.\end{gathered}\ ] ] the similar relation between the coefficients in wavelet expansions of @xmath3 and @xmath4 results in @xmath259\psi^{(p-1)}_{\gamma-1,j,{\left\{\alpha_k\right\}}}.\ ] ] notice , that substitution of the coefficients ( [ ctilde ] ) into the latter equation ( [ xwave ] ) and simultaneous direct multiplication of ( [ xwave ] ) on @xmath248 results in @xmath260\psi^{(p)}_{\gamma-1,j,{\left\{\alpha_k\right\}}}.\ ] ] it , together with ( [ lpzbar ] ) , gives the set of equations for the @xmath261 unknown coefficients @xmath262 @xmath263\psi^{(p)}_{p-1,j,{\left\{\alpha_k\right\}}},\ ] ] by analogy wit the above calculations , from ( [ ctilde ] ) we obtain the set of constraints for the coefficients @xmath262 @xmath264 the system of equations ( [ maineq ] ) , ( [ mainconstr ] ) is the result of the first step of the algorithm . the advantage of such reformulation of the original set of diophantine equations ( [ eq1 ] ) , ( [ initialization ] ) and ( [ condition ] ) is that now the whole set of equations , due to the special structure of wavelet basis , is divided by @xmath265 systems of @xmath27 linearly independent equations on @xmath22 variables containing also @xmath266 conditions . notice , that in case when @xmath267 one has to solve maximally @xmath5 system of equations , while the rest automatically generates trivial solutions only . for composite @xmath5 solution of equations ( [ maineq ] ) , ( [ mainconstr ] ) corresponding to the sequence with non - trivial period @xmath1 in case when @xmath268 results in vector @xmath228 , which non - zero components are all equal to @xmath1 and their total number is @xmath84 . in case @xmath269 the vector @xmath270 contains only four non - zero elements . there positions within the vector are marked by the indices @xmath271 , @xmath272 , @xmath273 , @xmath274 . after simplification of the left - hand - side of ( [ maineq ] ) for each @xmath275 we arrive to three linearly independent equations for four complex variables @xmath276 , @xmath277 , @xmath278 , @xmath279 : @xmath280 among all solutions of the above system one has to choose those that produce non - negative integer values of @xmath228 calculated from the formulae ( @xmath281 ) : @xmath282 the case @xmath94 is considered in details in section 3.4 . the system of @xmath29 linear equations ( [ maineq ] ) , ( [ mainconstr ] ) contains @xmath141 variables and has more then one solution . some of them can correspond to disjoint trajectories on the graph . therefore the second step of the algorithm : `` taking the intersection operation in @xmath232 '' means separation of the disjoint trajectories from the closed ones . let us , first , reproduce the analytic non - linear criterion suggested in @xcite . let @xmath283 denote scalar multiplication of vector @xmath228 and vector @xmath284 $ ] , let @xmath285 is a diagonal matrix of phases @xmath286 , and @xmath287 $ ] is a matrix obtained by multiplication @xmath288 , then @xmath228 corresponds to a cyclic sequence iff the integral @xmath289}\frac{d\phi_i}{2\pi}\right]\cdot e^{-{\textrm{i}}\;\bar z\cdot\phi}\;{\textrm{tr}\;}\big(a_{\ell^{p+1}}[\bar z]\lambda_{\phi}\big)^{n},\ ] ] is positive . indeed , the trace of matrix raised to the power @xmath5 is a sum of all possible products of @xmath5 matrix elements taken along the legitimate closed paths on the sub - graph isolated from the frame by the special choice of the adjacency matrix @xmath287 $ ] . the phases marks the individual paths on the sub - graph and the integral nullifies those that passes not all edges . moreover , the non - zero part of the integral , @xmath290 , gives the number of @xmath291-close cyclic sequences , which correspond to the frequency vector @xmath228 . this proofs the proposition 2 . the multidimensional integral ( [ integral ] ) is a good object for analytic study @xcite , while the practical calculation would require significant amount of resources . in applications it can be replaced by the direct test of the matrix @xmath287 $ ] . the trajectory is closed if the corresponding sub - graph is connected , i.e. the adjacency matrix @xmath287 $ ] consists of only one non - zero block . matrix @xmath287 $ ] has a special properties , which simplifies the testing : * if @xmath228 contains no zeros , then the trajectory passes through all vertices of the graph @xmath140 and the graph is obviously connected . * each row and each column of the matrix @xmath287 $ ] contains not more then @xmath22 non - zero elements or does not contain them at all . to demonstrate it let us represent the vector @xmath228 as a direct sum of its @xmath22 parts , such that first @xmath22 entries belong to the first part , the second @xmath22 entries belongs to the second part and so on ... , i.e. @xmath292 . the total size of non - zero blocks of the matrix @xmath287 $ ] coincides with the number of non - zero entries of the vector @xmath293 . other vertices of the graph @xmath140 are not visited . thus the total size of the non - zero matrix block ( blocks ) is not larger then @xmath5 . presence of a single block in the matrix @xmath287 $ ] on the graph s language means existence of at least one spanning tree @xcite . thus one can either check the non - nullification of the cofactor , see eq . ( [ cofactor ] ) , constructed for the graph with the adjacency matrix @xmath287 $ ] , or rearrange the vertices ( permute the rows and columns in the matrix ) to reveal the block structure explicitly . the latter strategy , out of its effectiveness in case of sparse matrices , has been utilized in the numerical construction of the clusters presented on fig . [ fig5 ] . the known generic results regarding the cluster structure are gathered in section 2.4 . here we discuss a few more observations for the case @xmath94 . the set of equations ( [ maineq ] ) , ( [ mainconstr ] ) in case @xmath94 takes the form @xmath294 for convenience we introduced the running index @xmath295 . let us consider level @xmath18 , here the frequency vector can be represented as @xmath296 $ ] . the symmetry of clusters allows us to define @xmath17 to be less then the integer part of @xmath297 , we also drop the trivial clusters , @xmath16 . parameter @xmath17 determines the number branches at the level @xmath18 ( see for example tree on the fig . [ fig5 ] ) . indeed , one can check that the number of solutions of equations ( [ y1])-([y2 ] ) corresponding to the closed graphs is equal to @xmath17 , see fig [ fig2 ] and the text in the caption . for instance , @xmath298 our numerical study of binary cyclic sequences ( see for example fig . [ fig5 ] ) shows that for prime @xmath5 the maximal level , where the tree is branching , is reached in the cluster starting from the frequency vector @xmath299 $ ] . this universal property can be explained within the proposed method of wavelet analysis by the following arguments . on the @xmath2th step of hierarchy one has to solve the system ( [ y1])-([y2 ] ) . on average all non - zero elements of @xmath300 are half of the elements of @xmath301 . this principle is violated if the difference between the nearest elements , @xmath302 , is large . an additional requirements is that the sequence should repeat itself as many times as possible , i.e. to have encounter of the largest length and order . therefore the second scenario , when @xmath303 , is the preferable one . on the level @xmath304 the resulting frequency vectors are @xmath305 $ ] . the procedure can be iterated till the level @xmath306 . thus @xmath12 and @xmath13 give the maximal number of iterations . the obtained after the iterations vectors are @xmath307 $ ] for @xmath12 and @xmath308 $ ] for @xmath13 . the latter vector still contains one more encounter , it corresponds to two choices of symbolic sequences : @xmath309 $ ] and @xmath310 $ ] ( note , that geometrically the first entry of @xmath4 means the number of visits of the loop @xmath311 on the graph @xmath213 , i.e. the left loops on the graphs plotted on fig . [ fig6 ] ) . thus the maximal level of hierarchy @xmath312 is achieved in the cluster marked by @xmath299 $ ] and it is equal to @xmath313 , @xmath19 for @xmath11 , see fig . [ fig5 ] . ) -([y2 ] ) . the functions @xmath314 are plotted in red , the black lines are functions @xmath315 . the grey rectangles shows the area of non - negative @xmath228 , the white dashed lines denote the possible integer values of @xmath228 . the feasible values of @xmath316 ( see blue marks ) can be found as coordinates of intersections of the integer - valued lines with the red and black lines within the intersection of grey areas ( dark grey area ) . , title="fig:",height=207 ] consider the problem of finding the number , @xmath317 of two - fold de bruijn sequences ( see definition 2 ) for the binary alphabet . in this section we propose a generic formula for counting of two - fold de bruijn sequences . let function @xmath318 counts the number of connected graphs obtained from de bruijn graph @xmath213 by removing @xmath31 pairs of edges , such that the removed edges are marked by pair of the basis vectors @xmath319 or @xmath320 , where @xmath275 takes @xmath31 different values from the set of integers @xmath321 . the number , @xmath322 of @xmath2-ary two - fold de bruijn sequences can be calculated by the formula @xmath323 where @xmath324 and @xmath325 is the multiplicity of the prime factor 2 in the integer index @xmath31 . below we give approval of the formula ( [ mainformula ] ) and analyze its ingredients in more details . any @xmath2-ary two - fold de bruijn sequence can be seen as a closed trajectory on graph @xmath326 , where each of @xmath327 edges is passed exactly two times . all entries of the corresponding frequency vector @xmath328 are equal to @xmath329 . then for each @xmath275 ( @xmath330 ) one gets from ( [ y1])-([y2 ] ) the system of equation @xmath331 the systems ( [ c1])-([c2 ] ) generate only three admissible solutions , they are @xmath332 , @xmath333 , @xmath334 . thus the corresponding values of the vector @xmath228 are @xmath335 the total number of elements in the set @xmath231 is @xmath336 . having obtained the solutions , first , one has to select from them the legitimate ones , i.e. those that produce connected graphs with the connectivity matrix @xmath337 $ ] ( see section 3.3 for definition ) . second , one has to calculate the number of eulerian cycles on the obtained graphs . note that some edges are doubled ( @xmath228 contains @xmath329 at corresponding position ) and the counting of the eulerian cycles with the help of best theorem , eq . ( [ eulercycles ] ) , aliquot increases the number of cycles , while the corresponding sequences are indistinguishable . to avoid this miscounting one can instead calculate the eulerian cycles on the graph minors obtained by contracting the doubled edges . the structure of the matrix @xmath337 $ ] ( see analysis in section 3.3 and also example in section 3.5 ) allows to conclude that the adjacency matrices of such minors depends only on the total number , @xmath31 , of solutions ( [ sol2 ] ) and ( [ sol3 ] ) entering the vector @xmath228 and can be represented in a generic form . the @xmath338 adjacency matrix @xmath339 of the graph @xmath213 minor formed by contracting of @xmath340 edges is defined by the formula @xmath341 where @xmath248 is the column vector of ones , @xmath342 ( compare with ( [ qellp ] ) ) . thus one can divide the set @xmath231 by @xmath343 subsets with respect to the number of solutions ( [ sol2 ] ) and ( [ sol3 ] ) entering the vectors . therefore the number of two - fold de bruijn sequences is a sum over @xmath344 of the cofactors , defining the number of eulerian cycles , multiplied by the function @xmath345 counting the number of connected graphs , @xmath346\cdot\phi_{2^p}(2^p-2k).\ ] ] it is natural to single out the terms corresponding to @xmath347 and @xmath348 . the cofactor in case @xmath347 should be explicitly defined as equal to one . thus the only contribution is given by the function @xmath349 . when all edges are doubled each of the corresponding two - fold de bruijn sequences is simply a combination of two copies of the single - folded de bruijn sequences , i.e. @xmath350 is the number of eulerian cycles on the graph @xmath326 or the number of hamiltonian cycles on @xmath213 . the last term , @xmath348 , can be estimated in the similar way . indeed , here the trajectory fully covers the graph @xmath213 , thus @xmath351 and the cofactor gives the number of eulerian cycles on @xmath213 , which is @xmath352 and we arrive to @xmath353\cdot\phi_{2^p}(2^p-2k)+2^{2^{p}-p-1}.\ ] ] finally , we derive explicit formula for the cofactor in ( [ mainformula2 ] ) . the secular equation for the matrix @xmath339 given by ( [ q2k ] ) has the form @xmath354=(\lambda-2)\lambda^{2k-1 } \prod_{j=0}^{\beta-1}\left(1-\lambda^{-1 } e^{\frac{2\pi { \textrm{i}}}{\beta } j}\right),\ ] ] where @xmath324 and parameter @xmath325 is the multiplicity of the prime factor 2 in the natural number @xmath31 . indeed , the eigenvalue @xmath329 of the matrix @xmath339 corresponds to the constant vector , the @xmath355 zero eigenvalues is the reflection of the projection properties of the matrix . it projects the span of the @xmath355 vectors from the wavelet basis ( [ wavelet1 ] ) with @xmath356 onto the complementary subspace . in this subspace the matrix @xmath339 becomes a matrix of cyclic permutations . to obtain expression for the sought cofactor one can use the following formula @xcite @xmath357=\frac{1}{2k}\left.\frac{d}{d\lambda}\det[\lambda -q'_{2k}]\right|_{\lambda=2}=\frac{2^{2k-\beta}}{4k } \prod_{j=0}^{\beta/2 - 1}\left(1 + 4\cos^2(\frac{2\pi}{\beta } j)\right)\ ] ] this finishes the proof of the proposition 3 . the function @xmath358 can be represented as a sum over permutations of @xmath31 solutions ( [ sol2 ] ) and ( [ sol3 ] ) in the vector @xmath228 . thus @xmath359\right ) , \quad \theta\left(a\right)=\begin{cases}1,&\mbox { $ a$ has only one block;}\\ 0,&\mbox { $ a$ has more then one block . } \end{cases}\ ] ] where @xmath360 denotes the permutations of three types of solutions ( [ sol1])-([sol3 ] ) entering @xmath228 over @xmath343 places , among them there are @xmath361 solutions of the sort ( [ sol2 ] ) and @xmath362 of the sort ( [ sol3 ] ) . @xmath363 denotes the vector from @xmath231 with particular permutation . the number of permutations of @xmath361 and @xmath362 elements over @xmath343 positions is known to be @xmath364 this is the number of graphs to be tested . it can be reduced by removing solutions which has @xmath329 at the first or / and at the last positions of the vector @xmath228 , since they correspond to disconnected graphs . analysis based on the similar combinatiorial arguments and the specifics of de bruijn graphs allows to find a few more explicit expressions for the function @xmath345 , they are : @xmath365 , @xmath366 , @xmath367 . below we calculate explicit numbers of two - fold de bruijn sequences by the formula ( [ mainformula ] ) for @xmath368 . for @xmath18 there are only two two - fold de bruijn sequences : @xmath369 $ ] and @xmath370 $ ] . consider the case @xmath304 . in this case we have to analyze only 4 vectors from the set @xmath231 , they are @xmath371,\right.\\\left . [ 1,1,0,2,1,1,2,0],\quad [ 0,2,0,2,2,0,2,0],\quad [ 0,2,1,1,2,0,1,1]\right\}.\end{gathered}\ ] ] note , that other vectors , which contain @xmath329 at the first and/or the last positions are thrown out ( see the previous paragraph ) . the first vector @xmath372 $ ] generates path on the full graph @xmath164 . there are two eulerian paths on this graph , which correspond to the sequences @xmath373 $ ] and @xmath374 $ ] . the third vector @xmath375 $ ] is a composition of two identical two - ary single - fold de bruijn sequences , i.e. the sequence @xmath376 $ ] . two other vectors correspond to the sub - graphs with the following adjacency matrices @xmath377=\left(\begin{array}{cccc } 1&0&1&0\\ 1&0&1&0\\ 0&0&0&2\\ 0&2&0&0 \end{array}\right)\quad a_{4}[0,2,1,1,2,0,1,1]=\left(\begin{array}{cccc } 0&0&2&0\\ 2&0&0&0\\ 0&1&0&1\\ 0&1&0&1 \end{array}\right).\ ] ] by contracting the doubled edges we obtain the minors of @xmath164 . in both cases the sub - graph s minors are described by one and the same adjacency matrix @xmath378 the number of eulerian cycles on the minors can be calculated by the formula ( [ cof ] ) , it equals one . thus we obtain @xmath379 . the two more sequences are @xmath380 $ ] , @xmath381 $ ] . the results for @xmath382 are gathered in the tables below l + l || c | c | c | c | c + @xmath31 & 0 & 1 & 2 & 3 & 4 + @xmath383 & 16 & 32 & 24 & 8 & 1 + @xmath384 & 2 & 8 & 11 & 6 & 1 + @xmath385 $ ] & 1 & 1 & 2 & 4 & 16 + + l || c | c | c | c | c | c | c | c | c + @xmath31 & 0 & 1 & 2 & 3 & 4 & 5 & 6 & 7 & 8 + @xmath383 & 256 & 1024 & 1792 & 1792 & 1120 & 448 & 112 & 16 & 1 + @xmath386 & 16 & 128 & 380 & 584 & 519 & 274 & 84 & 14 & 1 + @xmath385 $ ] & 1 & 1 & 2 & 4 & 16 & 48 & 128 & 448 & 2048 in present article we propose a method for studying of cluster structure of the set of cyclic symbolic sequences . the main analytic tool we use is analysis of properties of the operators acting as raising and lowering operators on the cluster tree . formalization of the operators allowed to find a proper basis , the wavelet basis , for optimization of the cluster analysis procedure and reduce the problem to two - stage algorithm including solution of linear diophantine equations and analysis of block structure of matrices . the proposed approach allowed to extend a number of solvable problems in combinatorial mathematics . in particular , we presented a formula ( [ mainformula ] ) for counting number of @xmath2-ary two - fold de bruijn sequences . note here , that the formula ( [ mainformula ] ) includes the counting function @xmath345 , which explicit form was found only for a few values of @xmath31 , some other values were tabulated in the previous section . we believe , however , that its behavior and asymptotic can be further studied with the help of the method of generating functions , widely used in combinatorics @xcite . the known generic results regarding the cluster structure of symbolic sequences are gathered in section 2.4 . our study allowed us to add the estimation of the maximal branching level of the tree , it is @xmath313 . though this result was rigorously obtained for prime @xmath5 only , there is no obstacles to get the same result for all @xmath5 by the same reasoning , our numeric study shows universality of this result . the wavelet analysis of cyclic symbolic sequences can be easily extended onto the case of open sequences of a finite length . by introducing an additional letter and closing the sequence by a long arc made up of these letters we can map the problem of open sequences back onto the problem of cyclic sequences . certainly the algorithms should be improved to take into account the artificial part of the sequence . we believe that the approach based on the wavelet analysis opens new possibilities for optimization of the dna sequence assembly algorithms . first , the wavelet transform of the frequency vectors allows to reformulate the problem in a way suitable for parallelization of the restoring procedures . moreover , in the frequency representation of symbolic sequences one has to save only addresses of the non - zero elements of the vector without saving the information about connectivity of the edges on de bruijn graph , which allows to optimize the data storage problem . note that the similar ideas , of representing de bruijn graph as a tree , was also discussed in literature @xcite . in the article we considered only the processes which conserve the length of the sequences , as soon as it is dictated by applications in theory of quantum chaos . in fact , the lowering operator , i.e. solution of diophantine equations , can be considered without the second part where one controls the connectivity of the sequence . then the resulting frequency vectors at @xmath387 should corresponds to a set of cyclic sequences , which total length is conserved , rather then to a single sequence . on the contrary , the raising operator acting on such frequency vectors can be seen as a gluing operator acting on several sequences . in such a way one can organize a stochastic process on a set of symbolic sequences . the process can be described as follows , after several subsequent rising of a starting long sequence and the same number of lowerings of the obtained frequency vector with a random choice of possible solutions one can get a solution which instead of one cyclic sequence contains many of them of a shorter lengths . this procedure of rising and lowering repeated several times can be considered as a stochastic process of `` random bubbling of symbolic sequences '' . to summarize , we can say that , generically , the idea of classification of sequences with respect to the local content is rather natural . first step in this direction was done in the work of hamming @xcite , who introduced a _ metric _ distance between sequences . the hamming distance between two given sequences , such that both have the same length and a fixed starting point , is defined as the total number of different symbols appearing in the identical positions . the metric space generated by the hamming distance on the set of binary sequences is known as hamming cube . the hamming distance , however does not take into account the surrounding symbols . on the contrary , the ultrametric distance is absolutely insensitive to the absolute position of symbols within the sequence , while it takes into account the local surrounding . thus even one different symbol in two sequences put them on the furthest distance from each other . this prompts us the next natural step in the direction of classification of symbolic sequences . this should be combination of both approaches together . note that first attempts of such kind are based not on the construction of distances , but on introduction of stochastic processes which take into account both positions of symbols and their local surrounding see the work @xcite and also works @xcite . * aknowlegements : * author thanks s.nechaev , p.braun , b.gutkin , t.guhr for useful discussion at the beginning of this work . the work was partially supported by kaw foundation . on the fig . [ fig10 ] we graphically represent structure of the factor set @xmath389 by a special choice of its representatives from the set @xmath390 . one can compare it with the structure of the same set ( see fig . [ fig5 ] ) revealed by the ultrametric distance ( [ ultrdist ] ) . the representatives of @xmath389 are chosen to maximize the number @xmath391 , i.e. the index of the basis vector ( [ trivultrametricity ] ) . for the case of prime @xmath5 one can offer a synthetic algorithm based on the following principles : * sequences are classified by @xmath392 , the lengths of the longest sub - string , @xmath393 $ ] , and by the frequency @xmath166 of appearance @xmath394 in the sequence . the integer partition of @xmath5 , then has the form @xmath395 let @xmath396 be the sequence @xmath394 followed by arbitrary string of the length @xmath397 , as on the scheme @xmath398 * since @xmath5 is prime , there are at least two non - equal @xmath399 entering the integer partition of @xmath5 . let @xmath400 , then there is at least one @xmath401 , such that @xmath402 is not divisible by @xmath403 . therefore one can organize a set of cyclic sequences constructed from two letters @xmath404 and @xmath405 , such that they have no other periods except @xmath406 , for instance @xmath407 * to obtain the representatives one should choose any suitable strategy of ordering of @xmath402 pieces between @xmath403 places , such that the resulting sequences of @xmath404 s and @xmath405 s can not be obtained one from the other by any rotation and , finally , one has to and fill up the yet empty places within the strings @xmath396 by all possible ways . by a binary tree . only half of the tree ( the sequences starting from @xmath138 ) is plotted . the representatives of the factor sets @xmath388 ( @xmath408 ) are marked by the circles on corresponding level of the tree . the nubers inside denote period of the corresponding sequence . ] m.sieber , k.richter , `` correlations between periodic orbits and their role in spectral statistics '' _ phys.scripta . _ * t90 * ( 2001 ) 128 s.heusler , s.mller , p.braun , f.haake `` universal spectral form factor for chaotic dynamics '' _ j.phys.a:math.gen . _ * 37 * ( 2004 ) l31 f.haake , `` quantuim signatures of chaos '' , 3-d edition ( springer , heidelberg , 2010 ) f.murtagh `` identifying and exploiting ultrametricity '' in `` data analysis , machine learning and applications '' ( springer , 2008 ) p.263 c. flye sainte - marie , `` solution to question nr . 48 '' _ lintermdiaire des mathmaticiens _ * 1 * ( 1894 ) 107 r.sharp `` degeneracy in the length spectrum for metric graphs '' _ geometriae dedicata volume _ * 149 * ( 2010 ) 177 g.tanner `` spectral statistics for unitary transfer matrices of binary graphs '' _ j. phys . a _ * 33 * ( 2000 ) 3567 s.kapoor , h.ramesh , `` an algorithm for enumerating all spanning trees of a directed graph '' _ algorithmica _ * 27 * ( 2000 ) 120

we present infrared 2.84.1 m slit spectra of 32 seyfert 2 galaxies in the cfa and 12 m samples . the 3.3 m polycyclic aromatic hydrocarbon ( pah ) emission feature was used to estimate the absolute magnitude of a compact nuclear starburst ( less than a few hundred parsecs in size ) that is presumed to have occurred in the outer region of an obscuring dusty molecular torus around a central supermassive black hole . we detected 3.3 m pah emission in 11 of the 32 seyfert 2 nuclei in our sample , providing evidence for the presence of compact nuclear starbursts in a significant fraction of seyfert 2 nuclei . however , the rest - frame equivalent widths of the 3.3 m pah emission , and the 3.3 m pah - to - infrared luminosity ratios measured in this study suggest that compact nuclear starbursts generally do not contribute significantly to the observed 34 m nuclear fluxes or to the infrared luminosities of seyfert 2 galaxies . absorption features at 3.4 m from bare dust were clearly detected in only two of the nuclei , and features at 3.1 m from ice - covered dust were detected in only one nucleus . if the dust properties in the direction of these seyfert 2 nuclei do not differ significantly from the galactic interstellar medium , then these small absorption optical depths suggest that dust extinction toward the 34 m continuum - emitting region in the innermost part of the obscuring dusty torus is modest : a 5060 mag . finally , the 3.3 m pah emission luminosities measured in this study were found to be significantly correlated with _ iras _ 12- and 25- m , and nuclear-band ( 10.6 m ) luminosities . if these three luminosities trace the power of the active galactic nucleus ( agn ) , then the luminosities of compact nuclear starbursts and agns are correlated . this correlation is in agreement with theories predicting that the presence of a compact nuclear starburst in the torus leads to an enhancement of the mass accretion rate onto the central supermassive black hole .

astro-ph0309083

seyferts are the most numerous class of active galactic nuclei ( agns ) . the two types of seyfert galaxies , type 1 ( with broad optical emission lines ) and type 2 ( with no such lines ) , can be unified by the combination of an obscuring dusty molecular torus around a central , accreting supermassive black hole , and different viewing angles @xcite . as dusty tori are rich in molecular gas , they are natural sites for starbursts to occur @xcite . for several reasons , such a _ nuclear _ starburst in the torus is likely to occur at the outer part , as originally proposed by @xcite . first , for a dusty molecular torus whose mass is significantly smaller than the central supermassive black hole @xcite , gas in the torus rotates in a keplerian manner , and the gravitational stability parameter q @xcite is described as q = @xmath4@xmath5/@xmath6g@xmath7 , where @xmath4 , @xmath5 , and @xmath7 are the gas velocity dispersion , angular velocity , and gas surface density , respectively . in the case of a power - law radial density distribution of gas ( @xmath7 @xmath8 r@xmath9 ; where @xmath10 is the distance from the central supermassive black hole ) in an isothermal keplerian disk ( @xmath5 @xmath8 r@xmath11 ) , q is proportional to r@xmath12 . if the radial density distribution of gas and dust in the torus is flatter than r@xmath11 , as is often assumed @xcite , then q decreases with radius , so that the gravitational collapse of molecular gas can occur more easily in the outer regions of the torus @xcite . in addition , star formation in the inner part of the torus could be suppressed due to strong x - ray radiation from the agn @xcite . finally , if there is molecular gas inflow from the host galaxy to the torus , gas compression will occur at the outer edges of the torus , encouraging star formation . such compact nuclear starbursts have been detected in some seyfert nuclei @xcite . however , their absolute luminosities and relation to the central agns are still unclear . infrared 2.84.1 @xmath0 m spectroscopy is one of the most powerful ways of detecting and quantitatively estimating the energetic importance of these compact nuclear starbursts @xcite . first , starburst emission is clearly distinguishable from emission from the agn , using the polycyclic aromatic hydrocarbon ( pah ) emission found at 3.3 @xmath0 m . in a _ normal _ starburst , pahs widely distributed in the interstellar medium are excited by far - uv photons from stars , and so strong 3.3 @xmath0 m pah emission is usually detected @xcite . close to an agn , where sufficient far - uv photons from the agn are available , x - ray radiation is also strong , which destroys pahs @xcite . at a certain distance from the central agn in the obscuring material , the x - ray flux is sufficiently attenuated to prevent this . however , as uv photons have a higher susceptibility to extinction than x - rays , far - uv photons from the agn are not expected to survive to these distances . thus , no pah emission is expected in a _ pure _ agn . the detection of 3.3 @xmath0 m pah emission from an agn requires far - uv photons from a starburst to be sufficiently shielded from agn emission , as is the case for a compact nuclear starburst occurring in the outer part of a dusty torus . the pah emission luminosity is therefore expected to roughly trace the magnitude of such a starburst . another advantage of infrared 2.84.1 @xmath0 m spectroscopy is that the flux attenuation is lower , as dust extinction at 2.84.1 @xmath0 m is much less than at uv or optical wavelengths . the absolute magnitude of a starburst in a seyfert can be determined from the _ observed _ 3.3 @xmath0 m pah emission luminosity @xcite . finally , since 3.3 @xmath0 m pah emission is intrinsically strong @xcite , even the signature of a weak starburst is detectable with a spectrum of typical signal - to - noise ratio . using infrared 2.84.1 @xmath0 m slit spectroscopy , it has been possible to detect and quantify previously undetected compact nuclear starbursts in seyferts @xcite . the physical scale of a compact nuclear starburst in the outer region of a dusty molecular torus is less than a few hundred parsecs , which corresponds to less than 12 arcsec in the majority of seyferts at @xmath13 0.007 . in seyferts , extended _ circumnuclear _ ring - shaped starbursts are found typically at distances of @xmath141 kpc from the nuclei @xcite , and are generally more powerful than the compact _ nuclear _ starbursts @xcite . hence , the use of a narrow ( @xmath212 arcsec ) slit is most effective to distinguish between extended starbursts and the compact nuclear starbursts that are of interest . therefore , we have performed ground - based infrared 2.84.1 @xmath0 m slit spectroscopy focusing initially on seyfert 2 nuclei to determine observational constraints on the properties of compact nuclear starbursts . throughout this paper , @xmath15 @xmath16 75 km s@xmath17 mpc@xmath17 , @xmath18 = 0.3 , and @xmath19 = 0.7 are adopted . seyfert 2 galaxies in the cfa @xcite and 12@xmath0 m ( rush , malkan , & spinoglio 1993 ) samples were selected based on their host galaxy magnitudes and _ iras _ 12 @xmath0 m fluxes , respectively . these samples are not expected to be biased toward or against the presence of compact nuclear starbursts , and is thus suitable for our program . the following sample selection criteria were employed : ( 1 ) in very nearby sources , slit spectroscopy may miss a significant fraction of emission from compact nuclear starbursts with a physical size scale of up to a few hundred pc . the slit width used , 0@xmath2091@xmath206 ( 3 ) , corresponds to 120220 pc at @xmath21 . seyfert 2s at @xmath22 @xmath2 0.007 are thus excluded . ( 2 ) in distant seyfert 2s , slit spectroscopy may be significantly contaminated by extended ( kpc scale ) powerful circumnuclear starbursts . the slit width used corresponds to 600 pc 1 kpc at @xmath23 . seyfert 2s at @xmath22 @xmath24 0.035 are excluded . ( 3 ) 12 @xmath0 m seyfert 2s at declination @xmath2535@xmath26 are excluded because they are not observable under good observing conditions from mauna kea , hawaii . ( 4 ) some 12@xmath0 m seyfert 2s in @xcite have been later re - classified as liners or hii - region galaxies based on higher quality optical spectra @xcite . these sources are excluded . all cfa seyfert 2s later re - observed by @xcite were re - classified as seyfert 1.8 , 1.9 , or 2 . these sources are included . these selection criteria result in 18 cfa seyfert 2s and 26 12@xmath0 m seyfert 2s , of which six sources are included in both samples . from the total sample of 38 seyfert 2s , 32 ( 32/38 = 84% ) sources have been observed so far . table [ tbl-1 ] summarizes the infrared emission properties of the 32 observed seyfert 2s . the six unobserved sources are ugc 1395 ( = 0152 + 06 ) , ngc 3362 , ugc 6100 ( = a1058 + 45 ) , ugc 8621 ( = 1335 + 39 ) , and ngc 5283 in the cfa sample , and mcg-2 - 8 - 39 in the 12@xmath0 m sample . although the sample is not statistically complete , major conclusions drawn from the 32 sources in this paper are unlikely to be affected significantly by omission of the six unobserved sources . the observing log is shown in table [ tbl-2 ] . observations were made with the ukirt cgs4 @xcite , the irtf nsfcam @xcite , the irtf spex @xcite , and the subaru ircs @xcite , on mauna kea , hawaii . details of the observations with the ukirt cgs4 ( for ngc 7172 ) and the irtf nsfcam ( for ngc 5256 and ngc 5135 ) were described by @xcite and @xcite , respectively , and are not repeated here . for the irtf spex observing run , the 1.94.2 @xmath0 m cross - dispersed mode with a 16 wide slit was employed . the achievable spectral resolution at 3.5 @xmath0 m is r @xmath14 450 . the sky conditions were photometric and the seeing sizes at @xmath27 ( 2.2 @xmath0 m ) were 0@xmath2060@xmath209 ( full - width at half - maximum ; fwhm ) throughout the observations . the position angle of the slit was set along the north - south direction . a standard telescope nodding technique ( abba pattern ) with a throw of 7.5 arcsec was employed along the slit to subtract background emission . as 34 @xmath0 m emission from a seyfert 2 galaxy is usually dominated by compact nuclear emission @xcite , this throw is believed to be sufficiently large . the telescope tracking was monitored with the infrared slit - viewer of spex . each exposure was 15 sec , and 2 coadds were made at each position . with this exposure time , signals at @xmath28 @xmath24 4.1 @xmath0 m ( in the observed frame ) exceed the linearity level of the spex array , and so data at @xmath28 @xmath24 4.1 @xmath0 m were removed . for the subaru ircs observing run , a 0@xmath209-wide slit and the @xmath29-grism were used with a 58-mas pixel scale . the achievable spectral resolution at 3.5 @xmath0 m is r @xmath14 140 . the sky was not completely photometric during the observations of some of the seyfert 2 nuclei ( mrk 334 , ngc 1144 , and ngc 1125 ) . the position angle of the slit was set along the east - west direction , except in the case of ngc 1144 where the angle was set along the north - south direction . the seeing at @xmath27 was 0@xmath2050@xmath208 in fwhm . a standard telescope nodding technique ( abba pattern ) , with a throw of 7 arcsec along the slit , was employed . each exposure was 2.03.0 sec , and 1530 coadds were made at each slit position . data at @xmath28 @xmath24 4.1 @xmath0 m were removed , due to signal levels above the linearity level of the ircs array . a- , f- , and g - type standard stars ( table [ tbl-2 ] ) were observed with airmass difference of @xmath20.1 to the individual seyfert 2 nuclei , to correct for the transmission of the earth s atmosphere . the @xmath29-band ( 3.5 @xmath0 m ) magnitudes of standard stars were estimated from their @xmath30-band ( 0.6 @xmath0 m ) magnitudes , adopting the @xmath31 colors appropriate to the stellar types of individual standard stars @xcite . standard data analysis procedures were employed using iraf . initially , bad pixels and pixels hit by cosmic rays were replaced with the interpolated values of the surrounding pixels . then , frames taken with an a ( or b ) beam were subtracted from frames subsequently taken with a b ( or a ) beam , and the resulting subtracted frames were added and were divided by a spectroscopic flat image . the seyfert 2 and standard star spectra were then extracted by integrating signals over 2@xmath2043@xmath200 and 1@xmath2041@xmath208 along the slit for the irtf spex and subaru ircs data , respectively . wavelength calibration was performed using the wavelength - dependent transmission of the earth s atmosphere . the seyfert 2 spectra were divided by those of the corresponding standard stars , and were multiplied by the spectra of blackbodies with temperatures corresponding to those of the individual standard stars ( table [ tbl-2 ] ) . all of the irtf spex observing runs and subaru ircs runs in 2003 march were made under photometric sky conditions , and so flux calibration was made based on the signal detected inside our slit . for the cfa sample seyfert 2s , @xmath29-band ( 3.5 @xmath0 m ) photometry taken with 1@xmath2058@xmath206 apertures was available @xcite . our flux calibration generally agrees with or is slightly ( @xmath21 mag ) fainter than their photometry , which is reasonable because 34 @xmath0 m continuum emission from a seyfert 2 galaxy is usually dominated by compact nuclear emission , but emission more extended than our slit size also exists @xcite . during the observations of mrk 334 , ngc 1125 , and ngc 1144 , the sky was not completely photometric due to the presence of thin cirrus . for mrk 334 and ngc 1144 , @xmath29-band photometry made with a 1.5 arcsec aperture by @xcite was used for the flux calibration of our slit spectra . for ngc 1125 , although no @xmath29-band photometry was available , the cirrus was thinner than during the observations of mrk 334 and ngc 1144 , and the flux discrepancy among independent data sets is small ( @xmath220% ) . the flux was thus calibrated using our slit spectra . appropriate spectral binning was applied for faint sources , particularly at @xmath28 @xmath2 3.3 @xmath0 m . in this wavelength range , the earth s atmospheric transmission is highly wavelength dependent , and even displacement at the sub - pixel level between a seyfert 2 and corresponding standard star along the wavelength direction could produce a spiky spectrum . binning to a spectral resolution of r @xmath2 100 can help reduce these spurious spikes @xcite . the targeted feature in this wavelength range is the 3.1 @xmath0 m absorption feature caused by ice - covered dust grains , which is sufficiently broad ( smith , sellgren , & tokunaga 1989 ) that a spectral resolution of r @xmath24 50 allows investigation of its properties . in the vicinity of the redshifted 3.3 @xmath0 m pah emission feature ( @xmath32 @xmath14 3.29 @xmath0 m ) , a spectral resolution of r @xmath24 100 was retained to properly trace its profile and estimate its flux reliably . for faint sources , data at the longer side of 3.3 @xmath0 m pah emission were also binned . the main targeted feature in this wavelength range is the absorption feature due to bare carbonaceous dust at @xmath32 @xmath14 3.4 @xmath0 m @xcite . as this feature is broad and weak , a spectral resolution of r @xmath24 100 is sufficient . figure [ fig1 ] shows flux - calibrated 2.84.1 @xmath0 m slit spectra . for ngc 5256 and ngc 5135 , the 3.04.0 @xmath0 m spectra were presented by @xcite . here , 2.84.1 @xmath0 m spectra are shown , which illustrate the properties of the broad 3.1 @xmath0 m absorption feature . the spectrum of ngc 7172 was taken with ukirt cgs4 , so the wavelength coverage is @xmath28 = 3.23.8 @xmath0 m . the spectrum presented by @xcite is shown again here . in the 2.84.1 @xmath0 m spectra shown in figure [ fig1 ] , we consider 3.3 @xmath0 m pah emission to be detected when at least two successive data points are significantly higher than the continuum level . using this criterion , 3.3 @xmath0 m pah emission was detected in 11 of the 32 observed seyfert 2 nuclei . the detection rates are 31% ( 4/13 ) and 36% ( 9/25 ) for cfa and 12 @xmath0 m seyfert 2s , respectively . the fluxes , luminosities , and rest - frame equivalent widths of the 3.3 @xmath0 m pah emission were estimated using the method described by @xcite and are summarized in table [ tbl-3 ] . if sufficiently obscured agn emission contributes significantly to an observed 34 @xmath0 m nuclear flux , the 3.4 @xmath0 m absorption feature due to bare carbonaceous dust grains @xcite and/or the 3.1 @xmath0 m absorption feature due to ice - covered dust grains @xcite should be detected . the 3.4 @xmath0 m absorption feature was clearly detected in only f04385@xmath330828 and ngc 7172 , with an optical depth of @xmath140.1 . the presence of the 3.1 @xmath0 m absorption feature was inferred only in ngc 4388 . in some bright nearby sources ( ngc 4388 , ngc 262 , ngc 1194 , ngc 4968 mcg-3 - 34 - 64 , and mcg-2 - 40 - 4 ) for which sufficient signal - to - noise ratios in the continuum are achieved even without spectral binning , [ siiv ] emission at @xmath32 = 3.94 @xmath0 m and br@xmath34 emission at @xmath32 = 4.05 @xmath0 m are detected . these emission lines are sufficiently strong in ngc 4388 and ngc 4968 that reliable flux estimates for these relatively narrow emission lines ( narrower than pah ) are possible even in our low - resolution spectra . for ngc 4388 , the observed [ siix ] and br@xmath34 emission fluxes are estimated to be 3 @xmath35 10@xmath36 w m@xmath37 and 4 @xmath35 10@xmath36 w m@xmath37 , respectively . for ngc 4968 , they are 1 @xmath35 10@xmath36 w m@xmath37 and 2 @xmath35 10@xmath36 w m@xmath37 , respectively . these fluxes are slightly higher than the estimates by @xcite , possibly due to our larger slit size . to investigate the contribution from compact nuclear starbursts to the observed 34 @xmath0 m fluxes and to the total infrared ( 81000 @xmath0 m ) luminosities , the rest - frame equivalent widths of 3.3 @xmath0 m pah emission features ( ew@xmath38 ) and 3.3 @xmath0 m pah - to - infrared luminosity ratios ( l@xmath38/l@xmath39 ) are shown in table [ tbl-3 ] , as described in @xcite . as an equivalent width ( @xmath40 the ratio of line to continuum flux ) is , by definition , robust to dust extinction , a large pah equivalent width ( @xmath14100 nm ; imanishi & dudley 2000 ) should always be detected in a starburst galaxy , regardless of dust extinction . all but ngc 5256 and mrk 938 show ew@xmath38 more than a factor of two smaller than @xmath14100 nm . therefore , although the compact nuclear starbursts are detected at 34 @xmath0 m in @xmath1430% ( 11/32 ) of seyfert 2s , their observed 34 @xmath0 m fluxes , except ngc 5256 and mrk 938 , should be dominated by a continuum that does not show a strong 3.3 @xmath0 m pah emission feature . agn - powered hot ( @xmath148001000k ) dust in the very inner part of an obscuring dusty torus @xcite close to the innermost dust sublimation radius ( which corresponds to 10001800 k dust ; granato et al . 1997 ) can emit 34 @xmath0 m continuum efficiently ( assuming a blackbody spectrum ) , without strong 3.3 @xmath0 m pah emission . thus , it can be concluded that the 34 @xmath0 m fluxes in the majority of the observed seyfert 2 nuclei originate in featureless continuum emission from hot dust powered by an agn . the l@xmath38/l@xmath39 ratios in all seyfert 2s ( table [ tbl-3 ] ) are a factor of @xmath412 smaller than the ratios for starburst - dominated galaxies ( @xmath141 @xmath35 10@xmath42 ; mouri et al . 1990 ; imanishi 2002 ) . thus , compact nuclear starbursts are unlikely to dominate the infrared luminosities of these seyfert 2 galaxies . the larger sample presented here confirms the small ew@xmath38 and l@xmath38/l@xmath39 found previously in a smaller sample of seyfert 2s @xcite . as it has been shown in @xmath435.1 that the observed 34 @xmath0 m fluxes in the majority of the observed seyfert 2 nuclei are dominated by agn - powered hot - dust emission in the innermost part of the obscuring dusty torus , the observed optical depths of the absorption features at 3.4 @xmath0 m and 3.1 @xmath0 m reflect the column density of obscuring dust toward the innermost 34 @xmath0 m continuum - emitting regions . in the galaxy , dust grains located deep inside molecular gas are covered with an ice mantle , if they are sufficiently shielded from ambient uv radiation @xcite . such dust grains show an absorption feature at 3.1 @xmath0 m @xcite , but no significant feature at 3.4 @xmath0 m @xcite . dust absorption at 3.4 @xmath0 m is detected in the diffuse interstellar medium @xcite where dust grains are not covered with an ice mantle ( that is , `` bare '' ) . as dust extinction toward the innermost 34 @xmath0 m continuum - emitting hot dust around an agn can be fitted with a foreground - screen dust model , the total column density of obscuring dust , including both bare and ice - covered dust grains , can be estimated in a fairly straightforward manner from the optical depths of these absorption features ( @xmath44 and @xmath45 ) . the @xmath45 values for the two detected sources ( f04385@xmath330828 and ngc 7172 ) are @xmath140.1 . conservative upper limits for @xmath45 in the remaining undetected sources are @xmath20.2 . the @xmath44 value for ngc 4388 is @xmath140.2 , some fraction of which may originate in dust in the edge - on host galaxy @xcite , rather than the dusty torus . conservative upper limits for @xmath44 for the remaining sources are @xmath20.3 . if the obscuring dust toward these seyfert 2 nuclei shows a dust extinction curve similar to that of the galaxy ( @xmath45/a@xmath1 @xmath16 0.0040.007 ; pendleton et al . 1994 , @xmath44/a@xmath1 @xmath14 0.06 ; smith , sellgren , & brooke 1993 ; tanaka et al . 1990 ; murakawa , tamura , & nagata 2000 ) , then the @xmath44 and @xmath45 values imply that a@xmath1 @xmath2 50 mag and @xmath2 5 mag for bare and ice - covered dust grains , respectively . hence , the total dust column density toward the innermost 34 @xmath0 m continuum - emitting hot dust around a central agn is modest , a@xmath1 @xmath2 5060 mag , in the majority of seyfert 2 nuclei . if the @xmath45/a@xmath1 ratios toward these seyfert 2 nuclei are substantially smaller than the galactic ratio , then the a@xmath1 value estimated in this way is underestimated in seyfert 2s . in the dusty torus around an agn , the dust size may be larger than in the galactic interstellar medium , due to dust coagulation @xcite , and the 9.7 @xmath0 m silicate dust feature may be weakened @xcite . it is possible that the 3.4 @xmath0 m absorption feature is also weakened and the @xmath45/a@xmath1 ratio toward an obscured agn is smaller than the galactic value . however , in a number of obscured agns , it has been found that the dereddened agn luminosity based on the galactic @xmath45/a@xmath1 ratio is comparable to the bolometric luminosity @xcite . if @xmath45/a@xmath1 ratios in these obscured agns are substantially smaller than the galactic ratio , then the dereddened agn luminosities would substantially exceed the bolometric luminosities . it is thus unlikely that the @xmath45/a@xmath1 ratio toward an obscured agn is substantially smaller than the galactic value , so that our estimate for dust extinction in these seyfert 2s appears reasonable . indeed , the estimated modest dust extinction toward the majority of seyfert 2 nuclei , excluding the dustiest , most infrared - luminous examples ( ultraluminous infrared galaxies ) , is generally in good agreement with other independent estimates . first , infrared spectral energy distributions of seyferts show similarly modest dust extinction @xcite . in addition , alonso - herrero , ward , & kotilainen ( 1997 ) found that dust extinction toward the innermost 3.5 @xmath0 m continuum - emitting regions is a@xmath1 @xmath2 5060 mag , based on a comparison of the luminosities of optical [ oiii ] emission lines from unobscured regions and 3.5 @xmath0 m continuum . finally , @xcite showed that the equivalent widths of 7.7 @xmath0 m pah emission measured with _ iso _ using a large aperture ( 24 @xmath35 24 arcsec@xmath46 ) are @xmath148 times larger in seyfert 2s than in seyfert 1s . if the 7.7 @xmath0 m pah emission luminosities , mostly originating from extended starbursts in the host galaxies , are intrinsically the same in seyfert 1s and 2s , then the larger equivalent widths in seyfert 2s can be explained by the attenuation of the 78 @xmath0 m continuum flux from the agn . adopting the dust extinction curve with a@xmath47/a@xmath1 @xmath14 0.040.05 derived by @xcite , the dust extinction toward 78 @xmath0 m continuum - emission regions in seyfert 2s is estimated to be a@xmath1 = 4555 mag . in fact , the extended pah emission luminosities in seyfert 2s are likely to be intrinsically larger than in seyfert 1s @xcite , so that the above dust extinction is probably an overestimate . although the dust extinction toward the innermost 34 @xmath0 m continuum - emitting region around an agn should be larger than that toward the outer 78 @xmath0 m continuum - emitting region , we find no significant difference . if the galactic n@xmath48/a@xmath1 ratio is assumed ( @xmath142 @xmath35 10@xmath49 cm@xmath37 mag@xmath17 ; predehl & schmitt 1995 ) , dust extinction of a@xmath1 @xmath2 5060 mag implies a hydrogen column density of n@xmath48 @xmath2 10@xmath51 cm@xmath37 in the majority of seyfert 2 nuclei , which is substantially smaller than measured values of n@xmath48 by x - rays toward agns in seyfert 2 nuclei @xcite . imaging observations have indicated substantially larger ratios of n@xmath48 ( columns of x - ray absorbing material)-to - a@xmath1 ( dust columns toward the innermost 3.5 @xmath0 m continuum emitting regions ) in seyfert 2s @xcite . our 2.84.1 @xmath0 m spectroscopy provided additional support for this . @xcite argued that there are two distinct types of seyfert 2 nuclei : one agn - powered and another powered by a compact nuclear starburst . however , @xcite did not find evidence for such distinct seyfert 2 populations . the l@xmath38 values measured from our slit spectra ( table [ tbl-3 ] ) are a good measure of the absolute magnitude of the compact nuclear starbursts @xcite , and _ iras _ 12 @xmath0 m and 25 @xmath0 m fluxes are often taken to be representative of the agn power in a seyfert 2 galaxy @xcite . ground - based @xmath3-band ( 10.6 @xmath0 m ) aperture photometry of seyfert 2 nuclei using a single - element bolometer can also be a good tracer of agn power @xcite as the contamination from extended starbursts is reduced . in figure [ fig2 ] , we compare l@xmath38 with _ iras _ 12- and 25-@xmath0 m , and nuclear @xmath3-band fluxes ( table [ tbl-1 ] ) . using the generalized kendall s rank correlation statistic @xcite we found the probability that a correlation is not present to be 0.007 , 0.002 , and 0.001 for fig . [ fig2]a , [ fig2]b , and [ fig2]c , respectively . thus , we suggest that the luminosities of agns and compact nuclear starbursts are correlated in seyfert 2 galaxies , so that more powerful agns tend to possess more powerful compact nuclear starbursts in the torus . it has been predicted that such compact nuclear starbursts in the dusty torus can increase the mass accretion rate onto the central supermassive black hole through enhancement of the turbulence of molecular gas in the torus @xcite and/or through radiation effects @xcite . the luminosity correlation between agns and compact nuclear starbursts is at least in qualitative agreement with these theoretical predictions . however , extended star formation , including circumnuclear ring - shaped starbursts and quiescent star formation in the host galaxies , may contribute to the _ iras _ 12 @xmath0 m and 25 @xmath0 m fluxes as well as to measurements from previous ground - based @xmath3-band photometry taken with a single element bolometer with @xmath24 3 arcsec apertures . @xmath3-band photometry of spatially - unresolved emission using recently available two - dimensional large - format mid - infrared arrays at large telescopes @xcite is undoubtedly a more reliable agn indicator . these photometric data , if available for our sample in the future , could be used to investigate the connection between agns and compact nuclear starbursts in much greater detail . a compact nuclear starburst can input energy into the torus and inflate it @xcite . as such an inflated torus has a large covering factor of gas and dust around the central agn , the equivalent width of the fe k@xmath34 emission line at 6.4 kev in x - rays ( ew@xmath52 ) can be large . the dependence of ew@xmath52 on the covering factor at various n@xmath48 has been calculated by several authors @xcite . both n@xmath48 and ew@xmath52 based on _ asca _ data are available for several sources in our sample @xcite . @xcite performed this calculation for the compton - thick case ( n@xmath48 @xmath24 1 @xmath35 10@xmath53 cm@xmath37 ) . four sources in our sample , ngc 7674 , ngc 1667 , ngc 4968 , and ngc 5135 , are classified as compton thick and the ew@xmath52s in ngc 7674 , ngc 1667 , ngc 4968 , and ngc 5135 are estimated to be 0.9 , @xmath23 , 1.18 , and 11.7 kev , respectively @xcite . all but ngc 1667 show a detectable compact nuclear starburst in the torus . the covering factor of the torus for ngc 7674 , ngc 4968 , and ngc 5135 is expected to be @xmath2440@xmath26 and could be as large as @xmath2470@xmath26 if n@xmath48 is close to 1 @xmath35 10@xmath53 cm@xmath37 ( fig . 3 of levenson et al . 2002 ) . however , the upper limit for ew@xmath52 in ngc 1667 , the source with an undetectable compact nuclear starburst , is so large ( @xmath23 kev ) that we are unable to draw meaningful conclusions about the dependence of the covering factor on the presence or absence of a compact nuclear starburst from these four sources . in addition , @xcite performed the same calculation for the compton - thin case ( n@xmath48 @xmath2 1 @xmath35 10@xmath53 cm@xmath37 ) . for the five compton - thin sources , ngc 4388 , ngc 5252 , ngc 5674 , ngc 262 , ngc 7172 , both n@xmath48 and ew@xmath52 are available @xcite . compact nuclear starbursts are undetected in all of these sources , but the covering factor is expected to be @xmath2460@xmath26 ( fig . 3 of ghisellini et al . 1994 ) . with the datasets currently available , no meaningful constraints can be obtained for the dependence of the torus covering factor on the presence of a compact nuclear starburst . however , if more accurate values of n@xmath48 and ew@xmath52 based on observations with _ chandra _ and _ xmm _ ( and _ astro - e2 _ for compton - thick sources ) become available for a significant number of our sample objects in the near future , then the dependence can be investigated in much more detail and the hypothesis that the compact nuclear starburst is the primary cause of inflation of the dusty molecular torus can be tested . we have presented infrared 2.84.1 @xmath0 m slit spectra of 32 seyfert 2 galaxies in the cfa and 12@xmath0 m samples , and made the following main conclusions . 1 . by using the 3.3 @xmath0 m pah emission luminosity as an indicator of a compact nuclear starburst in the dusty torus , we found evidence for the presence of such starbursts in @xmath1430% of seyfert 2s . the starbursts generally do not contribute significantly to either the observed 34 @xmath0 m nuclear flux or the total infrared luminosity of the observed seyfert 2 galaxies . 2 . the dust extinction in the direction of agn - powered 34 @xmath0m - continuum - emitting hot dust in the very inner part of the dusty torus , close to the dust sublimation radius , is modest ( a@xmath1 @xmath2 5060 mag ) in the majority of the observed seyfert 2 galaxies . provided that the _ iras _ 12- and 25-@xmath0 m , and nuclear @xmath3-band luminosities are a good measure of agn power , the agn and compact - nuclear - starburst luminosities are correlated . we are grateful to dr . j. rayner , dr . h. terada , dr . s. bus , p. sears , w. golisch , b. potter , d. scarla , and s. harasawa for their supports during our irtf spex and subaru ircs observing runs , and to drs . k. aoki , k. wada , y. taniguchi , and m. elitzur for useful discussions . this research has made use of the simbad database , operated at cds , strasbourg , france , and of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . alonso - herrero , a. , ward , m. j. , & kotilainen , j. k. 1997 , mnras , 288 , 977 alonso - herrero , a. , simpson , c. , ward , m. j. , & wilson , a. s. 1998 , apj , 495 , 196 alonso - herrero , a. , quillen , a. c. , simpson , c. , efstathiou , a. & ward , m. j. 2001 , aj , 121 , 1369 alonso - herrero , a. , ivanov , v. d. , jayawardhana , r. , & hosokawa , t. 2002 , apj , 571 , l1 alonso - herrero , a. , quillen , a. c. , rieke , g. h. , ivanov , v. , & efstathiou , a. 2003 , aj , 126 , 81 antonucci , r. 1993 , ara&a , 31 , 473 awaki , h. , koyama , k. , inoue , h. , & halpern , j. p. 1991 , pasj , 43 , 195 barvainis , r. 1987 , apj , 320 , 537 bassani , l. , dadina , m. , maiolino , r. , salvati , m. , risaliti , g. , della ceca , r. , matt , g. , & zamorani , g. 1999 , apjs , 121 , 473 carico , d. p. , sanders , d. b. , soifer , b. t. , elias , j. h. , matthews , k. , & neugebauer , g. 1988 , aj , 95 , 356 cid fernandes , r. jr . , & terlevich , r. 1995 , mnras , 272 , 423 cid fernandes , r. , heckman , t. , schmitt , h. , gonzalez delgado , r. m. , & storchi - bergmann , t. 2001 , apj , 558 , 81 clavel , j. , et al . 2000 , a&a , 357 , 839 devereux , n. a. , becklin , e. e. , & scoville , n. 1987 , apj , 312 , 529 dopita , m. a. , heisler , c. , lumsden , s. , & bailey , j. 1998 , apj , 498 , 570 edelson , r. a. , malkan , m. a. , & rieke , g. h. 1987 , apj , 321 , 233 efstathiou , a. , & rowan - robinson , m. 1995 , mnras , 273 , 649 fabian , a. c. , barcons , x. , almaini , o. , & iwasawa , k. 1998 , mnras , 297 , l11 fadda , d. , giuricin , g. , granato , g. l. , & vecchies , d. 1998 , apj , 496 , 117 ghisellini , g. , haardt , f. , & matt , g. 1994 , mnras , 267 , 743 gonzalez delgado , r. m. , heckman , t. , leitherer , c. , meurer , g. , krolik , j. , wilson , a. s. , kinney , a. , & koratkar , a. 1998 , apj , 505 , 174 gonzalez delgado , r. m. , heckman , t. , & leitherer , c. 2001 , apj , 546 , 845 granato , g. l. , & danese , l. 1994 , mnras , 268 , 235 granato , g. l. , danese , l. , & franceschini , a. 1997 , apj , 486 , 147 gu , q. , & huang , j. 2002 , apj , 579 , 205 heckman , t. m. , gonzalez delgado , r. , leitherer , c. , meurer , g. r. , krolik , j. , wilson , a. s. , koratkar , a. , & kinney , a. 1997 , apj , 482 , 114 hill , g. j. , becklin , e. e. , & wynn - williams , c. g. 1988 , apj , 330 , 737 huchra , j. , & burg , r. 1992 , apj , 393 , 90 imanishi , m. 2000 , mnras , 319 , 331 imanishi , m. 2001 , aj , 121 , 1927 imanishi , m. 2002 , apj , 569 , 44 imanishi , m. , sasaki , y. , goto , m. , kobayashi , n. , nagata , t. , & jones , t. j. 1996 , aj , 112 , 235 imanishi , m. , & dudley , c. c. 2000 , apj , 545 , 701 imanishi , m. , dudley , c. c. , & maloney , p. r. 2001 , apj , 558 , l93 imanishi , m. , & maloney , p. r. 2003 , apj , 588 , 165 isobe , t. , feigelson , e. d. , & nelson , p. i. 1986 , apj , 306 , 490 ivanov , v. d. , rieke , g. h. , groppi , c. e. , alonso - herrero , a. , rieke , m. j. , & engelbracht , c. w. 2000 , apj , 545 , 190 kobayashi , n. , et al . 2000 , proc . spie , 4008 , 1056 kohno , k. , matsushita , s. , vila - vilaro , b. , okumura , s. k. , shibatsuka , t. , okiura , m. , ishizuki , s. , kawabe , r. 2002 , in asp conf . 249 , the central kpc of starbursts and agn , 672 ( astro - ph/0206398 ) krabbe , a. , b@xmath54ker , t. , & maiolino , r. , 2001 , apj , 557 , 626 laor , a. , & draine , b. t. 1993 , apj , 402 , 441 le floch , e. , mirabel , i. f. , laurent , o. , charmandaris , v. , gallais , p. , sauvage , m. , vigroux , l. , & cesarsky , c. 2001 , a&a , 367 , 487 levenson , n. a. , krolik , j. h. , zycki , p. t. , heckman , t. m. , weaver , k. a. , awaki , h. , & terashima , y. 2002 , apj , 573 , l81 lutz , d. et al . 1996 , a&a , 315 , l269 lutz , d. , maiolino , r. , moorwood , a. f. m. , netzer , h. , wagner , s. j. , sturm , e. , & genzel , r. 2002 , a&a , 396 , 439 maiolino , r. , ruiz , m. , rieke , g. h. , & keller , l. d. 1995 , apj , 446 , 561 maiolino , r. , marconi , a. , salvati , m. , risaliti , g. , severgnini , p. , oliva , e. , la franca , f. , & vanzi , l. 2001 , a&a , 365 , 28 maiolino , r. , marconi , a. , & oliva , e. 2001 , a&a , 365 , 37 mennella , v. , munoz caro , g. m. , ruiterkamp , r. , schutte , w. a. , greenberg , j. m. , brucato , j. r. , & colangeli , l. 2001 , a&a , 367 , 355 moorwood , a. f. m. 1986 , a&a , 166 , 4 mountain , c. m. , robertson , d. j. , lee , t. j. , & wade , r. 1990 , proc . spie , 1235 , 25 mouri , h. , kawara , k. , taniguchi , y. , & nishida , m. 1990 , apj , 356 , l39 murakawa , k. , tamura , m. , & nagata , t. 2000 , apjs , 128 , 603 ohsuga , k. , umemura , m. , fukue , j. , & mineshige , s. 1999 , pasj , 51 , 345 oliva , e. , origlia , l. , maiolino , r. , & moorwood , a. f. m. 1999 , a&a , 350 , 9 osterbrock , d. e. , & martel , a. 1993 , apj , 414 , 552 pendleton , y. j. , sandford , s. a. , allamandola , l. j. , tielens , a. g. g. m. , & sellgren , k. 1994 , apj , 437 , 683 perez garcia , a. m. , & rogriguez espinosa , j. m. 2001 , apj , 557 , 39 pogge , r. w. , & martini , p. 2002 , apj , 569 , 624 predehl , p. , & schmitt , j. h. m. m. 1995 , a&a , 293 , 889 rawlings , m. g. , adamson , a. j. , & whittet , d. c. b. 2003 , mnras , 341 , 1121 rayner , j. t. , toomey , d. w. , onaka , p. m. , denault , a. j. , stahlberger , w. e. , vacca , w. d. , cushing , m. c. , & wang , s. 2003 , pasp , 115 , 362 rieke , g. h. 1978 , apj , 226 , 550 risaliti , g. , maiolino , r. , & salvati , m. 1999 , apj , 522 , 157 rodriguez - ardila , a. , & viegas , s. m. 2003 , mnras , 340 , l33 rodriguez espinosa , j. m. , & perez garcia , a. m. 1997 , apj , 487 , l33 rowan - robinson , m. 1995 , mnras , 272 , 737 rush , b. , malkan , m. a. , & spinoglio , l. 1993 , apjs , 89 , 1 sanders , d. b. , & mirabel , i. f. 1996 , ara&a , 34 , 749 shure , m. a. , toomey , d. w. , rayner , j. t. , onaka , p. , & denault , a. j. 1994 , proc . spie , 2198 , 614 smith , r. g. , sellgren , k. , & tokunaga , a. t. 1989 , apj , 344 , 413 smith , r. g. , sellgren , k. , & brooke , t. y. 1993 , mnras , 263 , 749 spinoglio , l. , & malkan , m. a. 1989 , apj , 342 , 83 storchi - bergmann , t. , wilson , a. s. , & baldwin , j. a. 1996 , apj , 460 , 252 storchi - bergmann , t. , rodriguez - ardila , a. , schmitt , h. r. , wilson , a. s. , & baldwin , j. a. 1996 , apj , 472 , 83 storchi - bergmann , t. , raimann , d. , bica , e. l. d. , & fraquelli , h. a. 2000 , apj , 544 , 747 tanaka , m. , sato , s. , nagata , t. , & yamamoto , t. 1990 , apj , 352 , 724 taniguchi , y. , & murayama , t. 1998 , apj , 501 , l25 tokunaga , a. t. 2000 , in allen s astrophysical quantities , ed . a. n. cox ( 4th ed ; berlin : springer ) , 143 toomre , a. , 1964 , apj , 139 , 1217 tran , h. d. 2001 , apj , 554 , l19 tran , h. d. 2003 , apj , 583 , 632 umemura , m. , fukue , j. , & mineshige , s. 1997 , apj , 479 , l97 umemura , m. , fukue , j. , & mineshige , s. 1998 , mnras , 299 , 1123 voit , g. m. 1992 , mnras , 258 , 841 von linden , s. , biermann , p. l. , duschl , w. j. , lesch , h. , & schmutzler , t. 1993 , a&a , 280 , 468 wada , k. , & norman , c. a. 2002 , apj , 566 , l21 whittet , d. c. b. , bode , m. f. , longmore , a. j. , adamson , a. j. , mcfadzean , a. d. , aitken , d. k. , & roche , p. f. 1988 , mnras , 233 , 321 lcrrrrrccc mrk 334 & 0.022 & 0.16 @xmath55 & 0.23 & 1.05 & 4.35 & 4.32 & 44.6 & cfa + mrk 993 & 0.015 & 0.018 @xmath56 & @xmath20.13 & @xmath20.13 & 0.30 & 1.32 & 43.243.6 & cfa + mrk 573 & 0.017 & 0.17 @xmath56 & @xmath20.25 & @xmath20.93 & 1.09 & 1.26 & 43.644.1 & cfa + ngc 1144 & 0.029 & @xmath20.035 @xmath57 & 0.28 & 0.63 & 5.30 & 11.3 & 45.0 & cfa , 12@xmath0 m + ngc 4388 & 0.008 & 0.40 @xmath58 & 1.00 & 3.46 & 10.2 & 18.1 & 44.2 & cfa , 12@xmath0 m + ngc 5252 & 0.023 & 0.028 @xmath55 & 0.08 & 0.14 & 1.85 & 1.13 & 44.2 & cfa + ngc 5256 ( mrk 266sw ) & 0.028 & 0.030 @xmath55 & 0.23 & 0.98 & 7.34 & 11.1 & 45.0 & cfa , 12@xmath0 m + ngc 5347 & 0.008 & 0.21 @xmath55 & 0.31 & 0.96 & 1.42 & 2.64 & 44.5 & cfa , 12@xmath0 m + ngc 5674 & 0.025 & 0.024 @xmath55 & 0.14 & 0.28 & 1.44 & 3.73 & 44.4 & cfa + ngc 5695 & 0.014 & @xmath20.01 @xmath55 & @xmath20.11 & 0.13 & 0.57 & 1.79 & 43.443.6 & cfa + ngc 5929 & 0.008 & 0.024 @xmath55 & 0.43 & 1.62 & 9.14 & 13.7 & 44.1 & cfa , 12@xmath0 m + ngc 7674 & 0.029 & 0.33 @xmath55 & 0.67 & 1.90 & 5.59 & 8.15 & 45.1 & cfa , 12@xmath0 m + ngc 7682 & 0.017 & 0.012 @xmath55 & 0.32 & 0.22 & 0.47 & 0.41 & 43.9 & cfa + mrk 938 & 0.019 & 0.24 @xmath55 & 0.40 & 2.37 & 16.6 & 17.2 & 45.0 & 12@xmath0 m + ngc 262 ( mrk 348 ) & 0.015 & 0.30 @xmath59 & 0.31 & 0.83 & 1.29 & 1.55 & 44.0 & 12@xmath0 m + ngc 513 & 0.020 & & 0.17 & 0.28 & 1.94 & 4.05 & 44.3 & 12@xmath0 m + f01475@xmath330740 & 0.017 & 0.21 @xmath60 & 0.31 & @xmath20.96 & 1.05 & 0.65 & 43.944.1 & 12@xmath0 m + ngc 1125 & 0.011 & & 0.17 & 0.83 & 3.31 & 3.94 & 43.9 & 12@xmath0 m + ngc 1194 & 0.013 & & 0.27 & 0.51 & 0.77 & @xmath20.93 & 43.7 & 12@xmath0 m + ngc 1241 & 0.014 & 0.08 @xmath55 & 0.24 & 0.45 & 3.56 & 10.3 & 44.2 & 12@xmath0 m + ngc 1320 ( mrk 607 ) & 0.010 & 0.028@xmath55 & 0.33 & 1.07 & 2.15 & 2.75 & 43.8 & 12@xmath0 m + f04385@xmath330828 & 0.015 & 0.27 @xmath55 & 0.42 & 1.62 & 2.77 & 2.53 & 44.3 & 12@xmath0 m + ngc 1667 & 0.015 & 0.005 @xmath55 & 0.43 & 0.68 & 5.95 & 14.7 & 44.5 & 12@xmath0 m + ngc 3660 & 0.012 & 0.033 @xmath55 & 0.19 & 0.22 & 1.87 & 4.54 & 43.8 & 12@xmath0 m + ngc 4501 & 0.008 & 0.006 @xmath55 & 1.02 & 1.28 & 13.7 & 54.7 & 44.4 & 12@xmath0 m + ngc 4968 & 0.010 & 0.12 @xmath55 & 0.39 & 1.05 & 2.38 & 2.94 & 43.8 & 12@xmath0 m + mcg-3 - 34 - 64 & 0.017 & 0.44 @xmath55 & 0.88 & 2.86 & 5.90 & 5.48 & 44.7 & 12@xmath0 m + ngc 5135 & 0.014 & 0.16 @xmath55 & 0.64 & 2.40 & 16.9 & 28.6 & 44.8 & 12@xmath0 m + mcg-2 - 40 - 4 & 0.024 & & 0.39 & 0.86 & 3.65 & 6.53 & 44.7 & 12@xmath0 m + f15480@xmath330344 & 0.030 & & 0.18 & 0.73 & 1.07 & @xmath24.05 & 44.544.6 & 12@xmath0 m + ngc 7172 & 0.009 & & 0.44 & 0.76 & 5.71 & 12.3 & 44.0 & 12@xmath0 m + mcg-3 - 58 - 7 & 0.032 & & 0.28 & 0.80 & 2.42 & 3.36 & 44.8 & 12@xmath0 m + llcclccc mrk 334 & 2002 aug 19 & subaru ircs & 20 & hr 8955 & 5.1 & f6v & 6400 + mrk 993 & 2002 aug 27 , 29 & irtf spex & 100 & hr 410 & 5.0 & f7v & 6240 + mrk 573 & 2002 aug 28 & irtf spex & 30 & hr 650 & 4.1 & f8v & 6000 + ngc 1144 & 2002 oct 24 & subaru ircs & 16 & hr 962 & 3.7 & f8v & 6000 + ngc 4388 & 2003 mar 18 & irtf spex & 30 & hr 4708 & 5.0 & f8v & 6000 + ngc 5252 & 2003 mar 19 & irtf spex & 40 & hr 5011 & 3.8 & g0v & 5930 + ngc 5256 & 2001 apr 9 & irtf nsfcam & 36 & hr 4767 & 4.8 & f8v g0v & 6000 + ngc 5347 & 2003 mar 18 & irtf spex & 40 & hr 5346 & 4.8 & f8v & 6000 + ngc 5674 & 2002 mar 28 & subaru ircs & 12 & hr 5386 & 5.1 & a0v & 9480 + ngc 5695 & 2003 mar 20 & irtf spex & 60 & hr 5630 & 4.9 & f8v & 6000 + ngc 5929 & 2002 aug 29 & irtf spex & 40 & hr 5728 & 4.5 & g3v & 5800 + ngc 7674 & 2002 aug 27 & irtf spex & 20 & hr 8653 & 4.6 & g8iv & 5400 + ngc 7682 & 2002 aug 27 , 29 & irtf spex & 160 & hr 8969 & 2.8 & f7v & 6240 + mrk 938 & 2002 aug 28 & irtf spex & 30 & hr 8917 & 4.9 & g0v & 5930 + ngc 262 & 2002 aug 28 & irtf spex & 30 & hr 410 & 5.0 & f7v & 6240 + ngc 513 & 2002 aug 28 & irtf spex & 30 & hr 410 & 5.0 & f7v & 6240 + f01475@xmath330740 & 2002 aug 27 & irtf spex & 30 & hr 466 & 4.9 & f7v & 6240 + ngc 1125 & 2002 oct 24 & subaru ircs & 8 & hr 784 & 4.5 & f6v & 6400 + ngc 1194 & 2002 aug 28 & irtf spex & 30 & hr 996 & 3.2 & g5v & 5700 + ngc 1241 & 2002 aug 29 & irtf spex & 50 & hr 784 & 4.5 & f6v & 6400 + ngc 1320 & 2002 aug 29 & irtf spex & 30 & hr 784 & 4.5 & f6v & 6400 + f04385@xmath330828 & 2002 aug 29 & irtf spex & 30 & hr 1536 & 4.4 & f8v & 6000 + ngc 1667 & 2003 mar 20 & irtf spex & 60 & hr 1536 & 4.4 & f8v & 6000 + ngc 3660 & 2003 mar 20 & irtf spex & 90 & hr 4529 & 4.9 & f7v & 6240 + ngc 4501 & 2003 mar 18 & irtf spex & 40 & hr 4708 & 5.0 & f8v & 6000 + ngc 4968 & 2003 mar 20 & irtf spex & 30 & hr 4935 & 4.2 & f7v & 6240 + mcg-3 - 34 - 64 & 2003 mar 19 & irtf spex & 40 & hr 4995 & 3.6 & g6v & 5620 + ngc 5135 & 2001 apr 8 & irtf nsfcam & 36 & hr 5212 & 4.8 & f7v & 6240 + mcg-2 - 40 - 4 & 2003 mar 19 & irtf spex & 40 & hr 5779 & 5.2 & f7v & 6240 + f15480@xmath330344 & 2002 aug 29 & irtf spex & 30 & hr 5779 & 5.2 & f7v & 6240 + ngc 7172 & 1999 sep 9 & ukirt cgs4 & 53.3 & hr 8087 & 5.3 & a0v & 9480 + mcg-3 - 58 - 7 & 2002 aug 28 & irtf spex & 20 & hr 8457 & 4.8 & f6v & 6400 + lcccc mrk 334 & [email protected] & 58@xmath611 & 0.1 & 11 + mrk 993 & @xmath22.2 & @xmath29.4 & @xmath20.6 & @xmath216 + mrk 573 & @xmath23.2 & @xmath218 & @xmath20.5 & @xmath26 + ngc 1144 & @xmath20.7 & @xmath212 & @xmath20.02 & @xmath27 + ngc 4388 & @xmath23.8 & @xmath24.7 & @xmath20.03 & @xmath25 + ngc 5252 & @xmath25.4 & @xmath256.4 & @xmath20.4 & @xmath212 + ngc 5256 & [email protected] & 133@xmath617 & 0.1 & 87 + ngc 5347 & @xmath26.4 & @xmath27.9 & @xmath20.3 & @xmath210 + ngc 5674 & @xmath21.5 & @xmath219 & @xmath20.08 & @xmath24 + ngc 5695 & @xmath21.6 & @xmath26.1 & @xmath20.3 & @xmath216 + ngc 5929 & @xmath21.4 & @xmath21.7 & @xmath20.02 & @xmath213 + ngc 7674 & [email protected] & 136@xmath618 & 0.1 & 7 + ngc 7682 & [email protected] & 11@xmath613 & 0.2 & 40 + mrk 938 & 41@xmath611 & 289@xmath613 & 0.3 & 75 + ngc 262 & @xmath28.6 & @xmath238 & @xmath20.4 & @xmath27 + ngc 513 & @xmath22.7 & @xmath222 & @xmath20.2 & @xmath26 + f01475 - 0740 & [email protected] & 22@xmath613 & 0.3 & 21 + ngc 1125 & [email protected] & 24@xmath613 & 0.3 & 50 + ngc 1194 & @xmath22.7 & @xmath28.8 & @xmath20.2 & @xmath24 + ngc 1241 & @xmath21.1 & @xmath24.1 & @xmath20.03 & @xmath212 + ngc 1320 & @xmath25.9 & @xmath212 & @xmath20.2 & @xmath28 + f04385@xmath330828 & [email protected] & 18@xmath611 & 0.1 & 4 + ngc 1667 & @xmath21.6 & @xmath27.1 & @xmath20.03 & @xmath216 + ngc 3660 & [email protected] & 11@xmath613 & 0.2 & 50 + ngc 4501 & @xmath23.8 & @xmath24.7 & @xmath20.02 & @xmath212 + ngc 4968 & [email protected] & 17@xmath612 & 0.2 & 18 + mcg-3 - 34 - 64 & @xmath23.3 & @xmath218 & @xmath20.04 & @xmath26 + ngc 5135 & 15@xmath613 & 58@xmath6111 & 0.09 & 24 + mcg-2 - 40 - 4 & @xmath27.1 & @xmath280 & @xmath20.2 & @xmath23 + f15480@xmath330344 & @xmath23.3 & @xmath259 & @xmath20.2 & @xmath211 + ngc 7172 & @xmath25.3 & @xmath28.4 & @xmath20.08 & @xmath23.5 + mcg-3 - 58 - 7 & @xmath23.8 & @xmath278 & @xmath20.2 & @xmath23 +

we investigate the problem of reheating and its effects on the evolution of the curvature perturbations in tachyonic inflationary models . we derive the equations governing the evolution of the scalar perturbations for a system consisting of a tachyon and a perfect fluid . assuming the perfect fluid to be radiation , we solve the coupled equations for the system numerically and study the evolution of the perturbations from the sub - hubble to the super - hubble scales . in particular , we analyze the effects of the transition from tachyon driven inflation to the radiation dominated epoch on the evolution of the large scale curvature and non - adiabatic pressure perturbations . we consider two different potentials to describe the tachyon and study the effects of three possible types of decay of the tachyon into radiation . we plot the spectrum of curvature perturbations at the end of inflation as well as at the early stages of the radiation dominated epoch . we find that reheating modifies the amplitude of the curvature perturbations _ only _ when the decay rate of the tachyon into radiation is explicitly time - dependent . we illustrate that it is the growth of the relative non - adiabatic pressure perturbation between the tachyon and radiation that is responsible for the change in the amplitude of the curvature perturbations during the transition . our results corroborate similar conclusions that have been arrived at earlier based on the study of the evolution of the perturbations in the super - hubble limit . in an appendix , we also discuss the corresponding results for the popular chaotic inflation model in the case of the canonical scalar field . we briefly comment on the possible implications of our results .

0902.1067

in the standard inflationary scenario , the scales of cosmological interest exit the hubble radius within the first few ( about @xmath0-@xmath1 ) @xmath2-folds of inflation and , hence , are outside the hubble scale during the later epochs . when comparing the predictions of the inflationary models with the cosmic microwave background ( cmb ) and the large scale structure data , it is often assumed that the amplitude of the curvature perturbations remain constant at super - hubble scales . in such a situation , the observed cmb anisotropies directly determine the amplitude and the shape of the perturbation spectrum imprinted on the modes when they left the hubble radius during inflation . typically , the amplitude of the perturbations constrain the parameters that describe the inflaton potential , while the shape of the spectrum limits its form ( see any of the standard texts @xcite or one of the following reviews @xcite ) . provided inflation is of the slow roll type for _ all _ of the required number of @xmath2-folds , it is indeed true that the amplitude of the curvature perturbations freeze at their value at hubble exit . however , if there is a period of deviation from slow roll inflation , then the asymptotic ( i.e. the extreme super - hubble ) amplitude of the modes that leave the hubble radius just before the deviation are enhanced when compared to their value at hubble exit @xcite . while modes that leave well before the deviation remain unaffected , it is found that there exists an intermediate range of modes whose amplitudes are actually suppressed at super - hubble scales @xcite . depending on the form of the departure from slow roll , these effects lead to certain features in the scalar perturbation spectrum . if such deviations occur either during or soon after the cosmological scales leave the hubble radius , then the cmb observations constrain the resulting features rather well ( for an inexhaustive list , see refs . but , at smaller scales , only theoretical tools are currently available to restrict the form of the primordial spectrum . these constraints are essentially based on the number density of primordial black holes that are formed towards the end of inflation ( for recent discussions in this context , see , for example , refs . @xcite ) since the cosmological scales are well outside the hubble radius by the early stages of inflation , clearly , the _ shape _ of the perturbation spectrum on such large scales is indeed unlikely to be affected by subsequent dynamics . but , over the past decade , it has been recognized that post - inflationary dynamics can alter the _ amplitude _ of the curvature perturbations at super - hubble scales , in particular , when more than one component of matter is present . preheating , the curvaton scenario and the modulated reheating mechanism are popular examples that illustrate the interesting possibilities of post - inflationary dynamics . preheating a mechanism that transfers energy from the inflaton to radiation through an explosive production of quanta corresponding to an intermediate scalar field is known to even lead to an exponential growth in the amplitude of the super - hubble perturbations ( for the earlier discussions , see refs . @xcite and , for more recent efforts , see , for instance , refs . @xcite ) . in the curvaton scenario , while the inflationary epoch still remains the source of the perturbations , these perturbations are amplified after inflation due to the presence of entropic perturbations ( see , for example , refs . the modulated reheating scenario is an extreme case wherein inflation is essentially required only to resolve the horizon problem , whereas the perturbations are generated due to an inhomogeneous decay rate when the energy is being transferred from the inflaton to radiation through other fields @xcite . these different alternatives indicate that the post - inflationary evolution of the large scale curvature perturbations can be highly model dependent and , therefore , requires a careful and systematic study . evidently , in these scenarios , the effects of post - inflationary dynamics have to be taken into account when constraining the inflationary models using the cmb observations ( in this context , see , for instance , ref . @xcite ) . with these motivations in mind , in this paper , we investigate the problem of the more conventional reheating scenario @xcite , and its effects on the evolution of the curvature perturbations in tachyonic inflationary models ( for the original discussions on the tachyon , see refs . @xcite ; for efforts on treating the tachyon as an inflaton , see , for instance , refs . @xcite ; for discussions on reheating in such inflationary models , see refs . we shall consider two types of potentials to describe the tachyon and construct scenarios of transition from inflation to radiation domination for the following three possible cases of the decay rate @xmath3 of the tachyon into radiation : ( i ) a constant , ( ii ) dependent on the tachyon , and ( iii ) explicitly dependent on time . we solve the coupled equations for the system numerically and study the evolution of the perturbations from the sub - hubble to the super - hubble scales . importantly , we shall consider the effects of the transition from inflation to the radiation dominated epoch on the evolution of the large scale ( i.e. those that correspond to cosmological scales today ) curvature and non - adiabatic pressure ( i.e. the intrinsic entropy as well as the relative entropy or the isocurvature ) perturbations . we shall evaluate the spectrum of curvature perturbations at the end of inflation as well as at the early stages of the radiation dominated epoch . as we shall illustrate , reheating affects the amplitude of the curvature perturbations _ only _ when the decay rate of the tachyon into radiation is explicitly time dependent . we shall also show that it is the growth of the isocurvature perturbation ( i.e. the relative non - adiabatic pressure perturbation between the tachyon and radiation ) that induces the change in the amplitude of the curvature perturbations during the transition . our results confirm similar conclusions that have been arrived at earlier in the literature based on the study of the evolution of the perturbations in the super - hubble limit @xcite . we shall also discuss similar effects in the case of the canonical scalar field . a few clarifying remarks are in order at this stage of the discussion . we should stress that the set up we are considering is the standard cold inflationary scenario , followed by an epoch of reheating achieved by the standard method of introducing a coarse - grained decay rate in the equation of motion describing the inflaton @xcite . recently , an analysis somewhat similar to what we shall consider here has been studied in the context of warm inflationary scenarios involving tachyonic fields @xcite . though there can exist similarities in the form of the equations in the cold and the warm inflationary scenarios , the physics in these two scenarios are rather different ( for a recent discussion , see , for example , ref . moreover , to analyze the effects of reheating on the large scale perturbations , one could have possibly studied the evolution of the perturbations in the super - hubble limit , say , the first order ( in time ) differential equations usually considered in the literature ( cf however , some concerns have been raised that the coupling between the inflaton and radiation may affect the amplitude as well as the scalar spectral index in certain situations @xcite . also , the modified background dynamics preceding reheating can effect the extent of small scale primordial black holes that are formed towards the end of inflation . if we are to address such issues , it requires that we study the evolution of the perturbations from the sub - hubble to the super - hubble scales . we shall comment further on some of these points in the concluding section . the remainder of this paper is organized as follows . in the following section , we shall summarize the essential background equations for the system consisting of the tachyon and a perfect fluid and set up scenarios of transition from inflation to radiation domination . in , we shall obtain the equations describing the scalar perturbations for the system of the tachyon and a perfect fluid . in , we shall evolve the coupled system of equations describing the perturbations and study the effects of the transition from inflation to the radiation dominated epoch on the super - hubble curvature perturbations . we shall plot the spectrum of curvature perturbations at the end of inflation as well as at the early stages of the radiation dominated epoch . we shall also explicitly illustrate that it is the growth of the isocurvature perturbations that acts as the source of change in the amplitude of the curvature perturbations during the transition . finally , in , we conclude with a summary and a discussion on the results we have obtained . in the appendix , we shall discuss the corresponding results for the chaotic inflation model in the case of the canonical scalar field . the notations and conventions we shall adopt are as follows . we shall set @xmath4 , but shall display @xmath5 explicitly , and define the planck mass to be @xmath6 . we shall work with the metric signature of @xmath7 . we shall express the various quantities in terms of the cosmic time @xmath8 , and we shall denote differentiation with respect to @xmath8 by an overdot . while we shall use the subscript @xmath9 to denote the spatial components of , say , the momentum flux associated with the matter fields , the subscripts @xmath10 and @xmath11 shall refer to the two components of matter . in this section , we shall discuss the background equations describing the evolution of the tachyon that is interacting with a perfect fluid . assuming the perfect fluid to be radiation , we shall construct scenarios of transition from tachyon driven inflation to radiation domination for the following three possible types of the decay rate @xmath3 [ cf . eq . below ] of the inflaton into radiation : ( i ) a constant , ( ii ) dependent on the tachyon , and ( iii ) explicitly dependent on time @xcite . consider a @xmath12-dimensional , spatially flat , smooth , expanding , friedmann universe described by the line element s^2 = dt^2-a^2(t)^2,[eq : frwle ] where @xmath8 denotes the cosmic time and @xmath13 is the scale factor . let @xmath14 and @xmath15 denote the total energy density and the total pressure of a system consisting of multiple components of fields and fluids that are driving the expansion . then , the einstein s equations corresponding to the above line - element lead to the following friedmann equations for the scale factor @xmath13 : h^2= ()= -(4 g ) (+ p),[eq : fe ] where @xmath16 is the hubble parameter . also , the conservation of the total energy of the system leads to the continuity equation + 3h (+ p ) = 0.[eq : tec ] the total energy density and the total pressure of the system can be expressed as the sum of the energy density @xmath17 and the pressure @xmath18 of the individual components as follows : = _ _ _ p = _ p__. if the different components of fields and fluids do not interact , then , in addition to the total energy density , the energy density of the individual components will be conserved as well . hence , in such a situation , the energy density @xmath17 of each component will individually satisfy the continuity equation + 3h ( _ _ + p _ _ ) = 0 . on the other hand , when the different components interact , the continuity equation for the individual components can be expressed as ( see , for instance , refs . @xcite ) + 3h ( _ _ + p _ _ ) = q__,[eq : ceic ] where @xmath19 denotes the rate at which energy density is transferred to the component @xmath10 from the other components . the conservation of the energy of the complete system then leads to the following constraint on the total rate of transfer of the energy densities : _ q__=0.[eq : c ] let us now consider the system of a tachyon ( which we shall denote as @xmath20 ) that is interacting with a perfect fluid ( referred to , hereafter , as @xmath21 ) , so that @xmath22 . we shall assume that the rate at which energy density is transferred to the perfect fluid , i.e. @xmath23 , is given by @xcite q__f= ^2 _ _ t,[eq : qf ] where @xmath24 is the energy density of the tachyon . such a transfer of energy is assumed to describe albeit , in a course grained fashion the perturbative decay of the tachyon into particles that constitute the perfect fluid . the quantity @xmath3 represents the corresponding decay rate and , as we shall discuss below , it can either be a constant , depend on the tachyon , or depend explicitly on time @xcite . the last possibility effectively describes a situation wherein the inflaton first decays into particles corresponding to an intermediate field before further decaying into particles constituting the perfect fluid . for the above @xmath23 , the continuity equation ( [ eq : ceic ] ) corresponding to the perfect fluid is given by _ _ f + 3h (1 + w__f ) _ _ f = ^2 _ _ t,[eq : cef ] where @xmath25 is the equation of state parameter describing the perfect fluid , which we shall assume to be a constant . also , it is evident from eq . ( [ eq : c ] ) that q__t =- q__f = - ^2 _ _ t . therefore , the continuity equation for the tachyon energy density @xmath24 reduces to _ _ t + 3h (__t + p__t ) = - ^2 _ _ t , [ eq : cet ] where @xmath26 is the pressure associated with the tachyon . given a potential @xmath27 describing the tachyon , the corresponding energy density @xmath24 and pressure @xmath26 are given by ( see , for instance , refs . @xcite ) _ _ t = ( ) p__t = -v(t ) . on substituting these two expressions in the continuity equation ( [ eq : cet ] ) , we arrive at the following equation of motion for the tachyon @xmath20 @xcite : ( ) + 3 h+ + ()=0 , [ eq : emt ] where @xmath28 . before we proceed , the following clarification on the choice of @xmath23 in eq . is in order at this stage of the discussion . recall that , in a situation wherein the canonical scalar field , say @xmath29 , drives inflation , it is common to introduce a @xmath30 term in the field equation to describe the perturbative decay of the inflaton [ see refs . @xcite , also see eq . ] . motivated by the canonical case , the choice of @xmath23 in eq . has been specifically made so as to lead to the @xmath31 term in the equation of motion ( [ eq : emt ] ) for the tachyon @xcite . needless to add , the choice ( [ eq : qf ] ) is but one of the many possibilities that can help in achieving reheating at the end of inflation . in this sub - section , assuming the perfect fluid to be radiation with @xmath32 , we shall construct specific scenarios of transition from tachyon driven inflation to an epoch of radiation domination . we shall consider two different types of potentials in order to describe the tachyon . * the first potential we shall consider is given by v_1(t)= ( ) , a potential that is well motivated from the string theory perspective @xcite . * our second choice will be the following phenomenologically motivated power law potential that has been considered earlier in the literature ( see , for example , ref . @xcite ) : v_2(t)= ( ) . in order to achieve the necessary amount of inflation and the correct amplitude for the scalar perturbations , suitable values for the two parameters @xmath33 and @xmath34 that describe the above potentials can be arrived at as follows . firstly , one finds that , in these potentials , inflation typically occurs around @xmath35 corresponding to an energy scale of about @xmath36 . secondly , it turns out that , the quantity @xmath37 has to be much larger than unity for the potential slow roll parameters to be small and thereby ensure that , at least , @xmath38 @xmath2-folds of inflation takes place . the cobe normalization condition for the scalar perturbations essentially constrains the value of @xmath33 . we shall then choose a sufficiently large value of @xmath39 , thereby determining the value of @xmath34 . in the absence of the fluid , we find that the two potentials @xmath40 and @xmath41 above allow about @xmath38 @xmath2-folds of slow roll inflation and lead to the correct cobe amplitude for the choice of the parameters and initial conditions listed in . 4 pt .[tab : beiaf]achieving inflation : the values for the parameters of the tachyon potential and the initial conditions for the inflaton ( the initial value @xmath42 of the tachyon and its initial velocity @xmath43 ) that we work with . these choices for the parameters and initial conditions lead to the required @xmath38 @xmath2-folds of slow roll inflation and the observed amplitude for the scalar perturbations . [ cols="^,^,^,^,^",options="header " , ] in the left column of , we have plotted the fractional contributions ( with respect to the total energy density ) of the energy densities of the tachyon and radiation , i.e. @xmath44 and @xmath45 , as a function of the number of @xmath2-folds @xmath46 for the potential @xmath40 . -100 true pt -220 true pt @xmath47 210 true pt @xmath48 105 pt -100 true pt -220 true pt @xmath47 210 true pt @xmath48 105 pt -100 true pt -220 true pt @xmath47 210 true pt @xmath48 85 pt 25 true pt @xmath46 215 true pt @xmath46 it is evident from the figure that the various choices for the decay rate @xmath3 ensure the completion of reheating ( i.e. @xmath49 ) within a couple of @xmath2-folds after the end of inflation . in the right column of the figure , we have plotted the corresponding effective equation of state parameter , viz . @xmath50 , and also the first hubble slow roll parameter , viz . @xmath51 , of the entire system , for the same tachyon potential . these plots , while confirming that inflation has indeed ended and reheating has been realized , also indicate the nature of the composite matter during the transition . we should add that a very similar behavior occurs for the potential @xmath41 . in this section , using the first order einstein s equations and the equations describing the conservation of the perturbed energy density of the tachyon and the perfect fluid , we shall arrive at the coupled , second order ( in time ) differential equations governing the scalar perturbations for the system consisting of the tachyon and the fluid . for convenience , we shall work in the uniform curvature gauge ( ucg ) , a gauge that is also referred to , often , as the spatially flat gauge . if we take into account the scalar perturbations to the spatially flat background metric ( [ eq : frwle ] ) , then , in the ucg , the friedmann line - element is given by @xcite s^2 = ( 1 + 2 a)dt^2 - 2 a ( _ i b ) dx_i dt - a^2(t)^2,[eq : frwlesp ] where @xmath52 and @xmath53 are functions that describe the two degrees of freedom associated with the perturbations . being a scalar field , the tachyon , evidently , does not possess any anisotropic stress at the linear order in the perturbations . we shall assume that the perfect fluid does not contain any anisotropic stress either . in such a case , at the first order in the perturbations , the einstein equations in the ucg can be written as @xcite -(3 h^2 ) a - ( ) ^2 b & = ( 4 g ) , [ eq : foee00 ] + h (_ia ) & = ( 4 g ) (_iq),[eq : foee0i ] + h a + (2h + 3 h^2 ) a & = ( 4 g ) p,[eq : foeeii ] where the quantities @xmath54 , @xmath55 , and @xmath56 denote the perturbations in the total energy density , the total momentum flux , and the total pressure of the complete system , respectively . the absence of the anisotropic stress leads to the following additional relation between the two functions @xmath52 and @xmath53 describing the scalar perturbations : a+a (b+2 h b)=0.[eq : foeeij ] for the system of our interest , viz . that of the tachyon and a perfect fluid described by a constant equation of state parameter @xmath57 , the quantities @xmath54 , @xmath58 and @xmath56 are given by = (__t + _ _ f ) = ( ) + ( ) (- a t ) + _ _ f,[eq : deltarho ] + q = (q__t + q__f ) = ( ) - _ _ f,[eq : deltaq ] + p = (p__t + p__f ) = -(v__t t ) + ( ) (- a t ) + w__f__f,[eq : deltap ] where @xmath59 denotes the perturbation in the tachyon and @xmath60 is proportional to the potential that determines the three ( i.e. the spatial ) velocity of the perfect fluid is written as @xmath61 $ ] , where @xmath62 is the potential whose spatial gradient describes the three velocity of the fluid ( see , for instance , refs . @xcite ) . for convenience , we have instead defined the entire quantity @xmath63 $ ] as @xmath64 @xcite . as we shall point out later , we have ensured that suitable sub - hubble initial conditions are imposed on @xmath60 . ] . in the ucg , at the first order in the perturbations , the equation describing the conservation of the energy density of the component @xmath10 is given by @xcite _ _ + 3h ( _ _ + p _ _ ) - ( ) ^2 [q _ _ + ( _ _ + p _ _ ) ab ] - q__a - q__=0,[eq : pece ] where the quantities @xmath65 , @xmath66 and @xmath67 denote the perturbations at the linear order in the energy density , the momentum flux , and the pressure of the particular component @xmath10 , while @xmath19 and @xmath68 indicate the rate at which the energy is transferred to the component @xmath10 and the perturbation in the rate , respectively . the equation of motion that governs the perturbation @xmath59 in the tachyon can now be obtained upon using the expressions for @xmath69 , @xmath70 and @xmath71 in above equation for the conservation of the perturbed energy density . we find that the equation of motion describing the perturbation @xmath59 is given by ( ) + [(3h+ ) (1 - 3t^2 ) -2 t ( ) ] ( ) + + [ ( ) - ( ) ^2 ] t + - ( ) + [6 h t^3 + 2 ( ) + t (1+t^2 ) ] ( ) + - ( ) ( ^2 b ) + t = 0,[eq : deltat ] where @xmath72 denotes the first order perturbation in the decay rate . upon using eq . , we find that the equation describing the conservation of the perturbed energy density of the fluid @xmath73 can be written as + 3 h ( 1+w__f ) + ()^2 _ _ f + - ( ) - ( ) t + ( ) a + - ( ) ^2 b - ( ) = 0 . [ eq : rhof ] also , on using the einstein s equations and , we obtain the following first order ( in time ) differential equation for the quantity @xmath60 that describes the spatial velocity of the fluid : _ _ f+3 h _ _ f + w__f__f + ( 1+w__f ) _ _ f a + ( ) t=0 . [ eq : psif ] we should emphasize here that the above equations take into account all the three possibilities of @xmath3 ( i.e. it can either be a constant , depend on the tachyon , or on time ) that we had discussed earlier . the variables describing the perturbations in the tachyon and the fluid that we shall eventually evolve numerically are @xcite _ _ t = t__f = ().[eq : calqf ] we have already obtained a second order differential equation ( in time ) describing the evolution of @xmath74 [ viz . . a similar equation for @xmath75 can be obtained by differentiating eq . with respect to time , and upon using eq . that describes the conservation of the perturbed energy density of the fluid then , on utilizing the first order einstein s equations , and to eliminate the scalar metric perturbations @xmath52 and @xmath53 , we can arrive at a set of coupled , second order ( in time , again ! ) differential equations for the quantities @xmath74 and @xmath75 . for simplicity , let us first discuss the case wherein the decay rate @xmath3 is a constant . we shall later indicate as to how certain coefficients in the differential equations change when @xmath3 is assumed to be either dependent on the tachyon or on time . upon suitably using the background equations , and , we find that the differential equations satisfied by the perturbation variables @xmath74 and @xmath75 can be written as _ _ + _ (f _ _ _ _ + g _ _ q _ _ ) = 0,[eq : ceqtqf ] where , as before , @xmath76 . the coefficients @xmath77 and @xmath78 appearing in the above equations are given by _ _ tt = (3h+ ) (1 - 3t^2 ) - 2 t ( ) , _ _ tf = ( ) [1- ( ) ] t , _ _ ft = ( ) \{ ( ) [1- ( ) ] . + .+[1+w__f ( ) ] ( ) } , _ _ ff = 3 h + ( ) ( ) , _ _ tt = (1-t^2 ) [()- ( ) ^2 ] + + ( ) ( ) [(2-t^2 ) ( ) + (2+t^2 ) (3h+ ) t. + -. ( ) (1-t^2 ) ] + - ()^2 \{ ( ) - ( ) [1+ ( ) ] (1+w__f)__f } , _ _ tf = ( ) \{- ( ) [1+ ( ) ] (1+w__f ) . + + ( ) [1- ( ) ] ( ) + -.(6 ht^3 + 2 ( ) + ( t ) (1+t^2 ) ) } + + ()^2 \{ ( ) + [1+()]t [(1+w__f)__f ] } , _ _ ft = - ( ) \{[1+ ( ) ] [(1+w__f)__f ] v__t . + .+ ( ) (1+w__f)^2 _ _ f - ( ) [1- ( ) ] } + ( ) + -()^2 \{ ( ) [1+ ( ) ] [(1+w__f)__f ] . + -. ( ) [(1+w__f ) _ _ f]^2 } ( ) + + ( ) [ ( ) (1-t^2 ) - ( ) w__f t^2 . + + . w__f t ] (),__ff = - ( ) ^2 + ( ) (1-w__f^2 ) + + ( ) \{ ( ) (1+w__f ) (1 + 3w__f ) _ _ f + ( ) ( ) (w__f-1 ) . + -. ( ) [1- ( ) ] } + -()^2 \{ ( ) [1+ ( ) ] [(1+w__f ) _ _ f ] . + + . 2 [(1+w__f ) _ _ f]^2 } + - ( ) \{ ( ) (2-t^2 ) + ( ) (1-t^2 ) ( ) . + -. ( ) (w__f+t^2 ) + ()}. as we had pointed out , the coefficients @xmath79 and @xmath80 above have been derived assuming that the decay rate @xmath3 is a constant . for the case wherein the decay rate is a function of the tachyon , the coefficients @xmath81 and @xmath82 are modified to _ _ tt & _ _ tt + (1-t^2 ) t _ _ t , + g__ft & _ _ ft + ( ) ( ) _ _ t , where @xmath83 . similarly , in the case wherein the decay rate is an explicit function of time , we find that these two coefficients are modified as follows : _ _ ft & _ _ ft + ( ) ( ) , + g__ff & _ _ ff + ( ) ( ) . the other coefficients remain unaffected in both these cases . we should mention here that we have checked that we indeed recover the standard equations in the different limiting cases from the coupled equations for @xmath74 and @xmath75 . for instance , when the fluid is ignored and the decay rate is assumed to vanish , the equation for the variable @xmath74 reduces to the equation that has been derived earlier in the literature for describing the perturbation in the tachyon ( see , for instance , refs . moreover , when the tachyon is assumed to vanish and its coupling to the fluid is also ignored , we find that we arrive at the well known equation governing the fluid perturbation , as required @xcite . in this section , we shall discuss the effects of reheating on the large scale curvature perturbations and the scalar power spectrum . we shall also investigate the behavior of the non - adiabatic pressure perturbations at large scales , when reheating is achieved through the three different choices for the decay rate . we shall solve the coupled equations for the two variables @xmath74 and @xmath75 numerically . imposing the standard bunch - davies initial conditions , we shall evolve the perturbations from the sub - hubble to the super - hubble scales , and analyse the effects of the transition from inflation to the radiation dominated epoch on the evolution of the large scale perturbations . we shall compute the scalar power spectrum at the end of inflation and soon after reheating is complete . let us begin by quickly summarizing the essential quantities of interest and the initial conditions that we shall be imposing . let @xmath84 and @xmath85 denote the curvature perturbations associated with the tachyon and the fluid , respectively . we find that these curvature perturbations are related to the variables @xmath74 and @xmath75 as follows @xcite : _ _ t & = ( ) t = ( ) q__t , + r__f & = - ( ) _ _ f = - ( ) q__f . the total curvature perturbation of the system , say , @xmath86 , can then be expressed as a weighted sum of the individual curvature perturbations in the following fashion ( see , for instance , refs . @xcite ) : = _ ( ) r__. therefore , for our system of interest , the total curvature perturbation is given by = h [ ( ) + [(1+w__f ) _ _ f]]^-1 + [ ( ) q__t -__f ] . the scalar power spectrum is then defined in terms of the fourier mode @xmath87 of the total curvature perturbation as _ _ s(k)= ( ) _ k^2 . the mukhanov - sasaki variables , say , @xmath88 and @xmath89 , associated with the tachyon and the fluid are related to the corresponding curvature perturbations as follows : v__t = (r__t z__t ) v__f = (r__f z__f ) , where the quantities @xmath90 and @xmath91 are given by @xcite z__t = ()^1/2 ( ) z__f = ( ) []^1/2 . as is usually done , we shall impose the following initial conditions on the fourier modes of the perturbation variables @xmath88 and @xmath89 : v_k = ()^1/2 _ k =- ( ) ()^1/2 , where _ k^2=((k [(z / z)+(h z / z ) ] ) , with @xmath92 in the case of the tachyon and @xmath93 in the case of the fluid . also , we shall impose these conditions when the modes are well inside the hubble radius during the inflationary epoch . we shall transform these initial conditions to the corresponding conditions on the variables @xmath74 and @xmath75 when numerically integrating the coupled equations . we shall focus on modes that correspond to cosmological scales today . we shall choose our parameters in such a fashion that modes spread over four orders of magnitude ( say , @xmath94 ) leave the hubble radius during the early stages of inflation ( they actually leave during @xmath95 ) . as we have discussed earlier , for our choice of the parameters and initial conditions on the background variables , reheating is achieved after about @xmath96 odd @xmath2-folds ( also see ) . moreover , the transition from inflation to radiation domination occurs within about @xmath97-@xmath98 @xmath2-folds after inflation has terminated . we evaluate the scalar power spectrum towards the end of inflation as well as soon after the transition to the radiation dominated epoch is complete . in the left column of , we have plotted the evolution of the amplitudes of the individual ( viz . @xmath99 and @xmath100 corresponding to the tachyon and radiation , respectively ) as well as the total curvature perturbation ( i.e. @xmath101 ) for a typical cosmological scale ( we have chosen @xmath102 ) across the transition from inflation to radiation domination . and , in the right column of the figure , we have plotted the scalar power spectrum ( constructed out of the total curvature perturbation ) evaluated before and after the transition . -104 true pt -235 true pt @xmath103 195 true pt @xmath104 108 pt -102 true pt -235 true pt @xmath103 195 true pt @xmath104 106 pt -109 true pt -235 true pt @xmath103 195 true pt @xmath104 95 pt 25 true pt @xmath46 215 true pt @xmath105 -2pt it is evident from the plots that the amplitude of the total curvature perturbation remains unaffected during reheating when the decay rate @xmath3 is either a constant or is a function of the tachyon . however , when the decay rate @xmath3 is an explicit function of time , for our choice of @xmath106 , we find that the amplitude of the total curvature perturbation is enhanced during reheating . these results confirm similar conclusions that have been arrived at earlier based on the first order ( in time ) super - hubble equations @xcite . as we shall illustrate in the next section , this enhancement can be attributed to the growth of the tachyon - fluid non - adiabatic pressure perturbations during the transition . in this section , we shall discuss the evolution of the entropic ( i.e. the non - adiabatic pressure ) perturbations across the transition from inflation to the radiation dominated epoch . we shall show that it is the growth of the isocurvature perturbations ( i.e. relative non - adiabatic pressure perturbation between the tachyon and the fluid ) that is responsible for the change in the amplitude of the large scale curvature perturbations during reheating . let us recall the key quantities and equations . in the ucg , the gauge invariant bardeen potential @xmath107 is given by = a+a(b+h b ) . upon using the first order einstein equations , and , it can be shown that the bardeen potential satisfies the differential equation @xcite + (4 + 3 c__a^2 ) h + [2 h+3 h^2 (1 + c__a^2 ) ] + - ( ) ^2 = (4 g ) p^_na , [ eq : emphi ] where @xmath108 denotes the adiabatic speed of sound in the composite system , and the quantity @xmath109 is the non - adiabatic pressure perturbation defined as p^_na = (p - c__a^2 ) . using the einstein equations and , the total curvature perturbation @xmath86 of the system can be written in terms of the bardeen potential @xmath107 as follows @xcite : = + ( ) ( ) . [ eq : r ] on substituting this expression for the curvature perturbation in the equation ( [ eq : emphi ] ) that governs the evolution of the bardeen potential @xmath107 and making use of the background equations , one can arrive at the well known equation = - ( ) [(4 g ) p^_na + ( ) ^2 ] . [ eq : emrdot ] this equation clearly suggests that , at super - hubble scales , any change in the amplitude of the curvature perturbations has to be due to the non - trivial evolution of the non - adiabatic pressure perturbation @xmath109 . the perfect fluid , by definition , does not possess any intrinsic entropy perturbation . therefore , the non - adiabatic pressure perturbation @xmath109 of our composite system depends on the intrinsic non - adiabatic pressure perturbation of the tachyon , say , @xmath110 , and the relative non - adiabatic pressure perturbation between the tachyon and the perfect fluid ( i.e. the isocurvature perturbation ) , which we shall denote as @xmath111 . hence , we can write p^_na = (p__t^_na+p__tf^_na ) , where the quantity @xmath112 is defined as ( see , for instance , ref . @xcite ) p__t^_na = (p__t - c__t^2 _ _ t ) with @xmath113 . the relative non - adiabatic pressure perturbation @xmath111 is then given by ( see , for instance , refs . @xcite ) p__tf^_na = (3h+ ) ( ) [(__f+p__f)-(/3 h ) (__t+p__t ) ] + (w__f - c__t^2 ) [ ( ) - ( ) ] . upon using the expressions and , and the einstein s equations , and , we find that we can write the quantities @xmath69 , @xmath71 and @xmath114 in terms of the variables @xmath74 and @xmath75 as follows : _ _ t = [ ( ) - ( ) ( ) ] q__t + ()__t + + ( ) ( ) q__f , p__t = -[v__t + ( ) ( ) ] q__t + ( ) _ _ t + + ( ) ( ) q__f , _ _ f = -[ ( ) ( ) (1+w__f ) ( ) + ( ) ] q__t + -[ ( ) (1-w__f ) + ( ) ()^1/2 . + -. ( ) ( ) [(1+w__f ) _ _ f]^3/2 ] q__f + - ( ) _ _ f . these three relations allow us to express the non - adiabatic pressure perturbations @xmath110 and @xmath111 in terms of @xmath74 , @xmath75 , and their time derivatives . in , we have plotted the amplitudes of the bardeen potential ( i.e. @xmath115 ) and the non - adiabatic pressure perturbations ( viz . @xmath116 and @xmath117 ) as a function of the number of @xmath2-folds . -118 true pt -250 true pt @xmath118 120 pt -118 true pt -250 true pt @xmath118 120 pt -120 true pt -250 true pt @xmath118 108 pt 15 true pt @xmath46 -8pt for convenience in numerical computation , we have chosen a very small scale mode ( corresponding to @xmath119 ) that leaves the hubble radius around @xmath120 @xmath2-folds or so during inflation ) prove to be very small when they are well outside the hubble radius . as a result , evolving these reliably until the transition to the radiation dominated epoch requires considerable numerical accuracy , which is difficult to achieve . for this reason , to illustrate the evolution of the non - adiabatic pressure perturbations , we choose to work with a very small scale mode that leaves the hubble radius close to the end of inflation . since the inflationary epoch is of the slow roll type for all the @xmath2-folds , the non - adiabatic pressure perturbations associated with the cosmological scales can be expected to behave in a fashion similar to that associated with such a small scale mode . ] . the following points are evident from the three plots . to begin with , barring a very small change in its amplitude which is expected during the transition from inflation to the radiation dominated epoch the bardeen potential @xmath107 remains a constant at super - hubble scales . secondly , it is clear from the plots that the intrinsic entropy perturbation associated with the tachyon decays as @xmath121 at super - hubble scales during inflation , a result that is well known in the literature ( see , for instance , refs . thirdly , and interestingly , we find that the isocurvature perturbation behaves in almost the same fashion as the intrinsic entropy perturbation both before and after the transition , a feature that has been noticed earlier @xcite . while they both behave as @xmath121 at super - hubble scales before the transition , they die down _ extremely _ rapidly ( roughly as @xmath122 ! ) during the radiation dominated epoch . lastly , it is also evident from the third plot that , when the decay rate is a function of time , while the intrinsic entropy perturbation continues to decay , the isocurvature perturbation exhibits a brief period of growth close to the transition before decaying . obviously , it is this growth of the isocurvature perturbations that is responsible for the amplification of the curvature perturbations during the transition . in this work , we have studied the evolution of perturbations in an interacting system consisting of a tachyon and radiation . treating the tachyon as an inflaton , we have investigated the effects of reheating i.e . the perturbative transfer of energy from the tachyon to radiation on the large scale curvature perturbations . we have shown that the transition does not alter the amplitude of the total curvature perturbation of the system when the rate describing the decay of the inflaton into radiation is either a constant or a function of the tachyon . however , we find that the total curvature perturbation is modified when the decay rate associated with the tachyon turns out to be explicitly dependent on time . we find that , for the specific functional form of the decay rate that we work with , the overall amplitude of the power spectrum is enhanced ( by a factor of about @xmath123 ) during the transition . we have also been able to explicitly illustrate that it is the growth of the isocurvature perturbations close to the transition is responsible for the amplification of the curvature perturbations . it may be considered an overkill to integrate the second order differential equations rather than work with the much simpler first order equations @xcite . our motivations were threefold . importantly , our effort allowed us to cross - check certain previous results @xcite . also , our aim is to later study the effects of deviations from slow roll inflation as well as the effects of the modified dynamics close to the transition to the radiation dominated epoch on the small scale perturbations . as we had mentioned in the introduction , these effects can have interesting implications on the number of primordial black holes that are formed towards the end of inflation @xcite . moreover , in possibilities such as the warm inflationary scenario , the decay rate , in addition to depending on the inflaton , can also depend on the temperature ( see , for example , refs . in such a case , the transfer of the energy from the inflaton to radiation may affect the amplitude and possibly the spectral index as well @xcite . we would like to conclude by commenting on our choices for @xmath124 and @xmath106 . as we had pointed our earlier , our choices were motivated by convenience in numerical evolution rather than physics . obviously , better motivated functional forms of these quantities need to be investigated . in particular , it will be interesting to understand whether there exist @xmath106 that suppress the amplitude of the curvature perturbations during the transition rather than enhance them . such a possibility is worth exploring as it can lead to an increase in the tensor - to - scalar ratio post inflation ( see ref . @xcite ; however , see also ref . we are currently investigating these issues . in this appendix , we shall illustrate the corresponding effects for the case of the canonical scalar field model that was considered recently in the literature @xcite . we shall rapidly summarize the essential equations describing the background evolution and the perturbations , and present the results . in the case of a canonical scalar field , say , @xmath29 , that is interacting with a perfect fluid , if one assumes that @xmath125 , then the equation that describes the conservation of the energy density of the fluid is given by _ _ f + 3 h (1 + w__f ) _ _ f = ^2 . then , eq . implies that @xmath126 and , hence , the continuity equation governing the energy density @xmath127 of the scalar field reduces to _ _ + 3h ( _ _ + p _ _ ) = - ^2,[eq : cephi ] where @xmath128 denotes the pressure of the field . the energy density and pressure associated with the canonical scalar field are given by _ _ = ( ) + v ( ) p _ _ = ( ) - v ( ) . upon using these expressions in the continuity equation , we arrive at the following equation of motion governing the scalar field @xcite : + 3 h+ + v_=0,[eq : emcsf ] where @xmath129 . upon using the expressions for the perturbed energy density , momentum flux and pressure of the scalar field in eq . , we can arrive at the equation of motion for the perturbation in the scalar field , say , @xmath130 , exactly as we did for the perturbation in the tachyon . in the ucg , we find that @xmath130 satisfies the differential equation + (3h + ) + [v _ _ - ()^2 ] + - a + (2 v__+ ) a - ( ) ( ^2 b ) + = 0,[eq : deltaphi ] where @xmath131 and , as before , @xmath72 denotes the first order perturbation in the decay rate . again , on using eq . , we find that the equation describing the conservation of the perturbed energy density of the fluid @xmath73 can be written as + 3h(1 + w__f ) _ _ f + ()^2 _ _ f + - 2 + ^2a - [ ] (^2 b ) -^2 = 0 . also , from the first order einstein s equations and , we can arrive at the following differential equation for the quantity @xmath60 that describes the spatial velocity of the fluid : _ _ f+3 h _ _ f + w__f _ _ f + (1+w__f ) _ _ f a + = 0 . the perturbation variables that we shall evolve numerically are @xmath132 and the quantity @xmath75 that we had introduced earlier [ cf . eq . ] . as in the case of the tachyon and the perfect fluid , using the background equations , the first order einstein s equations for the system and the energy conservation equations for the scalar field and the fluid , we can arrive at the coupled second order differential equations for @xmath133 and @xmath75 , with @xmath134 @xcite . for the case wherein the decay rate @xmath3 is a constant , the coefficients @xmath77 and @xmath78 are now given by _ _ = (3h + ) , _ _ f = ( ) ( ) , _ _ f = ( ) (1-w__f ) + (1 + 2w__f ) ( ) , _ _ ff = 3h + ( ) , _ _ = v _ _ -()^2 + ( ) [4v _ _ + 2 (3h+ ) ^2 - ( ) ] + -()^2 ^ 2 [2 ^ 2 + ()__f ] , _ _ f = ( ) [- ( ) ( ) . ( ) ( ) ^3 - 2 v _ _ - ] + + ()^2 [2 ^ 2 + ()__f ] , _ _ f = - ( ) [v _ _ (1+w__f)^2__f + 3h v _ _ (1+w__f)^2 _ _ f - ^3 ] ( ) + + ()^2 (^2 + 2__f ) ( ) + - ( ) (v__-3 h w__f + ) , _ _ ff = - ( ) ^2 + ( ) (1-w__f^2 ) + + ( ) [()(1+w__f ) (1 + 3w__f ) _ _ f + ( ) ^2 (w__f-1 ) - ^2 ] + - ()^2 [(1+w__f)^2 _ _ f ^2 + 2 ((1+w__f ) _ _ f)^2 ] + - ( ) [v__+ - ()w__f + ( ) ] . for the case wherein the decay rate is a function of the scalar field , the coefficients @xmath135 and @xmath136 are modified to _ _ & _ _ + _ _ , + g__f & _ _ f + ( ) ^2 _ _ , where @xmath137 . in a similar fashion , when the decay rate is an explicit function of time , we find that the above two coefficients are modified as follows : _ _ f & _ _ f + ( ) , + g__ff & _ _ ff + ( ) . the other coefficients remain unchanged . in , we have illustrated the transition from inflation to the radiation dominated epoch by plotting the dimensionless energy density parameters @xmath138 and @xmath139 for the three possible types of decay . we have also plotted the evolution of the first slow roll parameter @xmath140 and the equation of state parameter @xmath141 for the entire system . note that we have considered the quadratic potential @xmath142 , and have chosen suitable values for the parameters and initial conditions to achieve the desired behavior . -100 true pt -220 true pt @xmath143 210 true pt @xmath48 105 pt -100 true pt -220 true pt @xmath143 210 true pt @xmath48 105 pt -100 true pt -220 true pt @xmath143 210 true pt @xmath48 85 pt 25 true pt @xmath46 215 true pt @xmath46 and , in , as in the case of the tachyon discussed in the text , we have plotted evolution of the amplitude of the individual and the total curvature perturbations , as well as the scalar power spectrum before and after the transition . -105 true pt -240 true pt @xmath144 205 true pt @xmath104 110 pt -105 true pt -240 true pt @xmath144 205 true pt @xmath104 110 pt -105 true pt -240 true pt @xmath144 205 true pt @xmath104 90 pt 25 true pt @xmath46 215 true pt @xmath105 it is evident from these two figures that the conclusions we had arrived at for the tachyon apply equally well to the case involving the canonical scalar field as well . 99 e. w. kolb and m. s. turner , _ the early universe _ ( addison - wesley , redwood city , california , 1990 ) ; a. r. liddle and d. h. lyth , _ cosmological inflation and large - scale structure _ ( cambridge university press , cambridge , england , 1999 ) ; v. f. mukhanov , _ physical foundations of cosmology _ ( cambridge university press , cambridge , england , 2005 ) ; r. durrer , _ the cosmic microwave background _ ( cambridge university press , cambridge , england , 2008 ) . h. kodama and m. sasaki , prog . * 78 * , 1 ( 1984 ) ; v. f. mukhanov , h. a. feldman and r. h. brandenberger , phys . rep . * 215 * , 203 ( 1992 ) ; j. e. lidsey , a. liddle , e. w. kolb , e. j. copeland , t. barreiro and m. abney , rev . phys . * 69 * , 373 ( 1997 ) ; b. bassett , s. tsujikawa and d. wands , rev . mod . phys . * 78 * , 537 ( 2006 ) ; k. a. malik and d. wands , arxiv:0809.4944 [ astro - ph ] . s. m. leach and a. r. liddle , phys . d * 63 * , 043508 ( 2001 ) ; s. m. leach , m. sasaki , d. wands and a. r. liddle , phys . d * 64 * , 023512 ( 2001 ) . r. k. jain , p. chingangbam and l. sriramkumar , jcap * 0710 * , 003 ( 2007 ) . e. bugaev and p. klimai , phys . d * 78 * , 063515 ( 2008 ) . h. m. hodges , g. r. blumenthal , l. a. kofman and j. r. primack , nucl . b * 335 * , 197 ( 1990 ) ; v. f. mukhanov and m. i. zelnikov , phys . lett . b * 263 * , 169 ( 1991 ) ; a. a. starobinsky , sov . jetp lett . * 55 * , 489 ( 1992 ) ; j. a. adams , g. g. ross and s. sarkar , nucl . b * 503 * , 405 ( 1997 ) ; j. lesgourgues , nucl . b * 582 * , 593 ( 2000 ) ; j. barriga , e. gaztanaga , m. santos and s. sarkar , mon . not . roy.astron . soc . * 324 * , 977 ( 2001 ) ; nucl . * 95 * , 66 ( 2001 ) ; j. a. adams , b. cresswell , r. easther , phys . d * 64 * , 123514 ( 2001 ) ; b. feng and x. zhang , phys . lett . b * 570 * , 145 ( 2003 ) ; c. r. contaldi , m. peloso , l. kofman and a. linde , jcap * 0307 * , 002 ( 2003 ) ; j. m. cline , p. crotty and j. lesgourgues , jcap * 0309 * , 010 ( 2003 ) ; p. hunt and s. sarkar , phys . d * 70 * , 103518 ( 2004 ) ; _ ibid . _ * 76 * , 123504 ( 2007 ) ; j .- o . gong , jcap * 0507 * , 015 ( 2005 ) ; r. sinha and t. souradeep , phys . d * 74 * , 043518 ( 2006 ) ; l. covi , j. hamann , a. melchiorri , a. slosar and i. sorbera , phys . d * 74 * , 083509 ( 2006 ) ; j. hamann , l. covi , a. melchiorri and a. slosar , phys . d * 76 * , 023503 ( 2007 ) ; d. boyanovsky , h. j. de vega and n. g. sanchez , phys . d * 74 * , 123006 ( 2006 ) ; _ ibid _ * 74 * , 123007 ( 2006 ) ; b. a. powell and w. h. kinney , phys . d * 76 * , 063512 ( 2007 ) ; g. nicholson and c. r. contaldi , jcap * 0801 * , 002 ( 2008 ) ; m. joy , v. sahni , a. a. starobinsky , phys . rev . d * 77 * , 023514 ( 2008 ) ; m. joy , a. shafieloo , v. sahni , a. a. starobinsky , arxiv:0807.3334 [ astro - ph ] ; c. destri , h. j. de vega and n. g. sanchez , phys . d * 78 * , 023013 ( 2008 ) ; r. k. jain , p. chingangbam , j .- o . gong , l. sriramkumar and t. souradeep , jcap * 0901 * , 009 ( 2009 ) . i. zaballa , a. m. green , k. a. malik and m. sasaki , jcap * 0703 * , 010 ( 2007 ) ; s. chongchitnan and g. efstathiou , jcap * 0701 * , 011 ( 2007 ) ; k. kohri , d. h. lyth and a. melchiorri , jcap * 0804 * , 038 ( 2008 ) ; r. saito , j. yokoyama and r. nagata , jcap * 0806 * , 024 ( 2008 ) . j. h. traschen and r. brandenberger , phys . d * 42 * , 2491 ( 1990 ) ; a. d. dolgov and d. p. kirilova , sov . j. nucl . phys . * 51 * , 172 ( 1990 ) ; l. kofman , a. d. linde and a. a. starobinsky , phys . lett . * 73 * , 3195 ( 1994 ) ; y. shtanov , j. h. traschen and r. h. brandenberger , phys . d * 51 * , 5438 ( 1995 ) ; l. kofman , a. d. linde and a. a. starobinsky , phys . d * 56 * , 3258 ( 1997 ) ; b. a. bassett and f. tamburini , phys . lett . * 81 * , 2630 ( 1998 ) ; b. a. bassett , c. gordon , r. maartens and d. i. kaiser , phys . d * 61 * , 061302(r ) ( 2000 ) ; b. a. bassett and f. viniegra , _ ibid . _ * 62 * , 043507 ( 2000 ) ; b. a. bassett and s. tsujikawa , _ ibid . _ * 63 * , 123503 ( 2001 ) ; p. ivanov , phys . rev . d * 61 * , 023505 ( 1999 ) ; f. finelli and r. brandenberger , phys . 82 * , 1362 ( 1999 ) ; phys . d * 62 * , 083502 ( 2000 ) ; r. easther and m. parry , phys . d * 62 * , 103503 ( 2000 ) ; s. tsujikawa and b. a. bassett , phys . b * 536 * , 9 ( 2002 ) . t. suyama and s. yokoyama , class . * 24 * , 1615 ( 2007 ) ; p. r. anderson , c. molina - paris , d. evanicha and g. b. cook , arxiv:0801.0730 [ hep - ph ] ; d. battefeld and s. kawai , phys . d * 77 * , 123507 ( 2008 ) ; d. battefeld , arxiv:0809.3455 [ astro - ph ] . d. h. lyth and d. wands , phys . b * 524 * , 5 ( 2002 ) ; d. h. lyth , c. ungarelli and d. wands , phys . d * 67 * , 023503 ( 2003 ) . s. matarrese and a. riotto , jcap * 0308 * , 007 ( 2003 ) . a. mazumdar and m. postma , phys . lett . b * 573 * , 5 ( 2003 ) . s. tsujikawa , phys . d * 68 * , 083510 ( 2003 ) ; g. dvali , a. gruzinov and m. zaldarriaga , phys . d * 69 * , 023505 ( 2004 ) . k. dimopoulos and d. h. lyth , phys . d * 69 * , 123509 ( 2004 ) . a. albrecht , p. j. steinhardt , m. s. turner and f. wilczek , phys . * 48 * , 1437 ( 1982 ) ; m. s. turner , phys . d * 28 * , 1243 ( 1983 ) ; e. w. kolb , a. notari and a. riotto , phys . rev . d * 68 * , 123505 ( 2003 ) . f. di marco , f. finelli and a. gruppuso , phys . d * 76 * , 043530 ( 2007 ) . a. cerioni , f. di marco , f. finelli and a. gruppuso , phys . d * 78 * , 021301(r ) ( 2008 ) . a. sen , jhep * 9910 * , 008 ( 1999 ) ; m. garousi , nucl . b * 584 * 284 , ( 2000 ) ; e. bergshoeff , m. de roo , t. de wit , e. eyras and s. panda , jhep * 0005 * , 009 ( 2000 ) ; j. kluson , phys . d * 62 * , 126003 ( 2000 ) . a. sen , jhep * 0204 * , 048 ( 2002 ) ; jhep * 0207 * , 065 ( 2002 ) . m. sami , p. chingangbam and t. qureshi , phys . d * 66 * , 043530 ( 2002 ) . d. a. steer and f. vernizzi , phys . d * 70 * , 043527 ( 2004 ) . p. chingangbam , s. panda and a. deshamukhya , jhep * 0502 * , 052 ( 2005 ) . c. campuzano , s. del campo and r. herrera , phys . rev . d * 72 * , 083515 ( 2005 ) ; phys . b * 633 * , 149 ( 2006 ) ; jcap * 0606 * 017 , ( 2006 ) . v. h. cardenas , phys . d * 73 * , 103512 ( 2006 ) . r. herrera , s. del campo and c. campuzano , jcap * 0610 * , 009 ( 2006 ) ; s. del campo , r. herrera and j. saavedra , arxiv:0812.1081 [ gr - qc ] . m. bastero - gil and a. berera , arxiv:0902.0521 [ hep - ph ] . k. a. malik , d. wands and c. ungarelli , phys . d * 67 * , 063516 ( 2003 ) . k. a. malik , arxiv : astro - ph/0101563 . k. a. malik and d. wands , jcap * 0502 * , 007 ( 2005 ) . c. gordon , d. wands , b. a. bassett and r. maartens , phys . d * 63 * , 023506 ( 2001 ) . n. bartolo , p. s. corasaniti , a. r. liddle and m. malquarti , phys . d * 70 * , 043532 ( 2004 ) . a. deshamukhya and s. panda , arxiv:0901.0471 [ hep - th ] . t. multamaki , j. sainio and i. vilja , arxiv : astro - ph/0609019 . a. linde , v. mukhanov and m. sasaki , jcap * 0510 * , 002 ( 2005 ) .

the separation of the milky way disk into a thin and thick component is supported by differences in the spatial , kinematic and metallicity distributions of their stars . these differences have lead to the predominant view that the thick disk formed early via a cataclysmic event and constitutes fossil evidence of the hierarchical growth of the milky way . we show here , using-body simulations , how a double - exponential vertical structure , with stellar populations displaying similar dichotomies can arise purely through internal evolution . in this picture , stars migrate radially , while retaining nearly circular orbits , as described by sellwood & binney ( 2002 ) . as stars move outwards their vertical motions carry them to larger heights above the mid - plane , populating a thickened component . such stars found at the present time in the solar neighborhood formed early in the disk s history at smaller radii where stars are more metal - poor and-enhanced , leading to exactly the properties observed for thick disk stars . classifying stars as members of the thin or thick disk by either velocity or metallicity leads to an apparent separation in the other property as observed . this scenario is supported by the sdss observation that stars in the transition region do not show any correlation between rotational velocity and metallicity . although such a correlation is present in young stars because of epicyclic motions , the radial migration mixes stars , washing out the correlation . using the geneva copenhagen survey , we indeed find a velocity - metallicity correlation in the younger stars and none in the older stars . we predict a similar result when separating stars by [/fe ] . the good qualitative agreement between our simulation and observations in the milky way are especially remarkable because the simulation was not tuned to reproduce the milky way , hinting that the thick disk may be a ubiquitous galaxy feature generated by stellar migration . nonetheless , we can not exclude that some fraction of the thick disk is a fossil of a past more violent history , nor can this scenario explain thick disks in all galaxies , most strikingly those which counter - rotate with respect to the thin disk .

1009.5997

in the years since @xcite first proposed a two component structure to the milky way disk , a large body of observational work has provided supporting evidence for contrasting thin and thick disk attributes . structurally , the thin disk scale height is shorter than thick disk scale height ( for reviews see * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) , and the thick disk may have a longer scale length than the thin disk @xcite . kinematically , thick disk stars have larger velocity dispersions and lag the net rotation of the disk @xcite . additionally , thick disk stars are older and metal poor relative to their thin disk counterparts ( _ e.g. _ * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and at a given iron abundance thick disk stars are @xmath1-enhanced @xcite . moreoever , thin and thick disk attributes are not unique to the milky way but a ubiquitous feature for late type galaxies @xcite . recently , several sdss based studies have provided further strong observational constraints on the structural , kinematic and chemical properties of stars in the solar cylinder . * hereafter j08 ) used a photometric parallax method on sdss data to estimate distances to @xmath248 million stars and studied their spatial distribution . because sdss provides accurate photometry , which enables reasonably robust distances ( 1015% , @xcite ) , as well as faint magnitude limits ( @xmath3 ) and a large sky coverage ( 6500 deg@xmath4 ) , j08 were able to robustly constrain the parameters of a model for the global spatial distribution of stars in the milky way . the j08 model is qualitatively similar to previous work ( _ e.g . _ @xcite ) which identifies a clear change of slope in the counts of disk stars as a function of distance from the galactic plane ; this change in slope is usually interpreted as the transition from the thin to thick disk @xcite . * hereafter i08 ) further extended this global analysis of sdss data by developing a photometric metallicity estimator and by utilizing a large proper motion catalog based on sdss and palomar observatory sky survey data @xcite . i08 studied the dependence of the metallicity , [ fe / h ] and rotational velocity , @xmath5 , of disk stars on the distance from the galactic plane and detected gradients of both quantities over the distance ranging from several hundred parsecs to several kiloparsecs . such gradients would be expected in a thin / thick disk decomposition where the thick disk stars are a separate population defined by a bulk rotational velocity lag and a lower metallicity compared to those of the thin disk . however , such a model would also predict a correlation between the metallicity and the velocity lag , which is strongly excluded ( @xmath2@xmath6 level ) by the i08 analysis ( see figure 17 , i08 ) . in this work we turn to a more sophisticated galactic description an @xmath0body model to characterize stars within the sdss volume and solve this puzzle . over the past few decades , @xmath0body simulations have been used to provide supporting evidence for three distinct theories of thick disk formation : violent relaxation @xcite , substructure disruption @xcite , and heating by satellites @xcite . several works have recently redressed these ideas . @xcite and @xcite formed a thick disk in situ at high redshift during gas - rich mergers , where star formation is triggered by the rapid accretion of gas ; this result is consistent with the thick disk forming through violent relaxation of the galactic potential . in contrast , @xcite , @xcite , and @xcite investigated substructure disruption by using a cosmologically derived satellite accretion history to perturb a milky way - like disk ; subhalo - disk encounters increased the scale height of this disk at all radii effectively forming a thick disk . finally , @xcite showed that by tidally stripping / accreting satellites , the majority of the oldest stars in the thick disk could have formed externally rather than in situ . in this work , we study a new method of formation : radial migration . radial migration due to scattering from transient spirals was first described by @xcite . in this model energy and angular momentum changes occur from interactions with transient spiral arms , which move stars at the corotation resonance inward or outward in radius while preserving their nearly - circular orbits . * ; * ? ? ? * r08ab hereafter ) studied this phenomenon in n - body + smooth particle hydrodynamic ( sph ) simulations of disk formation , and showed that migrations are possible on short timescales . they explored the implications of radial mixing for stellar populations for a variety of stellar systems , including the solar neighborhood . here we extend their work by highlighting the vertical evolution that occurs as a result of migration . we note that in this paper , we are not testing the validity of the other models of formation . however , recently , @xcite proposed using the eccentricity of orbits of stars in the thick disk to constrain the thick disk s formation mechanism ; they presented the eccentricity distributions that result from four n - body simulations : @xcite , @xcite , r08b , and @xcite . they found that the distributions that result from heating , radial migration and mergers all had a strong peak at low eccentricity ( @xmath7 ) , while the distribution that results from accretion is centered at higher orbital eccentries ( @xmath8@xmath9@xmath10 ) . building on this , @xcite studied the eccentricity of orbits of stars in the thick disk observed in the radial velocity experiment ( rave ) @xcite and found these results to be inconsistent with expectations for the pure accretion simulation . @xcite also leveraged @xmath1 measurements from rave to conclude that the @xmath1 enhancement of the metal - poor thick disk implies that direct accretion of stars from dwarf galaxies did not play a major role in the formation of the thick disk . using sdss dr7 , @xcite showed that the eccentricity of orbits of stars in the thick disk implies the thick disk is unlikely to be fully populated by radially migrated stars . we note that we can not exclude that some fraction of the thick disk is a fossil of a past more violent history , nor can this scenario explain thick disks in all galaxies . however , in what follows , we show that a large fraction of the stars in the thick disk could have formed in situ and arrived at their present location via radial migration . the outline of this paper is as follows : in [ s : simul ] , we present two simulations , one with substantial migration and the other with relatively little migration . when we compare these two simulations we can show that migration can build a thick disk as first conceived by gilmore & reid ( 1983 ) : a component with a scale - height larger than that of the thin disk . in [ s : obs ] we qualitatively compare the milky way - like simulation ( with migration ) with the sdss observations to show that they match each other sufficiently well to pursue further comparison . in [ s : solar ] we present a detailed comparison between the simulation and the local sdss volume focusing on the reason for the lack of correlation between @xmath5 and [ fe / h ] ; in appendix [ s : spagna ] we reconsider recent observational claims concerning the lack of correlation between @xmath5 and [ fe / h ] . in [ s : obs_decomp ] we use the simulation as a proxy for the milky way to show that classifying stars as members of the thin or thick disk by either velocity or metallicity leads to an apparent separation in the other property as observed . in [ s : schoenrich ] we compare our results to recent theoretical work that used semi - analytics to investigate how the solar neighborhood could have been shaped by radial migration and chemical evolution effects . in [ s : predict ] we explore the correlation between [ @xmath1/fe ] and age to show the diagnostic power of [ @xmath1/fe ] as a stand - in for age . finally , in [ s : concl ] we summarize our results and conclusions . we analyze the results of an @xmath0body + sph simulation designed to mimic the quiescent formation and evolution of a milky way mass galaxy following the last major merger . the system is initialized as in @xcite and r08ab and consists of a rotating , pressure supported gas halo embedded in an nfw @xcite dark matter halo . this simulation was evolved for 10 gyr using the parallel @xmath0body@xmath11sph code , gasoline @xcite . as the simulation proceeds , the gas cools and collapses to the center of the halo , forming a thin disk from the inside out . gas is continually infalling from the hot halo onto the disk for the duration of the simulation . star formation and stellar feedback are modeled with subgrid recipes as described in @xcite . importantly , the stellar feedback prescriptions include sn ii , sn ia and agb metal production , as well as injection of supernova energy which impacts the thermodynamic properties of the disk interstellar medium ( ism ) . metal diffusion is calculated from a subgrid model of eddy turbulence based on the local smoothing length and velocity gradients @xcite . the simulation we utilize is nearly identical to r08ab ( see r08ab for further details ) , but with the addition of metal diffusion @xcite . no _ a priori _ assumptions about the disk s structure are made its growth and the subsequent evolution of its stellar populations are completely spontaneous and governed only by hydrodynamics / stellar feedback and gravity . although we do not account for the full cosmological context , merging in the @xmath12cdm paradigm is a higher order effect at the epochs in question @xcite . thus , our model galaxy lacks some structural components such as a stellar halo , which in @xmath12cdm is built up primarily during the merging process ( _ e.g . * ; * ? ? ? our focus here , however , is disk evolution ; by simplifying our assumptions , we are able to use much higher resolution and more easily study the impact of key dynamical effects on observational properties of stellar populations within the disk . based on such simulations , r08ab presented the implications of stellar radial migration resulting from the interactions of stars with transient spiral arms @xcite on the observable properties of disk stellar populations . radial migration efficiently mixes stars throughout the disk into the solar neighborhood , resulting in a flattened age - metallicity relation ( r08ab ) . figures [ f : rform ] , [ f : rform_cont ] , & [ f : rform_cont_comp ] illustrate the basic premise of this paper stars migrate radially and in the process rise out of the plane over time , so many stars are presently not near their birth place . previous studies have shown @xcite that the vertical evolution that results from radial migration can influence the characterization of the thick disk . in order to further illustrate the importance of radial migration within our adopted milky way ( mw ) simulation , we have repeated much of our analysis on a control case . the control simulation is a system with the same initial conditions as the mw simulation except for having a higher angular momentum content with a dimensionless spin parameter @xmath13 @xcite . this results in a more extended disk ( final disk scale - length @xmath14 @xmath15 , versus @xmath16 @xmath15 for the mw simulation ) , possibly similar to a low surface brightness galaxy ; we therefore refer to this simulation as the lsb simulation . due to its lower surface density , the disk forms weaker spirals and as a result the stellar populations at all radii are less affected by radial mixing . when we compare migration as a function of scale - lengths , we find that there is significantly less migration in the lsb galaxy than in the mw galaxy . .1 in .1 in .1 in .1 in .1 in .1 in the distribution of stellar mass away from the midplane is strongly affected by radial migration ; this can be seen in figure [ f : density_comp ] , which contrasts the mw simulation against the lsb case . here the normalized mass density distribution within four analogous cylindrical volumes drawn from a variety of radii are presented . at larger radii , the steepness of the profiles are quite different ; the lsb simulation has a constant slope while the mw simulation shows a transition from a steep to a shallow density distribution . thus the mw simulation can not be characterized by a single exponential or sech@xmath4 component in the vertical direction , as we show explicity in the following section and figure [ f : density_profiles ] . it is this double - component nature which first led to the identification of the thick disk @xcite ; we have thus shown that this feature need represent nothing more than internal evolution of the milky way . in the following section , we compare sdss observations with the mw simulation to demonstrate its usefulness as a model for understanding the milky way thick disk . here we study the stellar mass distribution , rotational velocity and metallicity as functions of distance from the galactic plane , @xmath17 , and galactocentric cylindrical radius , @xmath18 . we draw qualitative comparisons between the datasets by examining their mass weighted metallicity and kinematic distributions in this @xmath18@xmath17 space . .best fit parameters to radial bins sampled in figure [ f : density_profiles ] given by : thin disk normalization ( @xmath19 ) , thin disk scale height ( @xmath20 ) , thick disk normalization ( @xmath21 ) , & thick disk scale height ( @xmath22 ) . [ cols="<,^,^,^,>",options="header " , ] the observed milky way disk is best fit by a 2-component model that is exponential both in the @xmath18 and @xmath23 directions ( see table 10 , bias - corrected results , j08 ) . the top panel of figure [ f : color_maps ] shows the mass weighted density distribution of the entire mw simulation at its final timestep . this distribution is in qualitative agreement in both the @xmath18 and @xmath23 directions with j08 for up to @xmath2@xmath24 @xmath15 above the disk s plane and @xmath2@xmath25 @xmath15 from the galactic center . figure [ f : density_profiles ] shows our best fits to the vertical density profiles for the mw simulation for radial bins @xmath26 @xmath15 , @xmath27 @xmath15 , @xmath28 @xmath15 and @xmath29 @xmath15 . as in j08 , the innermost radial bin is best fit by a double exponential function ; however , all other bins are better fit by the sum of two double sech@xmath4 profiles , which is in agreement with the theoretical work by @xcite and observational results for other galaxies by @xcite . despite the small discrepancy between sech@xmath4 vs. exponential fits , the simulation is in good qualitative agreement with the sdss - based analysis of the milky way ( j08 ) . in addition , figure [ f : density_profiles ] shows the vertical density profiles as divided into two populations : radial migrators ( @xmath30 ) and an in situ population ( @xmath31 ) . we note the second component fit for radial bins @xmath28 @xmath15 and @xmath29 @xmath15 is entirely dominated by star particles that have migrated . we expect that stars that moved from elsewhere combined with stars that were born locally should not naturally conspire to produce a single continuous profile . for the solar cylinder , @xmath32 , we find that the model distribution of stars as a function of @xmath17 resembles a double sech@xmath4 profile , with the `` transition '' height of @xmath33@xmath34 , comparable to @xmath2@xmath35 found by j08 for the galactic disk . we have found the best - fit scale heights to be 381 pc and 913 pc , in qualitative agreement with the best - fit scale heights of 270 pc and 1200 pc for the sdss data ( j08 ) . the scale height ratio suggested by the simulation is slightly low @xmath2@xmath36 , instead of @xmath2@xmath37 from the data . moreover , the simulation s thick disk to thin disk normalization is slightly discrepant @xmath2@xmath38 , rather than @xmath2@xmath39 suggested by the sdss data . overall , we find that the thick disk in the mw simulation , formed through the process of radial migration , is qualitatively very similar to the observed galactic thick disk . as figure [ f : density_comp ] clearly attests , this conclusion does not apply to the lsb simulation . the median metallicity of the milky way disk exhibits a clear vertical gradient ( see figure 9 , bottom two panels , i08 ) . notably , the spatial variation of the median metallicity does not follow the distribution of the stellar number density ( i08 ) . the middle panel of figure [ f : color_maps ] shows that the mw simulation reproduces a qualitatively similar metallicity distribution ; we note that a constant additive offset of 0.2 dex has been applied to [ fe / h ] throughout the simulation so that the median metallicity in the plane of the disk in the solar cylinder matches observations . as expected , at low galactic latitudes and small radii , the volume is dominated by high ( near solar ) metallicities . at higher latitudes the volume is increasingly metal poor . as with metallicity , previous studies have found a gradient in the median @xmath5 with respect to @xmath23 in the milky way ( see figures 5 , 8 , and 9 in bond et al . ( 2010 , hereafter b10 ) , and references therein ) . these authors concluded that @xmath5 is also well characterized by a non - gaussian distribution ( see * ? ? ? * for the implied distribution function ) . the bottom panel of figure [ f : color_maps ] shows that the mw simulation reproduces qualitatively similar @xmath5 properties , including a strong @xmath17 gradient . thus on a gross scale the mw simulation qualitatively matches patterns observed in mass density , metallicity and rotational velocity in the milky way disk . we now look at the solar cylinder in greater detail . here we study in detail the distributions of age , stellar mass , rotational velocity and metallicity as functions of @xmath17 for the solar cylinder . to be consistent with the analysis of high galactic latitude sdss data by j08 and i08 , we select model particles from an annulus with 7 @xmath15 @xmath40 9 @xmath15 . this radial cut spans 2.22.8 scale lengths from the center of model galaxy ( scale length @xmath41 @xmath15 ) , and covers the sun s location of @xmath23 disk scale lengths from the milky way center ( scale length @xmath2@xmath42 ) ( see bias - corrected value , table 10 , j08 ) . the behavior of the mw simulation in this particular volume is illustrated in figure [ f : ringberg ] . the top panel shows the age distribution as a function of @xmath17 . as expected , near the midplane , the population is dominated by young stars , and the mean age monotonically increases with increasing distance from the midplane . only very old stars are found at large @xmath17 : by @xmath43@xmath44 the mean age is already @xmath25 gyr . recall that figure [ f : rform_cont ] shows that with increasing distance from the midplane , the stellar population becomes dominated by older stars that formed closer to the center of the disk . these two figures taken together give a coherent picture of the net dynamic effect on the system : _ on average stars move both radially outward and away from the midplane over time . _ radial migration is able to change the extent of a star s vertical oscillation , as a well as its rotational velocity ( middle panel , figure [ f : ringberg ] ) . in the mw simulation we find a vertical gradient of rotational velocity of -17 @xmath45 ( compared to a gradient of -30 @xmath45 found by i08 ) . note that stars which significantly lag the rotation of the disk in the midplane have traditionally been regarded as members of the thick disk . another observed trend that is physically well motivated by the net outward and upward movement of stars over time is the decline of metallicity with increased height ( bottom panel , figure [ f : ringberg ] ) . in the mw simulation , the metallicity distribution changes with @xmath17 with a best - fit gradient of @xmath2@xmath46 @xmath47 , again in qualitative agreement with the measured value for the milky way of @xmath2@xmath48 @xmath47 ( i08 ) . to understand how this trend arises , recall that the top panel of figure [ f : ringberg ] shows that the stellar population away from the plane is dominated by old stars . as figure [ f : rform_cont ] shows , the oldest stars at large @xmath17 mostly formed in the inner @xmath49 @xmath15 . moreover , at early times , the radial metallicity gradient was steep ( see figure 2 in r08b ) ; stars that are now in the thick disk in the solar cylinder in the mw simulation were once at the outer edge of the forming thin disk and hence formed at a low metallicity . as a result , stars that formed at a radius of @xmath49 @xmath15 early on in the galactic history are necessarily metal poor . these stars have been subsequently moved out and up over time . at later times , the metallicity gradient flattened out as the disk grew . again , looking at figure [ f : rform ] ( top right panel ) , one can see a significant fraction of young stars ( _ i.e . _ less than 4 gyr old ) formed at r @xmath26 @xmath15 where there was a relatively more metal rich ism . these stars had less time to migrate and as a result remain closer to the midplane of the disk . this complex co - dependence of radial migration , birth location , and metallicity gradient evolution then gives rise to a vertical metallicity gradient in the solar neighborhood . we turn our attention now to a particular region within the solar cylinder : @xmath50 @xmath15 , @xmath28 @xmath15 . this region in the simulation is analogous to the thin thick disk `` transition zone '' considered by i08 : within this volume a roughly equal number of thin and thick disk stars is expected ( see bottom left panel , figure [ f : density_profiles ] ) . the transition region within the mw simulation occurs in approximately the same place as in the data ( 0.75 @xmath15 versus @xmath51 @xmath15 ) which allows us to draw a direct comparison to the slice analyzed in i08 . we show here that all the trends observed in the sdss data can be explained by a continuous distribution rather than two distinct populations . in figure [ f : age_plots ] we show the distribution of observable properties as a function of age within this slice of the simulation . the top left panel of figure [ f : age_plots ] shows the distribution of ages in this thin volume slice : the region is well populated ( @xmath230,000 star particles ) and predominantly old ( @xmath52 older than 7 gyr , @xmath53 older than 5 gyr ) . the remaining panels in figure [ f : age_plots ] illustrate probability densities of formation radius , metallicity , and rotational velocity versus age . in all cases , the distributions do not suggest distinct populations . for the given @xmath17 slice , stars older than about 4 gyr are both numerous and formed significantly interior to their present location ( top right panel ) . these stars also show a rotational velocity lag of @xmath2@xmath54 , as shown in the bottom left panel . the oldest stars , those with ages @xmath55 gyr , have significantly lower metallicities than younger stars ( bottom right panel ) . and as noted previously , the oldest stars formed at a range of interior radii , when the ism metallicity was low and had a steep radial gradient . these effects drive the behavior of various observables , as discussed next . we have shown in section [ s : grad ] that the mw simulation yields vertical trends in metallicity and rotational velocity similar to those found in sdss observations presented in i08 & b10 . sdss data also revealed a surprising lack of correlation between @xmath5 and [ fe / h ] , contrary to the expectations based on a traditional two - disk model ( see appendix [ s : spagna ] for a discussion of competing observational claims ) . in the top left panel of figure [ f : vphi_feh_decomp ] , we show @xmath5 vs. [ fe / h ] for the vertical slice considered above ( @xmath56 ) corresponding to the transition between the thick and thin disk in the simulation . the mw simulation also yields a lack of correlation between these quantities : although both rotational velocity and metallicity show vertical gradients , when stars are selected from a thin @xmath23 slice , velocity and metallicity are not correlated . we can understand why there is no correlation between @xmath5 and [ fe / h ] if we consider the thin slice as occupied by an age ensemble that was brought together by radial migration . in this light , we can decompose the sample by age to see how the behavior of each population is modified with time ( see top left panel of figure [ f : age_plots ] for the age sub - samples considered ) . the top right panel of figure [ f : vphi_feh_decomp ] shows a strong correlation between rotational velocity and metallicity exists for young stars ; the gradient is @xmath57 @xmath58 . the young stars constitute a small fraction of the overall population in the selected volume ( @xmath59 ) and , as the top right and bottom right panels of figure [ f : age_plots ] illustrate , young stars have correlated metallicity and formation radius . thus if a young star has a metallicity different than the surrounding ism , it must originate from somewhere outside the solar cylinder and be at or near perigalacticon ( if @xmath60 @xmath61 @xmath62 @xmath15 , where @xmath63 is the radius of the stellar guiding center ) or at or near apogalacticon ( if @xmath60 @xmath64 @xmath65 @xmath15 ) ; consequently , it will either lead or lag the local standard of rest ( lsr ) . for increasingly older age bins , the gradient diminishes and ultimately fades away . the bottom left and right panels of figure [ f : vphi_feh_decomp ] show @xmath5 vs. [ fe / h ] for intermediate ( 4 gyr @xmath64 age @xmath64 6 gyr ) and old ( 8 gyr @xmath64 age @xmath64 10 gyr ) populations respectively . the intermediate age stars constitute @xmath66 of the stellar population in the thin slice and retain a slight gradient @xmath67 @xmath58 . the oldest stars make up @xmath68 of the stellar population and show a nearly flat / slightly positive slope of @xmath69 @xmath58 . the peak in the distribution of intermediate and oldest age star particles is offset to progressively lower metallicity and rotational velocity . the metallicity of intermediate age stars ranges from @xmath70 [ fe / h ] @xmath71 while the metallicity of old stars range from @xmath72 [ fe / h ] @xmath73 . we can think of these three age bins as dominating different portions of the overall @xmath5 vs. [ fe / h]space , with the space spanned by intermediate and old stars being perpendicular to the peak in the correlation in young stars . thus , for the data cut into slices of [ fe / h ] , the distributions of @xmath5 are offset relative to one another . for the highest density contours , the mean value of the oldest stars is [ fe / h]@xmath74 , @xmath75 ; this is significantly lower than the mean value in the highest density contour of the full age sample [ fe / h]@xmath76 , @xmath77 . the full age sample s peak mean value matches the intermediate age values , which is in turn lower than the peak of the youngest stars [ fe / h]@xmath78 , @xmath79 . when the entire population is considered as a whole , the correlations disappears . why does the @xmath5-[fe / h ] gradient diminish with increasing age ? to understand this , we return to figure [ f : age_plots ] and recall that most of the stars within this thin cut are intermediate to old age , and they did not form within @xmath65 @xmath15 @xmath64 @xmath18 @xmath64 @xmath62 @xmath15 . a range of formation radii corresponds to a range of formation environments ; this maps to a range of resulting metallicities for stars of a given old age bin . at the same time , depending on a star s formation location and subsequent migration and scattering off the disk substructure , it can end up with a range of possible rotational velocities ( see bottom left panel , figure [ f : age_plots ] ) . hence , even a single old age bin samples a wide range of formation environments and rotational velocities resulting from unique dynamical histories , which are not directly correlated . we stress the significant difference between these results and the traditional double - disk interpretation : in the latter case a correlation between @xmath5 and [ fe / h ] is expected as one moves from younger , metal rich thin disk population to an older , metal poor , thick disk population . in contrast , here we find that a trend is present in the young stars but absent in the older stars . the trend in the young stars can be easily understood as arising from epicyclic motions of stars with their birth radii imprinted into their metallicities . as migration moves the guiding centers of stars , this metallicity encoding is erased , and the correlation between @xmath5 and [ fe / h ] disappears . as we demonstrated above , we can understand why there is no correlation between @xmath5 and [ fe / h ] when we decompose the sample by age . older populations have had more time to radially mix ; thus in the older age bins the expected trend disappears . that is , the evolution of the trends shown in figure [ f : vphi_feh_decomp ] is a unique signature of radial mixing taking place in the disk . we can verify whether this is the case in the solar neighborhood by utilizing observational data taken from the geneva - copenhagen survey ( gcs ) @xcite . the gcs samples a wide range of stellar ages and is reasonably well populated with old stars to make qualitative comparsions to the mw simulation in the plane of the disk . while what we have shown in figure [ f : vphi_feh_decomp ] is the relationship at 0.5 - 1 @xmath15 above the plane , these trends evolve in the same way as a function of age in the mid - plane . we have selected the non - binary stars from the gcs and repeated the analysis we presented above . figure [ f : gc ] shows the results : the top panel shows that there is no net trend when the sample is considered as a whole . splitting the sample into two broad age bins ( 0 - 4 gyr and 4 - 15 gyr ) yields a trend only in the young stars the correlation is completely absent for the older stars . the evolution of the observed @xmath5 vs. [ fe / h ] trend in the gcs sample therefore matches our expectations based on the simulation and strongly suggests that the stellar populations in the solar neighborhood have been influenced by radial mixing . we now turn our attention to assigning thin or thick disk membership in a manner analogous to observational studies , i.e. either based on kinematics or metallicity . we find that classifying stars as members of the thin or thick disk by either velocity or metallicity leads to an apparent separation in the other property , as observed . one of the observational differences between _ kinematically selected _ thin and thick disk stars is that the latter have higher abundance of @xmath1 elements at a given [ fe / h ] ( * ? ? ? * ; * ? ? ? * and references therein ) . this difference is often interpreted as evidence for different formation histories . our simulations include a calculation of the oxygen abundance following the prescription by @xcite ; given that snii mostly yield oxygen @xcite , we use oxygen as a proxy for all @xmath1 elements . because metal yields are not precisely known , the normative offset for [ @xmath1/fe ] within the mw simulation does not match the milky way . however , in terms of chemical evolution , the mw simulation gives a good qualitative perspective on distributions in [ @xmath1/fe ] space within the solar cylinder . the top left panel of figure [ f : toomre ] shows the overall dependence of [ @xmath1/fe ] on [ fe / h ] for @xmath28 @xmath15 , @xmath80 @xmath15 ( _ i.e . _ the `` solar neighborhood '' ) . a similar behavior is seen for the stars at @xmath81@xmath15 in agreement with @xcite . when the same sample is separated by age , distinct portions of the parameter space are covered . in particular , at low [ fe / h ] , old stars show an enhancement of [ @xmath1/fe ] relative to young stars . locally , observed thick disk stars are selected kinematically , rather than by age @xcite . we reproduce the qualitative behavior of observations by following similar steps with the mw simulation . the top right panel in figure [ f : toomre ] shows a toomre diagram for simulated stars , with selection cuts motived by @xcite . the bottom left panel shows that these kinematically - selected thin and thick disk stars from the solar neighborhood show similar bifurcation of [ @xmath1/fe ] vs. [ fe / h ] behavior as young and old subsamples shown in top left panel . in the bottom right panel , stars from just the shaded [ fe / h ] cut in the bottom left plot are examined . stars falling within the kinematically selected thick disk region have a higher fraction of old stars relative to the overall population . these old stars are @xmath1-enhanced as they formed in the interior of the disk and radially migrated to their present location . thus kinematically dividing the stars locally biases the sample to an older , @xmath1-enhanced population . although the simulation and the data are not in detailed quantitative agreement , these qualitative results imply that the differences in [ @xmath1/fe ] vs. [ fe / h ] for kinematically selected thin and thick disk stars may be another consequence of mixing effects that result from radial migration . although not as commonly done , it is equally plausible to take the converse approach ; we can assign membership to the thin and thick disk based on an [ @xmath1/fe ] cut and then study the kinematic and [ fe / h ] distributions that result @xcite . here we follow the technique outlined by @xcite , who chemically divided the galactic disks using sdss segue-1 data ; we select star particles within @xmath82 @xmath15 , @xmath83 @xmath15 and split them so that stars with [ @xmath1/fe ] @xmath84 are assigned thick disk membership . assigning membership based on [ @xmath1/fe ] effectively divides the disk into two populations : old stars and young intermediate age stars , as can be seen clearly in the top left panel of figure [ f : connie ] . why is this the case ? star particles are born @xmath1-enhanced if they form in a region with a high local sfr and little snia pollution . since most old stars originated near the center of the disk , and that region is where the sfr was high , these stars are naturally @xmath1-enhanced . we compare our results to three trends discussed in @xcite : the radial metallicity gradient , distribution in @xmath5 and distribution in [ fe / h ] , which we show clockwise from top right in figure [ f : connie ] . the top right panel illustrates the best fit radial metallicity gradients as derived from the mass weighted mean value of [ fe / h ] . cutting by [ @xmath1/fe ] results in no trend ( slope @xmath2@xmath85 ) in the thick disk and a negative trend in the thin disk ( slope @xmath2@xmath86 ) ; this is similar to the thick disk slope @xmath87 and thin disk slope @xmath88 observed by @xcite . the bottom right panel illustrates the cumulative @xmath5 distributions that result for the mw simulation when stars are separated by [ @xmath1/fe ] : the thin and thick disk trends are offset by @xmath2@xmath89 , which is qualitatively similar to , if quantitatively smaller than , the @xmath2@xmath90 offset found by @xcite . finally the bottom left panel shows the cumulative [ fe / h ] distributions ; we find the thin and thick disk trends offset by @xmath2@xmath91 dex , not dissimilar from the observed @xmath2@xmath92 dex offset @xcite . therefore , by adopting a kinematic or [ @xmath1/fe ] selection criteria used by observers , we are able to reproduce an apparent separation in the other property , despite the fact that there are no distinct populations in the model galaxy . recently , @xcite investigated how radial migration and chemical evolution shape the solar neighborhood , by incorporating for the first time a prescription for radial migration in a semi - analytic model of galactic chemical evolution . their model represented a disk in which star formation commenced at all radii simultaneously ( _ i.e . _ without inside - out growth ) , with radially varying star formation rates set to yield a disk with an appropriate scale length . the guiding center radii of the stars in their model changed ( i.e. stars migrated radially ) according to a parametrized probabilistic prescription whose normalization was left as a free parameter in the model . the vertical structure of their disk was determined based on the assumption that coeval stars comprise an isothermal population with a velocity dispersion given by local observational constraints . because stars migrating from the inner disk retain their velocity dispersions but encounter a lower restoring potential in the outer disk , they populate the disk away from the plane . hence , @xcite showed that a thickened component may result simply by ( the inevitable ) radial migration , and by fitting the model they also reproduced many of the canonical features of the thick disk ( _ i.e . _ enhanced [ @xmath1/fe ] ratios , older ages , lower metallicites , rotational lag ) , similar to what we have shown in the previous section . it is therefore reassuring that our results presented here agree qualitatively with those of @xcite , given that our modeling methods are entirely different , and that our model was not specifically tuned to the galaxy . however , we also find subtle yet crucial differences . figure 4 of @xcite shows that due to radial mixing , a population that shares the same kinematics can show immense variations in its chemical composition . the top left panel of figure [ f : rok_plot ] shows that our model also yields stellar populations with the same kinematics but very different chemistry , consistent with recent observational work @xcite . however , we note the contours of mean @xmath5 in this plane are almost orthogonal in our simulation compared to those shown in fig . 4 of @xcite , even though the age structure is very similar ; at fixed [ fe / h ] age increases monotonically with rising [ @xmath1/fe ] ( middle panel of figure [ f : rok_plot ] and figure 5 of @xcite ) . a hint of a reason for this discrepancy is provided by scrutinizing the metal - rich end of these figures . in our model , the [ @xmath1/fe]-deficient metal - rich population originated in the interior of the disk ( see the rightmost panel of figure [ f : rok_plot ] ) and has migrated to the present radius without very much heating . this is apparent from the fact that the mean @xmath5 for this population is only very slightly lagging the lsr ( @xmath2@xmath93 ) . on the other hand , in the model of @xcite , that same population shows considerable lag from the lsr . this discrepancy implies that there are qualitative differences in the treatment of radial mixing between their prescription and our simulation . on the other hand , the differences in the velocity structure in this plane must also be a result of their assumption that the entire disk begins forming stars at once with a peak in star formation occurring @xmath210 gyr ago everywhere . in our model , it is impossible to have an old metal poor star which formed at the solar radius - _ all _ of these stars must have migrated to their present position because the solar neighborhood in our simulation does not exist @xmath2@xmath62 gyr ago . @xcite argued that the existing solar neighborhood samples ( _ e.g . _ the gcs ) show signatures of radial mixing as proposed by @xcite . in our sample shown in figure [ f : rok_plot ] , the metal - poor , low [ @xmath1/fe ] stars have high velocities and young ages , and are at or near peri - galactic passage from the outer disk into the solar neighborhood sample . this is consistent with the arguments put forth by @xcite as observational evidence of radial mixing . however , we point out that the large velocities of this tail in the distribution signify that their orbits have merely been heated , and these stars have not migrated via the corotation scattering mechanism over any significant distance . instead , it is the presence of significantly metal - enriched stars on kinematically inconspicuous orbits ( _ i.e . _ kinematically cool ) that should be considered as clear evidence of migration . we have demonstrated that we can understand the lack of a correlation between @xmath5 and [ fe / h ] as a consequence of radial migration effects ; the significance of these effects became clear when we decomposed the mw simulation within the solar cylinder by age . unfortunately age is not easily accessible observationally . however , @xmath1 is a directly measurable quantity and , as we have shown , for the oldest stars , @xmath1-enhancement is a reasonable proxy for age ( see top left panel of figure [ f : connie ] ) . thus we now reassess the relationship between @xmath5 and [ fe / h ] by decomposing figure [ f : vphi_feh_decomp ] in cuts of [ @xmath1/fe ] so that we can make a testable prediction for upcoming observational surveys . figure [ f : local_alpha_cuts ] shows @xmath5 @xmath94[fe / h ] for the `` transition zone '' considered in section [ s : solar ] ( @xmath95 @xmath15 , @xmath28 @xmath15 ) , as split by two broad bins in [ @xmath1/fe ] . below each @xmath5 @xmath94[fe / h ] plot is a histogram of the ages represented in the given bin . in the right panel , the high [ @xmath1/fe ] sample contains almost exclusively old stars ( @xmath96 older than 7 gyr ) ; the corresponding @xmath5 @xmath94[fe / h ] figure shows no trend between the quantities ( best fit slope @xmath2@xmath97 ) . in contrast to this , the left panel , the low [ @xmath1/fe ] content bin has few old stars ( @xmath98 older than 7 gyr ) while sampling young to intermediate aged stars relatively equally ( @xmath99 between 0 and 2 gyr old , and @xmath100 between 4 and 6 gyr old ) . here there is a trend between @xmath5 and [ fe / h ] , with an overall best fit value of @xmath2@xmath101 . we note that this [ @xmath1/fe ] decompositon works equally well in the solar neighborhood as the `` transition zone '' . for the low [ @xmath1/fe ] cut , we have fit the mass weighted mean values with a single linear fit ; however it is equally plausible to fit two lines here : @xmath102 dex [ @xmath103/fe ] @xmath104 dex and the other @xmath105 dex [ @xmath103/fe ] @xmath106 dex . in that case , the linear fit to low [ fe / h ] shows no trend while the linear fit to high [ fe / h ] shows a strong trend . notably , the portion spanning low [ fe / h ] is dominated by the intermediate aged stars while the portion spanning high [ fe / h ] is dominated by a younger population . this `` knee '' is a persistent feature ; when the upper limit on this [ @xmath1/fe ] cut is lowered , fewer intermediate age stars are sampled , and the knee in the trend shifts to lower [ fe / h ] . note , the trend is evident here because on average the stars in this [ fe / h ] space have experienced less radial mixing than older stars within the same spatial volume . thus , it is possible to recover this signature of radial mixing even in the absence of age estimates , but with knowledge of [ @xmath1/fe ] , [ fe / h ] , and @xmath5 for an unbiased population of stars located out of the midplane . to date , @xmath1 measurements have only been accessible to small targeted samples ; our initial comparison with a compilation of all currently available data is particularly encouraging @xcite . we note current work , like that of the sdss segue collaboration , aims to obtain a large , well - calibrated [ @xmath1/fe ] dataset . we eagerly anticipate the application of cuts on [ @xmath1/fe ] to an unbiased population of stars that fall within the region considered by i08 or indeed any large sample within the solar cylinder ; such an analysis would further elucidate whether radial mixing has played an important role in shaping the distribution of the milky way stars over time . we have used an n - body model designed to mimic the quiescent formation and evolution of a milky way type galactic disk ( rokar et al . 2008ab ) to interpret recent sdss - based observational constraints on the structure of the milky way disk . traditional simplistic decomposition of thin and thick disks as two distinct populations of metallicity and rotational velocity predicts a strong correlation of these quantities at @xmath21 kpc from the galactic plane . thanks to nearly complete flux - limited sdss stellar samples with both rotational and metallicity measurements , the expected correlation was strongly ruled out ( ivezi et al . 2008 ) . on the other hand , the mw simulation we presented here , while not a detailed model of the milky way , produces good qualitative agreement with the data and sheds new light on the origin and evolution of the observed disk structure . of particular importance is the role of radial migration in mixing stars born throughout the disk into the solar neighborhood . the properties of the rokar et al . model for the overall spatial , metallicity and kinematic distributions of the milky way stars are in qualitative agreement with the sdss data ( juri et al . 2008 ; ivezi et al . 2008 ; bond et al . while there are quantitative differences in the spatial gradients of these distributions , as well as in their detailed behavior , even this qualitative agreement is remarkable because the simulation was not fine tuned to match the milky way . not only does the model reproduce the observed change of slope in the counts of disk stars as a function of distance from the galactic plane ( the original motivation for introducing a separate thick disk component ) , but it also predicts the gradients in metallicity and rotational velocity . furthermore , the metallicity and rotational velocity are uncorrelated in this model for appropriately volume - selected subsamples of stars , in agreement with observations . the robust qualitative agreements between the data and model predictions motivate the use of model quantities inaccessible to observations , such as the stellar age and the ism metallicity at the time and position of stellar birth , to interpret recent sdss results . in particular , the lack of correlation between the metallicity and rotational velocity at @xmath21 kpc from the galactic plane can be understood as due to complex interplay between the ism metallicity at the time and position of stellar birth , and the subsequent secular evolution largely driven by spiral arms . no _ a priori _ assumptions about the disk s structure are incorporated in the model yet it reproduces the main observational results which motivated decomposition of the disk into two presumably distinct components . the absence of mergers in this model implies that they are not required to explain the overall disk structure . while merger remnants are detected within the milky way disk ( e.g. j08 ) , their influence is apparently well localized . instead , the milky way disk can be viewed as a single complex structure with an age distribution that is a strong function of position in the galaxy due to radial migration of stars . thus even if a primordial thick disk is present , having formed via accretion / external heating , it is likely to be substantially polluted by migrating disk stars . the same mixing effects are likely responsible for the observed differences in @xmath1 element abundance between _ kinematically selected _ thin and thick disk stars . by adopting kinematic selection criteria used by observers , we are able to reproduce distinctive [ @xmath1/fe ] vs. [ fe / h ] trends similar to those seen in the data , despite the fact that there are no distinct populations in the model galaxy . we look forward to the improved data derived from the emerging generation of surveys such as segue @xcite and apogee @xcite . key to garnering a deeper understanding of the importance of radial migration in the milky way evolution is gathering both precise age determinations and detailed chemical compositions . we are optimistic that this study will lead to further observational and theoretical work . we thank our numerous sdss ( www.sdss.org ) collaborators for their valuable contributions and helpful discussions . s. loebman and . ivezi acknowledge support by nsf grants ast-0707901 and ast-1008784 to the university of washington , and by nsf grant ast-0551161 to lsst for design and development activity . ivezi acknowledges support by the croatian national science foundation grant o-1548 - 2009 . this research was supported in part by the nsf through teragrid resources provided by tacc and psc . r. rokar and t. r. quinn were supported by the nsf itr grant phy-0205413 at the university of washington . , m. g. , navarro , j. f. , steinmetz , m. , & eke , v. r. 2003 , , 597 , 21 , f. , et al . 1999 , , 118 , 261 , c. , barklem , p. s. , lambert , d. l. , & cunha , k. 2004 , vizier online data catalog , 342 , 183 , j. n. , & soneira , r. m. 1980 , , 238 , l17 , t. , feltzing , s. , & lundstrm , i. 2003 , , 410 , 527 , t. , feltzing , s. , lundstrm , i. , & ilyin , i. 2005 , , 433 , 185 , j. 2010 , , 401 , 2318 , j. j. , munn , j. a. , hawley , s. l. , west , a. a. , covey , k. r. , & schneider , d. p. 2007 , , 134 , 2418 , n. a. , et al . 2010 , , 716 , 1 , f. , elmegreen , b. g. , & martig , m. 2009 , , 707 , l1 , m. m. 2004 , ph.d . thesis , the university of north carolina at chapel hill , c. b. , gibson , b. k. , martel , h. , & kawata , d. 2005 , , 630 , 298 , c. b. , kawata , d. , gibson , b. k. , & freeman , k. c. 2004 , , 612 , 894 , j. s. , dekel , a. , kolatt , t. s. , kravtsov , a. v. , klypin , a. a. , porciani , c. , & primack , j. r. 2001 , , 555 , 240 , j. s. , & johnston , k. v. 2005 , , 635 , 931 , d. 1979 , , 234 , 829 , r. , rong , j. , & karaali , s. 1999 , , 348 , 98 , d. , et al . 2010 , , 712 , 692 , j. 2009 , dissertation bsc ( hons ) , department of physics , university of malta , b. , et al . 2001 , , 553 , 184 , m. , & beers , t. c. 2000 , , 119 , 2843 , m. , klement , r. j. , rix , h. , & liu , c. 2010 , arxiv e - prints , s. 2006 , memorie della societa astronomica italiana , 77 , 1103 , s. , bensby , t. , & lundstrm , i. 2003 , , 397 , l1 , k. 1998 , , 338 , 161 , g. , & reid , n. 1983 , , 202 , 1025 , g. , wyse , r. f. g. , & norris , j. e. 2002 , , 574 , l39 , m. 2008 , , 388 , 1175 , r. d. , woosley , s. e. , weaver , t. a. , rauscher , t. , & thielemann , f. 1999 , , 521 , 735 , j. , nordstrm , b. , & andersen , j. 2009 , , 501 , 941 , . , et al . 2008 , , 684 , 287 , b. j. t. , & wyse , r. f. g. 1983 , , 120 , 165 , m. , et al . 2008 , , 673 , 864 , t. , mayer , l. , wadsley , j. , stadel , j. , & moore , b. 2007 , , 375 , 53 , s. , bullock , j. s. , zentner , a. r. , kravtsov , a. v. , & moustakas , l. a. 2008 , , 688 , 254 , j. a. , & humphreys , r. m. 2003 , , 125 , 1958 , y. s. , et al . 2010 , in bulletin of the american astronomical society , vol . 41 , bulletin of the american astronomical society , 250 , s. , et al . 2008 , in american institute of physics conference series , vol . 1082 , american institute of physics conference series , ed . c. a. l. bailer - jones , 238 , s. r. 1993 , , 31 , 575 , s. r. , wilson , j. c. , hearty , f. , schiavon , r. r. , & skrutskie , m. f. 2010 , in iau symposium , vol . 265 , iau symposium , ed . k. cunha , m. spite , & b. barbuy , 480 , t. v. , soubiran , c. , kovtyukh , v. v. , & korotin , s. a. 2004 , , 418 , 551 , j. a. , et al . 2004 , , 127 , 3034 , j. f. , abadi , m. g. , venn , k. a. , & freeman , k. c. 2010 , arxiv e - prints , j. f. , frenk , c. s. , & white , s. d. m. 1997 , , 490 , 493 , m. j. , sackett , p. d. , de marchi , g. , & paresce , f. 2002 , , 383 , 472 , p. e. 1995 , in iau symposium , vol . 164 , stellar populations , ed . p. c. van der kruit & g. gilmore , 109 , p. e. , & schuster , w. j. 2009 , in iau symposium , vol . 254 , iau symposium , ed . j. andersen , j. bland - hawthorn , & b. nordstrm , 103 , j. e. 1999 , , 265 , 213 , d. k. 2001 , , 322 , 426 , j. e. , humphreys , r. m. , & beers , t. c. 2004 , , 127 , 1567 , j. x. , naumov , s. o. , carney , b. w. , mcwilliam , a. , & wolfe , a. m. 2000 , , 120 , 2513 , p. j. , hernquist , l. , & fullagar , d. p. 1993 , , 403 , 74 , c. m. , villata , m. , gallino , r. , busso , m. , & cravanzola , a. 1999 , , 518 , l91 , b. e. , lambert , d. l. , & allende prieto , c. 2006 , , 367 , 1329 , b. e. , tomkin , j. , lambert , d. l. , & allende prieto , c. 2003 , , 340 , 304 , a. c. , haywood , m. , creze , m. , ojha , d. k. , & bienayme , o. 1996 , , 305 , 125 , c. , beers , t. c. , majewski , s. , schiavon , r. , & eisenstein , d. 2009 , in astronomy , vol . 2010 , astro2010 : the astronomy and astrophysics decadal survey , 14 , r. , debattista , v. p. , quinn , t. r. , stinson , g. s. , & wadsley , j. 2008a , , 684 , l79 , r. , debattista , v. p. , stinson , g. s. , quinn , t. r. , kaufmann , t. , & wadsley , j. 2008b , , 675 , l65 , g. r. , et al . 2010 , arxiv e - prints , l. v. , et al . 2009 , , 400 , l61 , r. , & binney , j. 2009 , , 396 , 203 , j. a. , & binney , j. j. 2002 , , 336 , 785 , b. , ivezi , . , & juri , m. 2008 , , 689 , 1244 , s. , wadsley , j. , & stinson , g. 2009 , arxiv e - prints , 0910.5956 , m. h. , majewski , s. r. , reid , i. n. , & thompson , i. b. 2002 , , 578 , 151 smagorinsky , j. 1963 , monthly weather review , 91 , 99 , c. , bienaym , o. , & siebert , a. 2003 , , 398 , 141 , a. , lattanzi , m. g. , re fiorentin , p. , & smart , r. l. 2010 , , 510 , l4 , l. , jr . 1942 , , 95 , 329 , t. s. 1988 , , 331 , 71 , m. , et al . 2006 , , 132 , 1645 , g. , seth , a. , katz , n. , wadsley , j. , governato , f. , & quinn , t. 2006 , , 373 , 1074 , g. , edvardsson , b. , tuominen , i. , & ilyin , i. 2001 , , 380 , 578 , p. c. , & searle , l. 1981 , , 95 , 105 , . , & helmi , a. 2008 , , 391 , 1806 , . , kazantzidis , s. , & helmi , a. 2010 , , 718 , 314 , j. w. , stadel , j. , & quinn , t. 2004 , new astronomy , 9 , 137 , j. w. , veeravalli , g. , & couchman , h. m. p. 2008 , , 387 , 427 , m. , et al . 2010 , arxiv e - prints , r. f. g. , gilmore , g. , norris , j. e. , wilkinson , m. i. , kleyna , j. t. , koch , a. , evans , n. w. , & grebel , e. k. 2006 , , 639 , l13 , p. 2007 thesis , university of washington , p. , & dalcanton , j. j. 2005 , , 624 , 701 , p. , & dalcanton , j. j. 2006 , , 131 , 226 , a. , willman , b. , brooks , a. m. , governato , f. , brook , c. b. , hogg , d. w. , quinn , t. , & stinson , g. 2009 , , 702 , 1058 recently , @xcite found that the rotational velocity for disk stars is correlated with metallicity for @xmath107 @xmath8 [ fe / h ] @xmath8 @xmath108 at @xmath109 @xmath15 @xmath8 @xmath23 @xmath8 @xmath37 @xmath15 . notably , they found a gradient within @xmath110 @xmath111 @xmath58 , such that more metal - poor stars rotate more slowly . their claim is in direct conflict with several other observational studies ( * ? ? ? * i08 , b10 ) . we reconsider the @xcite findings using sdss dr7 data to try to understand the differences between these observational findings . we note a direct comparison between i08 and @xcite is non - trivial as @xcite use their own proper motion measurements based on the gsc - ii catalog and do not provide a comparison to the @xcite proper motions on a star - by - star basis . additionally , their color selection is more generous ( @xmath112 ) than previous studies ( @xmath113 , i08 ) . this more generous selection lets in bhb stars , as well as red stars where [ fe / h ] reliability decreases . despite differences in selection criteria , we can reproduce @xcite results when using the sdss _ spectroscopic _ sample ( see left panel , figure [ f : spagna ] ) . while the left panel of figure [ f : spagna ] only considers a narrow bin of @xmath114 @xmath15 , we find the best fit lines are reproducible for other bins as well . notably , the median @xmath5 as a function of [ fe / h ] closely follows the @xcite result . we also reproduce a bimodal distribution of stars in the metallicity direction , with modes at [ fe / h]@xmath115 and [ fe / h ] @xmath116 ( see section 3.2 , * ? ? ? * ) . however , we detect no @xmath5 [ fe / h ] correlation when we consider a complete sample selected in the meridional plane ( @xmath117 or @xmath118 ) , where proper motion alone suffices to measure rotational velocity ( see right panel , figure [ f : spagna ] ) . this sample is essentially complete in selected color - distance limits and thus not subject to strong selection effects present in the sdss spectroscopic sample . this sample shows negligible dependence of the median @xmath5 on [ fe / h ] for [ fe / h]@xmath119 ( one can also see the bias due to halo stars for [ fe / h]@xmath120 ) . it is highly likely that the @xcite results are caused by selection biases in the sdss spectroscopic sample . notably , when we fit two gaussians ( one for the disk and halo ) to the @xmath5 and [ fe / h ] distributions , about @xmath121 of the spectroscopic sample are halo stars ( consistent with @xcite within @xmath122 errors ) , while halo stars make up only @xmath123 of the complete sample . and when we compare the two panels in figure [ f : spagna ] we are using exactly the same volume and exactly the same measurements : the only difference between the samples is that the spectroscopic sample includes only @xmath124 of all the stars , with the selection probability about 10 times higher for halo stars than for disk stars .

discrete approaches to gravity , both classical and quantum , are reviewed briefly , with emphasis on the method using piecewise - linear spaces . models of 3-dimensional quantum gravity involving 6j - symbols are then described , and progress in generalising these models to four dimensions is discussed , as is the relationship of these models in both three and four dimensions to topological theories . finally , the repercussions of the generalisations are explored for the original formulation of discrete gravity using edge - length variables .

gr-qc0012035

the original motivation for the development of a discrete formalism for gravity @xcite arose from a number of problems with the continuum formulation of general relativity . these included the difficulty of solving einstein s equations for general systems without a large degree of symmetry , the problems of representing complicated topologies and the need for considerable geometric insight and capacity for visualisation . it turned out , as we shall see , that the discretisation scheme to be described not only helped with these problems but also found a vital r^ ole in numerical relativity and in attempts at a formulation of quantum gravity . the related branches of mathematics which found their application to physics in this formulation of gravity are those of piecewise - linear spaces and topology and the geometric notion of intrinsic curvature on polyhedra . the immediate aim was to develop an approach to general relativity which avoided the use of coordinates , since the physical predictions of the theory are coordinate - independent . the basic idea of the approach , which has subsequently become known as _ regge calculus _ , is as follows . rather than considering spaces ( or space - times ) with continuously varying curvature , we deal with spaces where the curvature is restricted to subspaces of codimension two . this is achieved by considering collections of n - dimensional blocks , which are glued together by identification of their flat ( n-1)-dimensional faces . the curvature lies on the ( n-2)-dimensional subspaces , known as _ hinges _ or _ bones_. for technical reasons , it is convenient to use blocks which are _ simplices _ ( triangles , tetrahedra and their higher dimensional analogues ) . consider first the realisation of these ideas in two dimensions . here we have examples in everyday life , geodesic domes ; these consist of networks of flat triangles which are fitted together to approximate curved surfaces , usually parts of a sphere . since two triangles with a common edge can be flattened out without distortion , there is no curvature on the edges . however , when a collection of triangles meeting at a vertex is flattened , there will be a gap , indicating the presence of curvature at the vertex . the amount of curvature there depends simply on the size of the gap or _ deficit angle_. it is relatively simple to visualise the generalisation of a triangulated surface to three dimensions , where a collection of flat tetrahedra are glued together on their flat triangular faces . in general , the tetrahedra at an edge will not fit together exactly in flat space , so there will be a deficit angle at that edge giving a measure of the curvature there . in four dimensions , the curvature is restricted to the triangles between the tetrahedra where the four - simplices meet . and so on in higher dimensions . thus we have a set of flat simplices glued together to approximate a curved space . there is another way of viewing the scheme that has just been described . piecewise - flat spaces are interesting in their own right , so in addition to using them as an approximation scheme for some curved reality " , we may also study such spaces for their own sake . it has been argued ( for example by friedberg and lee @xcite ) that space - time is actually discrete at the smallest scales , so one could also regard curved spaces as approximations to a discrete reality . ` a chacun ses go^ uts ! in order for the piecewise - flat spaces to be of any practical use in relativity , beyond ease of visualisation , it must be possible to calculate geometric quantities like curvature and volume , and in particular to evaluate the einstein action of such a space . in @xcite it was shown heuristically that the analogue of the einstein action @xmath0 is given by @xmath1 where @xmath2 is the measure of a hinge @xmath3 and @xmath4 is the deficit angle there , equal to @xmath5 minus the sum of the dihedral angles between the faces of the simplices meeting at that hinge . rigorous justification for this formula followed in @xcite , where it was shown that it converges to the continuum form of the action , in the sense of measures , provided that certain conditions on the fatness of the simplices are satisfied . friedberg and lee @xcite approached the problem from the opposite direction , deriving the regge action from a sequence of continuum spaces approaching a discrete one . the reason for choosing the building blocks to be simplices is that the geometry of a flat simplex is completely determined by the specification of its edge lengths , so a simplicial space may be described exactly by these lengths without the need for any further variables like angles . this means that the simplest choice of variables for the discrete theory is the edge lengths ; clearly the action may be calculated once they are specified and they are also the obvious analogues of the metric tensor , which serves as variable in the continuum theory . there , an elegant way of deriving einstein s equations is from the principle of stationary action , varying @xmath6 with respect to the metric . the analogue in regge calculus is to vary @xmath7 with respect to the edge lengths , giving the simplicial equivalent of einstein s equations : @xmath8 where we have used the result in @xcite , that the variation of the angular terms gives zero when summed over each simplex ( schl " afli s differential identity ) . at first sight , it appears that there is one equation for each variable , promising the possibility of a complete solution for the edge lengths . however the situation is not as simple as that ; there are analogues of the bianchi identities in regge calculus , which in the case of flat space provide exact relations between sets of equations , and approximate relations in the nearly - flat case , so the equations may not provide sufficient information for a complete solution . in that case there is freedom to specify certain variables , in analogy with the freedom to specify lapse and shift in the 3 + 1 version of continuum general relativity . in the ten years after its formulation , regge calculus was applied almost exclusively to problems in classical relativity , in particular to the time development of simple model universes . ( rather than give a complete list of references here , we refer the reader to the bibliography @xcite which contains a comprehensive list for the first 20 years . ) the basic idea was really 3 + 1 in nature : take a triangulation of a 3-dimensional surface ( usually closed but not necessarily so ) to represent a hypersurface at a particular moment of time and join its vertices to the corresponding vertices of a second 3-dimensional triangulation , representing the same hypersurface at a later time . the edges used to join these vertices are taken to be timelike and the slice of 4-dimensional space - time between the two triangulations is then divided into 4-simplices by inserting appropriate diagonals . given the edge lengths on the first 3-d triangulation , and specifying the timelike edge lengths , the regge equations may in principle be solved for the edge lengths on the second 3-d triangulation . by repetition of this process , the classical evolution of the inital spacelike surface may be calculated . this sounds simple enough , but unless quite strong assumptions of symmetry are made , the numerical calculation , involving large sets of simultaneous equations for the edge lengths , can be very time - consuming and complicated . significant progress with this approach was made in the early nineties when , based on an idea of sorkin @xcite , it was realised that in general , the regge equations decouple into a collection of much smaller groups . these groups of equations can then be solved in parallel , which means that the computer time required for an equivalent calculation is much less . this parallelisable implicit evolution scheme is described in detail in @xcite and the basic mechanism is as follows . consider a single vertex in a triangulated 3-dimensional spacelike hypersurface and introduce a new vertex above `` this . connect the new vertex by a vertical '' edge to the chosen vertex , and by diagonal " edges to all the vertices in the original hypersurface to which the chosen vertex was joined . each tetrahedron in the original surface now has based on it a 4-simplex , with apex at the new vertex . note that there is one diagonal corresponding to each edge in the original vertex radiating from the chosen vertex . we now use the regge equations for these edges in the original surface and for the vertical edge ; the only unknown edges which these equations involve are the new vertical edge and the diagonal edges , and there is precisely the same number of equations as unknowns . thus , in principle , we can solve exactly for the unknown edge lengths . ( in practice , because of the approximate relationship between the equations from the bianchi identities , it is often more convenient to ignore some of the equations and instead specify conditions equivalent to the lapse and shift . ) we have described how to evolve vertices one - by - one in the sorkin evolution scheme , and the entire hypersurface can be evolved in this way . the method is very general and can be used for a hypersurface with arbitrary topology . however , advancing the vertices one - by - one will not ordinarily be the most efficient way of evolving a hypersurface . if any two vertices in a hypersurface are not connected by an edge , then they can be evolved to the next surface at the same time without interfering with each other , which is why the method is obviously parallelisable . the earliest application of regge calculus to quantum gravity was in three dimensions @xcite and involved 6j - symbols . this work , and subsequent developments along those lines , will be the subject of the next two main sections and we shall not discuss it further here . > from the early eighties onwards , there have been many attempts to formulate a theory of quantum gravity based on regge calculus , and we shall summarise the salient features of some of those approaches , both analytic and numerical . the first work on quantum regge calculus in four dimensions involved using a study of small perturbations about a flat background to relate the discrete variables with their continuum counterparts @xcite . the discrete propagator was derived in the euclidean case and shown to agree with the continuum propagator in the weak field limit . ( more details of this calculation will be given in the section on area regge calculus . ) the technique of weak field approximation has proved to be very useful not only for comparisons with the continuum theory but also as a guide in numerical calculations . the difficulties of analytic calculations in quantum regge calculus , coupled with the need for a non - perturbative approach and also the availability of sophisticated techniques developed in lattice gauge theories , have combined to stimulate numerical work in quantum grvity , based on regge calculus . one approach is to start with a regge lattice for , say , flat space , and allow it to evolve using a monte carlo algorithm ( see for example ) . random fluctuations are made in the edge lengths and the new configuration is rejected if it increases the action , and accepted with a certain probability if it decreases the action . the system evolves to some equilibrium configuration , about which it makes quantum fluctuations , and expectation values of various operators can be calculated . it is also possible to study the phase diagram and search for phase transitions , the nature of which will determine the vital question of whether or not the theory has a continuum limit . many of the simulations have involved an action with an extra term , quadratic in the curvature , to avoid problems of convergence of the functional integral ; some have included scalar fields coupled to gravity @xcite . recent work by riedler and collaborators in four dimensions describes evidence for a new continuous phase transition , essential for a continuum limit , at negative gravitational coupling @xcite . the choice of measure in the functional integral is still a matter for controversy , depending both on attitude to simplicial diffeomorphisms and also on the stage at which translation from the continuum to the discrete takes place . the numerical simulations just described mainly use a simple scale invariant measure @xcite . menotti and peirano have derived an expression for the functional measure in 2-dimensional regge gravity , starting from the dewitt supermetric , and giving exact expressions for the fadeev - popov determinant for both @xmath9 and @xmath10x@xmath10 topologies ( see @xcite and references therein ) . a rather different approach to numerical simulations of quantum gravity is that of _ dynamical triangulations_. ( for a review containing an extensive set of references , see @xcite ) . this also uses regge lattices and the regge action , but there are important differences . in the traditional approach , we are effectively integrating over the edge lengths in the functional integral , but in dynamical triangulations , the lattice is taken to be equilateral , with a certain length scale , and the summation is over different triangulations , which are generated by a set of _ ( k , l ) moves _ @xcite . in two dimensions , there are just two possible moves ( and their inverses ) : the reconnection of vertices in two triangles with a common edge , and the insertion of a vertex and edges in a triangle to divide it into three triangles ( 2 - 2 and 1 - 3 moves ) . there are straightforward generalisations of these moves to higher dimensions . the moves are ergodic in the sense that any combinatorially equivalent triangulation can be generated by a finite succession of these moves . it is argued that the restiction to equilateral triangulations is a way of avoiding over - counting gauge - related configurations . the approach has been very successful in two dimensions , where there are analytic results with which to compare the calculations . in three and four dimensions , there has been progress in , for example , deriving the crucially important exponential bound on the number of triangulations for a given number of vertices @xcite , but there are still open questions on the continuum limit , since the phase transition appears to be first order ( see the review by loll @xcite ) . recently a lorentzian version of dynamical triangulations has been formulated in ( 1 + 1)-dimensions @xcite . numerical simulations have revealed a new universality class for pure gravity , with haussdorf dimension two . discrete gravity has also proved very useful in calculations of the wave function of the universe @xcite . according to the hartle - hawking prescription , the wave function for a given 3-geometry is obtained by a path integral over all 4-geometries which have the given 3-geometry as a boundary . to calculate such an object in all its glorious generality is impossible , but one can hope to capture the essential features by integrating over those 4-geometries which might , for whatever reason , dominate the sum over histories . this has led to the concept of _ minisuperspace _ models , involving the use of a single 4-geometry ( or perhaps several ) . in the continuum theory , the calculation then becomes feasible if the chosen geometry depends only on a small number of parameters , but anything more complicated soon becomes extremely difficult . for this reason , hartle @xcite introduced the idea of summing over _ simplicial _ 4-geometries as an approximation tool in quantum cosmology . although this is an obvious way of reducing the number of integration variables , there are still technical difficulties : the unboundedness of the einstein action ( which persists in the discrete regge form ) leads to convergence problems for the functional integral , and it is necessary to rotate the integration contour in the complex plane to give a convergent result . in principle , the sum over 4-geometries should include not only a sum over metrics but also a sum over manifolds with different topologies . one then runs into the problem of classifying manifolds in four and higher dimensions , which led hartle @xcite to suggest a sum over more general objects than manifolds , _ unruly topologies_. schleich and witt @xcite have explored the possibility of using conifolds , which differ from manifolds at only a finite number of points , and this has been investigated in some simple cases . however , a sum over topologies is still very far from implementation . yet another area of application of regge calculus in quantum gravity involves the study of the simplicial supermetric , the metric on the space of 3-geometries . its signature is crucial for determining spacelike surfaces in superspace , which are important in dirac quantisation and in quantum cosmology . in the continuum , there are limited results on the signature and this led to the possibility of investigating it in the discrete case @xcite , where the analogue is the lund - regge supermetric @xcite . this supermetric was constructed for some simple manifolds ( @xmath11 and @xmath12 ) and its signature calculated . the results agreed with the continuum predictions and also showed that the supermetric can become degenerate . we still do not have a complete understanding of the division of the modes into vertical `` ( corresponding to metrics related by diffeomorphisms ) and horizontal '' ones . of course regge calculus is not the only way of setting up a theory of discretised general relativity . in this subsection , we shall describe some alternatives . one important class of schemes involves treating gravity as a gauge theory . for example , mannion and taylor @xcite defined a theory of gravity on a fixed hypercubic lattice , and kaku @xcite used a fixed random lattice . however a dynamical lattice seems more appropriate in a theory aiming to describe the quantum fluctuations of space - time , and this was used in much earlier work by weingarten @xcite . in an approach closely related to regge calculus , caselle , dadda and magnea @xcite defined a theory of gravity on the dual lattice , giving both first- and second - order formulations . the action they obtained was a compactified form of the regge action , involving the sine of the deficit angle . dadda and gionti showed @xcite that regge calculus is a solution of the first order formulation in the limit of small deficit angles . the action of caselle , dadda and magnea was also used by kawamoto and nielsen @xcite in their version of lattice gravity with fermions . immirzi investigated the links between canonical general relativity in the continuum , loop quantum gravity , and spin networks , in an attempt to formulate a quantised version of discrete gravity in the spirit of regge calculus but ran into problems over hermiticity @xcite . a totally different approach to discrete gravity is t hooft s polygon model in ( 2 + 1)-dimensions @xcite . this was introduced as a way to refute gott s claim of acausality in ( 2 + 1 ) gravity coupled to point particles @xcite . t hooft s method is to split space - time into the direct product of cosmological time and a cauchy surface tessellated by flat polygons . the local flatness of space - time in the pure gravity regime and the cone - like structure introduced by particles , as in regge calculus @xcite , are expressed in terms of conditions on the edges and vertices of the polygons . a local lorentz frame is attached to each polygon and two constraints imposed ; these are firstly that time runs at the same rate in each polygon ( which corresponds to a partial gauge fixing ) and secondly that all vertices are trivalent ( which is acceptable because higher order vertices can always be split into trivalent vertices connected by edges of zero length ) . the consequences of these conditions are that the length and velocity of an edge are the same in both polygons to which it belongs , and that the velocity of each edge is orthogonal to it in both frames . these facts result in transition rules for the vertices in the tessellation . the method for evolving such a space - time is as follows . initial data ( lengths and velocities ) , subject to consistency conditions , are assigned to the edges on a polygonally - tessellated hypersurface . the configuration evolves linearly until an edge collapses to zero length or a vertex crosses another edge . a transition , governed by the vertex conditions , then takes place to another configuration which will , in general , have different numbers of vertices , edges and polygons . the new data will still satisfy the consistency conditions and the process is then repeated . when there are particles present at the vertices , there are deficit angles proportional to their masses , and the transition rules are modified accordingly . it is not an easy task to follow the time - evolution of a ( 2 + 1)-dimensional model with particles , even though the system has a finite number of degrees of freedom . t hooft did numerical simulations on a small computer , with some unexpected predictions . his big - bang and big - crunch hypotheses were based on the evolution of a cauchy surface with @xmath9 or @xmath10x@xmath10 topology , tessellated by a single polygon @xcite . it would be interesting to test these predictions for more complex initial configurations , and as a means to this , there has been recent work @xcite in which the constraint equations have been interpreted in terms of hyperbolic geometry ( see also @xcite ) , and various consistent sets of initial data set up , but the evolution calculations have not yet been completed . part of the motivation for this work is to compare t hooft s method with other approaches to ( 2 + 1 ) gravity , in particular regge calculus . a ( 2 + 1)-dimensional code has been set up for regge space - times and a number of calculations performed @xcite , with a view to making detailed comparisons with the t hooft method . the ultimate aim is to understand the exact relationship between the two approaches , which seem rather different but have many concepts in common . based on the polygon approach , various toy models of ( 2 + 1)-dimensional gravity have been constructed , issues of topology been addressed and particle decay and space - time kinematics been investigated @xcite . t hooft himself has proposed quantised models of ( 2 + 1)-dimensional space - time @xcite , showing that gravitating particles live on a space - time lattice . for an @xmath9x@xmath10 topology , first quantisation of dirac particles is possible . waelbroeck has suggested a similar approach , using canonical quantisation in ( 2 + 1 ) dimensions @xcite . back at the classical level , brewin has formulated @xcite a discretisation of gravity which he feels is closer to the original theory of general relativity . preliminary calculations are encouraging . for other important work on lattice gravity by bander , jevicki and ninomiya , khatsymovsky and lehto , nielsen and ninomiya , we refer the reader to the regge calculus review and bibliography @xcite . we emphasise again that this paper is not meant to be an exhaustive review of the subject . after this rather rapid survey of applications of regge calculus , and some other approaches to discrete gravity , we shall now concentrate on one particular approach and show how it has led to exciting new developments in the search for a quantum theory of gravity . as promised , we now look in detail at the earliest link forged between regge calculus and quantum gravity , now known as the ponzano - regge model @xcite . this emerged from a paper on 6j - symbols and we will first give the background to these . 6j - symbols , which are generalisations of the more well - known clebsch - gordan coefficients , first arose as tools for the computation of matrix elements in the theory of complex spectra @xcite , and are now used routinely by atomic physicists and theoretical chemists in quantum mechanical calculations involving angular momentum . in particular , they relate the possible basis wave functions when three angular momentum are added : @xmath13 @xmath14 @xmath15 an alternative and useful definition involves the _ recoupling _ diagram : @xmath16 a graphical representation is obtained by associating a 6j - symbol with a tetrahedron : @xmath17 with the arguments of the 6j - symbol corresponding to the edge lengths of the tetrahedron . ( for technical reasons , it turns out to be more accurate to associate a symbol with arguments @xmath18 to a tetrahedron with edge lengths @xmath19 . ) for the 6j - symbol to be non - zero , the arguments have to satisfy the analogue of the triangle inequalities for each face of the tetrahedron : @xmath20 they can be evaluated from the formula @xmath21 @xmath22 @xmath23^{-1}\ ] ] where @xmath24^{-1}}.\ ] ] these 6j - symbols are based on the group @xmath25 , but as we shall see , it is also possible to have q - deformed 6j - symbols based on quantum groups . for example , define @xcite @xmath26 and @xmath27 } = { ( q^{n/2 } - q^{-n/2 } ) \over ( q^{1/2 } - q^{-1/2})}.\ ] ] then the q - deformed 6j - symbol for @xmath28 is defined in the same way as the undeformed one , with @xmath29 replaced everywhere by @xmath30 $ ] . note that @xmath30 \rightarrow n$ ] as @xmath31 and @xmath32 . the main purpose of the paper by ponzano and regge @xcite was to derive asymptotic formulae for classical ( ie undeformed ) 6j - symbols in the limit when certain arguments became large . the case most relevant to the exposition here is when all six parameters become large . the edge lengths of the corresponding tetrahedron are really related to @xmath33 and these quantities are kept finite as @xmath34 while @xmath35 so this process corresponds to the semi - classical limit . this asymptotic behaviour is given by @xmath36 where @xmath37 is the volume of the terahedron and @xmath38 is the exterior dihedral angle at edge @xmath39 ( ie the angle between the outward normals to the faces meeting there ) . this was recently proved rigorously by roberts @xcite . to see the connection between this formula and quantum gravity , consider the following _ state sum _ defined in @xcite . take a closed 2-dimensional surface , triangulate it and divide its interior into tetrahedra , possibly inserting internal vertices . label the internal edges by @xmath40 and the external ones by @xmath41 . define @xmath42 where the @xmath43 in the phase factor is a function of the edge lengths . although this expression is infinite in many cases , it has some extremely interesting properties . in particular , noting that in the sum over the internal edges , the large values dominate , we can replace the sum by an integral with respect to those edge lengths and use the asymptotic formula stated above . then the dominant contribution to the integral comes from the points of stationary phase , which are given by @xmath44 this means that the sum of the dihedral angles at each edge is @xmath45 , which is precisely the condition for local flatness in a 3-dimensional simplicial space . what is more , the state sum is given approximately by @xmath46 now this contains a term of the form @xmath47 @xmath48 which looks precisely like a feynman sum over histories with the regge calculus action in three dimensions : @xmath49 with @xmath50 where @xmath51 is the deficit angle at edge @xmath52 and @xmath53 is the measure on the space of edge lengths . this result was rather puzzling and , although hasslacher and perry @xcite emphasised the connection between spin networks and simplicial gravity , its significance was not fully appreciated until much later , when a very similar expression was written down in a different context . in the late eighties and early nineties , mathematicians put a lot of effort into searching for invariants of manifolds , the hope being , at least in part , that such quantities would help with the classification of manifolds . without being aware of the ponzano - regge work , turaev and viro @xcite defined a state sum for triangulated 3-manifolds , which in many aspects was identical to that of ponzano and regge . the main differences were that they gave formulae for closed manifolds as well as those with boundary , they showed explicitly that the quantity obtained was independent of triangulation , and finally , they used 6j - symbols for the quantum group @xmath54 . only some of the irreducible representations of this group , the ones with @xmath55 taking finite values , have suitable algebraic properties , which means that the edge lengths are not summed up to infinite values ; @xmath56 can take only integer and half - integer values from the set @xmath57 , with @xmath58 . a very important consequence of this is that the answer obtained is finite , and so the model appears to be a regularised version of the ponzano - regge model . the obvious question to ask is how the turaev - viro state sum is connected to quantum gravity . witten @xcite conjectured that it was equivalent to a feynman path integral with the chern - simons action for @xmath59 , and this and equivalent results were proved by a number of people . to see how this works , consider the chern - simons lagrangian for this group product : @xmath60 @xmath61 where @xmath62 with @xmath63 a basis of the @xmath25 lie algebra . making the change of variables @xmath64 where @xmath65 is the dreibein and @xmath66 with @xmath67 being the connection 2-form , we obtain @xmath68 which is the einstein - hilbert action for gravity with cosmological constant given by @xmath69 ( note that the @xmath70 here is equal to @xmath71 , where @xmath72 appears in the definition of @xmath73 . ) by taking the limit as @xmath74 , we obtain 3-dimensional gravity with zero cosmological constant ie the theory represented by the ponzano - regge model . this result is consistent with the fact that the @xmath31 limit of the tuaev - viro model is the ponzano - regge model . the properties of the turaev - viro state sum show that the formalism is an example of a _ topological quantum field theory _ ( see eg @xcite ) . this is perfectly appropriate for a theory of gravity in three dimensions where there are no local degrees of freedom . as for the ponzano - regge theory , the dominant classical configurations are locally flat ( recall that in chern - simons gravity , the solutions involve the space of flat connections . ) the relationship between the turaev - viro invariant and 3-dimensional quantum gravity is an extremely important one . it means that in three dimensions , we have in principle a way of calculating the partition function for triangulated manifolds . this has been done for many of the simpler 3-manifolds ( see for example ) . the turaev - viro expression can also be used for calculating topology - changing amplitudes in 3-dimensional gravity ; the method here is to construct a cobordism between two 2-dimensional triangulated surfaces and then use the turaev - viro expression for a manifold with boundaries to evaluate the transition probability @xcite . the turaev - viro expression is not the only method of calculating this particular invariant of 3-manifolds . various other prescriptions have been written down , and one that is worth describing at this stage is that using spin networks . these were invented by penrose @xcite who wanted to formulate a purely combinatorial approach to space - time . his networks had trivalent vertices and the edges of the graphs were labelled by spins . he developed a method of calculating the value of an arbitrary spin network and was able to show that this led to the usual angles of 3-dimensional space . penrose s spin networks were later generalised in a number of ways . the edges were labelled by representations of quantum groups and it was necessary to introduce intertwining operators or _ intertwiners _ at the vertices @xcite . in some cases a framing was introduced and the graphs became `` ribbon graphs '' @xcite . kauffman @xcite showed how to calculate the turaev - viro invariant by taking the graph dual to a triangulation to be a spin network ; the edges of the graph inherit the labels of the triangulation edges which they cross . spin networks have also been introduced into loop quantum gravity @xcite , where they are an important calculational tool , for instance in the derivation of the spectrum of the area and volume operators @xcite . ( note that freidel and krasnov also obtained a discrete spectrum for the volume operator in bf theory by differentiating the turaev - viro amplitude with respect to the cosmological constant @xcite . ) as we shall see , spin networks also play a r^ ole in recent attempts at formulations of 4-dimensional quantum gravity . after it was realised that the turaev - viro state sum provides a finite theory of 3-dimensional gravity , the search began for a generalisation to four dimensions . before this is described , we shall stop to ask what we hope to achieve by this . in classical general relativity , there are enormous qualitative differences between gravity in three and four dimensions . in particular , there are gravitons in four dimensions , but not in three , so although it seems reasonable to describe 3-dimensional gravity by a topological invariant of a manifold , it seems _ likely _ that an invariant of a 4-manifold might describe only some topological sector of gravity . we shall return to this point later . the obvious way of setting about extending the 3-dimensional model , based on 6j - symbols , to four dimensions is by using some 3nj - symbol for a value of @xmath29 larger than 2 . the 3nj - symbols in the state sum would then be expanded in terms of 6j - symbols , and the ponzano - regge formula for their asymptotic values inserted , the hope being that this would give an expression looking like a path integral with the 4-dimensional regge action . the problem with this is that the asymptotic formula involves the 3-dimensional dihedral angles and it is very difficult to relate these to 4-dimensional angles . this indicates that a more radical generalisation may be needed . we shall now describe some of the attempts at generalisation , leading up to some recent work which seems very promising . a source of inspiration for some generalisations of the ponzano - regge and turaev - viro models was boulatov s generalised matrix model @xcite , which involved a scheme for generating 3-dimensional simplicial complexes as terms in a perturbative expansion . the contribution from each simplicial - complex was weighted by its ponzano - regge or turaev - viro invariant , depending on the value of @xmath73 . boulatov s model was formulated in a way that it could be extended to higher dimensions , and the 4-dimensional case for @xmath75 was worked out by ooguri @xcite . the essential ingredients in ooguri s model are the assigning of group variables to the tetrahedra and spin j labels to the triangles in the triangulated 4-manifold . the terms in ooguri s action are of two types : the first is a product of two functions of the group variables , and this represents two glued tetrahedra ; the second is a product of five functions and represents the tetrahedra in a 4-simplex . a fourier decomposition is performed in terms of rotation matrices and the group variables are then integrated out , using the standard relationship between rotation matrices and 3j - symbols , and the invariant haar measure normalised to unity . the resulting expression has four 3j - symbols associated to each tetrahedron ; these may then be divided between the 4-simplices meeting on that tetrahedron , and then each 4-simplices ends up with ten 3j - symbols which can be combined to give a 15j - symbol . at first sight , it seems odd to associate a 15j - symbol with a 4-simplex which has only 10 triangles labelled by spin values . the way to interpret the symbol is to consider the dual graph , which has ten edges and five 4-valent vertices ( corresponding to each tetrahedon in the original triangulation ) . each of these 4-valent vertices can be split into two trivalent ones , and an extra spin label can be assigned to the edge joining them . this splitting sounds rather arbitrary but different splittings are related by 6j - symbols ( see the second diagram in the section on 6j - symbols ) and when all summations are performed , the result is independent of splitting . the partition function is calculated by integrating the exponential of minus the action over the fourier coefficients , and the resulting expression is @xmath76 with @xmath77 given by @xmath78 the summation in @xmath79 is over simplicial complexes @xmath80 , with @xmath81 being the rank of the symmetry group of @xmath80 , and @xmath82 the number of 4-simplices in @xmath80 . by writing the contributions from all the tetrahedra meeting on a particular triangle in terms of rotation matrices , one can show that the holonomy around any triangle is trivial . this ties up with the proposed link between ooguri s model and bf theory , as we shall see later . the extension of the ooguri model to general values of @xmath73 was worked out by various people . archer @xcite showed how to construct a q - deformed topological quantum field theory in general dimension , giving realisations in three and four dimensions based on the quantum group @xmath83 , and suggesting that his theory corresponded to bf theory with a cosmological constant . crane and yetter @xcite outlined the construction of a q - deformed version of ooguri s model and recognised its relationship with the work of roberts @xcite , who had defined a 4-dimensional generalisation of his own chain - mail " formulation of the turaev - viro invariant . roberts showed that his invariant for a 4-manifild @xmath84 depended on two simple functions of @xmath72 , one raised to the power of @xmath85 , the signature , and the other to the power of @xmath86 , the euler character . the result of roberts was disappointing but instructive for those trying to construct a theory of 4-dimensional quantum gravity by this method . since the models do not give any new information about 4-manifolds , it showed that a more radical generalisation was needed . an important step forward in these generalisation attempts has been taken recently with the formulation of the barrett - crane model . ( although the details of some aspects of the model , and other related models , have yet to be worked out , we consider the ideas sufficiently important to include in this review . ) first came the realisation that it made sense to generalise spin networks to _ relativistic spin networks _ appropriate to four dimensions @xcite . the symmetry group @xmath87 in three dimensions is replaced by @xmath88 in four dimensions , which has spin covering @xmath89 . barrett and crane therefore label the triangles by two spin labels rather than one . thus in a relativistic spin network , the edges ( dual to the triangles in the 4-complex ) carry labels @xmath90 and the vertices ( dual to tetrahedra ) carry the appropriate intertwiners . barrett and crane suggested that the two labels @xmath91 and @xmath92 should be equal to satisfy the constraints at the vertices , and reisenberger @xcite showed that this solution is unique . thus the barrett - crane model is a constrained doubling of the earlier attempts described in the previous subsections , which can thus be regarded as just describing the self - dual section of gravity . we now describe the barrett - crane model in a little more detail . consider a single 4-simplex , draw its dual graph and then split the vertices as described for the ooguri model . the first expression written down by barrett and crane for the amplitude of a 4-simplex was of the form @xmath93 where @xmath94 is a weight factor and the 15j - symbol is squared because of the @xmath95 labelling on each edge of the dual graph . it turned out to be very difficult to evaluate the asymptotic value of this expression , so barrett and crane tried a second approach . label the five tetrahedra in a 4-simplex by @xmath70 ; the spin label on the triangle where tetrahedra @xmath70 and @xmath52 meet is then denoted by @xmath96 . the matrix representing the element @xmath97 in the irreducible representation of spin @xmath96 is denoted by @xmath98 . variables @xmath99 are assigned to the tetrahedra and the invariant @xmath100 ( the second barrett - crane model ) is obtained by integrating a function of these variables over each copy of @xmath25 : @xmath101 the measure used is the haar measure normalised to unity . the next step is to relate this expression to the geometry of the 4-simplex @xcite . using the fact that @xmath25 is isomorhic to @xmath11 , and embedding @xmath11 in @xmath102 , we can regard the element @xmath103 as a unit vector in @xmath102 , normal to the 3-dimensional hyperplane in which tetrahedron @xmath70 lies . then according to a well - known formula in representation theory , @xmath104 where @xmath105 ie @xmath106 is the angle between the normals and thus the exterior angle between the two hyperplanes . note that the five hyperplanes define a 4-simplex up to translation and an overall scale . thus integration over the elements @xmath107 may be interpreted as integration over all possible 4-simplices . recalling the equivalence of the asymptotic value of the ponzano - regge model to a path integral with the 3-dimensional regge calculus action , we now look for a similar result here @xcite . we write @xmath108 in terms of exponentials and , for large @xmath55 , use the method of stationary phase to find the asymptotic value of the integral . setting @xmath109 , we write @xmath100 as @xmath110 which makes it clear that we need the stationary points of @xmath111 now the @xmath112 s for a 4-simplex are not independent variables ; as is shown in the original formulation of regge calculus @xcite , their variations are related by @xmath113 adding this constraint to @xmath6 with a lagrange multiplier @xmath114 , we find that for each triangle , @xmath115 the overall scale can then be fixed by taking @xmath116 . what has been established is that for a stationary phase point , then firstly , the angles @xmath112 are those of a geometric 4-simplex with triangle areas @xmath117 and secondly , the integrand is @xmath118 , with @xmath119 the regge calculus version of the einstein action for a 4-simplex , with @xmath116 . the formulation of this model is by no means complete . the next step is to sum over 4-simplices , which is likely to be more difficult than for the first model , where the extra labels on tetrahedra could provide links between neighbouring 4-simplices . the resulting expression will need to be regularised by passing to representations of the quantum group @xmath120 , as in the transition from the ponzano - regge state sum to that of turaev and viro . this analogy is not precise because the barrett - crane amplitude is _ not _ independent of triangulation . the covariance lost here may perhaps be restored by summing over triangulations using a generalised matrix model approach , as suggested by de pietri , freidel , krasnov and rovelli @xcite . ( note that these authors refer to what we have called the second barrett - crane model `` as their first version '' . ) we shall return to the interpretation of this model in the next subsection , but first note that the formulation described so far is euclidean . there have been lorentzian models proposed recently : in ( 2 + 1 ) dimensions , freidel @xcite has set up a version in which @xmath25 is replaced by @xmath121 , for which both discrete and continuous representations are used . this results in a model in which time is discrete and space continuous . the partition function requires summation over causal structures , which obviously has no analogue in the euclidean case . the 6j - symbols for the discrete series representation of @xmath121 were defined first by davids @xcite , who also obtained the analogous ponzano - regge formula , which here involves @xmath122 , where @xmath123 is the lorentzian regge action . in ( 3 + 1 ) dimensions , barrett and crane @xcite have proposed versions based on the classical lorentz group and on the quantum lorentz algebra , but the second of these is still at a preliminary stage . this is not the place for a review of bf theory , but let us briefly mention its relevant properties . it is a gauge theory which can be defined in any dimension and is background - free " in the sense that no pre - existing metric or other geometrical structure on space - time is needed . it is a theory with no local degrees of freedom . the action for bf theory in four dimensions is @xmath124 where @xmath125 is a lie algebra - valued 2-form , and @xmath126 , with @xmath127 the connection 1-form . it gives rise to the constraint @xmath128 , which means that the connection @xmath127 is flat . this ties up with the trivial holonomy around triangles in the ooguri model . the other constraint , @xmath129 , is the statement of a particular type of gauge symmetry in bf theory . to understand the relationship between general relativity and bf theory in four dimensions @xcite , consider the palatini formulation of general relativity , which has action @xmath130 with @xmath131 a 1-form on the manifold @xmath84 , and @xmath132 defined in terms of the connection as for bf theories . it is immediately apparent that there is a relationship between this palatini formulation , and bf theory with @xmath125 constrained to be of the form @xmath133 . there is a subtle difference between the equations of motion derived from the two actions : for general relativity , we have @xmath134 as compared with the bf equations @xmath135 thus the equations of general relativity are weaker here than those for bf theory , which , heuristically , is why general relativity in four dimensions is more general than a topological theory . we see that general relativity in four dimensions is equivalent to bf theory with an extra constraint @xmath136 ( giving rise to the paradoxical statement that adding a constraint produces a less restricted theory ! ) we see now a further justification of why , in the barrett - crane model , the two spin labels on each triangle should be equal ( ie we see the parallel between @xmath95 and @xmath133 ) . thus the constraint which reisenberger @xcite derived may be interpreted as equivalent to the constraint which relates bf theory to general relativity in four dimensions . reisenberger @xcite has explored further the relationship between the barrett - crane model and continuum theories , showing that the model corresponds to an @xmath88 bf theory in which the right- and left - handed areas , defined by the self - dual and anti - self - dual components of @xmath125 , are constrained to be equal . before considering an extension of bf theory in four dimensions , let us return to the case of three dimensions . it can be shown that 3-dimensional general relativity without matter is a special case of bf theory , where the equations of motion give simply that the connection is torsion - free and flat . adding an extra term to the bf lagrangian has a very interesting effect . starting from the modified action @xmath137 and making the transformation @xmath138 we can show that @xmath139 is equal to the difference of the two chern - simons actions as in section 2 . it was shown there that this was equivalent to 3-dimensional general relativity with a cosmological constant @xmath140 related to the deformation parameter @xmath73 , which gives a finite theory of quantum gravity in that dimension @xcite . thus a r^ ole of the cosmological constant is to regularise the theory . in four dimensions , the extra term that we need seems to be slightly different . the proposed modified action is @xmath141 the form of this extra term was first suggested by archer @xcite , whose contribution is described earlier . it has been discussed more recently by baez @xcite , who gives a very comprehensive discussion of bf theory and the discrete models of quantum gravity in three and four dimensions . ( reference @xcite is recommended strongly for fuller details of these issues . ) imposing the constraint @xmath142 as before , the action becomes that for the palatini formulation of general relativity with cosmological constant , @xmath143 this suggests the possibility of finding a regularised version of 4-dimensional quantum gravity by constructing a q - deformed version of the barrett - crane model , satisfying the relationship @xmath144 another possible ( and related ) way forward is through spin foam models , as described briefly in the next subsection . as mentioned in the section on 3-dimensional gravity , spin networks have played an important role in calculations of invariants of 3-manifolds , and in loop quantum gravity , where they provide a gauge - invariant basis of states @xcite . if we wish to describe _ space - time _ by this type of method , we need , as we have already remarked , an extension of the concept of spin networks . an alternative to the idea of relativistic spin networks is provided by what has been called _ spin foam _ @xcite , because one can think of a spin foam as a soap film connecting two spin networks at different times . sums over surfaces " formulations of loop quantum gravity have been given by reisenberger and rovelli @xcite , and iwaski @xcite has formulated the ponzano - regge model in terms of surfaces . turaev and viro @xcite formulated their theory not only in terms of a triangulation of the 3-manifold but also in terms of simple 2-polyhedra forming a 2-complex embedded in the manifold , and we can interpret this second method as the first example of a spin foam model ! the relationship between the evolution of spin networks and the approach using triangulated manifolds has been explored and illuminated by markopoulou @xcite . the theory of spin foam is a way of formalising the calculation of the partition function in bf theory by triangulating manifolds . recall that a spin network is a graph with edges labelled by irreducible representations and vertices by intertwiners . imagine moving such an object through space , or rather space - time , so that it traces out a 2-dimensional surface , a generic slice through which would be a spin network ; this , heuristically , is what we mean by a spin foam . it is a 2-complex , the faces of which are labelled by irreducible representations and the edges by intertwiners . the dual triangulation of a manifold is an example of such an object . baez @xcite has outlined how to calculate transition amplitudes in bf theory using sums over spin foams , and the derivation of the spin foam model from the classical action principle based on bf theory has been discussed by freidel and krasnov @xcite . it has already been shown @xcite that a particular type of spin network may be evaluated as a feynman graph , and the idea in the evaluation of spin foam sums is to use feynman s sum over histories approach , with bf theory playing the rle of the free theory and spin foams as 2-dimensional analogues of feynman diagrams . these techniques have produced agreement with the lowest order terms in the known state sum models @xcite . markopoulou and smolin @xcite have defined a model of the time evolution of spin networks based on local causality rules , which are equivalent to those for spin foams . recently smolin @xcite has suggested a connection between evolving spin networks , spin foam and such approaches related to loop quantum gravity , and string theory , where there are clearly intuitive similarities in the evolution of strings and membranes . any precise equivalence still needs to be worked out , but smolin s suggestion is typical of recent ideas in which a number of apparently unrelated approaches to quantum gravity seem at last to be coming together . it seems that those attempts at formulating a theory of quantum gravity in four dimensions described in the last section all need one ingredient to be at all successful ; this is the assignment of labelling to the triangles instead of ( or possibly as well as ) the edges . ( this fits in with work by birmingham and rakowski @xcite who constructed state sum models based on @xmath145 for 4-dimensional triangulated manifolds . when the colourings from @xmath145 were assigned only to the edges , the invariant depended only on the 3-dimensional boundary manifold , but when colourings were assigned also to the triangles , the invariant depended on the 4-dimensional structure . ) even the spin foam description fits into this pattern when one considers the triangulation to which it is dual . by considering the asymptotic value of the amplitude of a 4-simplex , we have seen that in this case , it appears to be related to the path integral with the regge calculus action but with the triangle areas playing the most important r^ ole , rather than the edge lengths . the idea that , in four dimensions , the triangle areas could be regarded as the basic variables in a modified form of regge calculus was first suggested by rovelli @xcite and the possibility was discussed in some detail in @xcite . in this section , we shall consider the advantages and disadvantages of the approach , and report on some progress in understanding the relationship between the two types of variable . a 4-simplex not only has ten edges , it also has ten triangles . thus at first sight , the change from edge lengths to triangles areas as basic variables looks very straightforward , but there are actually a number of problems @xcite . consider first a single 4-simplex . it is simple to express the triangle areas in terms of the edge lengths . however , to express the regge action in terms of the new variables , we need to invert the relationship between areas and edge lengths to be able to calculate the deficit angles . unfortunately the jacobian is singular in cases where a number of triangles are right - angled and there is not necessarily a unique set of edge lengths corresponding to a given set of areas . this means that , right from the start , certain regions in the space of edge lengths must be avoided . secondly , for a collection of 4-simplices joined together , there will not in general be equal numbers of edges and triangles so there may be ambiguity about which is the correct number of variables . thirdly , by considering two 4-simplices meeting on a tetrahedron with all triangle areas assigned , we can envisage the following bizarre situation . solve for the edge lengths of one of the 4-simplices in terms of its triangle areas . repeat this for the other 4-simplex . it is possible that the edge lengths of the common tetrahedron will differ according to the 4-simplex where the calculation was done ( see @xcite for an example ) . clearly there are difficulties in interpreting the edge lengths as real physical quantities in the usual sense . in this section , we shall now discuss possible theories in terms of equations of motion and then investigate the dynamical content of area regge calculus by studying the weak - field expansion about a flat background in terms of variations in the areas . the counting of degrees of freedom in a discrete theory is never completely straightforward . in a simplicial theory , the usual argument is that in @xmath29 dimensions , an n - simplex has @xmath146 edges , which corresponds to the number of independent degrees of freedom of the metric tensor in @xmath29 dimensions . if one thinks of these variables as being at some chosen point in each simplex , the counting becomes somewhat less clear when one realises that each of the edges is shared by a number of other simplices , so the number of variables per point is quite obscure . given this ambiguity , we can take two attitudes to the counting problem in area regge calculus . either we can take the areas as the fundamental variables , worrying about the different numbers of edge lengths only inasfar as we need them to calculate deficit angles or volumes , or we can regard some of the areas as redundant variables and aim to reduce their number to the number of edge lengths in the simplicial complex . in the theory where the areas are taken seriously as variables ( which is our principal interest here since we aim thereby to understand the the models described as 4-dimensional generalisations of the turaev - viro theory ) , we concentrate on the restricted class of metrics where the jacobian is non - singular . then the hyperdihedral angles are well - defined and the regge action may be written as @xmath147 where the sum is over triangles @xmath148 and @xmath149 is the deficit angle at triangle @xmath148 . variation of the action with respect to the area @xmath150 , use of the chain rule , an interchange of the orders of summation and use of the regge identity [ 1 ] leads to @xmath151 for details , see @xcite . since all deficit angles vanish , the space is locally flat ; the holonomy round any triangle is trivial . this agrees with ooguri s state sum model for bf theory @xcite . the interpretation of this result is not obvious and the investigation of such spaces using parallel transport is under way . the other possibility , that of regarding some of the areas as redundant variables , has been investigated by mkel @xcite . clearly in order to recover the conventional view of simplicial gravity where the edge lengths are real physical quantities , it is necessary to impose the condition that a given edge has the same length in whichever 4-simplex that length is calculated . this leads to a large number of constraints : for each edge , there is a constraint for each pair of 4-simplices meeting there . for a simplicial complex with @xmath152 edges and @xmath153 triangles , a total of @xmath154 of these constraints will be independent , but it is not easy to give any general rule for picking out which these are . ( an _ ad hoc _ rule has been formulated for a particular model and it is likely that there is some group - theoretic basis for the rule @xcite ) . mkel has shown that if the variations of the constraints are added in with lagrange multipliers to the variation of the regge action expressed in area variables , then the usual regge calculus equations of motion are recovered . restricting our attention now to the area variable theory without constraints , we investigate its dynamical content by performing a weak field expansion about a flat background @xcite . this is in analogy with the weak field expansion for edge length variables @xcite , which we now describe briefly . in the original calculation , a 4-dimensional hypercubic lattice is divided into simplices by drawing in various diagonals , giving fifteen edges per vertex . small variations of the edge lengths about their flat space values are made by setting @xmath155 with @xmath156 . the second variation of the regge action ( the first non - vanishing term ) is evaluated as a quadratic expression in the @xmath157 s , written as @xmath158 with @xmath159 a sparse infinite dimensional matrix . a fourier transform is then performed by relating @xmath157 in the @xmath29 direction and based at the lattice point @xmath160 steps in the @xmath161 directions from the origin ( see @xcite for details of the binary notation ) to the corresponding @xmath157 at the origin by @xmath162 with @xmath163 , where @xmath164 is the period in the @xmath114-direction . acting on periodic modes , @xmath84 reduces to a block diagonal matrix with 15x15 dimensional blocks , @xmath165 . this matrix @xmath165 has four zero modes , corresponding to periodic translations of points of the lattice , and a fifth zero mode corresponding to periodic fluctuations of the hyperbody diagonal . block diagonalising @xmath165 decouples four further modes ; they enter without @xmath166 s and so do not contribute to the dynamics at all . their equations of motion constrain them to vanish . we see from this that an apparent mismatch in the number of components ( fifteen per vertex ) is corrected by the dynamics of the theory , leaving ten degrees of freedom per vertex , as would be expected from the continuum theory . ( the zero modes correspond of course to gauge fluctuations . ) we now perform the analogous calculation with area variables . in this case , it is necessary to use a distorted " hypercubic lattice because the original one contains many right angles which lead to vanishing of the jacobian when transforming between areas and edge lengths . this is obtained by squeezing each unit hypercube along its hyperbody diagonal until it has length @xmath167 in lattice units , like the edges originally along the coordinate axes . the face and body diagonals then all have length @xmath168 . small variations of these edge lengths about their flat space values are then made and the second variation of the action within each 4-simplex calculated . these variations in edge lengths induce changes in the triangle areas represented by @xmath169 with @xmath170 . within each 4-simplex , the expressions for the @xmath171 s in terms of the @xmath172 s are inverted ( uniquely ) and the second variation of the action written in terms of the @xmath171 s . adding together the contributions from all 4-simplices gives @xmath173 with @xmath174 again a sparse infinite dimensional matrix . a fourier transform is then performed as in the edge - length variable case , and @xmath175 reduces to a block diagonal matrix with 50x50 dimensional blocks @xmath176 ( note that there are 50 triangles based at each vertex ) . the size of @xmath176 makes it necessary to investigate the modes numerically , and , somewhat contrary to our original expectations , it turns out that the number of dynamical modes is exactly the same as in the edge length case . there are again four zero modes , corresponding to periodic fluctuations of the lattice , and six further modes scaling with @xmath177 , where @xmath70 is the momentum in the fourier transform . the remaining forty modes enter non - dynamically ( they are massive and do not scale with momentum ) and are constrained to vanish by their equations of motion . thus the theory with area variables is equivalent to the edge length variable theory from the point of view of dynamical content . this is very encouraging and gives impetus to the search for the exact correspondence between the variables in models like that of barrett and crane , the variables of regge calculus and ultimately the variables of conventional general relativity . that search continues the authors thank john barrett and radu ionicioiu for help with the preparation of this paper . the work was supported in part by the uk particle physics and astronomy research council . w. a. miller , geometric computation : null - strut geometrodynamics and the inchworm algorithm " , in _ dynamical spacetimes and numerical relativity _ , edited by j. centrella ( cambridge university press , 1986 ) . g. ponzano and t. regge , semi - classical limit of racah coefficients " , in _ spectroscopic and group theoretical methods in physics _ , edited by f. block , s. g. cohen , a. deshalit , s. sambursky and i. talmi ( north holland , amsterdam , 1968 ) . h. w. hamber , simplicial quantum gravity " , in _ critical phenomena , random systems , gauge theories _ , edited by k. osterwalder and r. stora ( north holland , amsterdam , 1986 ) ; nucl . phys . a ( proc . suppl . ) * 25 * , 150 ( 1991 ) . m. kaku , generally covariant lattices , the random calculus and the strong coupling approach to the renormalisation of gravity " , in _ quantum field theory and quantum statistics _ , edited by i. a. batalin , c. j. isham and g. a. vilkovisky ( adam hilger , bristol , 1987 ) . a. n. kirillov and n. yu . reshetikhin , representations of the algebra @xmath178 , q - orthogonal polynomials and invariants of links " , in _ infinite dimensional lie algebras and groups _ , edited by v. g. kac ( world scientific , singapore , 1989 ) . j. baez , an introduction to spin foam models of bf theory and quantum gravity " , gr - qc/9905087 , to appear in _ geometry and quantum physics _ , edited by h. gausterer and h. grosse , lecture notes in physics , ( springer - verlag , berlin ) .

our goal is to resolve a problem proposed by fernholz and karatzas ( 2008 ) : to characterize the minimum amount of initial capital with which an investor can beat the market portfolio with a certain probability , as a function of the market configuration and time to maturity . we show that this value function is the smallest nonnegative viscosity supersolution of a non - linear pde . as in fernholz and karatzas ( 2008 ) , we do not assume the existence of an equivalent local martingale measure but merely the existence of a local martingale deflator .

1006.3224

in this paper we consider the quantile hedging problem when the underlying market does not have an equivalent martingale measure . instead , we assume that there exists a _ local martingale deflator _ ( a strict local martingale which when multiplied by the asset prices yields a positive local martingale ) . we characterize the value function as the smallest nonnegative viscosity supersolution of a fully non - linear partial differential equation . this resolves the open problem proposed in the final section of @xcite ; also see pages 61 and 62 of @xcite . our framework falls under the umbrella of the stochastic portfolio theory of fernholz and karatzas , see e.g. @xcite , @xcite , @xcite ; and the benchmark approach of platen @xcite . in this framework , the linear partial differential equation that the superhedging price satisfies does not have a unique solution ; see e.g. @xcite , @xcite , @xcite , and @xcite . similar phenomena occur when the asset prices have _ bubbles _ : an equivalent local martingale measure exists , but the asset prices under this measure are strict local martingales ; see e.g. @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite . a related series of papers @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite addressed the issue of bubbles in the context of stochastic volatility models . in particular , @xcite gave necessary and sufficient conditions for linear partial differential equations appearing in the context of stochastic volatility models to have a unique solution . in contrast , we show that the quantile hedging problem , which is equivalent to an optimal control problem , is the smallest nonnegative viscosity supersolution to a fully non - linear pde . as in the linear case , these pdes may not have a unique solution , and , therefore , an alternative characterization for the value function needs to be provided . recently , @xcite , @xcite , and @xcite also considered stochastic control problems in this framework . the first reference solves the classical utility maximization problem , the second one solves the optimal stopping problem , whereas the third one determines the optimal arbitrage under model uncertainty , which is equivalent to solving a zero - sum stochastic game . the structure of the paper is simple : in section [ eq : model ] , we formulate the problem . in this section we also discuss the implications of assuming the existence of a local martingale deflator . in section [ sec : quantile - hedging ] , we generalize the results of @xcite on quantile hedging , in particular the neyman - pearson lemma . we also prove other properties of the value function such as convexity . section [ sec : pde - characterization ] is where we give the pde characterization of the value function . we consider a financial market with a bond which is always equal to @xmath0 , and @xmath1 stocks @xmath2 which satisfy @xmath3 where @xmath4 is a @xmath1-dimensional brownian motion . following the set up in ( * ? ? ? * section 8) , we make the following assumption . [ as : fassp ] let @xmath5 and @xmath6 be continuous functions . set @xmath7 and @xmath8 , which we assume to be invertible for all @xmath9 . we also assume that has a weak solution that is unique in distribution for every initial value . let @xmath10 denote the probability space specified by a weak solution . another assumption we will impose is that @xmath11 where @xmath12 , @xmath13 . we will denote by @xmath14 the right - continuous version of the natural filtration generated by @xmath15 , and by @xmath16 the @xmath17-augmentation of the filtration @xmath18 . thanks to assumption [ as : fassp ] , the brownian motion @xmath19 of is adapted to @xmath16 ( see e.g. ( * ? ? ? * section 2 ) ) , every local martingale of @xmath18 has the martingale representation property , i.e. it can be represented as a stochastic integral , with respect to @xmath20 , of some @xmath16-progressively measurable integrand ( see e.g. the discussion on p.1185 in @xcite ) , the solution of takes values in the positive orthant , and the exponential local martingale @xmath21 the so - called _ deflator _ is well defined . we do not exclude the possibility that @xmath22 is a strict local martingale . let @xmath23 be the set of @xmath16-progressively measurable processes @xmath24 , which satisfies @xmath25 in which @xmath26 and @xmath27 with @xmath28 , @xmath29 , and @xmath30 . at time @xmath31 , an investor invests @xmath32 proportion of his wealth in the @xmath33 stock . the proportion @xmath34 gets invested in the bond . for each @xmath35 and initial wealth @xmath36 the associated wealth process will be denoted by @xmath37 . this process solves @xmath38 it can be easily seen that @xmath39 is a positive local martingale for any @xmath35 . let @xmath40 be a measurable function satisfying @xmath41<\infty,\ ] ] and define @xmath42 thanks to assumption [ as : fassp ] , we have that @xmath43 $ ] ; see e.g. ( * ? ? ? * section 10 ) . note that if @xmath44 has linear growth , then is satisfied since the process @xmath45 is a positive supermartingale . although , we do not assume the existence of equivalent local martingale measures , we assume the existence of a local martingale deflator . this is equivalent to the _ no - unbounded - profit - with - bounded - risk _ ( nupbr ) condition ; see ( * ? ? ? * theorem 4.12 ) . nupbr is defined as follows : a sequence @xmath46 of admissible portfolios is said to generate a upbr if @xmath47>0 $ ] . if no such sequence exists , then we say that nupbr holds ; see ( * ? ? ? * proposition 4.2 ) . in fact , the so - called _ no - free - lunch - with - vanishing - risk _ ( nflvr ) is equivalent to nupbr plus the classical _ no - arbitrage _ assumption . thus , in our setting ( since we assumed the existence of local martingale deflators ) , although arbitrages exist they remain on the level of cheap thrills " , which was coined by @xcite . ( note that the results of karatzas and kardaras @xcite also imply that one does not need nflvr for the portfolio optimization problem of an individual to be well - defined . one merely needs the nupbr condition to hold . ) the failure of no - arbitrage means that the money market is not an optimal investment and is dominated by other investments . it follows that a short position in the money market and long position in the dominating assets leads one to arbitrage . however , one can not scale the arbitrage and make an arbitrary profit because of the admissibility constraint , which requires the wealth to be positive . this is what is contained in nupbr , which holds in our setting . also , see @xcite , where these issues are further discussed . in this section , we develop new probabilistic tools to extend results of f@xmath48llmer and leukert @xcite on quantile hedging to settings where equivalent martingale measures need not exist . this is not only mathematically intriguing , but also economically important because it admits arbitrage in the market , which opens the door to the notion of optimal arbitrage , recently introduced in fernholz and karatzas @xcite . the tools in this section facilitate the discussion of quantile hedging under the context of optimal arbitrage , leading us to generalize the results of @xcite on this sort of probability - one outperformance . we will try to determine @xmath49 for @xmath50 $ ] . note that the set on which we take infimum in is nonempty . indeed , under the condition , there exists @xmath51 such that @xmath52 a.s . , where @xmath53 $ ] ; see e.g. ( * ? ? ? * section 10 ) . it follows that for any @xmath54 $ ] , @xmath55 also observe that @xmath56 when @xmath57 , observe that @xmath58 is equal to equation ( 6.1 ) of @xcite , the smallest relative amount to beat the market capitalization @xmath59 . clearly , @xmath60 by analogy with @xcite , we shall present a probabilistic characterization of @xmath61 . first , we will generalize the neyman - pearson lemma ( see e.g. ( * ? ? ? * theorem a.28 ) ) in the next result . [ t - pchar ] suppose that assumption [ as : fassp ] holds and @xmath44 satisfies . let @xmath62 satisfy @xmath63 then @xmath64.\ ] ] furthermore , if @xmath62 satisfies with equality and @xmath65 then @xmath66 satisfies with equality . under assumption [ as : fassp ] , since @xmath67 satisfies condition , it is replicable with initial capital @xmath68 $ ] ; see e.g. section 10.1 of @xcite . that is , there exists @xmath51 such that @xmath69 a.s . now if @xmath70 , we have @xmath71 . then it follows from that @xmath72 $ ] . now , take an arbitrary pair @xmath73 of initial capital and admissible portfolio that replicates @xmath74 with probability greater than or equal to @xmath75 , i.e. @xmath76 let @xmath77 satisfy @xmath78 and . to prove equality in , it is enough to show that @xmath79,\ ] ] which can be shown as follows : @xmath80 = \mathbb{e}[z(t ) y^{y_0 , \pi_0}(t)1_b ] + \mathbb{e}[z(t)y^{y_0 , \pi_0}(t)1_{b^c } ] \\ & \ge \mathbb{e } [ z(t ) g(x(t ) ) 1_b ] = \mathbb{e } [ z(t ) g(x(t ) ) 1_{a \cap b } ] + \mathbb{e } [ z(t ) g(x(t ) ) 1_{a^c \cap b } ] \\ & \ge \mathbb{e } [ z(t ) g(x(t ) ) 1_{a\cap b } ] + \mathbb{p}(a^c\cap b ) \ , { \rm ess\,inf}_{a^c \cap b } \{z(t ) g(x(t))\ } \\ & \ge \mathbb{e } [ z(t ) g(x(t ) ) 1_{a\cap b } ] + \mathbb{p}(a\cap b^c)\ , { \rm ess\,sup}_{a\cap b^c } \{z(t ) g(x(t))\ } \\ & \ge \mathbb{e } [ z(t ) g(x(t ) ) 1_{a \cap b } ] + \mathbb{e } [ z(t ) g(x(t ) ) 1_{a \cap b^c } ] \\ & = \mathbb{e } [ z(t ) g(x(t ) ) 1_{a } ] , \end{array}\ ] ] where in the fourth inequality we use the following two observations : first , @xmath81 . second , @xmath82 in which the second inequality follows from . let @xmath83 be the cumulative distribution function of @xmath84 and for any @xmath85 define @xmath86 and let @xmath87 denote @xmath88 ; that is , @xmath89 taking @xmath90 in lemma [ t - pchar ] , we see that is satisfied . it follows that @xmath91.\ ] ] on the other hand , taking @xmath92 , we see that is again satisfied . we therefore obtain @xmath93.\ ] ] the last two equalities imply the following relationship @xmath94 next , we will determine @xmath61 for @xmath95 when @xmath96 . [ c - pchar ] suppose assumption [ as : fassp ] holds . fix an @xmath97 $ ] . 1 . there exists @xmath77 satisfying with equality and . as a result , holds with equality . 2 . if @xmath98 , then letting @xmath99 we have @xmath100 * ( i ) * if there exists @xmath101 such that either @xmath102 or @xmath103 , then we can take @xmath104 or @xmath105 , thanks to and . in the rest of the proof we will assume that @xmath106 let @xmath107 be a brownian motion with respect to @xmath18 and define @xmath108 . let us define @xmath109 by @xmath110 . the function @xmath111 satisfies @xmath112 and @xmath113 . moreover , the function @xmath109 is continuous and nondecreasing . right continuity can be shown as follows : for @xmath114 @xmath115 the right continuity follows from observing that the last expression goes to zero as @xmath116 . one can show left continuity of @xmath109 in a similar fashion . since @xmath117 , thanks to the above properties of @xmath111 there exists @xmath118 satisfying @xmath119 . define @xmath120 . observe that @xmath121 and that @xmath66 satisfies . * ( ii ) * this follows immediately from ( 1 ) : @xmath122 \\ & = \mathbb{e}[z(t)g(x(t ) ) 1_{a_a } ] + \mathbb{e}[z(t)g(x(t ) ) 1_{\partial a_a \cap b_{b^ * } } ] \\ & = v(t , x , f(a- ) ) + a \mathbb{p}(\partial a_a \cap b_{b^ * } ) \\ & = v(t , x , f(a- ) ) + a(p - f(a- ) ) . \end{array}\ ] ] [ rem : rafl ] note that when @xmath123 is a martingale , using the neyman - pearson lemma , it was shown in @xcite that @xmath124 = { \mathbb{e}}[z(t ) g(x(t ) ) \varphi^*],\ ] ] where @xmath125 \bigg| \mathcal{f}_t \hbox { measurable } , \mathbb{e } [ \varphi ] \geq p\right\}.\ ] ] the randomized test function @xmath126 is not necessarily an indicator function . using lemma [ t - pchar ] and the fine structure of the filtration @xmath127 , we provide in proposition [ c - pchar ] another optimizer of which is an indicator function . [ c - convex ] suppose assumption [ as : fassp ] holds . then , the map @xmath128 is convex and continuous on the closed interval @xmath129 $ ] . hence , @xmath130 for all @xmath131 $ ] . by proposition [ c - pchar ] , for any @xmath54 $ ] there exists @xmath132 such that @xmath133 \le \mathbb{e}[z(t)g(x(t ) ) ] < \infty.\ ] ] then thanks to a theorem by ostroski ( see @xcite ) , to show the convexity it suffices to demonstrate the midpoint convexity @xmath134 denote @xmath135 . it follows from proposition [ c - pchar ] that there exist @xmath136 with @xmath137 satisfying , @xmath138 , \ i = 1,2,\ ] ] and @xmath139.\ ] ] by , @xmath140 which implies that @xmath141 \ge \mathbb{e } [ z(t)g(x ( t ) ) 1_{\tilde a \cap a_1^c } ] .\ ] ] as a result , @xmath142 - \mathbb{e}[z(t)g(x(t ) ) 1_{\tilde a } ] \ge \mathbb{e } [ z(t)g(x(t ) ) 1_{\tilde a } ] - \mathbb{e } [ z(t)g(x(t ) ) 1_{a_1 } ] , \end{array}\ ] ] which is equivalent to . now thanks to convexity , we immediately have that @xmath128 is continuous on @xmath143 . it remains to show that it is continuous from the left at @xmath144 ; but this is indeed true because @xmath145\\ & = & \mathbb{e}[z(t)g(x(t ) ) ] = v(t , x,1),\end{aligned}\ ] ] where the second equality is due to the dominated convergence theorem . [ e - bessel ] consider a market with a single stock , whose dynamics follow a three - dimensional bessel process , i.e. @xmath146 and let @xmath147 . in this case @xmath148 , which is the classical example for a strict local martingale ; see @xcite . on the other hand , @xmath149 is a martingale . thanks to proposition [ c - pchar ] there exists a set @xmath150 with @xmath151 such that @xmath152 = p x.\ ] ] in @xcite , the following result was proved when @xmath123 is a martingale . here , we generalize this result to the case where @xmath123 is only a local martingale . [ p - urep ] under assumption [ as : fassp ] @xmath153,\ ] ] where @xmath154 is defined in . thanks to proposition [ c - pchar ] there exists a set @xmath150 satisfying @xmath155 and such that @xmath156 $ ] . since @xmath157 , clearly @xmath158.\ ] ] for the other direction , it is enough to show that for any @xmath159 , we have @xmath160 \le \mathbb{e}[z(t ) g(x(t ) ) \varphi].\ ] ] indeed , since the left hand side is actually @xmath61 , we can get the desired result by taking infimum on both sides over @xmath161 . letting @xmath162 , we observe that @xmath163 - \mathbb{e}[z(t ) g(x(t ) ) 1_a ] \\ \quad = \mathbb{e}[z(t ) g(x(t ) ) \varphi 1_a ] + \mathbb{e}[z(t ) g(x(t ) ) \varphi 1_{a^c } ] - \mathbb{e}[z(t ) g(x(t))1_a ] \\ \quad = \mathbb{e}[z(t ) g(x(t ) ) \varphi 1_{a^c } ] - \mathbb{e}[z(t ) g(x(t ) ) 1_a ( 1 - \varphi ) ] \\ \quad \ge { \rm ess\,inf}_{a^c } \{z(t)g(x(t))\}\mathbb{e}[\varphi 1_{a^c}]-m\mathbb{e}[1_a ( 1-\varphi)]\\ \quad \ge m \mathbb{e}[\varphi 1_{a^c } ] - m \mathbb{e}[1_a ( 1-\varphi ) ] \quad \hbox{(by \eqref{eq : t - pchar1 } ) } \\ \quad = m\mathbb{e}[\varphi]-m\mathbb{e}[1_a]\geq 0 . \end{array}\ ] ] for @xmath54 $ ] , we introduce an additional controlled state variable @xmath165,\ ] ] where @xmath166 is a @xmath167progressively measurable @xmath168-valued process satisfying the integrability condition @xmath169 a.s . such that @xmath170 takes values in @xmath129 $ ] . we will denote the class of such processes by @xmath171 . note that @xmath171 is nonempty , as the constant control @xmath172 obviously lies in @xmath171 . the next result obtains an alternative representation for @xmath164 in terms of @xmath173 . [ t - control ] under assumption [ as : fassp ] , @xmath174<\infty.\ ] ] the finiteness follows from . define @xmath175 \bigg| \mathcal{f}_t \hbox { measurable } , \mathbb{e } [ \varphi ] = p\right\}.\ ] ] thanks to proposition [ c - pchar ] , there exists a set @xmath150 satisfying @xmath155 and such that @xmath176 \ge \inf_{\varphi \in \widetilde{{\mathcal{m } } } } \mathbb{e}[z(t ) g(x(t ) ) \varphi].\ ] ] since the opposite inequality follows immediately from proposition [ p - urep ] , we conclude that @xmath177.\ ] ] therefore , it is enough to show that @xmath178 satisfies @xmath179 the inclusion @xmath180 is clear . to show the other inclusion we will use the martingale representation theorem : for any @xmath181 there exists a @xmath167progressively measurable @xmath168-valued process @xmath182 satisfying @xmath183 a.s . such that @xmath184 = p + \int_0^t \psi(s ) ' d w(s),\ t\in[0,t].\ ] ] note that since @xmath185 takes values in @xmath129 $ ] , so does @xmath186 $ ] for all @xmath187 $ ] . then we see that @xmath186 $ ] satisfies with @xmath188 . we denote by @xmath189 the solution of starting from @xmath190 at time @xmath31 and by @xmath191 the solution of @xmath192 define the process @xmath193 by @xmath194 then we see from that @xmath195 satisfies @xmath196 we then introduce the value function @xmath197,\ ] ] where @xmath154 is defined in . note that the original value function @xmath164 can be written in terms of @xmath198 as @xmath199 . we also consider the legendre transform of @xmath198 with respect to the @xmath75 variable . to make the discussion clear , however , let us first extend the domain of the map @xmath200 from @xmath129 $ ] to the entire real line @xmath201 by setting @xmath202 then the legendre transform of @xmath198 with respect to @xmath75 is well - defined @xmath203}\{pq - u(t , x , p)\ } , & \text{if } ~q\ge 0 . \end{array}\right.\ ] ] > from proposition [ c - convex ] , we already know that @xmath200 is convex and continuous on @xmath129 $ ] . since @xmath204 , we see from and that @xmath200 is continuous on @xmath205 $ ] and lower semicontinuous on @xmath201 . moreover , considering that @xmath200 is increasing on @xmath129 $ ] , we conclude that @xmath200 is also convex on @xmath201 . now thanks to @xcite , the convexity and the lower semicontinuity of @xmath200 on @xmath201 imply that the double transform of @xmath198 is indeed equal to @xmath198 itself . that is , for any @xmath206\times(0,\infty)^d\times\mathbb{r}$ ] , @xmath207 where the second equality is a consequence of . in this section , we also consider the function @xmath208=\mathbb{e}[(q - z^{t , x,1}(t)g(x^{t , x}(t)))^+ ] , \ ] ] for any @xmath209\times(0,\infty)^d\times ( 0,\infty)$ ] . we will show that @xmath210 and derive various properties of @xmath211 . > from the definition of @xmath211 in , @xmath211 is the upper hedging price for the contingent claim @xmath212 , and potentially solves the linear pde @xmath213 this is not , however , a traditional black - scholes type equation because it is degenerate on the entire space @xmath214 . consider the following function @xmath215 which takes values in the space of @xmath216 matrices : @xmath217\ ] ] degeneracy can be seen by observing that @xmath218 is only positive semi - definite for all @xmath219 . or , one may observe degeneracy by noting that there are @xmath220 risky assets , @xmath221 and @xmath222 , with only @xmath1 independent sources of uncertainty , @xmath223 . as a result , the existence of classical solutions to can not be guaranteed by standard results for parabolic equations . indeed , under the setting of example [ e - bessel ] , we have @xmath224=(q - x)^+,\ ] ] which is not smooth . in this subsection , we will approximate @xmath211 by a sequence of smooth functions @xmath225 , constructed by elliptic regularization . we will then derive some properties of @xmath225 and investigate the relation between @xmath211 and @xmath225 . finally , we will show that @xmath226 , which validates the construction of @xmath225 . to perform elliptic regularization under our setting , we need to first introduce a product probability space . recall that we have been working on a probability space @xmath227 , given by a weak solution to the sde . now consider the sample space @xmath228;\mathbb{r})$ ] and the canonical process @xmath229 . let @xmath230 be the filtration generated by @xmath231 and @xmath232 be the wiener measure on @xmath233 . we then introduce the product probability space @xmath234 , with @xmath235 , @xmath236 and @xmath237 . for any @xmath238 , we write @xmath239 , where @xmath240 and @xmath241 . also , we denote by @xmath242 the expectation taken under @xmath234 . for any @xmath114 , introduce the process @xmath243 which satisfies the following dynamics @xmath244 then under the probability space @xmath234 , we have @xmath220 risky assets , the @xmath1 stocks @xmath245 and @xmath246 . define @xmath247,\ \bar{b}:= \left[\begin{array}{c } b_{1}\\ \vdots\\ b_{d}\\ \hline \end{array}\right],\ ] ] and @xmath248.\ ] ] since we assume that the matrix @xmath249 has full rank ( assumption [ as : fassp ] ) , @xmath250 has full rank by definition . it follows that @xmath251 is positive definite . now we can define the corresponding market price of risk under @xmath234 as @xmath252 , and the corresponding deflator @xmath253 under @xmath234 as the solution of @xmath254 where @xmath255 is a @xmath256-dimensional brownian motion . observe that @xmath257 \left[\begin{array}{c } b\\ \end{array}\right ] = \left[\begin{array}{c } \theta\\ 0 \end{array}\right].\ ] ] this implies that coincides with . thus , we conclude that @xmath258 . finally , let us introduce the function @xmath259,\ ] ] for any @xmath209\times(0,\infty)^d\times(0,\infty)$ ] . by and , we see that the processes @xmath260 and @xmath195 have the following relation @xmath261.\ ] ] it then follows from , the fact that @xmath258 , and the definition of @xmath225 that @xmath262.\ ] ] since @xmath251 is positive definite and continuous , it must satisfy the following ellipticity condition : for every compact set @xmath270 , there exists a positive constant @xmath271 such that @xmath272 for all @xmath273 and @xmath274 ; see e.g. ( * ? ? ? * lemma 3 ) . under assumption [ as : loc lips ] and , the smoothness of @xmath225 and the pde follow immediately from ( * ? ? ? * theorem 4.2 ) . finally , note that @xmath225 satisfies the boundary condition by definition . we will first compute @xmath280 , and then show that it is strictly increasing in @xmath281 from @xmath282 to @xmath0 . let @xmath283 and @xmath284 for @xmath285 . fix an arbitrary @xmath286 . for any @xmath287 , define @xmath288 note that by construction , @xmath289 and @xmath290 are disjoint , and @xmath291 . it follows that @xmath292\\ & = \frac{1}{\delta}\left\{\bar{\mathbb{e}}\left[\left((q+\delta)l_\varepsilon(t , t)-z^{t , x,1}(t)g(x^{t , x}(t))\right)\text{1}_{\widetilde{a}_{q+\delta}}\right]-\bar{\mathbb{e}}\left[\left ( ql_\varepsilon(t , t)-z^{t , x,1}(t)g(x^{t , x}(t))\right)\text{1}_{\widetilde{a}_q}\right]\right\}\\ & = \frac{1}{\delta}\bigg\{\bar{\mathbb{e}}\left[\left((q+\delta)l_\varepsilon(t , t)-z^{t , x,1}(t)g(x^{t , x}(t))\right)\text{1}_{\widetilde{a}_q}\right]+\bar{\mathbb{e}}\left[\left((q+\delta)l_\varepsilon(t , t)-z^{t , x,1}(t)g(x^{t , x}(t))\right)\text{1}_{e^\delta}\right]\\ & \hspace{0.5 in } -\bar{\mathbb{e}}\left[(ql_\varepsilon(t , t)-z^{t , x,1}(t)g(x^{t , x}(t)))\text{1}_{\widetilde{a}_q}\right]\bigg\}\\ & = \bar{\mathbb{e}}[l_\varepsilon(t , t)1_{\widetilde{a}_q}]+\frac{1}{\delta}\bar{\mathbb{e}}\left[\left((q+\delta)l_\varepsilon(t , t)-z^{t , x,1}(t)g(x^{t , x}(t))\right)\text{1}_{e^\delta}\right ] . \end{split}\ ] ] by the definition of @xmath290 , @xmath293\leq\frac{1}{\delta}\bar{\mathbb{e}}[\delta l_\varepsilon(t , t ) \text{1}_{e^\delta}]\\ & = & \bar{\mathbb{e}}[l_\varepsilon(t , t)1_{e^\delta}]\to 0,\ \text{as}\ \delta\downarrow 0 , \end{aligned}\ ] ] where we use the dominated convergence theorem . we therefore conclude that @xmath294}=\bar{\mathbb{e}}[l_\varepsilon(t , t)1_{\widetilde{a}_q}].\ ] ] thanks to the dominated convergence theorem again , we have @xmath295=0\ \text{and}\ \lim_{q\to\infty}\bar{\mathbb{e}}[l_\varepsilon(t , t)1_{\widetilde{a}_q}]=\bar{\mathbb{e}}[l_\varepsilon(t , t)]=1.\ ] ] it remains to prove that @xmath296 $ ] is strictly increasing in @xmath281 . note that it is enough to show that the event @xmath290 has positive probability for all @xmath287 . under the integrability condition , the deflator @xmath22 is strictly positive with probability 1 ; see e.g. ( * ? ? ? * section 6 ) . it follows from our assumptions on @xmath44 ( see and the line before it ) that @xmath297 this implies that @xmath298 now , from and the definitions @xmath290 and @xmath299 , we see that @xmath300 equals to the probability of the event @xmath301 thanks to fubini s theorem , this probability is strictly positive . * for any @xmath209\times ( 0,\infty)^d\times ( 0,\infty)$ ] , @xmath302 * for any compact subset @xmath303 , @xmath225 converges to @xmath211 uniformly on @xmath304\times(0,\infty)^d\times e$ ] . moreover , for any @xmath209\times ( 0,\infty)^d\times ( 0,\infty)$ ] @xmath305 * ( i ) * by , we observe that @xmath306}z^{t , x,1}(t)q^{t , x , q}_\varepsilon(t)\right]=\bar{\mathbb{e}}\left[\sup_{\varepsilon\in(0,1]}q\exp\left\{-\frac{1}{2}\varepsilon^2(t - t)+\varepsilon(b(t)-b(t))\right\}\right]\nonumber\\ & \le & q \bar{\mathbb{e}}\left[\sup_{\varepsilon\in(0,1]}\exp\left\{\varepsilon ( b(t)-b(t))\right\}\right]\nonumber\\ & \le & q \bar{\mathbb{e}}\left[\sup_{\varepsilon\in(0,1]}\exp\left\{\varepsilon ( b(t)-b(t))\right\}1_{\{b(t)-b(t)\ge 0\}}\right]+ q \bar{\mathbb{e}}\left[\sup_{\varepsilon\in(0,1]}\exp\left\{\varepsilon ( b(t)-b(t))\right\}1_{\{b(t)-b(t ) < 0\}}\right]\nonumber\\ & \le & q\bar{\mathbb{e}}\left[\exp\left\{b(t)-b(t)\right\}\right]+q = q\left(\exp\left\{\frac{1}{2}(t - t)\right\}+1\right ) < \infty.\end{aligned}\ ] ] then it follows from the dominated convergence theorem that @xmath307\nonumber\\ & = & \bar{\mathbb{e}}[(q - z^{t , x,1}(t)g(x^{t , x}(t)))^+]\nonumber\\ & = & \mathbb{e}[(q - z^{t , x,1}(t)g(x^{t , x}(t)))^+]= \widetilde{w}(t , x , q),\end{aligned}\ ] ] where the third equality is due to the fact that @xmath308 depends only on @xmath309 . * ( ii ) * from , , and the observation that @xmath310 for any @xmath311 , @xmath312\\ & = q\left[\left(1+\phi({\varepsilon}\sqrt{t - t})-\phi(-{\varepsilon}\sqrt{t - t})\right ) e^{{\varepsilon}^2 ( t - t)}-1\right]\\ & \le q\left[\left(1+\phi({\varepsilon}\sqrt{t})-\phi(-{\varepsilon}\sqrt{t})\right ) e^{{\varepsilon}^2 t}-1\right ] , \end{split}\ ] ] where @xmath313 is the cumulative distribution function of the standard normal distribution . note that the second line of follows from the inequality @xmath314 for @xmath315 ; this inequality holds because if @xmath316 , @xmath317 and if @xmath318 , @xmath319 . we can then conclude from that @xmath225 converges to @xmath211 uniformly on @xmath304\times(0,\infty)^d\times e$ ] , for any compact subset @xmath320 of @xmath277 . now , by lemma [ lem : smooth twe ] @xmath225 is continuous on @xmath321 . then as a result of uniform convergence , @xmath211 must be continuous on the same domain . noting that @xmath322 we see that follows from the continuity of @xmath211 and the uniform convergence of @xmath225 to @xmath211 on @xmath304\times(0,\infty)^d\times e$ ] for any compact subset @xmath320 of @xmath277 . by lemmas [ lem : smooth twe ] and [ lem : tw twe ] ( ii ) , the viscosity solution property follows as a direct application of ( * ? ? ? * proposition 2.3 ) . and the boundary condition holds trivially from the definition of @xmath211 . now we want to relate to @xmath211 to @xmath325 . given @xmath275\times(0,\infty)^d$ ] , recall the notation in section 3 : for any @xmath326 , @xmath327 ; also , @xmath83 again denotes the cumulative distribution function of @xmath308 . we first present another representation for @xmath211 as follows . now consider @xmath334 . set @xmath335 . observing that @xmath332 and @xmath336 are disjoint , and @xmath337 , we have @xmath338\\ & = & \mathbb{e}[(q - z^{t , x,1}(t)g(x^{t , x}(t)))\text{1}_{\bar{a}_q}]+\mathbb{e}[(q - z^{t , x,1}(t)g(x^{t , x}(t)))\text{1}_{f}]\\ & \le & \mathbb{e}[(q - z^{t , x,1}(t)g(x^{t , x}(t)))\text{1}_{\bar{a}_q } ] = \widetilde{w}(t , x , q),\end{aligned}\ ] ] where the inequality is due to the fact that @xmath339 on @xmath336 . given @xmath54 $ ] , there exists @xmath285 such that @xmath341 . we can take two nonnegative numbers @xmath342 and @xmath343 with @xmath344 such that @xmath345 observe that @xmath346 . plugging this into the first line of , we get @xmath347 also note from that @xmath348 plugging this back into , we obtain @xmath349 it then follows from and that @xmath350+\lambda_2[f(a-)q - u(t , x , f(a-))]\nonumber\\ & \le & \max\left\{f(a)q - u(t , x , f(a)),f(a-)q - u(t , x , f(a-))\right\}.\end{aligned}\ ] ] choose a sequence @xmath351 such that @xmath352 from the left as @xmath353 . thanks to proposition [ c - convex ] , @xmath200 is continuous on @xmath129 $ ] . we can therefore select a subsequence of @xmath354 ( without relabelling ) such that for any @xmath355 , @xmath356 it follows that for any @xmath355 @xmath357 which yields @xmath358 combining and , we obtain @xmath359}f(\delta)q - u(t , x , f(\delta))\le \sup_{\delta\ge 0}f(\delta)q - u(t , x , f(\delta)).\ ] ] this implies @xmath360}\{pq - u(t , x , p)\}\le\sup_{a\ge0}\{f(a)q - u(t , x , f(a))\}.\ ] ] since @xmath361 $ ] for all @xmath285 , the opposite inequality is trivial . we therefore conclude @xmath362}\{pq - u(t , x , p)\}=\sup_{a\ge0}\{f(a)q - u(t , x , f(a))\}.\ ] ] now , thanks to , we have @xmath363\nonumber\\ & = & \mathbb{e}[(q - z^{t , x,1}(t)g(x^{t , x}(t)))\text{1}_{\bar{a}_a}].\end{aligned}\ ] ] it follows from , and lemma [ lem : max = tw ] that @xmath364=\widetilde{w}(t , x , q).\ ] ] let us extend the domain of the map @xmath276 from @xmath277 to the entire real line @xmath201 by setting @xmath366 and @xmath367 for @xmath368 . in this subsection , we consider the legendre transform of @xmath225 with respect to the @xmath281 variable @xmath369 we will first show that @xmath370 is a classical solution to a nonlinear pde . then we will relate @xmath370 to @xmath198 and derive the viscosity supersolution property of @xmath198 . under assumption [ as : loc lips ] , we have that @xmath371 and satisfies the equation @xmath372+\inf_{a\in\mathbb{r}^d}\left((d_{xp}u_\varepsilon)'\sigma a + \frac{1}{2}|a|^2d_{pp}u_\varepsilon-\theta'ad_pu_\varepsilon\right ) + \inf_{b\in\mathbb{r}^d}\left(\frac{1}{2}|b|^2d_{pp}u_\varepsilon-\varepsilon d_pu_\varepsilon{\bf 1}'b\right),\ ] ] where @xmath373 , with the boundary condition @xmath374 moreover , @xmath375 is strictly convex in the @xmath75 variable for @xmath376 , with @xmath377 since from proposition [ prop : strict convex ] the function @xmath278 is strictly increasing on @xmath277 with @xmath378 its inverse function @xmath379 is well - defined on @xmath380 . moreover , considering that @xmath381 is smooth on @xmath365 , @xmath375 is smooth on @xmath382 and can be expressed as @xmath383 see e.g. @xcite . by direct calculations , we have @xmath384 in particular , we see that @xmath375 is strictly convex in @xmath75 for @xmath376 and satisfies . now by setting @xmath385 , we deduce from that @xmath386-\frac{1}{2}(|\theta|^2+\varepsilon^2)q^2d_{qq}\widetilde{w}_\varepsilon - qtr[\sigma\theta d_{xq}\widetilde{w}_\varepsilon]\\ & = \partial_t u_\varepsilon+\frac{1}{2}tr[\sigma \sigma ' d_{xx}u_\varepsilon]-\frac{1}{2d_{pp}u_\varepsilon}tr[\sigma \sigma ' ( d_{px}u_\varepsilon)(d_{px}u_\varepsilon)']-\frac{1}{2}(|\theta|^2+\varepsilon^2)\frac{(d_p u_\varepsilon)^2}{d_{pp}u_\varepsilon}\\ & \hspace{0.2in}+\frac{d_p u_\varepsilon}{d_{pp}u_\varepsilon}tr[\sigma\theta d_{px}u_\varepsilon]\\ & = \partial_t u_\varepsilon+\frac{1}{2}tr[\sigma \sigma ' d_{xx}u_\varepsilon]+\left((d_{xp}u_\varepsilon)'\sigma a^*+\frac{1}{2}|a^*|^2d_{pp}u_\varepsilon-\theta'a^*d_p u_\varepsilon\right)+\left(\frac{1}{2}|b^*|^2d_{pp}u_\varepsilon-\varepsilon d_pu_\varepsilon { \bf 1}'b^*\right)\\ & = \partial_t u_\varepsilon+\frac{1}{2}tr[\sigma \sigma ' d_{xx}u_\varepsilon]+\inf_{a\in\mathbb{r}^d}\left((d_{xp}u_\varepsilon)'\sigma a+\frac{1}{2}|a|^2d_{pp}u_\varepsilon-\theta'a d_p u_\varepsilon\right ) + \inf_{b\in\mathbb{r}^d}\left(\frac{1}{2}|b|^2d_{pp}u_\varepsilon-\varepsilon d_pu_\varepsilon{\bf 1}'b\right ) , \end{split}\ ] ] where the minimizers @xmath387 and @xmath388 are defined by @xmath389 finally , observe that for any @xmath376 , the maximum of @xmath390 is attained at @xmath391 . therefore , by @xmath392 as a consequence of lemma [ lem : tw twe ] ( ii ) , @xmath381 is continuous at @xmath394\times(0,\infty)^d\times(0,\infty)$ ] . this implies that @xmath395 is lower semicontinuous at @xmath396\times(0,\infty)^d\times\mathbb{r}$ ] . it follows that @xmath397 where the second equality follows from proposition [ prop : w = tw ] . before we state the supersolution property for @xmath198 , let us first introduce some notation . for any @xmath398 , define @xmath399 we also consider the lower semicontinuous envelope of @xmath400 @xmath401 observe that by definition , @xmath402 [ prop : viscosity u ] under assumption [ as : loc lips ] , @xmath198 is a lower semicontinuous viscosity supersolution to the equation @xmath403+g_*(x , d_{p}u , d_{pp}u , d_{xp}u),\ ] ] for @xmath404 , with the boundary condition @xmath405 note that the lower semicontinuity of @xmath198 is a consequence of lemma [ lem : u = liminfue ] , and the boundary condition comes from the fact that @xmath210 and the definition of @xmath211 as the following calculation demonstrates : @xmath406 let us now turn to the pde characterization inside the domain of @xmath198 . set @xmath407 . let @xmath185 be a smooth function such that @xmath408 attains a local minimum at @xmath409 and @xmath410 . note from that as @xmath411 , we must have @xmath412 . thus , the viscosity supersolution property is trivially satisfied . we therefore assume in the following that @xmath413 . let @xmath414 denote the right hand side of . observe from the calculation in that as @xmath415 , @xmath416-\frac{1}{2\gamma}tr[\sigma(x)\sigma(x)'\lambda\lambda']-\frac{\beta^2}{2\gamma}(|\theta(x)|^2+\varepsilon^2)+\frac{\beta}{\gamma}tr[\sigma(x)\theta(x)\lambda].\ ] ] this shows that @xmath417 is continuous at every @xmath418 as long as @xmath415 . it follows that for any @xmath419 with @xmath415 , we have @xmath420+\inf_{a\in\mathbb{r}^d}\left(\lambda'\sigma(x)a+\frac{1}{2}|a|^2\gamma-\theta(x)'a\beta\right).\ ] ] since we have @xmath421 from lemma [ lem : u = liminfue ] , we may use the same argument in ( * ? ? ? * proposition 2.3 ) and obtain that @xmath422 considering that @xmath423 , we see from and that this is the desired supersolution property . results similar to proposition [ prop : viscosity u ] were proved by @xcite , with stronger assumptions ( such as the existence of an equivalent martingale measure and the existence of a unique strong solution to ) , using the stochastic target formulation . here , we first observe that the legendre transform of @xmath198 is equal to @xmath211 and that @xmath211 can be approximated by @xmath225 , which is a classical solution to a linear pde and is strictly convex in @xmath281 ; then , we apply the legendre duality argument , as carried out in @xcite , to show that @xmath370 , the legendre transform of @xmath225 , is a classical solution to a nonlinear pde . finally , the stability of viscosity solutions leads to the viscosity supersolution property of @xmath198 . instead of relying on the legendre duality we could directly apply the dynamic programming principle of @xcite for weak solutions to the formulation in section [ sec : stoc - cont ] . the problem with this approach is that it requires some growth conditions on the coefficients of , which would rule out the possibility of arbitrage , the thing we are interested in and want to keep in the scope of our discussion . * let us consider the pde satisfied by the superhedging price @xmath424 : @xmath425 @xmath426 unless additional boundary conditions are specified , this pde may have multiple solutions . the role of additional boundary conditions in identifying @xmath427 as the unique solution of the above cauchy problem is discussed in section 4 of @xcite . also see @xcite for a similar discussion on boundary conditions for degenerate parabolic problems on bounded domains . + even when additional boundary conditions are specified , the growth of @xmath428 might lead to the loss of uniqueness ; see for example @xcite and theorem 4.8 of @xcite which give necessary and sufficient conditions on the uniqueness of cauchy problems in one and two dimensional setting in terms of the growth rate of its coefficients . we also note that @xcite develops necessary and sufficient conditions for uniqueness , in terms of the attainability of the boundary of the positive orthant by an auxiliary diffusion ( or , more generally , an auxiliary it@xmath429 ) process . * let @xmath430 be the difference of two solutions of - . then both @xmath431 and @xmath432 are solutions of ( along with its boundary conditions ) . as a result , whenever and has multiple solutions , so does the pde for the value function @xmath198 . we intend to characterize @xmath370 as the smallest solution among a particular class of functions , as specified below in proposition [ prop : charac ue ] . then , considering that @xmath433 from lemma [ lem : u = liminfue ] , this gives a characterization for @xmath198 . in determining @xmath198 numerically , one could use @xmath434 as a proxy for @xmath198 for small enough @xmath435 . additionally , we will characterize @xmath198 as the smallest nonnegative supersolution of in proposition [ prop : charac u ] . [ prop : charac ue ] suppose that assumption [ as : loc lips ] holds . let @xmath436\times(0,\infty)^d\times[0,1]\mapsto[0,\infty)$ ] be of class @xmath437 such that @xmath438 and @xmath439 is strictly convex in @xmath75 for @xmath376 with @xmath440 if @xmath441 satisfies the following partial differential inequality @xmath442+\inf_{a\in\mathbb{r}^d}\left((d_{xp}u)'\sigma a + \frac{1}{2}|a|^2d_{pp}u-\theta'ad_p u\right)+\inf_{b\in\mathbb{r}^d}\left(\frac{1}{2}|b|^2d_{pp}u-\varepsilon d_p u{\bf 1}'b\right),\ ] ] where @xmath373 , with the boundary condition @xmath443 then @xmath444 . let us extend the domain of the map @xmath445 from @xmath129 $ ] to the entire real line @xmath201 by setting @xmath446 for @xmath447 and @xmath448 for @xmath449 . then , we can define the legendre transform of @xmath441 with respect to the @xmath75 variable @xmath450}\{pq - u(t , x , p)\}\ge 0,\ \text{for}\ q\ge0 , \end{aligned}\ ] ] where the positivity comes from the condition @xmath438 . first , observe that since @xmath441 is nonnegative , we must have @xmath451}pq = q,\ \text{for any}\ q\ge0.\ ] ] next , we derive the boundary condition of @xmath452 from as @xmath453}\{pq - u(t , x , p)\}=\sup_{p\in[0,1]}\{pq - pg(x)\}=(q - g(x))^+.\ ] ] now , since @xmath439 is strictly convex in @xmath75 for @xmath376 and satisfies , we can express @xmath452 as @xmath454 where @xmath455 is the inverse function of @xmath456 . we can therefore compute the derivatives of @xmath457 in terms of those of @xmath458 , as carried out in . we can then perform the same calculation in ( but going backward ) , and deduce from that for any @xmath268 , @xmath459+\frac{1}{2}(|\theta|^2+\varepsilon^2)q^2d_{qq}w^u+qtr[\sigma\theta d_{xq}w^u].\ ] ] define the process @xmath460 for @xmath461 $ ] . observing that @xmath462 we conclude that @xmath463 is a martingale with @xmath464=q$ ] and @xmath465 for all @xmath461 $ ] , and satisfies the following sde @xmath466,\ \text{and}\ y(t)=q.\ ] ] thanks to the burkholder - davis - gundy inequality , there exists a constant @xmath467 such that @xmath468\le c\bar{\mathbb{e}}\left[\int_{t}^t\varepsilon^2y^2(s)ds\right]\\ = c\varepsilon^2\int_t^t q^2(e^{\varepsilon^2(s - t)}-1)+q^2\ ds\\ < \infty . \ ] ] for each @xmath355 , define the stopping time @xmath469 . by applying the product rule to the process @xmath470 and using , we get @xmath471,\ \text{for}\ n\in\mathbb{n}.\ ] ] now , observe from that @xmath472 for any @xmath461 $ ] . then from , we may apply the dominated convergence theorem to and obtain @xmath473\\ & = \bar{\mathbb{e}}[z^{t , x,1}(t)(q^{t , x , q}_\varepsilon(t)-g(x^{t , x}(t)))^+ ] = \widetilde{w}_\varepsilon(t , x , q ) , \end{split}\ ] ] where the first equality is due to . it follows that @xmath474 [ prop : charac u ] suppose assumption [ as : loc lips ] holds . let @xmath436\times(0,\infty)^d\times[0,1]\mapsto[0,\infty)$ ] be such that @xmath438 , @xmath439 is convex in @xmath75 , and the legendre transform of @xmath441 with respect to the @xmath75 variable , as defined in the proof of proposition [ prop : charac ue ] , is continuous on @xmath304\times(0,\infty)^d\times(0,\infty)$ ] . if @xmath441 is a lower semicontinuous viscosity supersolution to on @xmath382 with the boundary condition @xmath475 , then @xmath476 . let us denote by @xmath452 the legendre transform of @xmath441 with respect to @xmath75 . by the same argument in the proof of proposition [ prop : charac ue ] , we can show that , and are true . moreover , as demonstrated in ( * ? ? ? * section 4 ) , by using the supersolution property of @xmath441 we may show that @xmath452 is an upper semicontinuous viscosity subsolution on @xmath365 to the equation @xmath477 let @xmath478 be a nonnegative @xmath479 function supported in @xmath480,|(x , q)|\le 1\}$ ] with unit mass . without loss of generality , set @xmath481 for @xmath482\times(0,\infty)^d\times(0,\infty)\right)^c$ ] . then for any @xmath483 , define @xmath484 by definition , @xmath485 is @xmath479 . moreover , it can be shown that @xmath485 is a subsolution to on @xmath365 ; see e.g. ( 3.23)-(3.24 ) in ( * ? ? ? * section 3.3.2 ) and ( * ? ? ? * lemma 2.7 ) . set @xmath486 . by , we see from the definition of @xmath485 that @xmath487 also , the continuity of @xmath452 implies that @xmath488 for every @xmath209\times(0,\infty)^d\times(0,\infty)$ ] . considering that @xmath485 is a classical subsolution to , we have @xmath489,\ \text{for}\ n\in\mathbb{n},\ ] ] where @xmath490 . for each fixed @xmath355 , thanks to we may apply the dominated convergence theorem as we take the limit @xmath491 in . we thus get @xmath492.\ ] ] now by applying the reverse fatou s lemma ( see e.g. @xcite ) to , we have @xmath493\\ & \le&\mathbb{e}[z^{t , x,1}(t)w^u(t , x^{t , x}(t),q^{t , x , q}(t))]\\ & \le&\mathbb{e}[z^{t , x,1}(t)(q^{t , x , q}(t)-g(x^{t , x}(t)))^+]=w(t , x , q ) , \end{aligned}\ ] ] where the second inequality follows from the upper semicontinuity of @xmath452 and the third inequality is due to . finally , we conclude that @xmath494 where the first equality is guaranteed by the convexity and the lower semicontinuity of @xmath441 . one should note that @xmath434 and @xmath198 satisfy the assumptions stated in propositions [ prop : charac ue ] and [ prop : charac u ] , respectively . therefore , one can indeed see these results as pde characterizations of the functions @xmath434 and @xmath198 . in this paper , under the context where equivalent martingale measures need not exist , we discuss the quantile hedging problem and focus on the pde characterization for the minimum amount of initial capital required for quantile hedging . an interesting problem following this is the construction of the corresponding quantile hedging portfolio . we leave this problem open for future research .

we determine the structure of the fundamental group of the regular leaves of a closed singular riemannian foliation on a compact , simply connected riemannian manifold . we also study closed singular riemannian foliations whose leaves are homeomorphic to aspherical or to bieberbach manifolds . these foliations , which we call _ a - foliations _ and _ b - foliations _ , respectively , generalize isometric torus actions on riemannian manifolds . we apply our results to the classification problem of compact , simply connected riemannian- and-manifolds with positive or nonnegative sectional curvature . the author is part of sfb 878 : _ groups , geometry & actions _ , at the university of mnster . ] [ multiblock footnote omitted ]

1302.4593

the study of effective smooth torus actions on compact , smooth manifolds has a rich and long tradition in the theory of smooth transformation groups ( cf . @xcite ) . in riemannian geometry , starting with hsiang and kleiner s topological classification of compact riemannian @xmath0-manifolds of positive ( sectional ) curvature with an effective isometric circle action @xcite , isometric actions of tori have been successfully used to obtain classification results on compact riemannian manifolds with positive or nonnegative curvature and large isometry groups ( cf . @xcite ) . the present paper s main contribution is the observation that several results on smooth torus actions on compact smooth simply connected manifolds , and on isometric torus actions on compact simply connected riemannian manifolds with positive or nonnegative curvature , hold under less restrictive conditions which do not involve the existence of a group action . indeed , many of these results do not hold because of the presence of a torus action , but rather because the orbit decomposition of the manifold has the structure of a _ singular riemannian foliation _ whose leaves are diffeomorphic to flat tori of possibly different dimensions . to make this statement precise , we introduce a special class of singular riemannian foliations , _ b - foliations _ , which generalize isometric torus actions on complete riemannian manifolds . roughly speaking , a b - foliation @xmath2 is a partition of a complete riemannian manifold @xmath3 into connected closed submanifolds , called the _ leaves _ of @xmath4 , all of which are homeomorphic to some flat manifold and are at a constant distance from each other . more generally , b - foliations are a particular instance of singular riemannian foliations whose leaves are homeomorphic to some closed aspherical manifold . we will call such singular riemannian foliations _ a - foliations_. the fact that a - foliations are more general than isometric torus actions is clear for several reasons . on the one hand , the leaves need not be tori . on the other hand , even when the leaves of an a - foliation @xmath2 on a complete riemannian manifold @xmath3 are diffeomorphic to standard tori , the foliation may not be homogeneous , i.e. there might not be a global torus action on @xmath3 inducing the given singular riemannian foliation @xmath4 . this occurs , for example , when the distribution of the tangent spaces of the torus leaves is not orientable . moreover , we assume the leaves of an a - foliation to be homeomorphic to aspherical manifolds , but not necessarily diffeomorphic . therefore , our main results hold when the leaves carry exotic smooth structures , e.g. in the case of exotic tori . nevertheless , we do not know of any non - trivial examples of singular riemannian foliations whose leaves are exotic tori . it would be interesting to find b - foliations by exotic tori on simply connected manifolds and , in particular , on spheres . in this paper we focus our attention on a- and b - foliations on compact simply connected riemannian manifolds . although every aspherical manifold can appear as the regular leaf of an a - foliation , our first result implies that the simply connected case is considerably more rigid . [ t : fund_gp ] let @xmath2 be a closed singular riemannian foliation on a compact , simply connected riemannian manifold @xmath3 . if @xmath5 is a regular leaf of @xmath4 , then @xmath6 is isomorphic to @xmath7 , where @xmath8 is abelian and @xmath9 is a finite @xmath10 step nilpotent @xmath10-group . in particular , we have the following corollary : [ c : torus_leaves]let @xmath2 be an a - foliation on a compact riemannian manifold @xmath3 . if @xmath3 is simply connected , then the regular leaves are homeomorphic to tori . observe that @xmath10 step nilpotent @xmath10-groups already appear as fundamental groups of regular leaves of codimension one singular riemannian foliations and hence can not be avoided in the statement of theorem a ( cf . * table 1.4 ) ) . this occurs , for example , in the cohomogeneity one @xmath11 action on @xmath12 . with theorem [ t : fund_gp ] and corollary [ c : torus_leaves ] in place , we extend to a- and b - foliations several basic results on smooth effective torus actions on smooth compact manifolds ( cf . we prove : [ t : kobayashi ] let @xmath2 be a b - foliation on a compact riemannian manifold @xmath3 and let @xmath13 denote the stratum of @xmath14-dimensional leaves . then @xmath15 . [ t : codim_1_bfol_sc ] the only codimension @xmath16 a - foliations on compact , simply connected riemannian manifolds are the homogeneous singular riemannian foliations @xmath17 and @xmath18 . [ t : bfol_codim2 ] let @xmath19 be a codimension @xmath10 a - foliation on a compact , simply connected riemannian manifold @xmath20 with @xmath21 . then either @xmath22 and @xmath4 is given by a weighted hopf action , or the following hold : + 1 . the leaf space @xmath23 is homeomorphic to a @xmath10-disk , the interior of @xmath24 is smooth , and the boundary @xmath25 consists of at least @xmath26 totally geodesic segments meeting at an angle of @xmath27 . let @xmath28 be a generic leaf and @xmath29 a singular leaf . then there is a submersion @xmath30 , with fiber @xmath31 if @xmath29 belongs to a geodesic in @xmath25 , or with fiber @xmath32 , if @xmath29 belongs to a vertex of @xmath25 . another generalization of isometric torus actions are the so - called _ f - structures _ , introduced by cheeger and gromov @xcite . these structures are , roughly speaking , generalized local torus actions and play a central role in the cheeger - fukaya - gromov theory of collapsed riemannian manifolds with bounded sectional curvature ( cf . the so - called _ pure _ f - structures ( see @xcite ) give rise to b - foliations with leaves diffeomorphic to flat manifolds . recall that , by work of cheeger , fukaya and gromov , there exists a constant @xmath33 , such that any compact riemannian manifold @xmath34 with curvature @xmath35 , @xmath36 and @xmath37 admits a pure f - structure ( see @xcite and references therein ) . therefore , @xmath34 is b - foliated . although b - foliations resemble f - structures , the two concepts are independent . a b - foliation on a riemannian manifold does not necessarily correspond to an f - structure since , for instance , b - foliations with exotic torus leaves can not be generated by f - structures . on the other hand , certain f - structures among those that are not pure do not generate a b - foliation . + as an application of our results , we extend work in @xcite on positively and nonnegatively curved compact , simply connected riemannian manifolds with large effective isometric torus actions to the case of a - foliations . recall that a riemannian manifold @xmath38 is said to have _ quasi - positive curvature _ if @xmath38 has nonnegative ( sectional ) curvature and a point with strictly positive curvature . [ t : qpc_d4_codim2 ] let @xmath39 be a compact , simply connected riemannian @xmath26-manifold with quasi - positive curvature supporting a codimension @xmath10 a - foliation . if @xmath40 , then @xmath41 is diffeomorphic to @xmath12 or @xmath42 . + 2 . if @xmath43 , then @xmath44 is diffeomorphic to @xmath45 . [ t : nnc_d4_codim2 ] let @xmath39 be a compact , simply connected riemannian @xmath26-manifold with nonnegative curvature and a codimension @xmath10 a - foliation . + 1 . if @xmath40 , then @xmath41 is diffeomorphic to @xmath12 , @xmath42 or homeomorphic to @xmath46 or @xmath47 . + 2 . if @xmath43 , then @xmath44 is diffeomorphic to @xmath45 or to one of the two @xmath48-bundles over @xmath49 . [ t : d4_s1fol ] let @xmath3 be a compact , simply connected riemannian @xmath0-manifold with a singular riemannian foliation by circles . then the foliation is induced by a smooth circle action and the following hold : + 1 . if @xmath3 has positive curvature , then @xmath3 is diffeomorphic to @xmath12 or @xmath42 . + 2 . if @xmath3 has nonnegative curvature , then @xmath3 is diffeomorphic to @xmath12 , @xmath42 , @xmath46 or @xmath47 . + in part ( 1 ) of theorem [ t : nnc_d4_codim2 ] the classification is only up to homeomorphism because we do not know whether the a - foliation is induced by a smooth torus action . we call a b - foliation _ euclidean _ if its regular leaves are flat with the induced riemannian metric and define the _ euclidean rank _ of a riemannian manifold @xmath38 as the maximum dimension of euclidean b - foliations on @xmath3 compatible with the fixed metric @xmath50 . this invariant generalizes the symmetry rank of @xmath38 ( cf . it follows from otsuki s lemma ( * ? ? ? * lemma 3.3 , p. 224 ) ( cf . also an argument due to wilking found in @xcite ) that the euclidean rank of a compact , quasi - positively curved riemannian @xmath26-manifold is less than or equal to @xmath51 . in dimension @xmath52 , it is easy to show , following the comparison arguments in the proof of theorem [ t : nnc_d4_codim2 ] , that the euclidean rank of a compact , simply connected riemannian @xmath26-manifold of nonnegative curvature is bounded above by @xmath53 . these bounds coincide with the corresponding bounds for the symmetry rank ( cf . @xcite ) . our paper is structured as follows . in section [ basic ] we recall some basic facts on singular riemannian foliations . in section [ s : b_fol ] we introduce a- and b - foliations and show that the infinitesimal foliation at any point of a manifold with an a- or a b - foliation is also an a- or a b - foliation , respectively . sections [ s : fundgp_leaves ] through [ s : codim_2 ] contain the proofs of theorems [ t : fund_gp ] through [ t : bfol_codim2 ] . section [ s : curv_1 ] contains the proof of theorems [ t : qpc_d4_codim2 ] and [ t : nnc_d4_codim2 ] . finally , in section [ s : curv_2 ] we prove theorem [ t : d4_s1fol ] . throughout our paper we will assume all manifolds to be connected and without boundary , unless stated otherwise . the second named author thanks xiaoyang chen for some initial conversations . both authors wish to acknowledge the hospitality of the mathematisches forschungsinstitut oberwolfach , where the work presented in this paper was initiated . the authors also thank wolfgang ziller for comments on a first version of this paper , alexander lytchak for suggestions which led to the proofs of theorems [ t : fund_gp ] and [ t : fibration - flat ] , and the referee for suggesting improvements to our original results . in this section we collect some background material on singular riemannian foliations . we refer the reader to @xcite for further results on the theory . a _ transnormal system _ @xmath54 on a complete riemannian manifold @xmath3 is a decomposition of @xmath3 into smooth , complete , injectively immersed connected submanifolds , called _ leaves _ , such that every geodesic emanating perpendicularly to one leaf remains perpendicular to all leaves . a transnormal system @xmath54 is called a _ singular riemannian foliation _ if there are smooth vector fields @xmath55 on @xmath3 such that , for each point @xmath56 , the tangent space @xmath57 to the leaf @xmath58 through @xmath59 is given as the span of the vectors @xmath60 . we will call the quotient space @xmath61 the _ leaf space _ , and will also denote it by @xmath62 . we will let @xmath63 be the leaf projection map . the pair @xmath2 will denote a singular riemannian foliation @xmath4 on a complete riemannian manifold @xmath3 . slightly abusing notation , we will also refer to the pair @xmath2 as a singular riemannian foliation . a singular riemannian foliation @xmath4 will be called _ closed _ if all its leaves are closed in @xmath3 ; the foliation will be called _ locally closed _ at @xmath64 if , for some neighborhood @xmath65 of @xmath66 , the restriction of @xmath4 to @xmath65 is closed , i.e. connected components of the intersection of the leaves of @xmath4 with @xmath65 are closed in @xmath65 . if @xmath4 is locally closed at @xmath66 , then the local quotient @xmath67 is a well defined alexandrov space of curvature locally bounded from below . similarly , if @xmath4 is closed , then the quotient space @xmath68 is an alexandrov space of curvature locally bounded below . we will henceforth only consider closed singular riemannian foliations as group actions will appear throughout our work , let us fix some notation before proceeding . given a lie group @xmath69 acting ( on the left ) on a smooth manifold @xmath3 , we denote by @xmath70 the _ isotropy group _ at @xmath56 and by @xmath71 the _ orbit _ of @xmath59 . the _ ineffective kernel _ of the action is the subgroup @xmath72 . we say that @xmath69 acts _ effectively _ on @xmath3 if @xmath73 is trivial . the action is _ free _ if every isotropy group is trivial . given a subset @xmath74 , we will denote its image in @xmath75 under the orbit projection map by @xmath76 . when convenient , we will also denote the orbit space @xmath75 by @xmath62 . perhaps the most familiar example of a singular riemannian foliation is the one induced by an ( effective ) isometric action of a lie group @xmath69 on a complete riemannian manifold @xmath3 . in this case , the foliation is given by the orbits of the action , and we say that the foliation is a _ homogeneous foliation_. if @xmath69 is compact , then the foliation is closed , and it is locally closed if and only if all the slice representations @xmath77 have compact image . we will sometimes denote a homogeneous foliation , given by the action of a lie group @xmath69 , by @xmath78 , provided the @xmath69-action is understood . let @xmath3 be a complete riemannian manifold with a closed singular riemannian foliation @xmath54 . the _ dimension _ of @xmath54 , denoted by @xmath79 , is the maximal dimension of its leaves . the _ codimension _ of @xmath54 is , by definition , @xmath80 for @xmath81 , define @xmath82 every connected component @xmath83 of the set @xmath84 is an embedded ( possibly non complete ) submanifold of @xmath3 and the restriction of @xmath54 to @xmath83 is a riemannian foliation . given @xmath56 , let @xmath85 be the connected component of @xmath84 through @xmath59 , where @xmath86 . we will refer to the decomposition of @xmath3 into the submanifolds @xmath85 as the _ canonical stratification _ of @xmath3 . the subset @xmath87 is open , dense and connected in @xmath3 ; it is called the _ regular stratum _ of @xmath3 . it will be denoted by @xmath88 and its points will be called _ regular points_. if @xmath89 we say that the foliation is _ regular_. all other strata @xmath90 have codimension at least @xmath10 in @xmath3 and are called _ singular strata_. for any singular stratum @xmath85 , we have @xmath91 let @xmath3 be a complete riemannian manifold with a closed singular riemannian foliation @xmath54 . given a point @xmath56 and some small @xmath92 , let @xmath93 be a _ slice _ through @xmath59 , where @xmath94 is the distance ball of radius @xmath95 around @xmath59 . the foliation @xmath54 induces a foliation @xmath96 on @xmath97 by letting the leaves of @xmath96 be the connected components of the intersection between @xmath97 and the leaves of @xmath54 . the foliation @xmath98 may not be a singular riemannian foliation with respect to the induced metric on @xmath97 . nevertheless , the _ pull - back _ foliation @xmath99 is a singular riemannian foliation on @xmath100 equipped with the euclidean metric ( cf . * proposition 6.5 ) ) , and it is invariant under homotheties fixing the origin ( cf . * lemma 6.2 ) ) . in particular , it is possible to extend @xmath101 to all of @xmath102 , giving rise to a singular riemannian foliation @xmath103 called the _ infinitesimal foliation _ of @xmath54 at @xmath59 . notice that @xmath104 is always a leaf of the infinitesimal foliation @xmath105 . by definition , leaves stay at a constant distance from each other , in particular every leaf stays in some distance sphere around the origin , and it makes sense to consider the infinitesimal foliation restricted to the unit sphere . since the infinitesimal foliation is invariant under homothetic transformations , it can be reconstructed from its own restriction to the unit sphere . taking this into account , we will sometimes refer to @xmath106 also as the infinitesimal foliation at @xmath59 and shall write @xmath107 . given two points @xmath108 in some leaf @xmath5 , the corresponding infinitesimal foliations @xmath109 , @xmath110 are foliated isometric , in the sense that there is a ( non - canonical ) linear isometry @xmath111 preserving the foliation . moreover , these foliations can be glued together to give a foliation on @xmath112 in the following sense : if one identifies @xmath112 via the normal exponential map with @xmath113 , the boundary of an @xmath95-tubular neighborhood of @xmath5 , then the intersections of leaves in @xmath4 with with @xmath114 are exactly the leaves in @xmath115 . in particular , if @xmath116 is a leaf in @xmath117 , then @xmath116 is a union of infinitesimal leaves . moreover , if @xmath118 and @xmath119 can be written as @xmath120 , @xmath121 , then the connected components of a fiber of @xmath59 under the metric projection @xmath122 ( which is a submersion , cf . * lemma 6.1 ) ) are given by @xmath123 , where @xmath124 are diffeomorphic to the infinitesimal leaf of @xmath114 passing through @xmath125 . therefore , there is a fibration @xmath126 for some finite cover @xmath127 . [ r : space_dir ] let @xmath2 be a closed singular riemannian foliation . as recalled in section [ ss : srf ] , the leaf space @xmath62 is an alexandrov space of curvature locally bounded below . let us quickly recall the procedure to compute the space of directions @xmath128 at a point @xmath129 . let @xmath56 be a point in the preimage of @xmath130 and let @xmath131 be the quotient of the infinitesimal foliation @xmath132 . the fundamental group @xmath133 acts on @xmath134 by isometries via the so - called _ holonomy action _ and @xmath128 is isometric to @xmath135 . given @xmath121 with image @xmath136 , let @xmath137 be the subgroup of @xmath133 fixing @xmath138 . then , in fibration , the cover @xmath139 is @xmath140 , where @xmath141 is the universal cover of @xmath58 . let @xmath78 be a homogeneous foliation . given a point @xmath56 , the connected component @xmath142 of the isotropy group @xmath143 acts on @xmath144 by isometries , via the so - called _ slice representation_. in this case , the infinitesimal foliation @xmath105 is the homogeneous foliation given by the orbits of @xmath142 on @xmath144 . given @xmath145 close to @xmath59 , with isotropy @xmath146 , the projection is the projection @xmath147 where @xmath148 is a cover of the orbit @xmath149 though @xmath59 . we conclude this section by recalling the main properties of the so - called _ molino bundle _ * proposition 4.1 ) ) . we let @xmath2 be a closed singular riemannian foliation of codimension @xmath150 on a compact riemannian manifold @xmath3 . since the foliation on the regular stratum @xmath88 is regular , there exists a principal @xmath151-bundle @xmath152 , called the _ molino bundle _ , and a foliation @xmath153 such that the leaves of @xmath154 are galois covers of the leaves of @xmath4 . moreover , the leaves of @xmath154 are actually diffeomorphic to those of @xmath4 on an open dense set . in addition , since @xmath4 is closed , the leaves of @xmath154 are given by the fibers of a submersion @xmath155 , where @xmath156 is the frame bundle of the orbifold @xmath157 . in particular , @xmath156 is a manifold with an almost free smooth @xmath151-action and @xmath158 is @xmath151-equivariant . let @xmath159 and @xmath160 be the borel constructions of @xmath161 and @xmath156 , respectively . then @xmath162 is homotopy equivalent to @xmath88 and @xmath158 induces a fibration @xmath163 with the same fibers as @xmath155 . furthermore , the space @xmath164 coincides with haefliger s classifying space of the orbifold @xmath157 ( cf . therefore , up to homotopy , there is a fibration @xmath165 where @xmath5 is a regular leaf of @xmath4 . we now introduce _ a - foliations _ and _ b - foliations _ , which are the main object of study in our paper . a closed singular riemannian foliation @xmath2 is an _ a - foliation _ if every leaf is an aspherical manifold . a closed singular riemannian foliation @xmath2 is a _ b - foliation _ if every leaf is homeomorphic to some bieberbach manifold . recall that a _ bieberbach manifold _ is a manifold diffeomorphic to @xmath166 , where @xmath69 is a discrete group of euclidean isometries acting freely and cocompactly on @xmath167 . these groups are called _ bieberbach groups_. abstractly , bieberbach groups can be characterized as torsion - free groups with a normal finite index abelian subgroup ( cf . in particular , every subgroup of a bieberbach group is a bieberbach group . every bieberbach manifold is compact , has no boundary and admits a flat riemannian metric . [ r : borel_conj ] we shall use the fact that any aspherical manifold @xmath168 with fundamental group isomorphic to a bieberbach group @xmath69 must be homeomorphic to a bieberbach manifold . to see this , observe first that @xmath168 must be homotopy equivalent to @xmath166 , since both are models for @xmath169 . it then follows from the solution of the borel conjecture for flat manifolds ( cf . * section 4 ) and @xcite ) that @xmath168 and @xmath170 must be homeomorphic . the manifolds @xmath168 and @xmath166 may not be diffeomorphic , as illustrated by the existence of exotic tori , which appear already in dimension @xmath1 ( cf . @xcite ) . [ r : borel_conj ] in the preceding definition we do not assume that the leaves are flat with the induced riemannian metric . every isometric torus action on a complete riemannian manifold induces a ( homogeneous ) b - foliation . bundles whose fibers are homeomorphic to bieberbach manifolds are also examples of b - foliations . the simplest way to construct non - homogeneous b - foliations with regular leaves homeomorphic to tori is to take the riemannian product of a complete riemannian manifold and an exotic torus . as the leaves are exotic tori , they can not correspond to the orbits of an isometric torus action . one can also construct non - homogeneous b - foliations whose leaves are diffeomorphic to flat tori in the following way . let @xmath24 be a compact smooth manifold with non - trivial fundamental group , let @xmath171 be a standard @xmath26-dimensional torus , and @xmath172 a homomorphism . let @xmath173 be the universal cover of @xmath24 and let @xmath174 act diagonally on the product @xmath175 . this action is free and , taking the quotient , we obtain a fiber bundle @xmath176 the total space @xmath177 is b - foliated by the fibers of the bundle . if the b - foliation is homogeneous , then the bundle is principal , and the structure group reduces to a subgroup of @xmath178 . in particular , the structure group is contained in the identity component @xmath179 of @xmath180 . thus , to construct a non - homogeneous b - foliation , it is enough to consider a homomorphism @xmath172 whose image is not entirely contained in @xmath179 . as a concrete example , if @xmath181 , @xmath182 , and @xmath183 is an orientation reversing diffeomorphism , then @xmath184 is a klein bottle @xmath73 , and the foliation is given by the fibers of the submersion @xmath185 . in the case of a- and b - foliations , the total space and the base in fibration are homeomorphic , respectively , to aspherical and bieberbach manifolds . the following general result shows that these classes of manifolds are well - behaved with respect to fibrations . [ t : fibration - flat ] let @xmath186 , @xmath3 , and @xmath168 be topological manifolds and let @xmath187 be a fibration . 1 . if @xmath3 is aspherical , then @xmath186 and @xmath168 are aspherical . 2 . if @xmath3 is homeomorphic to a bieberbach manifold , then @xmath186 and @xmath168 are homeomorphic to bieberbach manifolds . we first prove part ( 1 ) . consider the fibration between the universal covers @xmath188 , with fiber @xmath137 . since @xmath189 is contractible and @xmath190 is simply connected , we can apply the serre spectral sequence with integral coefficients , and from it we obtain that @xmath137 and @xmath190 are contractible . in fact , if @xmath191 has cohomological dimension @xmath192 , and @xmath193 has cohomological dimension @xmath194 , then @xmath195 has cohomological dimension @xmath196 and this has to be @xmath14 . in particular , @xmath190 is aspherical . therefore , @xmath168 is aspherical and , from the long exact sequence in homotopy , so is @xmath186 . now we prove part ( 2 ) . by remark [ r : borel_conj ] it suffices to show that @xmath186 and @xmath168 are aspherical and @xmath197 and @xmath198 are bieberbach groups . since @xmath3 is homeomorphic to a bieberbach manifold , it follows from part ( 1 ) that @xmath186 and @xmath168 are aspherical . from the long exact sequence of the fibration , we have @xmath199 where @xmath200 is a bieberbach group , i.e. a torsion free group with a finite index normal abelian subgroup . since @xmath197 is a subgroup of a bieberbach group , it is again a bieberbach group and therefore @xmath186 is homeomorphic to a flat manifold . we now prove that @xmath198 is a bieberbach group . first , we show that @xmath198 is torsion free . suppose this is not the case . then there is a finite cyclic subgroup @xmath201 acting freely on the contractible manifold @xmath190 . it follows that @xmath202 is a @xmath203 , which contradicts the fact that @xmath203 has infinite cohomological dimension . finally , let us show that @xmath198 contains a finite index normal abelian subgroup . since @xmath200 is a bieberbach group , there exists a finite index normal subgroup @xmath204 . the image of @xmath205 in @xmath198 is a finitely generated normal torsion free abelian group @xmath8 . since the map @xmath206 is surjective , @xmath8 has finite index in @xmath198 . therefore @xmath198 is a bieberbach group , and therefore @xmath168 is homeomorphic to a bieberbach manifold . theorem [ t : fibration - flat ] can be applied to fibration to obtain the following corollaries . let @xmath2 be a closed singular riemannian foliation . 1 . if a regular leaf of @xmath2 is aspherical , then @xmath2 is an a - foliation . 2 . if a regular leaf of @xmath2 is homeomorphic to a flat manifold , then @xmath2 is a b - foliation . [ c : inf_b_fol ] 1 . the infinitesimal foliations of an a - foliation are again a - foliations . the infinitesimal foliations of a b - foliation are again b - foliations . since b - foliations generalize torus actions , it is natural to ask when such a foliation is homogeneous . we will now show that closed singular riemannian foliations by circles on compact , simply connected riemannian manifolds are homogeneous , answering the simplest instance of this question . we first prove the following general lemma . [ l : orient_fol ] let @xmath2 be a closed , singular riemannian foliation on a compact , simply connected riemannian manifold @xmath3 . then the foliation @xmath2 restricted to the regular part @xmath88 is orientable . by ( * proposition 3.7 ) , the leaf space @xmath157 is an orbifold . lytchak showed in ( * ? ? ? * corollary 5.3 ) that @xmath207 , i.e. the classifying space @xmath24 of @xmath157 is simply connected . in particular , @xmath24 is orientable , which implies the result . [ t : homogeneity ] let @xmath2 be a closed , singular riemannian foliation on a compact , simply connected riemannian manifold . if the regular leaves of the foliation are circles , then the foliation is homogeneous . by lemma [ l : orient_fol ] , the foliation @xmath4 restricted to the regular part @xmath88 is orientable . hence @xmath208 is given by a circle action . since the singular strata of the foliation are smooth closed embedded submanifolds , the action can be extended to the singular strata by radially extending it on small tubular neighborhoods around each component of the singular strata . in the subsequent sections we will assume all manifolds to be compact , unless stated otherwise . let @xmath2 be a closed singular riemannian foliation on a compact , simply connected riemannian manifold . observe that @xmath3 remains simply connected if we discard all the singular strata of codimension at least @xmath209 . therefore , we may assume that @xmath210 where @xmath211 are the connected components of codimension @xmath10 of the singular stratum and @xmath88 is the regular stratum of @xmath4 . we will assume that @xmath212 is the empty set if there are no codimension @xmath10 strata . + for each @xmath213 , let @xmath214 be a small tubular neighborhood of @xmath211 with foot - point projection @xmath215 . the restriction of @xmath216 to @xmath217 is a circle bundle . the fibers of this circle bundle define a free homotopy class @xmath218 $ ] of loops in @xmath88 . moreover , if @xmath219 , then @xmath220 is entirely contained in @xmath214 and the restriction of @xmath216 to @xmath220 is a circle bundle @xmath221 let us fix a regular leaf @xmath28 in @xmath88 and a point @xmath222 . for @xmath223 , fix a horizontal curve @xmath224\to m_0 $ ] from @xmath225 to some @xmath219 . this curve induces an isomorphism @xmath226 , given by @xmath227 , where @xmath228 is the only horizontal curve ending at @xmath59 whose projection to @xmath157 coincides with the projection of @xmath229 . let @xmath230 be a representative of the free homotopy class @xmath218 $ ] defined in the preceding paragraph , passing through @xmath231 . the element @xmath232 is uniquely determined , up to a sign , by @xmath229 and @xmath230 . let @xmath233 be the group generated by the elements @xmath234 . if @xmath235\to m_0 $ ] is a second horizontal curve from @xmath225 to @xmath236 , a different homeomorphism @xmath237 is induced , and we obtain a different element @xmath238 . letting @xmath239 denote the inclusion of @xmath28 in @xmath88 , the elements @xmath240 and @xmath241 in @xmath242 are conjugate by an element of @xmath242 . + recall from section [ ss : molino ] that , up to homotopy , there is a fibration @xmath243 where @xmath244 is the inclusion and @xmath24 is haefliger s classifying space of the orbifold @xmath157 . in particular , by definition , @xmath245 for all @xmath246 . let @xmath137 be the image of the boundary map @xmath247 in the homotopy exact sequence of fibration . there is an exact sequence @xmath248 to prove theorem [ t : fund_gp ] , we will proceed in three steps : * @xmath249 maps surjectively onto @xmath242 under @xmath250 . * @xmath73 splits as a product @xmath251 , where @xmath252 is abelian and @xmath9 is a finite @xmath10 step nilpotent @xmath10-group . * @xmath253 is central . by the first step , @xmath254 is generated by @xmath137 and @xmath73 . by the second step , @xmath73 splits as a product @xmath251 , and by the third step @xmath255=\{e\}$ ] . let @xmath8 denote the group generated by @xmath252 and @xmath137 . then @xmath8 and @xmath9 generate @xmath254 , @xmath256=0 $ ] and @xmath257 . therefore @xmath258 splits as @xmath7 . + _ proof of step 1 . _ let @xmath259 be the smallest normal subgroup containing @xmath260 , i.e. the group generated by all the conjugates of the elements @xmath240 , @xmath213 . we first prove that @xmath261 . let @xmath262 be a loop in @xmath88 . since @xmath3 is simply connected , there exists a disk @xmath263 bounding @xmath262 . we can choose @xmath264 so that it intersects the strata @xmath211 transversally , in a finite number of points . for each point @xmath265 we can produce a curve @xmath266 in @xmath264 going around @xmath265 only . the curve @xmath262 is homotopic to the product of these @xmath266 and , by the discussion at the beginning of the proof , every such @xmath266 is conjugate to some @xmath234 or @xmath267 . since @xmath268 by ( * ? ? ? * corollary 5.3 ) , the exact sequence in homotopy of fibration implies that @xmath250 is surjective . since @xmath73 is normal in @xmath254 , so is @xmath260 and therefore @xmath269 . + _ proof of step 2 . _ suppose that @xmath270 do not commute or , equivalently , that @xmath271 consider the circle bundle @xmath272 where @xmath273 is the map defined in . equation says that @xmath274 is not orientable along a curve representing the class @xmath234 . therefore , we must have @xmath275 or , equivalently , @xmath276=k_j^{-2}$ ] . from it follows that @xmath277 commutes with all @xmath234 , and is therefore central in @xmath73 . by exchanging the roles of @xmath234 and @xmath278 in , we similarly obtain @xmath276=k_i^{-2}$ ] . thus @xmath279 for all non commuting @xmath280 , @xmath281 . since @xmath234 and @xmath278 do not commute , @xmath234 and @xmath282 do not commute either and therefore @xmath283 . in particular , @xmath284 unless @xmath234 is central in @xmath73 . the center of @xmath73 splits uniquely as @xmath285 , where @xmath286 is the sylow @xmath10-subgroup of @xmath287 . let @xmath288 and @xmath289 . by the computations above , @xmath290 $ ] is generated by the squares of the generators @xmath234 in @xmath168 and therefore it is central . hence , it is abelian , and @xmath9 is @xmath10-step nilpotent . on the other hand , @xmath290 $ ] is finitely generated and every generator has order @xmath10 . therefore , @xmath290 $ ] is a finite @xmath10-group . it follows from the short exact sequence @xmath291\rightarrow k_2\rightarrow k_2/[k_2,k_2]\rightarrow 0\ ] ] that @xmath9 is a finite @xmath10-group . clearly , @xmath252 is abelian , @xmath292=0 $ ] and @xmath293 . thus @xmath294 . + _ proof of step 3 . _ the map @xmath295 in sequence can be seen as the map @xmath296 induced by the fibration @xmath297 , where @xmath298 is the loop space of @xmath24 . the map @xmath299 extends to an action of @xmath298 on @xmath28 via the holonomy of the fibration @xmath300 . we will denote this action by `` @xmath301 '' . the existence of this action implies that @xmath302 is central in @xmath254 . indeed , given @xmath303 and @xmath304 , the homotopy @xmath305 is a homotopy between @xmath306 and @xmath307 . @xmath308 + the following corollary is a direct consequence of the proof of theorem a. [ c : ref_cor ] let @xmath2 be a closed singular riemannian foliation on a compact , simply connected riemannian manifold @xmath3 . let @xmath88 be the regular part of @xmath4 , let @xmath28 be a regular leaf and let @xmath309 be the number of codimension @xmath10 strata . if @xmath310 , then @xmath311 is generated by at most @xmath309 elements . in the proof of theorem [ t : fund_gp ] , the non - abelian part @xmath9 is of a very particular type . one can prove that there exists a surjective homomorphism @xmath312 , where @xmath313 is isomorphic to a finite product of groups @xmath314 that are central extensions @xmath315 , for some @xmath316 . the only instances known to us of such groups are @xmath317 , @xmath318 , the quaternion group @xmath319 and products of these groups . let @xmath156 be a small tubular neighborhood @xmath320 of @xmath321 and let @xmath322 . as @xmath323 is an open cover of @xmath3 , we have @xmath324 & = & \chi(\sigma_0)+\chi(v)-\chi(w\cap v).\end{aligned}\ ] ] observe that @xmath325 and @xmath326 are saturated submanifolds without @xmath14-dimensional leaves . we will now show that @xmath327 , which proves the theorem . notice that @xmath328 retracts to @xmath329 , which is a compact saturated submanifold of @xmath3 . recall that @xmath330 admits a finite good open cover @xmath331 , i.e. every finite intersection @xmath332 is contractible ( cf . @xcite ) . for @xmath333 , we let @xmath334 be the preimage of @xmath335 under the leaf projection map . we can write @xmath336 where the sum is finite . observe that every finite intersection retracts to a leaf . since these leaves are homeomorphic to nontrivial bieberbach manifolds , their euler characteristic is @xmath14 . in particular , @xmath337 . we will now show that @xmath338 . observe first that @xmath339 , the closure of @xmath325 , has boundary @xmath340 . the double @xmath341 is a compact manifold which admits a b - foliation without @xmath14-dimensional leaves and , as before , @xmath342 . on the other hand , @xmath343 , so @xmath338 . we first prove the following lemma . [ l : cod1_sing ] let @xmath344 be a compact , simply connected @xmath345-manifold . if @xmath346 is a codimension one closed singular riemannian foliation , then the foliation can not be regular . suppose that @xmath347 is a regular foliation . since @xmath344 is simply connected , it follows from work of molino @xcite that @xmath347 must be a simple foliation , i.e. it is given by the fibers of a riemannian submersion . hence there is a fibration @xmath348 . since @xmath344 is simply connected , the long exact sequence in homotopy for the fibration yields a contradiction . we now prove theorem [ t : codim_1_bfol_sc ] . let @xmath3 be a compact , simply connected manifold and @xmath2 a codimension @xmath16 a - foliation of @xmath3 . by lemma [ l : cod1_sing ] , @xmath4 is singular . therefore , the leaf space @xmath62 is homeomorphic to a closed interval @xmath349 $ ] . in particular , @xmath350 is a contractible manifold and there are at most two strata of codimension @xmath10 . by corollary [ c : ref_cor ] , the fundamental group of a regular leaf has at most two generators . by corollary [ c : torus_leaves ] , the regular leaves are homeomorphic to tori . therefore , a regular leaf must be diffeomorphic to @xmath351 or @xmath32 . it follows that @xmath3 is either @xmath10- or @xmath209-dimensional and , since @xmath3 is simply connected , it must be diffeomorphic to @xmath49 or to @xmath48 . in the case of @xmath49 , it follows from theorem [ t : homogeneity ] that the foliation comes from a smooth circle action . suppose now that @xmath3 is diffeomorphic to @xmath48 . in this case , the regular leaves are @xmath10-dimensional , the singular leaves @xmath352 are @xmath16-dimensional , hence they are circles , and the regular leaves fiber over @xmath352 with fiber a circle . by corollary [ c : torus_leaves ] , the regular leaves are diffeomorphic to a @xmath10-torus . we can therefore describe @xmath48 as a double disk bundle @xmath353 where @xmath354 is a disk bundle over @xmath352 and @xmath355 is a diffeomorphism . moreover , the foliation @xmath4 consists of the distance tubes to the zero section ( with respect to some euclidean structure on the disk bundles ) . notice that @xmath354 are solid tori and @xmath356 are tori . since @xmath3 is diffeomorphic to @xmath48 , the gluing map @xmath357 is unique up to isotopy . in particular , since the foliation @xmath2 is uniquely determined up to foliated diffeomorphism by the isotopy type of @xmath357 . it follows that there is only one foliated diffeomorphism type of codimension one foliation in @xmath48 that decomposes @xmath48 into two full tori , and this must be the one given by the standard linear @xmath32-action . @xmath308 throughout this section we let @xmath2 be an a - foliation of codimension @xmath10 on a compact simply connected riemannian manifold @xmath3 . we first consider the case where @xmath2 is regular . if @xmath4 is regular , then @xmath3 is diffeomorphic to @xmath48 and the generic leaf is diffeomorphic to @xmath351 . since @xmath2 is regular and has codimension @xmath10 , it follows from ( * ? ? ? * theorem 1.6 ) that the quotient @xmath68 is a compact simply connected orbifold without boundary . in particular , @xmath68 is homeomorphic to @xmath49 . in our case , the fibration is given by @xmath358 , where @xmath5 is a regular leaf of @xmath4 and @xmath24 is the classifying space of the orbifold @xmath68 . therefore , there is a rational homotopy equivalence @xmath359 and , as a consequence , @xmath360 . tensoring the long exact homotopy sequence of the fibration @xmath358 with @xmath361 , we get @xmath362 by corollary [ c : torus_leaves ] , @xmath5 is a torus . therefore , the sequence above implies that @xmath5 is diffeomorphic to @xmath351 . since @xmath2 has codimension @xmath10 , @xmath3 is @xmath209-dimensional . by perelman s proof of the poincar conjecture , @xmath3 must be diffeomorphic to @xmath48 . by theorem [ t : homogeneity ] , a regular foliation on @xmath48 by circles must be homogeneous . moreover , the circle action must be equivalent to a linear circle action on @xmath48 ( cf . @xcite ) . we now consider the case where @xmath2 is singular . + since the foliation is not regular , by @xcite there are no exceptional leaves and the leaf space @xmath62 is homeomorphic to a @xmath10-dimensional orbifold @xmath24 with non - empty boundary corresponding to singular strata . as in the case of group actions , the fundamental group of @xmath3 surjects onto the fundamental group of the leaf space ( cf . * chapter ii , theorem 6.2 and corollary 6.3 ) ) . therefore , the leaf space is simply connected and hence it is homeomorphic to a disk . the boundary of @xmath24 consists of the union of geodesic arcs . the points in the interior of these arcs correspond to leaves which we will call _ least singular leaves _ , while the vertices of the leaf space , i.e. the points where two geodesic arcs in the boundary meet , correspond to leaves which we will call _ most singular leaves _ ( see figure [ f : lspace_labels ] ) . a - foliation . ] let @xmath5 be a singular leaf and fix @xmath118 . by corollary [ c : inf_b_fol ] , the infinitesimal foliation @xmath363 is a codimension one a - foliation , whose quotient is a closed interval . by theorem [ t : codim_1_bfol_sc ] , the infinitesimal foliation @xmath363 must be one of the homogeneous foliations @xmath17 or @xmath18 . since @xmath364 is a round unit sphere , it follows from the main theorem in @xcite that the foliation is isometric to one of the homogeneous foliations @xmath17 or @xmath18 induced by orthogonal actions on round unit spheres . in the first case , the leaf space of the homogeneous foliation is isometric to a closed interval of length @xmath365 and corresponds to the infinitesimal foliation of a least singular leaf . in the second case , the leaf space of the homogeneous foliation is isometric to a closed interval of length @xmath27 and corresponds to the infinitesimal foliation of a most singular leaf . since there are no exceptional leaves , by remark [ r : space_dir ] , the holonomy action is trivial . hence the space of directions at any point @xmath129 is isometric to the leaf space of the infinitesimal foliation at @xmath130 . in particular , the angle between geodesic arcs meeting at vertices of @xmath62 is @xmath27 and fibration yields the desired metric fibrations in part ( 2 ) of theorem [ t : bfol_codim2 ] . finally , since the regular leaves are homeomorphic to @xmath26-tori and the edges of the leaf space correspond to codimension @xmath10 strata , corollary [ c : ref_cor ] implies that there must be at least @xmath26 edges . this concludes the proof of theorem [ t : bfol_codim2 ] . we prove assertions ( 1 ) and ( 2 ) separately . + _ proof of assertion ( 1 ) . _ suppose @xmath40 . then the leaf space is a @xmath10-dimensional riemannian manifold of nonnegative curvature , homeomorphic to a @xmath366-disk , with polyhedral boundary and with positive curvature in an open subset . by theorem [ t : bfol_codim2 ] , the leaf space @xmath62 has @xmath367 vertices where boundary edges meet at an angle of @xmath27 . these points , which we will denote by @xmath368 , correspond to @xmath14-dimensional leaves @xmath231 in @xmath3 . since @xmath3 is quasipositively curved , so is @xmath62 . by the gauss - bonnet theorem , @xmath369 , which implies that @xmath370 or @xmath371 . in either case , @xmath3 decomposes as a union @xmath372 , where @xmath373 is diffeomorphic to a small distance ball around @xmath374 and @xmath375 is diffeomorphic to a tubular neighborhood around either @xmath376 , if @xmath370 , or to the closure of the stratum opposite to @xmath374 , if @xmath371 . therefore , @xmath377 . if @xmath370 , there exist diffeomorphisms @xmath378 where @xmath379 is a unit ball in @xmath380 , and @xmath3 is diffeomorphic to @xmath381 , where @xmath382 . by hatcher s proof of the smale conjecture @xcite , @xmath357 is isotopic to @xmath383 , and thus @xmath384 . if @xmath371 , the closure @xmath168 of the stratum opposite to @xmath374 is a @xmath10-dimensional smooth submanifold of @xmath3 and @xmath4 restricts to a codimension one foliation with two singular leaves . hence , @xmath168 is diffeomorphic to @xmath49 and , since @xmath385 , there is a diffeomorphism @xmath386 , where @xmath24 is a tubular neighborhood of a totally geodesic @xmath49 in @xmath387 . again , there is a diffeomorphism @xmath388 , and @xmath3 is diffeomorphic to @xmath389 where @xmath390 is defined using @xmath391 , as before . once again , @xmath390 is isotopic to @xmath383 , and @xmath392 . + _ proof of assertion ( 2 ) . _ by the work of barden and smale @xcite , it suffices to verify that @xmath393 . the leaf space @xmath62 is homeomorphic to a disk and it has at least three vertices , by theorem [ t : bfol_codim2 ] . a comparison argument as in part ( 1 ) implies that there are exactly three vertices in @xmath62 . let @xmath394 be the preimage of an edge of @xmath62 , and let @xmath395 be the preimage of the opposite vertex . the preimage @xmath396 is smooth closed submanifold of @xmath3 without boundary , and it is a deformation retract of @xmath397 . since @xmath398 , @xmath399 and therefore @xmath400 . it follows that @xmath3 admits a decomposition as a double disk bundle @xmath401 where @xmath402 is a distance tube around @xmath396 and @xmath403 is an orientable circle bundle . in particular , @xmath404 and from the meyer - vietoris sequence applied to the double disk decomposition we obtain @xmath405 from which it follows easily that @xmath393.@xmath308 we prove parts ( 1 ) and ( 2 ) separately . + _ proof of part ( 1 ) . _ by the gauss - bonnet theorem , the leaf space @xmath62 has at most four vertices . on the other hand , by theorem [ t : bfol_codim2 ] , @xmath62 has at least two vertices . hence , by theorem [ t : kobayashi ] , the euler characteristic of @xmath41 is @xmath10 , @xmath209 or @xmath0 , and it follows from the work of freedman @xcite that @xmath41 is homeomorphic to @xmath12 , @xmath46 or @xmath47 . if @xmath62 has two or three vertices , then the leaf space structure is the same as the one in the proof of part ( 1 ) of theorem [ t : qpc_d4_codim2 ] . it follows that @xmath41 is diffeomorphic to @xmath12 , if @xmath62 has two vertices , or to @xmath42 , if @xmath62 has three vertices . + _ proof of part ( 2 ) . _ by theorem [ t : bfol_codim2 ] , the leaf space @xmath62 has at least three vertices . on , the other hand , since the leaf space is nonnegatively curved , it may have at most four vertices . if there are exactly three , then , proceeding as in the proof of theorem [ t : qpc_d4_codim2 ] , we conclude that @xmath3 is diffeomorphic to @xmath45 . therefore , we can restrict our attention to the case in which @xmath62 has four vertices , so that @xmath62 is isometric to a flat rectangle @xmath349\times[-1,1]$ ] . to conclude that @xmath3 is diffeomorphic to one of the two @xmath48-bundles over @xmath49 , it suffices to prove that @xmath406 and then appeal to the work of barden @xcite and smale @xcite . + let @xmath407 and @xmath408 be the preimage of @xmath409\in m^*$ ] and @xmath410 $ ] , respectively . similarly , define @xmath411 to be the preimages of @xmath349\times\{\pm 1\}$ ] , @xmath349\times\{0\}$ ] . they are all smooth closed submanifolds of @xmath3 without boundary , there are maps @xmath412 which are circle bundles , and @xmath3 can be written as a double disk bundle @xmath413 let @xmath168 be the preimage of the point @xmath414 . the leaf @xmath168 is diffeomorphic to @xmath351 , and its normal bundle @xmath415 has rank @xmath0 and is orientable . since the universal cover of @xmath168 is contractible , the structure group of @xmath416 is completely determined by an isometry @xmath417 of @xmath418 at some point @xmath419 . since the structure group preserves the foliation @xmath4 , the isometry @xmath417 preserves the infinitesimal foliation @xmath420 , which is isometric to the foliation induced by the isometric @xmath32 action on @xmath48 . since @xmath417 also preserves the orientation of @xmath48 , @xmath421 . @xmath422 . @xmath417 belongs to @xmath423 if and only if it preserves the orientation of both singular leaves @xmath424 of the infinitesimal foliation at @xmath59 , otherwise it reverses both . if @xmath425 are the ( singular ) leaves containing @xmath426 , @xmath427 respectively , they are 2-dimensional and they are both tori if and only if @xmath422 , otherwise they are both klein bottles . it follows that either all the 2-dimensional leaves are orientable ( if @xmath422 ) or none of them is . moreover , since the generic leaf in @xmath428 is orientable if and only if @xmath396 is ( and the same holds for @xmath429 ) it follows that @xmath430 are all orientable if and only if @xmath422 , otherwise none of them is . suppose now that @xmath431 . as we said , it follows that @xmath432 are non - orientable , and since @xmath402 is always orientable , the circle bundles @xmath433 are non - orientable . it follows from equation and ( * ? ? ? * table 1.4 ) that @xmath434 is finite . on the other hand , @xmath402 itself can be written as a double disk bundle @xmath435 where @xmath436 are klein bottles . from the mayer - vietoris sequence , it follows that @xmath434 can not be finite , and this provides a contradiction . the following statements immediately follow from the proof of the lemma : * every @xmath10-dimensional leaf is a torus . * the manifolds @xmath437 , @xmath438 are orientable . * the bundles @xmath439 are orientable , and therefore principal @xmath351-bundles . from the facts listed above , @xmath396 can be decomposed as a union of two solid tori . therefore , @xmath396 is diffeomorphic to either @xmath440 or to a lens space @xmath441 . since @xmath442 is a principal bundle , @xmath402 is homotopy equivalent to either @xmath443 ( only if @xmath444 ) or to @xmath445 . if @xmath446 , consider the homotopy fibration @xmath447 . since @xmath448 are orientable , it follows from ( * ? ? ? * table 1.4 ) that @xmath449 , and from the long exact sequence in homotopy we obtain @xmath450 therefore @xmath451 . the only possibility left is that @xmath452 , and @xmath453 . applying the meyer - vietoris sequence to the double disk bundle decomposition we obtain @xmath454 it follows immediately that @xmath455 . moreover , the map @xmath456 is explicitly computable and its cokernel is @xmath457 . therefore @xmath458 in this case as well . throughout this section we let @xmath2 be a singular riemannian foliation by circles on a compact , simply connected riemannian @xmath0-manifold . by theorem [ t : homogeneity ] , @xmath2 is a homogeneous foliation , i.e. it is induced by a smooth effective circle action on @xmath3 . by work of fintushel @xcite , pao @xcite , and perelman s proof of the poincar conjecture , a compact , simply connected smooth @xmath0-manifold with a smooth effective circle action is diffeomorphic to a connected sum of copies of @xmath12 , @xmath459 and @xmath460 . it follows from theorem [ t : homogeneity ] that a compact , simply connected @xmath0-manifold with a singular riemannian foliation by circles is diffeomorphic to a connected sum of copies of @xmath12 , @xmath459 and @xmath460 . the leaf space structure of @xmath2 corresponds to the orbit space structure of a smooth circle action on a compact , simply connected smooth @xmath0-manifold ( cf . in particular , the leaf space @xmath62 is a simply connected topological @xmath209-manifold , possibly with boundary , the components of the @xmath14-dimensional stratum are homeomorphic to @xmath10-spheres or isolated points and the boundary components of @xmath62 are @xmath10-sphere components in the @xmath14-dimensional stratum . with these preliminary remarks in place , we prove the rest of theorem [ t : d4_s1fol ] . by poincar duality , @xmath461 . by the discussion in the first paragraph of the proof , to prove parts ( 1 ) and ( 2 ) of the theorem it suffices to show that @xmath462 , when @xmath3 is positively curved , and @xmath463 , when @xmath3 is nonnegatively curved . by the results in the preceding subsection , the leaf space @xmath62 is a simply connected topological manifold with an alexandrov space structure in particular , @xmath62 is positively curved if @xmath3 has positive curvature , and @xmath62 is nonnegatively curved if @xmath3 has nonnegative curvature . since the proof follows as in the proof of an isometric circle action , via comparison arguments already found in the literature ( cf . @xcite ) , we only indicate the necessary steps . + _ positively curved case_. suppose first that @xmath464 is not empty , and let @xmath465 be a connected component of @xmath464 . then , by the soul theorem for alexandrov spaces , there exists a unique point @xmath466 at maximal distance from @xmath465 and all the points between @xmath465 and @xmath466 must correspond to regular leaves . in particular , the boundary of @xmath62 is always connected . the point @xmath466 is either a regular leaf or an isolated point in @xmath321 . therefore @xmath462 . suppose now that @xmath464 is empty . then @xmath321 consists only of isolated points . the space of directions at an isolated point in the @xmath14-dimensional stratum is isometric to the quotient of a round unit @xmath48 by an isometric circle action without fixed points . by a triangle comparison argument as in @xcite , there can be at most three such points . therefore , @xmath467 . + _ nonnegatively curved case . _ suppose first that @xmath464 has at least two components , @xmath468 and @xmath469 . then @xmath62 is isometric to @xmath470 $ ] and there are no isolated points in @xmath321 . hence @xmath471 . suppose now that @xmath464 is connected and let @xmath472 be the set at maximal distance from @xmath464 in @xmath62 . there can be at most @xmath10 isolated points in @xmath321 contained in @xmath472 , so @xmath473 . finally , suppose that @xmath464 is empty and @xmath321 consists only of isolated points . as in the positively curved case , a triangle comparison argument as in @xcite implies that there can be at most four such points , so @xmath463.@xmath308 grove , k. , _ developments around positive sectional curvature _ , surveys in differential geometry . geometry , analysis , and algebraic geometry : forty years of the journal of differential geometry , 117133 , surv . geom . , * 13 * , int . press , somerville , ma , 2009 .

we present a perturbation theory for studying the instabilities of non - axisymmetric gaseous discs . we perturb the dynamical equations of self - gravitating fluids in the vicinity of a non - axisymmetric equilibrium , and expand the perturbed physical quantities in terms of a complete basis set and a small non - axisymmetry parameter . we then derive a linear eigenvalue problem in matrix form , and determine the pattern speed , growth rate and mode shapes of the first three unstable modes . in non - axisymmetric discs the amplitude and phase angle of travelling waves are functions of both the radius and the azimuthal angle . this is due to the interaction of different wave components in the response spectrum . we demonstrate that wave interaction in unstable discs with small initial asymmetries , can develop dense clumps during the phase of exponential growth . local clumps , which occur on the major spiral arms , can constitute seeds of gas giant planets in accretion discs . [ firstpage ] accretion discs hydrodynamics instabilities galaxies : structure galaxies : spiral

astro-ph0609055

unstable density waves inside self - gravitating gaseous media play an important role in structuring of spiral galaxies and accretion discs . prior works in the literature deal with the instability of differentially rotating discs , which are initially circular . aoki et al . ( 1979 ) carried out a global analysis of spiral modes in the soft - centred , gaseous kuzmin ( 1956 ) disc using hunter s ( 1963 ) device . lemos et al . ( 1991 ) studied the unstable axisymmetric modes of the more puzzling scale - free discs and solved the dispersion relation to determine the sufficient conditions for ring formation and for the existence of breathing modes . recently goodman & evans ( 1999 , hereafter ge ) constructed the normal modes of the scale - free mestel ( 1963 ) disc in the cowling approximation . for self - gravitating modes they computed the ratio of growth rate to pattern speed by solving a recurrence relation in the mellin - transform space . density waves in the presence of an external potential field , like that of a massive central object , have also been extensively studied using semi - analytical and finite difference techniques ( e.g. , papaloizou & lin 1989 ; savonije & heemskerk 1990 ; papaloizou & savonije 1991 ) . the assumption of axisymmetry of the equilibrium ( or initial ) state , however , imposes significant simplifications on the solution of dynamical equations . in fact , unstable modes of circular discs are decoupled because the only @xmath2-dependent term , @xmath3 ( @xmath4 ) , is factored out of the linearized poisson and hydrodynamic equations . here @xmath2 is the azimuthal angle and @xmath5 is the wave number . such a factorization is impossible if the unperturbed state is non - axisymmetric . azimuthal modes of perturbed non - axisymmetric discs can interact with the fourier components of the equilibrium state and result in more complex patterns . in this paper we attempt to formulate the instability problem of non - axisymmetric discs based on a perturbation theory , and present a mechanism for the generation of dense clumps through _ wave interaction _ in _ linear regime_. that occurs when a non - axisymmetric disc is destabilized . we apply our formulation to the simple discs of jalali & abolghasemi ( 2002 , hereafter ja ) , which reduce to mestel s disc in axisymmetric limit . we have chosen ja models as our unperturbed state because of two reasons . firstly , closed form ( exact ) expressions are available for physical quantities like the potential density pair , pressure , sound speed and velocity components . secondly , both the potential and the surface density have a single fourier component . this minimizes the number of interacting terms in our subsequent analysis . we review steady - state , non - axisymmetric , scale - free discs of ja in section [ sec : non - axi - symmetric - discs ] and introduce a cutout function that helps us handle the singularity at the centre . we present a first - order perturbation theory for non - axisymmetric gaseous discs in section [ sec : perturbation - formulation ] , where we also introduce a basis set by which we expand the perturbed physical quantities . we then derive an eigenvalue problem for determining the normal modes of non - axisymmetric discs and discuss on the algorithms used for solving the resulting eigensystem . in sections [ sec::modes - of - circular - discs ] and [ sec : modes - nonaxi - discs ] , we study the unstable modes of circular and non - axisymmetric discs for the fundamental wave numbers @xmath6 , 1 and 2 , and discuss on the results and their physical implications in section [ sec : discussion ] . there are few papers in the literature that address non - axisymmetric equilibria of gaseous discs ( syer & tremaine 1996 ; galli et al . 2001 ; ja ) . in this section we briefly review one of _ exact _ non - axisymmetric models of ja whose surface density consists of a simple harmonic function of the azimuthal angle . we define @xmath7 as the usual polar coordinates and assume a self - gravitating , inviscid , compressible fluid in a steady two dimensional state so that the surface density distribution and its self - consistent potential are scale - free as @xmath8,~~v_0 ^ 2=2\pi g\sigma_s r_0 . \label{eq : equilibrium - potential}\end{aligned}\ ] ] here @xmath9 is the constant of gravitation , and @xmath10 and @xmath11 are the normalizing constants of the length and surface density , respectively . our discs are non - axisymmetric because @xmath12 and @xmath13 depend on the azimuthal angle @xmath2 . due to the continuity of physical quantities in the azimuthal direction , the functions @xmath14 and @xmath15 are @xmath16-periodic in @xmath2 . for the sake of simplicity , we work with non - axisymmetric discs of the form @xmath17 where @xmath18 . the @xmath2-dependent part of the potential reads @xmath19 for @xmath20 , our discs reduce to the isothermal , gaseous mestel disc . let us respectively define @xmath21 , @xmath22 and @xmath23=@xmath24 as the sound speed , velocity of circular orbits and mach number in mestel s disc . the components of the velocity field associated with ( [ eq : equilibrium - density ] ) and ( [ eq : equilibrium - potential ] ) are obtained after solving the hydrodynamic equations as ( ja ) @xmath25 the sound speed @xmath26 and the pressure distribution @xmath27 are then given by @xmath28 @xmath29 reduces to @xmath21 in the limit of @xmath30 . from ( [ eq : equilibrium - potential ] ) and ( [ eq : nonaxi - pressure ] ) , and equation ( 2 ) in ja , one can verify that @xmath31=@xmath32 . sound waves are determinate if the pressure is positive , i.e. , @xmath33 . a necessary and sufficient condition for the positiveness of @xmath34 is @xmath35 here @xmath36 is the minimum of @xmath37 over the interval @xmath38 . inequality ( [ con : positivespeed ] ) restricts model parameters . ja models allow for transonic flows because the sound speed depends on @xmath2 and it varies along a given streamline . in principle , transition from subsonic to supersonic speeds can take place smoothly but the reverse phenomenon is impossible without a shock wave . to avoid shock waves , we follow ja and exclude transonic flows from our calculations and investigate supersonic and subsonic discs . in this paper we choose a bisymmetric equilibrium with @xmath39 . the allowable region of the parameter space @xmath40 is determined based on two requirements . firstly the square of the sound speed must be positive . secondly , transonic flows are excluded . [ fig : fig1 ] shows the admissible zone of the parameter space . note that for @xmath41 the mach number is a function of the azimuthal angle : @xmath42 in the limit of circular discs ( @xmath30 ) we obtain @xmath43 as expected . we are studying scale - free gaseous discs whose surface density and potential diverge near the center according to ( [ eq : equilibrium - density ] ) and ( [ eq : equilibrium - potential ] ) . the linear stability problem of scale - free discs is ill - posed unless one decides on the way that inward traveling waves are reflected off the cusp . several methods have been devised to tackle this problem . for scale - free stellar discs zang ( 1976 ) ( see also evans & read 1998a , b ) used an inner cutout , which carves out the density function and deactivates the central region of the stellar system . in this way , stars with high orbital frequencies do not participate in perturbations . for the gaseous mestel disc ge used a hard internal boundary that fixes the phase with which waves are reflected from the centre . in this paper we apply evans & read s ( 1998a ) method to our cuspy gaseous discs of [ sec::ja - models ] and deactivate some parts of the disc mass by the cutout function @xmath44}{r_0^{n_{\rm out}}\over \left [ ( \gamma r)^{n_{\rm out}}+r_0^{n_{\rm out}}\right ] } . \label{eq : cutout - fnc}\ ] ] we then analyze the perturbations of the _ active _ density @xmath45 here @xmath46 and @xmath47 are the inner and outer cutout radii , respectively . since the cutout is not applied to the potential , the immobile mass will resemble a rigid bulge / halo component that does not participate in the disc dynamics , but it contributes to the force field . instability properties are not sensitive to the choice of @xmath48 . we apply the outer cutout for computational convenience and use a large @xmath48 far beyond the region that instabilities occur . the nature of functions by which we will expand perturbed fields , constrains the feasible values of @xmath49 and @xmath50 . we have used @xmath49= @xmath51 throughout this paper . we define @xmath52 and consider the first order eulerian perturbations of physical quantities as @xmath53 the superscript @xmath54 denotes transpose and the subscripts 0 and 1 refer to the equilibrium and perturbed states , respectively . we assume that the perturbed quantities , @xmath55@xmath56@xmath57 , are composed of a discrete set of normal modes , each of the form @xmath58 where @xmath59 , @xmath60 and @xmath61 are respectively the eigenfrequency , pattern speed and growth rate of the @xmath5th mode . it is remarked that the real part of @xmath62 gives the physical solution . we use the equilibrium fields of [ sec : non - axi - symmetric - discs ] and substitute from ( [ eq::perturbations - for - all ] ) into the continuity and momentum equations of self - gravitating fluids [ equations ( 1e-3 ) and ( 6 - 20 ) in binney & tremaine 1987 ] . after collecting the first - order terms in perturbed variables , we obtain @xmath63 with @xmath64 . the elements of the linear operator @xmath65 $ ] , ( @xmath66 ) , have been given in appendix a. @xmath67 and @xmath68 should also satisfy poisson s integral @xmath69 modes of non - axisymmetric discs bifurcate from those of circular discs once we let @xmath0 grow from zero . therefore , we seek for a perturbation solution for @xmath62 in terms of the non - axisymmetry parameter @xmath0 . moreover , continuity of physical quantities in the azimuthal direction implies the periodicity conditions @xmath70 these requirements suggest to represent @xmath71 by the fourier series @xmath72 and propose the _ ansatz _ @xmath73 using equations ( [ eq::equilibrium - fields - expansion ] ) and ( [ eq::perturbed - fields - expansion ] ) and substituting from ( [ eq::discrete - modes ] ) in ( [ eq : linearized - perturbed - equations ] ) , we obtain the determining equation of the @xmath5th mode as @xmath74 which is a system of linear ordinary differential equations with respect to @xmath1 . here @xmath75s are linear differential operators ( they are square matrices of dimension @xmath76 ) whose elements have been given in appendix b for @xmath77 . the second term on the left side of ( [ eq::perturbed - vec - m - mode ] ) shows that how different azimuthal waves interact . the interacting terms rapidly decay proportional to @xmath78 for @xmath79 . in the axisymmetric limit @xmath80 , equation ( [ eq::discrete - modes ] ) reduces to @xmath81 so that @xmath59 is associated with a single azimuthal wave component . modal decomposition of this kind is impossible for initially non - axisymmetric discs , for the equilibrium fields include the powers of @xmath82 that cause wave interaction . one way for solving ( [ eq::poisson - integral - equation ] ) and ( [ eq::perturbed - vec - m - mode ] ) is through expanding @xmath83 [ see equation ( [ eq::perturbed - fields - expansion ] ) ] in terms of some basis , preferably biorthogonal , functions . clutton - brock ( 1972 ) found such a set , which was then recalculated by aoki & iye ( 1978 ) in terms of associated legendre functions . we follow them and expand the functions @xmath84 as @xmath85 where @xmath86=@xmath87 is a to - be - determined vector of complex coefficients and @xmath88 , \\ \xi \!\!\ ! & = & \!\!\ ! { r^2-r_c^2\over r^2+r_c^2},~w_1=\left ( 1-\xi \over 2\right ) ^{\frac 32},~w_2=\left ( 1-\xi \over 1+\xi \right ) ^{\frac 12}. \nonumber\end{aligned}\ ] ] @xmath89 are the normalized associated legendre functions defined by @xmath90^{\frac 12 } p^{|k|}_j\left(\xi\right).\ ] ] we set the scale length of clutton - brock functions , @xmath91 , to the normalizing length of ( [ eq : equilibrium - density ] ) and ( [ eq : equilibrium - potential ] ) . i.e. , @xmath92 . the criterion in choosing the _ weight _ function @xmath93 was the convergence of proceeding definite integrals that constitute the coefficients of the matrix @xmath94 in ( [ eq : eigenvalue - equation ] ) . if @xmath95 , the functions @xmath96 give the factor @xmath97^{|k|/2}$ ] . therefore , the expansions used for velocity components are not singular at the origin ( when @xmath98 ) . for @xmath99 , the perturbed velocity components become singular for @xmath100 . nonetheless , the energy and angular momentum are not singular there and the continuity equation is satisfied . we now substitute from ( [ eq : bi - orthogonal - potential ] ) and ( [ eq : bi - orthogonal - u - vector ] ) into ( [ eq::perturbed - vec - m - mode ] ) , left - multiply both sides of the resulting equation by @xmath101 ( @xmath102 ) , and integrate over @xmath103 . using the orthogonality of the associated legendre functions , we obtain a system of linear equations for the coefficients @xmath104 , @xmath105 and @xmath106 as @xmath107 it is seen that @xmath108 is an eigenvalue of @xmath94 and @xmath109 is its corresponding eigenvector . we find that the elements of @xmath94 are linear combinations of the following definite integrals @xmath110 where @xmath111 and @xmath112 are integer numbers of the form @xmath113 and @xmath114 , respectively . the integrals in ( [ eq : integrals ] ) are evaluated using gaussian quadratures . we transform @xmath94 to hessenberg form and obtain its corresponding qr transform ( press et al . we then compute the eigenvalues and eigenvectors of the linear system ( [ eq : eigenvalue - equation ] ) . consequently , the surface density of the @xmath5th mode is determined from ( [ eq::perturbed - fields - expansion ] ) and ( [ eq : bi - orthogonal - u - vector ] ) as @xmath115 , \label{eq::mode - nonaxi}\\ a_m(r,\phi,\epsilon ) \!\!\!\!\ ! & = & \!\!\!\!\ ! \sqrt { \sigma^{(m)}_1(r,\phi,\epsilon)\bar\sigma^{(m)}_1(r,\phi,\epsilon ) } , \label{eq::amplitude - nonaxisymmetric}\\ \theta_m(r,\phi,\epsilon ) \!\!\!\!\ ! & = & \!\!\!\!\ ! \arctan \left [ -{\rm i}{\sigma^{(m)}_1 -\bar\sigma^{(m)}_1 \over \sigma^{(m)}_1 + \bar\sigma^{(m)}_1}\right ] -m\phi , \label{eq::phase - nonaxisymmetric}\end{aligned}\ ] ] where a bar denotes complex conjugate . one can expand @xmath116 and @xmath117 in the taylor series of @xmath0 as @xmath118 so that @xmath119 are @xmath16-periodic in @xmath2 . the leading zeroth order terms of ( [ eq::amplitude - expansion ] ) , @xmath120 , determine the overall shape of the perturbed density , which is usually an @xmath5-fold trailing spiral . the higher - order terms ( @xmath121 ) lead to density fluctuations along the major spiral arm(s ) . both @xmath94 and @xmath109 are infinite dimensional however , the non - axisymmetry parameter @xmath0 is small and its powers , which serve as the coefficients of interacting terms , fall off rapidly . therefore , it is reasonable to truncate the series of the interacting modes in ( [ eq::perturbed - vec - m - mode ] ) at some @xmath122 . we also truncate the series in ( [ eq : bi - orthogonal - potential ] ) and ( [ eq : bi - orthogonal - u - vector ] ) at @xmath123 . we begin with @xmath124 and @xmath125 , and increase them until we acquire a relative accuracy of 1% in computing @xmath59 . in the axisymmetric limit , @xmath94 becomes a block diagonal matrix and we obtain the following eigenvalue problem for the @xmath5th mode @xmath126 with @xmath127 being a single @xmath128 block of @xmath94 . the perturbed surface density corresponding to ( [ eq : axisymmetric - eigenvalue - equation ] ) becomes @xmath129 . \label{eq : mode - density - axisymmetric - disc}\ ] ] the matrix @xmath130 has @xmath131 eigenvalues , which lie on several branches in the complex @xmath132-plane . however , eigenvalues of most branches diverge as @xmath133 is increased . they are virtual roots of the determinantal equation @xmath134 with @xmath135 being the unit matrix . physical eigenvalues of unstable discs ( if they exist ) belong to a convergent complex branch . the convergent eigenvalue with the largest growth rate corresponds to the fundamental eigenmode . the ideal scale - free limit will be attained if we let both @xmath136 and @xmath137 tend to zero . our calculations show that the stability / instability properties do not change substantially by decreasing @xmath137 . the choice of @xmath138 gives accurate and robust results in most cases . so we need to play with @xmath136 . since the scale length of clutton - brock functions is fixed , by decreasing @xmath136 we require too many terms in the expansion of @xmath83 to get the series converged . the smallest value of @xmath136 that we could approach , while assuring computational accuracy , was @xmath139 . to this end , we study instabilities of cutout mestel discs , which are obtained from the models of [ sec : non - axi - symmetric - discs ] by setting @xmath140 . calculation of unstable modes of circular discs is straightforward because the equilibrium fields do not depend on the azimuthal angle @xmath2 . we set @xmath6 , 1 and 2 in ( [ eq : axisymmetric - eigenvalue - equation ] ) and calculate @xmath59 over the interval @xmath141 . this covers all subsonic ( hot ) and a wide range of supersonic ( cold ) discs . we calculate the normal modes for @xmath136=0.5 , 0.1 and 0.03 , which respectively require @xmath142 , 100 and 200 for a computational accuracy of 1% . top row in fig . [ fig : fig2 ] shows the eigenfrequency spectrum of the @xmath6 mode for @xmath23=0.7 , 2.5 , 5 and @xmath143 . most eigenfrequencies of the model with @xmath23=0.7 are real ( fig . [ fig : fig2]_a _ ) . there are a few divergent complex eigenfrequencies whose magnitudes are increased by increasing @xmath133 . the complex eigenfrequencies completely disappear for @xmath144 ( fig . [ fig : fig2]_b _ ) . a convergent complex branch occurs for @xmath145 whose eigenfrequencies correspond to non - rotating unstable ring modes with @xmath146 ( fig . [ fig : fig2]_c _ ) . fundamental _ eigenfrequency , denoted by @xmath147 , is located at the end of the complex branch of the spectrum . [ fig : fig3]_a _ shows the variation of the fundamental growth rate with respect to @xmath148 and @xmath136 . it is seen that for @xmath149 our results do not vary that much as we change @xmath136 . the upper limit of @xmath148 in stable discs is @xmath150 for @xmath143 and @xmath151 for @xmath139 . our results for @xmath152 can be compared with the critical @xmath153 computed from equation ( 6.2 ) of lemos et al . their @xmath154 and @xmath155 for the @xmath156 models are our @xmath22 and @xmath21 , respectively . by keeping the minus sign before the square root in equation ( 6.2 ) of lemos et al . ( 1991 ) ( this is to guarantee ring formation ) , and setting @xmath124 , @xmath156 and @xmath157 , one finds @xmath158 . the difference between this critical number and @xmath151 of our @xmath139 model is only 0.66% , which validates the accuracy of our matrix formulation based on the expansions introduced in equations ( [ eq : bi - orthogonal - potential ] ) and ( [ eq : bi - orthogonal - u - vector ] ) . we note that the choice of @xmath139 is quite enough for understanding the behavior in the scale - free limit . we have also found that unstable breathing modes do exist in subsonic discs with @xmath143 if @xmath159 . for @xmath139 we find @xmath160 , which coincides with the upper limit of @xmath153 calculated in lemos et al . ( 1991 ) for the existence of breathing modes . the eigenfrequency spectra of the @xmath161 modes have three branches in the complex @xmath132-plane . subsonic discs have no convergent unstable ( complex ) branch . they are stable under @xmath161 excitations ( figs [ fig : fig2]_d _ and _ g _ ) . a continuum of convergent real eigenfrequencies exists on the @xmath60-axis , which survives for almost all @xmath23 , although it may be shrunk to a small interval of positive pattern speeds . aoki et al . ( 1979 ) have encountered similar eigenfrequencies in studying the gaseous kuzmin disc . convergent real eigenfrequencies correspond to pure oscillatory waves . here we are interested in growing modes and we will overlook the real branch of the spectrum . as we increase @xmath23 , a convergent unstable branch , with the largest @xmath162=@xmath163 , occurs in the spectrum . one can readily identify the fundamental eigenfrequency @xmath147 on this newly born branch . in a sufficiently cold disc , the convergent branch attracts more eigenfrequencies of the linear system ( [ eq : axisymmetric - eigenvalue - equation ] ) and @xmath164 becomes the global maximum ( see figs [ fig : fig2]_f _ and [ fig : fig2]_i _ ) . in figs [ fig : fig3]_b _ and _ c _ we have illustrated @xmath162=@xmath165 ( @xmath161 ) in terms of @xmath148 and for several choices of @xmath136 . these figures show that the growth rate tends to zero faster than the pattern speed while the disc is stabilized . instabilities are suppressed in discs with @xmath166 because pressure waves travel faster than gravitationally generated waves . figs [ fig : fig3]_b _ and _ c _ can be compared with fig . 3 in ge . the graphs of @xmath167 have similar trends in our and their papers : for @xmath168 , @xmath162 has a local minimum , while it is an almost monotonic function of @xmath23 for cold discs and for @xmath169 . the major discrepancy between our and ge results occurs when @xmath170 . ge predicted a rapid growth of @xmath167 contrary to the abrupt drop in our calculations . we ascribe the existing discrepancies to the type of imposed internal boundary conditions . we use a cutout , which disables the medium needed for the propagation of incoming waves . ge , however , fix @xmath171 to make a reflective boundary at @xmath172 . fig . [ fig : fig4 ] displays the unstable @xmath6 , 1 and 2 modes of a circular disc with @xmath173 and @xmath143 . we have plotted the positive part of @xmath174 $ ] . gray scale contours mark 10 to 100 percent of the maximum perturbed density . the @xmath168 and @xmath169 modes are tightly wound spirals whose amplitude functions , @xmath175 and @xmath176 , have only one peak . there is no local density concentration along the spiral arms and the wave amplitude slowly decays well within the corotation radius . we have also plotted @xmath177 versus @xmath178 below each mode shape in fig . [ fig : fig4 ] . the plots show a rapid convergence of the sequence @xmath177 . as it can be seen in mode shapes , unstable waves can not penetrate into the centre . this is due to the inner cutout , which damps the impinging waves by immobilizing the mass of central regions . we now apply the formulation of [ sec : perturbation - formulation ] and calculate the @xmath6 , 1 and 2 modes of our non - axisymmetric discs . we set @xmath179 , which keeps @xmath180=7 interacting modes . such a truncation has enough accuracy for the models with @xmath181 , which mostly have @xmath182 ( see fig . [ fig : fig1 ] ) . the major limiting factor in the normal mode calculation of non - axisymmetric discs is the choice of @xmath136 whose small values need too many terms in ( [ eq : bi - orthogonal - potential ] ) and ( [ eq : bi - orthogonal - u - vector ] ) to ensure the convergency . taking into account the fact that the size of @xmath94 in non - axisymmetric discs increases by a factor of @xmath183 compared to the axisymmetric limit , the run time of our eigenvalue and eigenvector solver rises drastically . to optimize the cost of numerical computations , we use @xmath143 in our non - axisymmetric discs because it requires only @xmath142 to achieve a relative accuracy of 1% . this is a legitimate choice because according to the graphs of fig . [ fig : fig3 ] the qualitative features of normal modes do not change by varying @xmath136 . fig . [ fig : fig5 ] shows the evolution of the fundamental eigenfrequencies with respect to the variations in @xmath0 for two choices of @xmath23 for each mode number @xmath5 . left panels in fig . [ fig : fig5 ] correspond to a marginal @xmath23 , which is chosen slightly larger than the upper limit of mach number for stable circular discs . right panels in fig . [ fig : fig5 ] are for a cold disc with @xmath184 . left panels in fig . [ fig : fig6 ] display the @xmath6 , 1 and 2 modes of non - axisymmetric discs that bifurcate from the modes of fig . [ fig : fig4 ] . gray scale contours show the positive part of @xmath185.\ ] ] right panel to each mode shape shows the absolute values of the scaled expansion coefficients , @xmath186=@xmath187 ( @xmath188 ) , in terms of @xmath189 and @xmath178 . our computations show that non - axisymmetric discs with @xmath190 are stable for @xmath5=0 excitations independent of the magnitude of @xmath0 . in discs with @xmath191 a tangent bifurcation occurs in the complex @xmath132-plane at the location of each eigenfrequency ( including @xmath147 ) and two sets of eigenfrequencies with non - zero pattern speeds come to existence . they are of the form @xmath192=@xmath193 . this is an interesting result : we have found an unstable , ring - shaped density perturbation that can rotate . our computations show that the models with @xmath194 are stabilized by increasing @xmath0 . [ fig : fig5]_a _ shows that how a model with @xmath23=2.8 is eventually stabilized for @xmath195 . the pattern speed of the @xmath6 mode takes both positive and negative values because equation ( [ eq::perturbed - vec - m - mode ] ) is invariant under the reflections @xmath196 and @xmath197 for @xmath6 . according to fig . [ fig : fig1 ] , the upper limit of @xmath0 decreases for our cold bisymmetric discs , and therefore , models with large @xmath23 can not be stabilized only by varying the non - axisymmetry parameter . nonetheless , the placement of eigenfrequencies in the complex plane is similar for all unstable discs : the pattern rotates faster and grows slower as @xmath0 increases . [ fig : fig6]_a _ displays the growing @xmath5=0 mode of a non - axisymmetric disc with @xmath40=@xmath198 . the corresponding @xmath199 ( fig . [ fig : fig6]_b _ ) tend to zero as @xmath200 and @xmath178 are increased . thus , the perturbation expansion given in ( [ eq::perturbed - fields - expansion ] ) is convergent both in the azimuthal and in the radial directions . our eigenfrequency calculations for @xmath168 has been summarized in figs [ fig : fig5]_c _ and _ d _ for @xmath2011.7 and @xmath202 , respectively . the unstable @xmath168 mode has a definite sense of rotation and its pattern speed decreases as @xmath0 is increased . for the disc with @xmath203 , @xmath59 steeply falls off and ultimately vanishes for @xmath204 . this property is shared by all models in the range @xmath205 . models with @xmath206 are stable for @xmath168 excitations regardless of the magnitude of @xmath0 . the fundamental eigenfrequency of the models with @xmath207 does not vary considerably as @xmath0 is increased in the admissible zone of fig . [ fig : fig1 ] . however , the mode shape is highly influenced by the non - axisymmetry . [ fig : fig6]_c _ shows the unstable @xmath168 mode of a non - axisymmetric disc with @xmath40=@xmath198 . although the overall shape is a single armed spiral , which corresponds to the @xmath208 wave component of equation ( [ eq::perturbations - for - all ] ) , a number of dense clumps occur along the major spiral arm . these local clumps are generated by the @xmath209 wave components of the perturbed density expansion . [ fig : fig6]_d _ shows the variation of @xmath210 versus @xmath189 and @xmath178 for @xmath168 . the sequence nicely converges as @xmath211 is increased . it is seen that the choice of @xmath212 provides enough accuracy both for @xmath6 and for @xmath168 . finding the eigenfrequencies of the @xmath169 mode for @xmath213 is troublesome because the convergent complex branch emerges very close to divergent branches ( see fig . [ fig : fig2]_h _ ) . we have written a program that computes the full eigenvalue spectrum for several choices of @xmath133 and searches between all eigenvalues for the most convergent one . according to our program , convergent unstable @xmath169 modes exist for @xmath214 . numerical results displayed in fig . [ fig : fig5]_e _ and _ f _ show that the fundamental eigenfrequency is robust to variations in @xmath0 . the corresponding mode shapes are double armed spirals with some sort of dense clumps distributed along the major arms . the pitch angle of the global arms deceases as the disc is cooled ( large @xmath23 ) . fig . [ fig : fig6]_e _ displays the @xmath169 mode of a non - axisymmetric disc with @xmath40=@xmath215 . the perturbation series ( [ eq::perturbed - fields - expansion ] ) slowly converges in the @xmath2-direction for @xmath169 . so we kept more azimuthal wave components and used @xmath179 ( fig . [ fig : fig6]_f _ ) . the classical procedure in the instability analysis of gaseous ( and stellar ) discs is to start with an axisymmetric equilibrium and then perturb the governing equations in the vicinity of this equilibrium state . the subsequent eigenvalue problem [ e.g. , equation ( [ eq : axisymmetric - eigenvalue - equation ] ) ] results in a complex frequency @xmath216 , which implies an @xmath5-fold barred / spiral pattern @xmath217 $ ] that rotates by the angular velocity @xmath218 and exponentially grows by the rate @xmath61 . nevertheless , most events of physical significance happen in nonlinear regime when the modes begin to interact . in such a circumstance , more than one azimuthal wave component is present in the response spectrum . the most general form of a density wave in nonlinear regime is @xmath219 } , \label{eq::general - density - wave}\ ] ] which means that the amplitude and phase angle of @xmath68 are functions of all spatial and time variables . a primary result of ( [ eq::general - density - wave ] ) is the generation of dense clumps at the locations of the maxima of @xmath68 . clumps may undergo gravitational collapse , merge and give birth to gas giants . this is a widely accepted formation scenario of protoplanets ( e.g. , boss 1997 ; mayer et al . 2004 ) because it can take place in short time scales . asymmetries like the presence of a double star ( boss 2006 ) can also generate dense clumps . according to ( [ eq::mode - nonaxi ] ) and calculations of [ sec : modes - nonaxi - discs ] , unstable density waves of non - axisymmetric discs have a rich structure ( in linear regime ) even for very small non - axisymmetry parameter @xmath0 . although the temporal evolution of the @xmath5th mode is governed by the simple function @xmath220 , the eigenfunction associated with @xmath59 is by no means simple and it contains infinite number of azimuthal wave components , which in turn are capable of generating local clumps . since realistic accretion discs have not initially the _ nice _ axisymmetry feature , their instability can give birth to protoplanets sooner than what nonlinear hydrodynamical simulations predict . a protoplanet , as a possible product of the gravitational collapse of a dense clump , is separated from the rest of the evolving gas and migrates until it is captured by a resonant island in the phase space . our perturbation theory can be applied to any non - axisymmetric configuration regardless of the nature of the centre . cutouts , which are special treatments designed for scale - free models , are not needed in cored systems with finite total mass . the most important factor in success of our perturbation theory is the smallness of the non - axisymmetry parameter @xmath0 , which prohibits the interaction of azimuthal wave components in the zeroth order terms . our equilibrium density has a single azimuthal fourier component [ see equation ( [ eq : simple - disc ] ) ] . accordingly , the harmonic numbers of wave components differ in steps of @xmath221 ( @xmath222 ) . for bisymmetric discs studied in this paper , this means that wave components have only even or odd harmonic numbers . we will have the most complete spectrum if the equilibrium fields include fourier terms with both even and odd harmonic numbers , or if the disc is lopsided with @xmath223 . we presented three length scales to the perturbed equation ( [ eq::perturbations - for - all ] ) . the first two are the cutout radii , @xmath224 and @xmath48 , and the third one is the length scale of the basis functions , @xmath91 . at a first look , such a treatment of scale - free systems seems unfavorable because in this way we destroy their scale - invariance and cuspy features . however , with appropriate selection of the exponents of the cutout functions , @xmath49=@xmath50=2 , and the weight function of the velocity components , @xmath93 , the coefficient matrix @xmath94 of our eigenvalue problem became a function of the two dimensionless parameters @xmath136 and @xmath137 . we easily decreased @xmath137 and noticed that unstable waves do not extend to distant regions although the mass of our equilibrium models ( with flat rotation curves ) is infinite at @xmath225 . we then let @xmath136 tend to zero with the aim of removing the inner cutout . this needed more terms of the series of associated legendre functions , but the qualitative features of our results remained unchanged as long as mode shapes and @xmath167 were concerned . there was an excellent match between our results and those of lemos et al . ( 1991 ) who used logarithmic spirals to model their perturbed quantities . logarithmic spirals do not introduce any length scale to variational equations . the critical value of @xmath23 , below which our bisymmetric discs are stable , is a function of the fundamental wave number @xmath5 . numerical results show that the models with @xmath226 are stable for @xmath6 excitations when @xmath227 . this limit approximately becomes @xmath228 and @xmath229 for @xmath168 and 2 , respectively . the most interesting property that we have found for unstable non - axisymmetric discs is that the @xmath6 and @xmath168 modes can be stabilized by increasing the non - axisymmetry parameter @xmath0 . this happens for @xmath6 when @xmath194 and for @xmath168 when @xmath230 . there are some remarkable differences between the stabilized @xmath6 and @xmath168 modes . the stabilized @xmath6 mode has the form of a ring , which rotates according to the results displayed in fig . [ fig : fig5]_a_. it occurs when the wave components with the azimuthal harmonic numbers @xmath231 and @xmath232 cancel each other for @xmath121 . the stabilized @xmath168 mode , however , is a non - spiral , equiangular neutral mode . it emerges for a sufficiently large @xmath0 , when the wave component with the harmonic number @xmath233 ( @xmath234 ) cancels the fundamental component with @xmath235 ( @xmath208 ) . it is worth noting that the @xmath169 mode can not be stabilized by increasing @xmath0 , because the interacting wave components of @xmath236 have the harmonic numbers @xmath237 ( @xmath234 ) and @xmath238 ( @xmath239 ) . neither of these first - order components is capable of suppressing the effect of the fundamental component with the harmonic number @xmath240 . on the other hand , the wave component with the harmonic number @xmath241 is of @xmath242 , which is too weak to compensate the zeroth order @xmath169 component . based on the above reasonings , we speculate that the @xmath169 mode can be stabilized in discs whose surface density includes the fourth harmonic of the azimuthal angle or when @xmath243 in the simple discs of ja . furthermore , the unstable @xmath168 and @xmath169 modes of lopsided discs ( with @xmath223 ) can not be stabilized by increasing the non - axisymmetry parameter . during the first 2 years of his phd program , nma received a research grant from the aerospace research institute through a collaboration scheme between the iasbs and ari . nma also thanks prof . mohsen bahrami for insightful discussions . 99 aoki s. , iye m. 1978 , pasj , 30 , 519 aoki s. , noguchi m. , iye m. 1979 , pasj , 31 , 737 binney j. , tremaine s. , 1987 , galactic dynamics , princeton university press , princeton boss a.p . , 1997 , science , 276 , 1836 boss a.p . , 2006 , apj , 641 , 1148 clutton - brock m. , 1972 , ap&ss , 16 , 101 evans n.w . , read j.c.a . , 1998a , mnras , 300 , 83 evans n.w . , read j.c.a . , 1998b , mnras , 300 , 106 galli d. , shu f.h . , laughlin g. , lizano s. , 2001 , apj , 551 , 367 goodman j. , evans n.w . , 1999 , mnras , 309 , 599 ( ge ) hunter c. , 1963 , mnras , 126 , 299 jalali m.a . , abolghasemi m. , 2002 , apj , 580 , 718 ( ja ) kuzmin g.g . , 1956 , astr . , 33 , 27 lemos j.p.s . , kalnajs a.j . , lynden - bell d. , 1991 , apj , 375 , 484 mayer l. , quinn t. , wadsley j. , stadel j. 2004 , apj , 609 , 1045 mestel l. , 1963 , mnras , 126 , 553 papaloizou j.c . , lin d.n.c . , 1989 , apj , 344 , 645 papaloizou j.c . , savonije g.j . , 1991 , mnras , 248 , 353 press w.h . , teukolsky s.a . , vetterling w.t . , flannery b.p . 1997 , numerical recipes in c 2nd ed . , cambridge university press , cambridge savonije g.j . , heemskerk m.h.m . , 1990 , a&a , 240 , 191 syer d. , tremaine s. , 1996 , mnras , 281,925 zang t.a . , 1976 , ph.d thesis , massachusetts institute of technology the linear operator @xmath244 appeared in equation ( [ eq : linearized - perturbed - equations ] ) is in terms of the active surface density @xmath245 , and the velocity field @xmath246,\ ] ] of the equilibrium state . the elements of @xmath244 are @xmath247 , \\ l_{13 } \!\!\!\ ! & = & \!\!\!\ ! { 1\over r } \left [ \partial _ \phi + \partial _ \phi \left ( \ln \sigma _ { \rm act } \right ) \right ] , \\ l_{21 } \!\!\!\ ! & = & \!\!\!\ ! { c^2(\phi)\over \sigma_{\rm act } } \left [ \partial _ r - \partial _ r \left ( \ln \sigma _ { \rm act } \right ) \right ] , \\ l_{22 } \!\!\!\ ! & = & \!\!\!\ ! { \partial _ t}+\partial _ r v_{r0 } + v_{r0 } \partial _ r + { v_{\phi 0}\over r}\partial _ l_{23 } \!\!\!\ ! & = & \!\!\!\ ! { 1\over r}\partial _ \phi v_{r0 } - { 2 v_{\phi 0 } \over r } , \\ l_{31 } \!\!\!\ ! & = & \!\!\!\ ! { 1\over \sigma_{\rm act } } \left \ { \frac 1r \partial _ \phi c^2(\phi ) + { c^2(\phi)\over r } \left [ \partial _ \phi - \partial _ \phi \left ( \ln \sigma_{\rm act } \right ) \right ] \right \ } , \\ l_{32 } \!\!\!\ ! & = & \!\!\!\ ! \partial _ r v_{\phi 0}+ { v_{\phi 0}\over r } , \\ l_{33 } \!\!\!\ ! & = & \!\!\!\ ! { \partial _ t } + \frac 1{r}\partial _ \phi v_{\phi 0}+{v_{r 0}\over r}+v_{r 0}\partial _ r + { v_{\phi 0}\over r } \partial _ \phi.\end{aligned}\ ] ] the operators @xmath249=@xmath250 $ ] ( @xmath66 ) appear in the interacting terms of ( [ eq::perturbed - vec - m - mode ] ) . they are functions of @xmath1 and @xmath251 . let us define @xmath188 and introduce @xmath252\epsilon^2 , \nonumber\\ \!\!\!&{}&\!\!\ ! \mu_2 = \frac{4 + 3\epsilon^2 + 8n(1 - 2n)\left ( 1+\epsilon^2 \right ) } { 8(1+n)(-1 + 4n^2 ) } \nonumber \\ \!\!\ ! & { } & \!\!\ ! -\frac{16n^3\left ( 2 + 3\epsilon^2 \right ) } { 8(1+n)(-1 + 4n^2 ) } \nonumber\\ \!\!\!&{}&\!\!\!+{m}^2\left[\frac{2+n}{2(1+n)}+ \frac{72n^3 + 104n^2 - 22n-27}{8(1+n)(-1 + 4n^2 ) } \ \epsilon^2\right ] , \nonumber\\ \!\!\!&{}&\!\!\ ! \mu_3 = \frac{12n^3 + 6n^2-n-1}{4(1+n)(-1 + 4n^2 ) } \nonumber\\ \!\!\!&{}&\!\!\!-{m}^2\frac{12n^3 + 18n^2 - 5n-5}{4(1+n)(-1 + 4n^2 ) } , \nonumber\\ \!\!\!&{}&\!\!\ ! \mu_4 = \frac{-16n^3 - 8n^2 + 1}{8(1+n)(-1 + 4n^2 ) } \nonumber\\ \!\!\!&{}&\!\!\!+{m}^2\frac{24n^3 + 32n^2 - 10n-9}{8(1+n)(-1 + 4n^2)}. \nonumber\end{aligned}\ ] ] for @xmath253 , we have found @xmath254\nonumber \\ \!\!\!&{}&\!\!\!+h(r)\left(\frac{r_0 ^ 2}{r}\right)\frac{d}{dr } , \nonumber\\ d_{0,13}\!\!\!&=&\!\!\ ! \textrm{i } k h(r)\left(\frac{r_0}{r}\right)^2 , \nonumber\\ d_{0,21}\!\!\!&=&\!\!\ ! \mu_1 h_{1}(r)+\mu_1 \frac{r}{h(r ) } \frac{d}{dr } , \nonumber\\ d_{0,22}\!\!\!&=&\!\!\ ! \textrm{i } k { m}\left(\frac{r_0}{r}\right)\left ( 1+\frac12 \epsilon^2 \right ) , \nonumber \\ -2 { m}\left(\frac{r_0}{r}\right)\left ( 1+\frac12 \epsilon^2 \right ) , \nonumber \\ d_{0,31}\!\!\!&=&\!\!\ ! \textrm{i } k \mu_1 \frac{1}{h(r)},\nonumber \\ d_{0,32}\!\!\!&=&\!\!\ ! { m}\left(\frac{r_0}{r}\right)\left ( 1+\frac12 \epsilon^2 \right ) , \nonumber\\ d_{0,33}\!\!\!&=&\!\!\ ! \textrm{i } k { m}\left(\frac{r_0}{r}\right)\left ( 1+\frac12 \epsilon^2 \right ) , \nonumber\\ d_{+1,11}\!\!\!&=&\!\!\ ! -\textrm{i } \frac{1+{m}^2}{2{m}(1+n ) } \nonumber \\ \!\!\ ! & { } & \!\!\ ! \times \left[h_1(r)h(r)\left(\frac{r_0}{r}\right)+r_0\frac{d}{dr}\right ] \nonumber\\ \!\!\!&{}&\!\!\!-\textrm{i } k{m}\left ( \frac{1}{2}+\frac38 \epsilon^2 \right ) \left(\frac{r_0}{r}\right ) , \nonumber\\ d_{+1,12}\!\!\!&=&\!\!\ ! ^ 2}{r}\right)\frac{d}{dr } , \nonumber\\ d_{+1,13}\!\!\!&=&\!\!\ ! \frac12\textrm{i}k h(r)\left(\frac{r_0}{r}\right)^2 , \nonumber\\ d_{+1,21}\!\!\!&=&\!\!\ ! \mu_2 h_1(r ) + \mu_2 \frac{r}{h(r ) } \frac{d}{dr } , \nonumber\\ d_{+1,22}\!\!\!&=&\!\!\ ! -\textrm{i } \frac{1+{m}^2}{2{m}(1+n)}r_0\frac{d}{dr}\nonumber\\ \!\!\!&{}&\!\!\!-\textrm{i } { m}(k+n)\left ( \frac{1}{2}+\frac38 \epsilon^2 \right ) \left(\frac{r_0}{r}\right ) , \nonumber\\ d_{+1,23}\!\!\!&=&\!\!\ ! \left[\frac{{m}^2(2+n)-n}{2{m}(1+n)}+\frac34{m}\epsilon^2\right ] \left(\frac{r_0}{r}\right ) , \nonumber\\ d_{+1,31}\!\!\!&=&\!\!\ ! \textrm{i}k \mu_2 \frac{1}{h(r ) } , \nonumber\\ d_{+1,32}\!\!\!&=&\!\!\ ! -{m}\left ( \frac{1}{2}+\frac38 \epsilon^2 \right ) \left(\frac{r_0}{r}\right ) , \nonumber\\ d_{+1,33}\!\!\!&=&\!\!\ ! -\textrm{i } \frac{1+{m}^2}{2{m}(1+n)}\left[r_0\frac{d}{dr}+ \left(\frac{r_0}{r}\right)\right]\nonumber\\ \!\!\!&{}&\!\!\!-\textrm{i } k { m}\left ( \frac{1}{2}+\frac38 \epsilon^2 \right ) \left(\frac{r_0}{r}\right ) , \nonumber\\ d_{-1,11}\!\!\!&=&\!\!\ ! \textrm{i } \frac{1+{m}^2}{2{m}(1+n ) } \nonumber \\ \!\!\ ! & { } & \!\!\ ! \times \left[h_1(r)h(r)\left(\frac{r_0}{r}\right)+r_0\frac{d}{dr}\right ] \nonumber\\ \!\!\!&{}&\!\!\!-\textrm{i } k{m}\left ( \frac{1}{2}+\frac38 \epsilon^2 \right ) \left(\frac{r_0}{r}\right ) , \nonumber\\ d_{-1,12}\!\!\!&=&\!\!\ ! h^2(r)\left[\alpha^2\left(\frac{r_0}{r}\right)^4-\gamma^2\right]\nonumber\\ \!\!\!&{}&\!\!\!+\frac12h(r)\left(\frac{r_0 ^ 2}{r}\right)\frac{d}{dr } , \nonumber\\ d_{-1,13}\!\!\!&=&\!\!\ ! \frac12\textrm{i}k h(r)\left(\frac{r_0}{r}\right)^2 , \nonumber\\ d_{-1,21}\!\!\!&=&\!\!\ ! \mu_2 h_1(r ) + \mu_2 \frac{r}{h(r ) } \frac{d}{dr } , \nonumber\\ d_{-1,22}\!\!\!&=&\!\!\ ! \frac{1+{m}^2}{2{m}(1+n)}r_0\frac{d}{dr}\nonumber\\ \!\!\!&{}&\!\!\!-\textrm{i } { m}(k - n ) \left ( \frac{1}{2}+\frac38 \epsilon^2 \right ) \left(\frac{r_0}{r}\right ) , \nonumber\\ d_{-1,23}\!\!\!&=&\!\!\ ! \left[\frac{{m}^2(2+n)-n}{2{m}(1+n)}+\frac34{m}\epsilon^2\right ] \left(\frac{r_0}{r}\right ) , \nonumber \\ \frac{1}{h(r ) } , \nonumber \\ d_{-1,32}\!\!\!&=&\!\!\ ! -{m}\left ( \frac{1}{2}+\frac38 \epsilon^2 \right ) \left(\frac{r_0}{r}\right ) \nonumber \\ d_{-1,33}\!\!\!&=&\!\!\ ! \textrm{i } \frac{1+{m}^2}{2{m}(1+n)}\left[r_0\frac{d}{dr}+\left(\frac{r_0}{r}\right)\right]\nonumber\\ \!\!\!&{}&\!\!\!-\textrm{i } k { m}\left ( \frac{1}{2}+\frac38 \epsilon^2 \right ) \left(\frac{r_0}{r}\right ) , \nonumber \\ d_{+2,11}\!\!\!&=&\!\!\ ! \textrm{i } \frac{1+{m}^2}{4{m}(-1 + 4n^2 ) } \nonumber \\ \!\!\ ! & { } & \!\!\ ! \times \left[h_1(r)h(r)\left(\frac{r_0}{r}\right)+r_0\frac{d}{dr}\right ] \nonumber \\ \!\!\!&{}&\!\!\ ! + \frac{1}{4}\textrm{i } k { m}\left(\frac{r_0}{r}\right ) , \nonumber \\ d_{+2,12}\!\!\!&=&\!\!\ ! 0 \nonumber \\ d_{+2,13}\!\!\!&=&\!\!\ ! 0 \nonumber \\ \mu_3 h_1(r ) + \mu_3 \frac{r}{h(r)}\frac{d}{dr } \nonumber \\ d_{+2,22}\!\!\!&=&\!\!\ ! \textrm{i}\frac{1+{m}^2}{4{m}(-1 + 4n^2)}r_0\frac{d}{dr}\nonumber \\ \!\!\ ! & { } & \!\!\ ! + \textrm{i}\frac{{m}(k+2n)}{4}\left(\frac{r_0}{r}\right ) \nonumber \\ d_{+2,23}\!\!\!&=&\!\!\ ! \frac{n+{m}^2(1+n-4n^2)}{2{m}(-1 + 4n^2)}\left(\frac{r_0}{r}\right ) \nonumber \\ \textrm{i}k \mu_3 \frac{1}{h(r ) } \nonumber \\ \frac{1}{4}{m}\left(\frac{r_0}{r}\right ) \nonumber \\ d_{+2,33}\!\!\!&=&\!\!\ ! \textrm{i}\frac{1+{m}^2}{4{m}(-1 + 4n^2 ) } \left[\left(\frac{r_0}{r}\right)+r_0\frac{d}{dr}\right]\nonumber\\ \!\!\!&{}&\!\!\!+\frac{1}{4}\textrm{i}k { m}\left(\frac{r_0}{r}\right ) \nonumber \\ d_{-2,11}\!\!\!&=&\!\!\ ! -\textrm{i } \frac{1+{m}^2}{4{m}(-1 + 4n^2 ) } \nonumber \\ \!\!\ ! & { } & \!\!\ ! \times \left[h_1(r)h(r)\left(\frac{r_0}{r}\right)+r_0\frac{d}{dr}\right ] \nonumber \\ \!\!\!&{}&\!\!\ ! + \frac{1}{4}\textrm{i } k { m}\left(\frac{r_0}{r}\right ) , \nonumber \\ d_{-2,12}\!\!\!&=&\!\!\ ! 0 , \nonumber \\ d_{-2,13}\!\!\!&=&\!\!\ ! 0 , \nonumber \\ \mu_3 h_1(r ) + \mu_3 \frac{r}{h(r)}\frac{d}{dr } , \nonumber \\ d_{-2,22}\!\!\!&=&\!\!\ ! -\textrm{i}\frac{1+{m}^2}{4{m}(-1 + 4n^2)}r_0\frac{d}{dr}\nonumber \\ \!\!\ ! & { } & \!\!\ ! + \textrm{i}\frac{{m}(k-2n)}{4 } \left(\frac{r_0}{r}\right ) , \nonumber \\ \frac{n+{m}^2(1+n-4n^2)}{2{m}(-1 + 4n^2)}\left(\frac{r_0}{r}\right ) , \nonumber\\ d_{-2,31}\!\!\!&=&\!\!\ ! \textrm{i}k \mu_3 \frac{1}{h(r ) } , \nonumber\\ d_{-2,32}\!\!\!&=&\!\!\ ! \frac{1}{4}{m}\left(\frac{r_0}{r}\right ) , \nonumber\\ d_{-2,33}\!\!\!&=&\!\!\!-\textrm{i}\frac{1+{m}^2}{4{m}(-1 + 4n^2 ) } \left[\left(\frac{r_0}{r}\right)+r_0\frac{d}{dr}\right]\nonumber\\ \!\!\!&{}&\!\!\!+\frac{1}{4}\textrm{i}k { m}\left(\frac{r_0}{r}\right ) , \nonumber\\ d_{+3,11}\!\!\!&=&\!\!\ ! -\frac18\textrm{i } k { m}\left(\frac{r_0}{r}\right ) , \nonumber\\ d_{+3,12}\!\!\!&=&\!\!\ ! 0 , \nonumber\\ d_{+3,13}\!\!\!&=&\!\!\ ! 0 , \nonumber\\ d_{+3,21}\!\!\!&=&\!\!\!\mu_4 h_1(r ) + \mu_4 \frac{r}{h(r)}\frac{d}{dr } , \nonumber\\ d_{+3,22}\!\!\!&=&\!\!\ ! -\frac18\textrm{i}{m}(k+3n)\left(\frac{r_0}{r}\right ) , \nonumber\\ d_{+3,23}\!\!\!&=&\!\!\ ! \frac{1}{4}{m}\left(\frac{r_0}{r}\right ) , \nonumber\\ d_{+3,31}\!\!\!&=&\!\!\ ! \textrm{i}k \mu_4 \frac{1}{h(r ) } , \nonumber\\ d_{+3,32}\!\!\!&=&\!\!\ ! -\frac{1}{8}{m}\left(\frac{r_0}{r}\right ) , \nonumber\\ d_{+3,33}\!\!\!&=&\!\!\ ! -\frac{1}{8}\textrm{i}k{m}\left(\frac{r_0}{r}\right ) , \nonumber\\ d_{-3,11}\!\!\!&=&\!\!\ ! -\frac{1}{8}\textrm{i}k{m}\left(\frac{r_0}{r}\right ) , \nonumber\\ d_{-3,12}\!\!\!&=&\!\!\ ! 0 , \nonumber\\ d_{-3,13}\!\!\!&=&\!\!\ ! 0 , \nonumber\\ d_{-3,21}\!\!\!&=&\!\!\ ! \mu_4 h_1(r ) + \mu_4 \frac{r}{h(r)}\frac{d}{dr } , \nonumber\\ d_{-3,22}\!\!\!&=&\!\!\ ! -\frac{1}{8}\textrm{i}{m}(k-3n ) \left(\frac{r_0}{r}\right ) , \nonumber\\ d_{-3,23}\!\!\!&=&\!\!\ ! \frac{1}{4 } { m}\left(\frac{r_0}{r}\right ) , \nonumber\\ d_{-3,31}\!\!\!&=&\!\!\ ! \textrm{i}k \mu_4 \frac{1}{h(r ) } , \nonumber\\ d_{-3,32}\!\!\!&=&\!\!\ ! -\frac{1}{8}{m}\left(\frac{r_0}{r}\right ) , \nonumber\\ d_{-3,33}\!\!\!&=&\!\!\ ! -\frac{1}{8}\textrm{i}k{m}\left(\frac{r_0}{r}\right ) . \nonumber\\\end{aligned}\ ] ]

first order invariants of generic immersions of manifolds of dimension into manifolds of dimension , are constructed using the geometry of self - intersections . the range of one of these invariants is related to bernoulli numbers . as by - products some geometrically defined invariants of regular homotopy are found .

math9904031

an immersion of a smooth manifold @xmath3 into a smooth manifold @xmath4 is a smooth map with everywhere injective differential . two immersions are regularly homotopic if they can be connected by a continuous 1-parameter family of immersions . an immersion is generic if all its self - intersections are transversal . in the space @xmath5 of immersions @xmath6 , generic immersions form an open dense subspace . its complement is _ the discriminant hypersurface _ , two generic immersions belong to the same path component of @xmath8 if they can be connected by a regular homotopy , which at each instance is a generic immersion . we shall consider the classification of generic immersions up to regular homotopy through generic immersions . it is similar to the classification of embeddings up to diffeotopy ( knot theory ) . in both cases , all topological properties of equivalent maps are the same . an _ invariant of generic immersions _ is a function on @xmath8 which is locally constant . the value of such a function along a path in @xmath5 jumps at intersections with @xmath9 . invariants may be classified according to the complexity of their jumps . the most basic invariants in this classification are called _ first order invariants _ ( see section [ mainres ] ) . in @xcite , arnold studies generic regular plane curves ( i.e. generic immersions @xmath10 ) . he finds three first order invariants @xmath11 , @xmath12 , and @xmath13 . in @xcite , the author considers the case @xmath14 . two first order invariants @xmath15 and @xmath16 are found . in both these cases , the only self - intersections of generic immersions are transversal double points and in generic 1-parameter families there appear isolated instances of self - tangencies and triple points . the values of the invariants @xmath17 and @xmath15 change at instances of self - tangency and remain constant at instances of triple points . the invariants @xmath13 and @xmath16 change at triple points and remain constant at self - tangencies . in this paper we shall consider high - dimensional analogs of these invariants . the most straightforward generalizations arise for immersions @xmath18 , where self - tangencies and triple points are the only degeneracies in generic 1-parameter families . the above range of dimensions is included in a 2-parameter family , @xmath19 , @xmath2 , where generic immersions do not have @xmath20-fold self - intersection points if @xmath21 and in generic 1-parameter families there appear isolated instances of @xmath22-fold self - intersection . under these circumstances , we find first order invariants of generic immersions . ( for precise statements , see theorem [ thml ] and theorem [ thmj ] in section [ mainres ] , where the main results of this paper are formulated . ) in particular , if @xmath23 is even and @xmath24 is orientable and satisfies a certain homology condition ( see theorem [ thml ] ( b ) ) then there exists an integer - valued invariant @xmath16 of generic immersions @xmath25 which is an analog of arnold s @xmath13 : it changes under instances of @xmath22-fold self - intersection , does not change under other degeneracies which appear in generic 1-parameter families , and is additive under connected sum . the value of @xmath16 at a generic immersion @xmath26 is the linking number of a copy of the set of @xmath27-fold self - intersection points of @xmath26 , shifted in a special way , with @xmath28 in @xmath29 . ( see definition [ dfnl ] . ) for immersions of odd - dimensional spheres in codimension two there appear restrictions on the possible values of @xmath16 ( see theorem [ thmdivl ] ) . this phenomenon is especially interesting in the case of immersions @xmath30 . using a result of hughes and melvin @xcite we prove that the range of @xmath16 is related to bernoulli numbers : for any generic immersion @xmath31 , @xmath32 is divisible by @xmath33 , where @xmath34 is a prime and @xmath35 is an integer such that @xmath36 divides @xmath37 and @xmath38 does not divide @xmath39 for some integer @xmath20 . here @xmath39 is the denominator of @xmath40 , where @xmath41 is the @xmath42 bernoulli number . in general , it is not known if the above result gives all restrictions on the range of @xmath16 . however , in special cases it does . for example , for immersions @xmath14 the range of @xmath16 is @xmath43 ( see @xcite and also remark [ lkrmk ] ) and for immersions @xmath44 it is @xmath45 ( see section [ rmksonl ] ) . in @xcite , the author gives a complete classification of generic immersions @xmath46 , @xmath47 , @xmath48 up to regular homotopy through generic immersions ( under some conditions on the lower homotopy groups of @xmath3 ) . here , the class of a generic immersion is determined by its self - intersection with induced natural additional structures ( e.g. spin structures ) . the existence of invariants such as @xmath16 mentioned above implies that the corresponding classification in other dimensions is more involved ( see remark [ nonloc ] ) . the invariants of generic immersions in theorems [ thml ] and [ thmj ] give rise to invariants of regular homotopy which take values in finite cyclic groups ( section [ rhinv ] ) . we construct examples showing that , depending on the source and target manifolds , the regular homotopy invariants may or may not be trivial ( section [ expbm ] ) . it would be interesting to relate these invariants to invariants arising from the cobordism theory of immersions ( see for example eccles @xcite ) . before stating the main results we define the notions of invariants of orders zero and one . they are similar to knot invariants of finite type introduced by vassiliev in @xcite . as in the introduction , let @xmath5 denote the space of immersions and let @xmath7 denote the discriminant hypersurface . the set @xmath49 of non - generic immersions which appear at isolated instances in generic 1-parameter families ( see lemmas [ vdef1sm ] and [ vdef1big ] for descriptions of such immersions @xmath19 ) is a smooth submanifold of @xmath5 of codimension one . ( @xmath5 is an open subspace of the space of smooth maps , thus an infinite dimensional manifold and the notion of codimension in @xmath5 makes sense . ) if @xmath50 then there is a neighborhood @xmath51 of @xmath52 in @xmath5 cut in two parts by @xmath53 . a coherent choice of a positive and negative part of @xmath51 for each @xmath50 is a _ coorientation _ of @xmath9 . ( the discriminant hypersurface in the space of immersions @xmath19 will be cooriented in section [ codiscr ] . ) let @xmath54 be an invariant of generic immersions and let @xmath50 . define the jump @xmath55 of @xmath54 as @xmath56 where @xmath57 and @xmath58 are generic immersions in the positive respectively negative part of @xmath51 . then @xmath55 is a locally constant function on @xmath53 . an invariant @xmath54 of generic immersions is a _ zero order invariant _ if @xmath59 . let @xmath60 be the set of all immersions which appear in generic 2-parameter families but can be avoided in generic 1-parameter families ( see lemma [ vdef2 ] for the case @xmath19 ) . then @xmath61 is a smooth codimension two submanifold of @xmath5 . an invariant @xmath54 of generic immersions is a _ first order invariant _ if @xmath62 for any immersions @xmath63 which can be joined by a path in @xmath64 such that at intersections with @xmath60 its tangent vector is transversal to the tangent space of @xmath60 . [ thml ] let @xmath65 and @xmath66 be integers and let @xmath24 be a closed manifold . * if @xmath27 is even and @xmath67 then there exists a unique ( up to addition of zero order invariants ) first order @xmath68-valued invariant @xmath69 of generic immersions @xmath70 satisfying the following conditions : it jumps by @xmath71 on the part of @xmath53 which consists of immersions with one @xmath22-fold self - intersection point and does not jump on other parts of @xmath53 . * if @xmath23 is even , @xmath3 is orientable , and @xmath72 then there exists a unique ( up to addition of zero order invariants ) first order integer - valued invariant @xmath16 of generic immersions @xmath70 satisfying the following conditions : it jumps by @xmath73 on the part of @xmath53 which consists of immersions with one @xmath22-fold self - intersection point and does not jump on other parts of @xmath53 . theorem [ thml ] is proved in section [ pfthml ] . the invariants @xmath69 and @xmath16 are defined in definition [ dfnlambda ] and definition [ dfnl ] , respectively . if appropriately normalized , @xmath69 and @xmath16 are additive under connected summation of generic immersions ( see section [ csgi ] ) . [ thmdivl ] let @xmath74 be a generic immersion . * if @xmath75 then @xmath37 divides @xmath32 . * if @xmath76 then @xmath77 divides @xmath32 . * if @xmath78 then @xmath33 divides @xmath32 for every prime @xmath34 and integer @xmath35 such that @xmath36 divides @xmath37 and @xmath38 does not divide @xmath39 for some integer @xmath20 . here @xmath39 is the denominator of @xmath40 , where @xmath41 is the @xmath42 bernoulli number . we prove theorem [ thmdivl ] in section [ pfthmdivl ] . [ thmj ] let @xmath65 and @xmath66 be integers and let @xmath24 be a closed manifold and let @xmath79 be a manifold . * if @xmath23 is odd then , for each integer @xmath80 such that @xmath81 is even , there exist a unique ( up to addition of zero order invariants ) first order integer - valued invariant @xmath82 , of generic immersions @xmath19 satisfying the following conditions : the invariant jumps by @xmath83 on the part of @xmath53 which consists of immersions with one degenerate @xmath35-fold self - intersection point and does not jump on other parts of @xmath53 . * if @xmath84 then there exists a unique ( up to addition of zero order invariants ) first order integer - valued invariant @xmath15 , of generic immersions @xmath85 satisfying the following conditions : the invariant jumps by @xmath71 on the part of @xmath53 which consists of immersions with one degenerate @xmath27-fold self - intersection point and does not jump on other parts of @xmath53 . theorem [ thmj ] is proved in section [ pfthmj ] . if appropriately normalized , @xmath82 and @xmath15 are additive under connected summation of generic immersions ( see section [ csgi ] ) . the value of @xmath82 on a generic immersion is the euler characteristic of its resolved @xmath35-fold self - intersection manifold , the value of @xmath15 is the number of components in its @xmath27-fold self - intersection ( which is a closed 1-dimensional manifold ) . in this section we define generic immersions and describe the immersions corresponding to non - generic instances in generic 1- and 2-parameter families . we also describe the versal deformations of these non - generic immersions . let @xmath3 and @xmath4 be smooth manifolds and let @xmath86 be an immersion . a point @xmath87 is a _ @xmath20-fold self - intersection point of @xmath26 _ if @xmath88 consists of exactly @xmath20 points . let @xmath89 denote the set of @xmath20-fold self - intersection points of @xmath26 and let @xmath90 denote its preimage . note that @xmath91 is a @xmath20-fold covering . [ dfngi ] let @xmath65 and @xmath66 . an immersion @xmath92 is _ generic _ if * @xmath93 is empty for @xmath21 , and * if @xmath94 @xmath95 then for any @xmath96 , @xmath97 @xmath98 a standard application of the jet - transversality theorem shows that the set of generic immersions is open and dense in the space of all immersions @xmath5 . we remark that the set of @xmath20-fold self - intersection points @xmath93 of a generic immersion @xmath92 is a smooth submanifold of @xmath4 of dimension @xmath99 . moreover , the deepest self - intersection @xmath100 is a closed manifold . lemma [ giloc ] below gives a local coordinate description of a generic immersion close to a @xmath20-fold self - intersection point . to state it we introduce some notation : let @xmath101 , we write @xmath102 , where @xmath103 and @xmath104 , for @xmath105 . similarly , we write @xmath106 as @xmath107 , where @xmath108 and @xmath109 , for @xmath110 . [ giloc ] let @xmath92 be a generic immersion and let @xmath111 be a @xmath20-fold self intersection point of @xmath26 . then there are coordinates @xmath112 in @xmath113 centered at @xmath114 and coordinates @xmath115 on @xmath116 centered at @xmath117 , @xmath118 such that @xmath26 is given by @xmath119 ( that is , if @xmath120 then @xmath121 , @xmath122 for @xmath123 , @xmath124 , and @xmath125 for @xmath126 . ) the proof is straightforward . when the source and target of a generic immersion are oriented and the codimension is even then there are induced orientations on the self intersection manifolds : let @xmath127 , @xmath65 , @xmath128 , and let @xmath129 be a generic immersion . orientations on @xmath130 and @xmath131 induce an orientation on @xmath93 . let @xmath132 denote the normal bundle of the immersion . the decomposition @xmath133 induces an orientation on @xmath132 . if @xmath134 , @xmath111 then @xmath135 the orientation on @xmath132 induces an orientation on @xmath136 . since the dimension of the bundle @xmath132 is @xmath137 which is even , the orientation on the sum is independent on the ordering of the summands . hence , the decomposition above induces a well - defined orientation on @xmath138 . our next result describes ( the 1-jets of ) immersions in @xmath53 ( see section [ mainres ] ) . [ 1jetsi1 ] if @xmath139 is an immersion in @xmath53 then * g1 * and * g2 * of definition [ dfngi ] holds , except at one @xmath20-fold @xmath140 self - intersection point @xmath141 , where , * if @xmath142 , @xmath143 or * if @xmath144 , for @xmath145 @xmath146 and @xmath147 we have to show that degeneracies as above appears at isolated parameter values in generic 1-parameter families , and that further degeneracies ( of the 1-jet ) may be avoided . this follows from the jet - transversality theorem applied to maps @xmath148\to w^{n(m+1)-1}$ ] . a @xmath20-fold self - intersection point @xmath114 of an immersion @xmath139 where * g1 * and * g2 * of definition [ dfngi ] does not hold will be called a _ degenerate self - intersection point of @xmath52_. if @xmath128 we say that @xmath114 is a _ @xmath20-fold self - tangency point of @xmath52 _ if @xmath149 we say that @xmath114 is a _ @xmath22-fold self - intersection point of @xmath52_. recall that a deformation @xmath150 of a map @xmath26 is called _ versal _ if any deformation of @xmath26 is equivalent ( up to left - right action of diffeomorphisms ) to one induced from @xmath150 . let @xmath52 be an immersion in @xmath53 . then its versal deformation @xmath151 is a 1-parameter deformation . in other words , it is a path @xmath152 in @xmath5 which intersects @xmath53 transversally at @xmath52 and thus , @xmath151 are generic immersions for small @xmath153 . next , we shall describe immersions @xmath50 in local coordinates close to their degenerate self intersection point . self - tangency points and @xmath22-fold self - intersection points are treated in lemma [ vdef1sm ] and lemma [ vdef1big ] , respectively . to accomplish this we need a more detailed description of coordinates than that given in lemma [ giloc ] , we use coordinates as there with one more ingredient : we write ( when necessary ) @xmath154 and @xmath155 , @xmath156 as @xmath157 and @xmath158 , where @xmath159 and @xmath160 . ( greek letters for scalars and roman for vectors ) . [ vdef1sm ] let @xmath139 be an immersion in @xmath53 and let @xmath141 be a point of degenerate @xmath20-fold self intersection @xmath161 . then there are coordinates @xmath112 on @xmath162 centered at @xmath114 and coordinates @xmath115 on @xmath163 centered at @xmath117 , @xmath118 such that in these coordinates the versal deformation @xmath151 , @xmath164 of @xmath52 is constant outside of @xmath165 and in neighborhoods of @xmath166 it is given by * @xmath167 + ( that is , if @xmath168 then _ for @xmath169 _ , @xmath121 , @xmath170 , @xmath122 for @xmath123 , @xmath124 , and @xmath125 for @xmath126 . ) * @xmath171 + where @xmath172 is a nondegenerate quadratic form in the @xmath173 variables @xmath174 . it is straightforward to see that we can find coordinates @xmath115 on @xmath175 , @xmath176 , so that up to first order of approximation @xmath177 is given by the expressions in ( a ) and the first term in ( b ) above . we must consider also second order terms : let @xmath132 be a linear subspace in the coordinates @xmath112 transversal to @xmath178 . let @xmath179 be orthogonal projection onto @xmath132 . then @xmath180 and @xmath181 is 1-dimensional . the second derivative @xmath182 must be a non - degenerate quadratic form , otherwise , we can avoid @xmath52 in generic 1-parameter families . ( this is a consequence of the jet transversality theorem . ) the morse lemma then implies that , after possibly adjusting the coordinates in @xmath183 by adding quadratic expressions in @xmath174 to @xmath184 , @xmath185 , there exists coordinates for @xmath52 as stated . finally , we must prove that the deformation @xmath151 as given above is versal . a standard result in singularity theory says that it is enough to prove that the deformation is infinitesimally versal . this is straightforward . in lemma [ vdef1big ] below we will use coordinates as at generic @xmath27-fold self intersection points . that is , @xmath186 and @xmath187 , where @xmath188 and @xmath189 for @xmath190 . [ vdef1big ] let @xmath139 be an immersion in @xmath53 and let @xmath191 be a point of @xmath22-fold self - intersection . then there are coordinates @xmath112 on @xmath162 centered at @xmath114 and coordinates @xmath115 on @xmath163 centered at @xmath117 , @xmath118 such that in these coordinates the versal deformation @xmath151 , @xmath164 of @xmath52 is constant outside of @xmath165 and in neighborhoods of @xmath166 it is given by * @xmath192 + ( that is , if @xmath168 then _ for @xmath193 _ , @xmath121 , @xmath122 for @xmath194 , @xmath195 , and @xmath125 for @xmath196 . ) * @xmath197 the proof is similar to the proof of lemma [ vdef1sm ] , but easier . let @xmath198 be an immersion in @xmath60 . then either ( a ) or ( b ) below holds . * @xmath199 has two distinct degenerate self - intersection points @xmath200 and @xmath201 . locally around @xmath202 , @xmath203 @xmath199 is as in lemma [ vdef1sm ] or lemma [ vdef1big ] . the versal deformation of @xmath199 is a product of the corresponding 1-parameter versal deformations . * @xmath199 has one degenerate @xmath20-fold @xmath128 self - intersection point @xmath114 . there are coordinates @xmath112 centered at @xmath114 and coordinates @xmath115 centered at @xmath117 , @xmath118 such that in these coordinates the versal deformation @xmath204 , @xmath205 of @xmath199 is constant outside of @xmath165 and in a neighborhood of @xmath166 it is given by * * @xmath206 + that is , if @xmath207 then _ for @xmath169 _ , @xmath121 , @xmath170 , @xmath122 for @xmath123 , @xmath124 , and @xmath125 for @xmath126 . * * @xmath208 + where @xmath172 is a nondegenerate quadratic form in the @xmath99 variables @xmath209 . the jet - transversality theorem applied to maps of @xmath210 ^ 2 $ ] into @xmath131 shows that immersions with points of @xmath20-fold self - intersection @xmath211 , as well as immersions with points of @xmath20-fold self - intersection points @xmath212 at which the 1-jet has further degenerations than the 1-jets of lemma [ 1jetsi1 ] can be avoided in generic 2-parameter families . assume that @xmath199 has a degenerate @xmath20-fold self - intersection point . if @xmath149 it is easy to see that @xmath199 has the same local form as the map in lemma [ vdef1big ] . so , immersions with @xmath22-fold self - intersection points appears along 1-parameter subfamilies in generic 2-parameter families . if @xmath213 we proceed as in the proof of lemma [ vdef1sm ] and construct the map @xmath214 . applying the jet transversality theorem once again we see that if the rank of @xmath215 is smaller than @xmath99 then @xmath199 can be avoided in generic 2-parameter families . if the rank of @xmath215 equals @xmath99 then the third derivative of @xmath214 in the direction of the null - space of @xmath215 must be non - zero otherwise @xmath199 can be avoided in generic 2-parameter families . with this information at hand we can find coordinates as claimed . finally , as in the proof of lemma [ vdef1sm ] , we must check that the deformation given is infinitesimally versal . this is straightforward . pictures ( a ) and ( b ) in figure [ figcod2 ] correspond to cases ( a ) and ( b ) in theorem [ vdef2 ] . the codimension two parts of the discriminant are represented by points . in ( a ) two branches of the discriminant , which consist of immersions with one degenerate @xmath35-fold and one degenerate @xmath20-fold self - intersection point respectively , intersect in @xmath60 . in ( b ) the smooth points of the semi - cubical cusp represents immersions with one degenerate @xmath20-fold self - intersection point . the singular point represents @xmath60 . if an immersion in @xmath9 above the singular point ( @xmath60 ) is moved below it then the index of the quadratic form @xmath172 in local coordinates close to the degenerate self - intersection point ( see lemma [ vdef1sm ] ) changes . in this section we define the invariants @xmath82 , @xmath15 , @xmath16 and @xmath69 . to this end , we describe resolutions of self - intersections ( for the definitions of @xmath82 and @xmath15 ) and we compute homology of image - complements of and of normal bundles ( for the definitions of @xmath16 and @xmath69 ) . for generic immersions @xmath92 let @xmath216 , for @xmath217 and let @xmath218 . resolving @xmath219 we obtain a smooth manifold @xmath220 : [ resol ] let @xmath65 and @xmath66 and @xmath217 be integers . let @xmath92 be a generic immersion . then there exists closed @xmath221-manifolds @xmath222 and @xmath220 , unique up to diffeomorphisms , and immersions @xmath223 and @xmath224 such that the diagram @xmath225 commutes . the maps @xmath226 and @xmath227 are surjective , have multiple points only along @xmath228 and @xmath229 respectively , and @xmath34 is a @xmath230-fold cover . this is immediate from lemma [ giloc ] : close to a @xmath20-fold self intersection point @xmath93 looks like the intersection of @xmath20 @xmath231-planes in general position in @xmath29 . [ dfnj_r ] let @xmath65 and @xmath66 be integers and assume that @xmath23 is odd . for a generic immersion @xmath92 and an integer @xmath80 such that @xmath81 is even , define @xmath232 where @xmath233 denotes euler characteristic . [ dfnj ] let @xmath65 be an integer . for a generic immersion @xmath234 , define @xmath235 [ wdj ] the functions @xmath82 and @xmath15 are invariants of generic immersions . let @xmath151 , @xmath236 be a regular homotopy through generic immersions . if @xmath237\to w\times[0,1]$ ] is the immersion @xmath238 then @xmath239 . it follows that @xmath240 is diffeomorphic to @xmath241 . [ hcom ] let @xmath242 be a generic immersion . assume that @xmath3 is closed . then * @xmath243 and * if @xmath3 is oriented then @xmath244 alexander duality implies that @xmath245 . by lemma [ giloc ] , we can choose triangulations of @xmath3 and @xmath28 such that the map @xmath246 is a bijection on the @xmath247-skeleta for @xmath248 . [ idhcom ] in case ( a ) of lemma [ hcom ] we will use the unique @xmath68-orientation of @xmath3 and the duality isomorphism to identify @xmath249 with @xmath68 . in case ( b ) of lemma [ hcom ] , a @xmath43-orientation of @xmath3 determines an isomorphism @xmath250 . we shall use this and the duality isomorphism to identify @xmath251 with @xmath43 . a small ( @xmath252)-dimensional sphere going around a fiber in the normal bundle of @xmath253 generates these groups . consider an immersion @xmath92 . let @xmath132 denote its normal bundle . then @xmath132 is a vector bundle of dimension @xmath23 over @xmath3 . choose a riemannian metric on @xmath132 and consider the associated bundle @xmath254 of unit vectors in @xmath132 this is an ( @xmath252)-sphere bundle over @xmath3 . let @xmath255 denote a fiber of @xmath254 . [ hfib ] let @xmath92 be an immersion . let @xmath256 denote the inclusion of the fiber . * if @xmath257 then @xmath258 is an isomorphism . * if @xmath3 and @xmath4 are oriented and @xmath259 then @xmath260 is an isomorphism . this follows from the leray - serre spectral sequence . [ idhfib ] in case ( a ) of lemma [ hfib ] we use the isomorphism @xmath261 and a canonical generator of @xmath262 to identify @xmath263 with @xmath68 . in case ( b ) of lemma [ hfib ] , orientations of @xmath3 and @xmath4 induce an orientation of the fiber sphere @xmath264 . we use this orientation and @xmath261 in lemma [ hfib ] to identify @xmath265 with @xmath43 . let @xmath92 be a generic immersion . recall that the set of @xmath27-fold self - intersection points of @xmath26 is a closed submanifold @xmath100 of @xmath79 of dimension @xmath252 and its preimage @xmath266 is a closed submanifold of the same dimension in @xmath24 . [ sctn ] if @xmath92 is an immersion then there exists a smooth section @xmath267 , where @xmath254 is the unit sphere bundle of the normal bundle of @xmath26 . this is immediate : the base space of the @xmath23-dimensional vector bundle @xmath268 has dimension @xmath252 . let @xmath214 denote the canonical bundle map over @xmath92 , @xmath269 and let @xmath270 be as in lemma [ sctn ] . we define a vector field @xmath271 along @xmath100 : for @xmath272 let @xmath273 it is easy to see that @xmath274 is well - defined and smooth . we now restrict attention to the case where the target manifold is euclidean space . [ shif ] for a generic immersion @xmath242 , let @xmath275 be the submanifold of @xmath29 obtained by shifting @xmath100 a small distance @xmath276 along @xmath274 . [ skeps ] note that for @xmath276 small enough @xmath277 . moreover , if @xmath278 is such that for all @xmath279 the corresponding @xmath275 is disjoint from @xmath28 then @xmath275 and @xmath280 are homotopic in @xmath281 . [ dfnlambda ] let @xmath242 be a generic immersion . assume that @xmath3 satisfies @xmath282 define @xmath283 as @xmath284-s_\ast[{\widetilde}{\gamma}_m(f)]\in{{\mathbb{z}}}_2,\ ] ] where @xmath276 is very small and @xmath285\in h_{n-1}({{\mathbb{r}}}^{n(m+1)-1}-f(m);{{\mathbb{z}}})\cong{{\mathbb{z}}}_2 $ ] ( see remark [ idhcom ] ) and @xmath286\in h_{n-1}(\partial n;{{\mathbb{z}}})\cong{{\mathbb{z}}}_2 $ ] ( see remark [ idhfib ] ) . [ dfnl ] let @xmath242 be a generic immersion . assume that @xmath23 is even and that @xmath3 is oriented and satisfies @xmath287 define @xmath288 as @xmath289-s_\ast[{\widetilde}{\gamma}_m(f)]\in{{\mathbb{z}}},\ ] ] where @xmath276 is very small and @xmath290\in h_{n-1}({{\mathbb{r}}}^{2m(j+1)-1}-f(m);{{\mathbb{z}}})\cong{{\mathbb{z}}}$ ] ( see remark [ idhcom ] ) and @xmath286\in h_{n-1}(\partial n;{{\mathbb{z}}})\cong{{\mathbb{z}}}$ ] ( see remark [ idhfib ] ) . if @xmath291 and @xmath3 satisfies condition @xmath292 in definition [ dfnl ] then by the universal coefficient theorem @xmath3 also satisfies condition @xmath293 in definition [ dfnlambda ] . it follows from remarks [ idhcom ] , [ idhfib ] that in this case @xmath294 . [ wdl ] @xmath69 and @xmath16 are well - defined . that is , they do neither depend on the choice of @xmath270 , nor on the choice of @xmath276 . we shall often drop the awkward notation @xmath275 and simply write @xmath295 . this is justified by lemma [ wdl ] . the independence of @xmath276 follows immediately from remark [ skeps ] . we will therefore not write all @xmath296 s out in the sequel of this proof . let @xmath297 and @xmath298 be two homotopic sections of @xmath299 . let @xmath300 and @xmath301 be the corresponding normal vector fields along @xmath100 . a homotopy @xmath302 between @xmath297 and @xmath298 induces a homotopy @xmath303 between @xmath300 and @xmath301 and hence between @xmath304 and @xmath305 in @xmath281 . this shows that @xmath69 and @xmath16 only depend on the homotopy class of @xmath270 . to see that @xmath69 and @xmath16 are independent of the vector field we introduce the notion of adding a local twist : let @xmath306 be a section and let @xmath307 . choose a neighborhood @xmath308 of @xmath34 in @xmath266 and a trivialization @xmath309 such that @xmath310 , where @xmath311 is a point in @xmath312 . let @xmath313 be disk inside @xmath308 and let @xmath314 be a smooth map of degree @xmath315 such that @xmath316 in a neighborhood of @xmath317 . let @xmath318 be the section which is @xmath270 outside of @xmath319 and @xmath226 in @xmath319 . we say that @xmath318 is the result of adding a local twist to @xmath270 . ( in the case when @xmath3 is oriented and @xmath23 is even , the local twist is said to be positive if the degree of @xmath226 is @xmath320 and negative if the degree of @xmath226 is @xmath321 . ) let @xmath318 be the vector field obtained by adding a twist to @xmath270 . let @xmath322 and @xmath274 be the vector fields along @xmath100 obtained from @xmath318 and @xmath270 respectively . clearly , @xmath323=s_\ast[\gamma_m(f)]\pm 1 \quad\text { and } \quad [ \gamma_m'(f , v^{\rm tw})]=[\gamma_m'(f , v)]\pm 1.\ ] ] hence , @xmath69 and @xmath16 are invariant under adding local twists . a standard obstruction theory argument shows that if @xmath270 and @xmath324 are two sections of @xmath325 then by adding local twists to @xmath324 we can obtain a section @xmath326 which is homotopic to @xmath270 . hence , @xmath69 and @xmath16 are independent of the choice of section . @xmath69 and @xmath16 are invariants of generic immersions . let @xmath151 , @xmath236 be a regular homotopy through generic immersions . consider the induced map @xmath327 where @xmath328 is the unit interval @xmath329 $ ] . shifting @xmath330 off @xmath331 using a suitable vector field in @xmath332 it is easy to see that @xmath333 . in this section we study how our invariants behave under two natural operations on generic immersions : connected sum and reversing orientation . for ( oriented ) manifolds @xmath3 and @xmath334 of the same dimension , let @xmath335 denote the ( oriented ) connected sum of @xmath3 and @xmath334 . let @xmath242 and @xmath336 be two generic immersions . we shall define the connected sum @xmath337 of these . it will be a generic immersion @xmath338 . let @xmath339 , @xmath340 and @xmath341 be coordinates on @xmath29 . composing the immersions @xmath26 and @xmath342 with translations we may assume that @xmath343 and @xmath344 . choose a point @xmath345 and a point @xmath346 such that there is only one point in @xmath347 and in @xmath348 . pick an arc @xmath349 in @xmath29 connecting @xmath350 to @xmath351 and such that @xmath352 . moreover , assume that @xmath349 meets @xmath28 and @xmath353 transversally at its endpoints . let @xmath132 be the normal bundle of @xmath349 . pick a ( oriented ) basis of @xmath354 and an ( anti - oriented ) one of @xmath355 . these give rise to @xmath0 vectors over @xmath356 in @xmath132 . extend these vectors to @xmath0 independent normal vector fields along @xmath349 . using a suitable map of @xmath132 into a tubular neighborhood of @xmath349 these vector fields give rise to an embedding @xmath357 , where @xmath319 denotes a disk of dimension @xmath0 , such that @xmath358 is an ( orientation preserving ) embedding into @xmath28 and @xmath359 is an ( orientation reversing ) embedding into @xmath353 . the tube @xmath360 now joins @xmath361 to @xmath362 . smoothing the corners we get a generic immersion @xmath363 . let @xmath364 be generic immersions . the connected sum @xmath337 is independent up to regular homotopy through generic immersions of both the choices of points @xmath350 and @xmath351 and the choice of the path @xmath349 used to connect them . this is straightforward . ( note that the preimages of self - intersections has codimension @xmath66 . ) if @xmath365 and @xmath366 , @xmath2 are regularly homotopic through generic immersions then @xmath337 and @xmath367 are regularly homotopic through generic immersions . this is straightforward [ csj ] the invariants @xmath82 and @xmath15 are additive under connected summation . @xmath368 . note that if @xmath24 and @xmath369 are manifolds which both satisfy condition @xmath293 in definition [ dfnlambda ] or condition @xmath292 in definition [ dfnl ] then so does @xmath335 . [ csl ] let @xmath242 and @xmath370 be generic immersions . * if @xmath3 and @xmath334 both satisfy condition @xmath293 then @xmath371 * if @xmath23 is even and @xmath3 and @xmath334 are both oriented and both satisfy condition @xmath292 then @xmath372 note that @xmath373 . consider case ( b ) . choose @xmath137-chains @xmath319 and @xmath374 in @xmath375 and @xmath376 bounding @xmath100 and @xmath377 respectively and disjoint from the arc @xmath349 , used in the construction of @xmath337 . then @xmath378 case ( a ) is proved in exactly the same way . the invariants @xmath82 , @xmath15 , and @xmath69 are clearly orientation independent . in contrast to this , the invariant @xmath16 is orientation sensitive . to have @xmath16 defined , let @xmath379 and consider an oriented closed manifold @xmath130 which satisfies condition ( c@xmath16 ) . assume that there exists an orientation reversing diffeomorphism @xmath380 . let @xmath381 be a generic immersion . then @xmath382 is a generic immersion and @xmath383 note that @xmath384 . the orientation of @xmath385 is induced from the decomposition @xmath386 the immersions @xmath26 and @xmath382 induces opposite orientations on each @xmath387 hence the orientations induced on @xmath385 agrees if @xmath27 is even and does not agree if @xmath27 is odd . let @xmath319 be a @xmath137-chain bounding @xmath385 , with its orientation induced from @xmath26 . if @xmath27 is even then @xmath388 if @xmath27 is odd then @xmath389 let @xmath390 be reflection in a hyperplane . let @xmath391 be a generic immersion . then @xmath392 is a generic immersion and @xmath393 note that the oriented normal bundle of @xmath394 is @xmath395 . so the correctly oriented @xmath137-chain bounding @xmath396 is @xmath397 . thus , @xmath398 in this section we prove theorems [ thml ] and [ thmj ] . to do that we need a coorientation of the discriminant hypersurface in the space of immersions . [ nabla0 ] let @xmath54 be an invariant of generic immersions . if @xmath54 is @xmath68-valued then @xmath55 ( see section [ mainres ] ) is well - defined without reference to any coorientation of @xmath9 . moreover , if @xmath54 is integer - valued and @xmath399 is a union of path components of @xmath53 then the notion @xmath400 is well - defined without reference to a coorientation of @xmath9 . we shall coorient the relevant parts ( see remark [ nabla0 ] ) of the discriminant hypersurface . that is , we shall find coorientations of the parts of the discriminant hypersurface in the space of generic immersions where our invariants have non - zero jumps . these coorientations will be _ continuous _ ( see @xcite , section 7 ) . that is , the intersection number of any generic small loop in @xmath5 and @xmath53 vanishes and the coorientation extends continuously over @xmath60 . [ cojr ] let @xmath65 and @xmath66 . assume that @xmath23 is odd . let @xmath401 and assume that @xmath81 is even . let @xmath402 be a generic immersion in @xmath53 with one degenerate @xmath35-fold self - intersection point . let @xmath151 be a versal deformation of @xmath52 . we say that @xmath403 is on the positive side of @xmath53 and @xmath404 on the negative side if @xmath405 . note that by lemma [ vdef1sm ] @xmath406 is obtained from @xmath407 by a morse modification . since the dimension of @xmath406 is @xmath408 is even the euler characteristic changes under such modifications . [ coj ] let @xmath65 . let @xmath409 be a generic immersion in @xmath53 with one degenerate @xmath27-fold self - intersection point . let @xmath151 be a versal deformation of @xmath52 . we say that @xmath403 is on the positive side of @xmath53 and @xmath404 on the negative side if the number of components in @xmath410 is larger than the number of components in @xmath411 . note that by lemma [ vdef1sm ] @xmath410 is obtained from @xmath411 by a morse modification . since these are 1-manifolds the number of components change . the construction of the relevant coorientation for @xmath16 is more involved . it is based on a high - dimensional counterpart of the notion of over- and under - crossings in classical knot theory . let @xmath412 be an immersion of oriented manifolds . assume that @xmath52 has an ( @xmath73)-fold self - intersection point , @xmath191 . let @xmath151 be a versal deformation of @xmath52 . let @xmath175 be small neighborhoods of @xmath117 and let @xmath413 denote the oriented sheet @xmath414 . note that @xmath415 and that @xmath416 if @xmath153 . let @xmath417 . let @xmath418 be a line transversal to @xmath419 at @xmath114 . for small @xmath153 both @xmath413 and @xmath420 intersects @xmath418 . orienting the line from @xmath413 to @xmath420 gives a local orientation @xmath421 of @xmath422 . comparing it with the standard orientation of @xmath422 we get a sign @xmath423 , where @xmath424 denotes the sign of the orientation . note that , if we orient the line from @xmath420 to @xmath413 we get the opposite orientation @xmath425 of @xmath311 and @xmath426 since @xmath420 and @xmath413 are both odd - dimensional . note also that @xmath427 ( see lemma [ vdef1big ] ) . we next demonstrate that @xmath428 for all @xmath429 : let @xmath430 be the oriented normal bundle of @xmath413 . let @xmath311 be a vector from @xmath431 to @xmath432 , transversal to both @xmath433 and @xmath434 . orient it from @xmath431 to @xmath432 . then @xmath435 and hence @xmath436 gives a normal bundle @xmath437 and by the convention used to orient the normal bundle @xmath438 . in a similar way it follows that @xmath439 . now , by definition @xmath440}= -\sigma_t(2){\operatorname{or}}(d_1^t,-\vec w , d_2^t , n_3^t,\dots , n_m^t)=\\ = & \sigma_t(2)\sigma_t(1){\operatorname{or}}(n_1^t , d_2^t , n_3^t,\dots , n_m^t ) = \sigma_t(1)\sigma_t(2).\end{aligned}\ ] ] hence , @xmath441 as claimed . [ col ] let @xmath412 be an immersion of oriented manifolds . assume that @xmath52 has an @xmath22-fold self - intersection point . let @xmath151 be a versal deformation of @xmath52 . we say that @xmath403 is on the positive side of @xmath53 at @xmath52 if @xmath442 the following obvious lemma will be used below . [ jump ] any first order invariant of generic immersions is uniquely ( up to addition of zero order invariants ) determined by its jump . we know from lemma [ wdj ] that @xmath15 and @xmath82 are invariants of generic immersions . we must calculate their jumps . we start in case ( a ) : assume that @xmath23 is odd . let @xmath443 , @xmath444 $ ] be a versal deformation of @xmath50 and fix @xmath35 , @xmath80 such that @xmath81 is even . if @xmath52 has a degenerate @xmath20-fold intersection point @xmath212 and @xmath445 then @xmath446 is diffeomorphic to @xmath447 : if @xmath448 then @xmath446 is not affected at all by the versal deformation . if @xmath449 then the immersed submanifold @xmath450 of @xmath446 is changed by surgery ( or is deformed by regular homotopy if @xmath149 ) under the versal deformation . this does not affect the diffeomorphism class of @xmath446 . if @xmath52 has a degenerate @xmath35-fold self - intersection point then @xmath451 is obtained from @xmath452 by a surgery ( see lemma [ vdef1sm ] ) . since the dimension of @xmath446 is @xmath408 which is even this changes the euler characteristic by @xmath453 . according to our coorientation conventions @xmath454 if @xmath52 has a degenerate @xmath35-fold self - intersection point . also , @xmath455 if @xmath52 has any other degeneracy . the jump of @xmath15 in case ( b ) is @xmath71 by the same argument . it is evident from lemma [ vdef2 ] ( see figure [ figcod2 ] ) that @xmath15 and @xmath456 are first order invariants . the theorem now follows from lemma [ jump ] . we start with ( b ) : let @xmath457 , @xmath444 $ ] be a generic one - parameter family intersecting @xmath53 in @xmath52 . if the degenerate intersection point of @xmath52 is a @xmath20-fold intersection point with @xmath458 then clearly @xmath459 , since the @xmath27-fold self - intersection is not affected under such deformations . assume that @xmath52 has a degenerate @xmath27-fold self - intersection point . without loss of generality we may assume that @xmath151 is a deformation of the form in lemma [ vdef1sm ] ( we use coordinates as there ) : let @xmath238 . shifting @xmath460 we obtain a @xmath137-chain bounded by @xmath461 in @xmath462-f(m\times[-\delta,\delta])$ ] . it follows that @xmath463 . if @xmath52 has an @xmath22-fold self - intersection then the discussion preceding definition [ col ] shows that @xmath464 : at an @xmath22-fold self intersection point @xmath73 crossings are turned into crossings of opposite sign . hence , @xmath465 if @xmath52 has an @xmath22-fold self - intersection point and @xmath466 if @xmath52 has any other degeneracy . the calculation of @xmath467 in ( a ) is analogous . let us just make a remark about the parity of @xmath27 : at an @xmath22-fold self - intersection point @xmath73 crossings are changed . hence , at instances of @xmath22-fold self - intersection the invariant @xmath69 changes by @xmath468 if @xmath73 is odd and does not change if @xmath73 is even . it is immediate from lemma [ vdef2 ] , see figure [ figcod2 ] that @xmath69 and @xmath16 are first order invariants . the theorem now follows from lemma [ jump ] . a function of immersions @xmath6 which is constant on path components of @xmath5 will be called an _ invariant of regular homotopy_. our geometrically defined invariants of generic immersion give rise to torsion invariants of regular homotopy . let @xmath23 be odd . let @xmath92 be an immersion . let @xmath469 be a generic immersion regularly homotopic to @xmath26 . for @xmath80 such that @xmath81 is even , define @xmath470 as @xmath471 [ rhj ] the function @xmath472 is an invariant of regular homotopy . clearly it is enough to show that for any two regularly homotopic generic immersions @xmath52 and @xmath473 , @xmath474 . let @xmath151 be a generic regular homotopy from @xmath52 to @xmath473 . then @xmath151 intersects @xmath9 transversally in a finite number of points in @xmath53 . it follows from theorem [ thmj ] that @xmath472 remains unchanged at such intersections . hence , @xmath474 . let @xmath23 be even . assume that @xmath24 is a manifold which satisfy condition @xmath292 . let @xmath475 be an immersion . let @xmath469 be a generic immersion regularly homotopic to @xmath26 . define @xmath476 as @xmath477 [ rhl ] the function @xmath478 is an invariant of regular homotopy . the proof is identical to the proof of proposition [ rhj ] . let @xmath65 be odd . assume that @xmath24 is a manifold which satisfy condition @xmath293 . let @xmath475 be an immersion . let @xmath469 be a generic immersion regularly homotopic to @xmath26 . define @xmath479 as @xmath480 the function @xmath481 is an invariant of regular homotopy . this follows from the proof of theorem [ thml ] , where it is noted that when @xmath27 is odd @xmath69 remains constant when a regular homotopy intersects @xmath53 . in this section theorem [ thmdivl ] will be proved . to do that we will first discuss the classifications of sphere - immersions and sphere - embeddings up to regular homotopy . in smale s classical work @xcite it is proved that there is a bijection between the set of regular homotopy classes @xmath482 of immersions @xmath483 and the elements of the group @xmath484 , the @xmath485 homotopy group of the stiefel manifold of @xmath20-frames in @xmath486-space . if @xmath487 is an immersion we let @xmath488 denote its smale invariant . via @xmath489 we can view @xmath482 as an abelian group . if the codimension is two the groups appearing in smale s classification are easily computed : the exact homotopy sequence of the fibration @xmath490 implies that @xmath491 . bott - periodicity then gives : @xmath492 [ geomope ] it is possible to identify the group operations in @xmath493 geometrically . kervaire @xcite proves that the smale invariant @xmath489 is additive under connected sum of immersions . this gives the geometric counterpart of addition . if @xmath494 is an immersion , @xmath495 is an orientation reversing diffeomorphism , and @xmath496 then @xmath497 . ( see @xcite for the case @xmath14 , the other cases are analogous ) . this gives the geometric counterpart of the inverse operation in @xmath498 for @xmath496 . it is interesting to note that for immersions @xmath499 , @xmath500 : the smale invariant can in this case be computed as the algebraic number of self - intersection points . this number is clearly invariant under reversing orientation . ( the same is true also for immersions @xmath501 , @xmath502 . ) [ lhom ] the regular homotopy invariant @xmath478 induces a homomorphism @xmath503 note that the dimensions are such that @xmath16 is defined and spheres certainly satisfy condition @xmath292 . the invariant @xmath16 is additive under connected summation and changes sign if an immersion is composed on the left with an orientation reversing diffeomorphism . hence , @xmath478 induces a homomorphism . let @xmath504 be the set of regular homotopy classes which contain embeddings . by remark [ geomope ] @xmath505 is a subgroup of @xmath498 . a result of hughes and melvin @xcite states that @xmath506 is a subgroup of index @xmath39 . here @xmath507 , where @xmath41 is the @xmath42 bernoulli number and @xmath508 and @xmath39 are coprime integers . by lemma [ lhom ] , @xmath509 is a homomorphism and clearly , @xmath510 . in case ( b ) @xmath511 and hence the image of @xmath478 is zero , which proves that @xmath32 is always divisible by @xmath512 . in case ( a ) hence , for any immersion @xmath26 , @xmath514 . thus , @xmath515 is divisible by @xmath516 , which implies that @xmath32 is divisible by @xmath37 . in case ( c ) @xmath517 as a subgroup of @xmath518 . hence , for any immersion @xmath26 , @xmath519 divides @xmath520\dots{\operatorname{\ensuremath{\sharp}}}f)=\mu_jl(f).\ ] ] thus , if if @xmath34 is a prime and @xmath521 are integers such that @xmath36 divides @xmath37 and @xmath38 does not divide @xmath39 then @xmath522 divides @xmath32 . in this section we construct examples of 1-parameter families of immersions which shows that all the first order invariants we have defined are non - trivial . we also discuss some problems in connection with the regular homotopy invariants defined in section [ rhinv ] . using our local coordinate description of immersions with one degenerate self - intersection point we can construct examples showing that the invariants @xmath15 , @xmath82 , @xmath69 and @xmath16 are non - trivial . we start with @xmath69 and @xmath16 : choose @xmath27 standard spheres @xmath523 of dimension @xmath231 intersecting in general position in @xmath29 so that @xmath524 . pick a point @xmath525 and let @xmath526 be another standard @xmath231-sphere intersecting @xmath527 at @xmath34 so that in a neighborhood of @xmath34 the embeddings are given by the expressions in lemma [ vdef1big ] and so that @xmath34 is the only degenerate intersection point . let @xmath528 be the immersions which is the connected sum of @xmath529 . let @xmath151 be a versal deformation of @xmath26 . then , after possibly reversing the direction of the versal deformation we have @xmath530 , if @xmath23 is odd or @xmath531 if @xmath23 is even . in the latter case we would get the other sign of @xmath532 if the orientation of @xmath526 in the construction above is reversed . to distinguish these two immersions denote one by @xmath533 and the other by @xmath534 , so that @xmath535 . then it is an immediate consequence of theorem [ thml ] that : the proof that the invariants @xmath15 and @xmath82 are nontrivial are along the same lines : construct a sphere - immersion into euclidean space with one degenerate self- intersection point as in the local models in lemma [ vdef1sm ] . then embed this euclidean space with immersed sphere into any target manifold and use connected sum together with lemma [ csj ] . [ lrange ] applying connected sum to the sphere - immersion constructed above in the case when it is an immersion @xmath537 shows that for any integer @xmath20 there exists a generic immersions @xmath74 such that @xmath538 . let @xmath541 and @xmath542 . then if @xmath543 are small regular neighborhoods of @xmath544 and @xmath545 then there exist diffeomorphisms @xmath214 and @xmath546 such that the following diagram commutes @xmath547 that is , the local properties of a generic immersion close to its self - intersection does not change at instances of @xmath22-fold self - intersection points in generic 1-parameter families . assume that @xmath3 fulfills the requirements of theorem [ thml ] . then @xmath69 or @xmath16 is defined and @xmath548 or @xmath549 , respectively . this implies that @xmath404 and @xmath403 are not regularly homotopic through generic immersions . hence , knowledge of the local properties of a generic immersion @xmath25 close to its self - intersection is not enough to determine it up to regular homotopy through generic immersions . theorem [ thmdivl ] gives information on the possible range of @xmath478 for sphere - immersions in codimension two . we consider the most interesting cases of immersions @xmath30 , @xmath550 ( theorem [ thmdivl ] ( c ) ) . in the first case @xmath14 , @xmath478 is non - trivial . this was shown in @xcite . hence , for any integer @xmath551 there are generic immersions @xmath552 such that @xmath553 . [ lkrmk ] the invariant @xmath478 is called @xmath481 in @xcite and is the @xmath554 reduction of an integer - valued invariant called @xmath555 . the definition of @xmath555 given in @xcite differs slightly from the definition of @xmath16 given here . there is an easy indirect way to see that , nonetheless , the two invariants are the same : due to theorem [ thml ] and theorem 2 in @xcite , @xmath556 is an invariant of regular homotopy which is @xmath557 on embeddings and additive under connected summation . assume that @xmath558 for some immersion . then , since @xmath559 , the connected sum of @xmath560 copies of any immersion is regularly homotopic to an embedding . hence , @xmath561 and therefore @xmath562 . thus , @xmath563 . on the other hand , if none of the prime factors of @xmath37 divides @xmath39 , theorem [ thmdivl ] implies that @xmath478 is trivial and in such cases remark [ lrange ] allows us to determine the range of @xmath16 which is @xmath564 . the first two cases where this happens are @xmath44 and @xmath565 , where @xmath37 equals @xmath566 and @xmath567 and @xmath39 equals @xmath568 and @xmath569 , respectively . theorem 7.30 in @xcite shows that the self - intersection surface of a generic immersions of @xmath573 must have _ even _ euler characteristic ( even though it may be non - orientable ) . thus , in this case @xmath570 is trivial . ( note that @xmath574 so that there are two regular homotopy classes of immersions @xmath573 . hence , it is non - trivial to see that @xmath478 is trivial in this case . ) consider three hyperplanes @xmath579 , @xmath580 , @xmath581 in general position in @xmath577 . fix a point @xmath582 . note that @xmath583 is a non - orientable line bundle over @xmath584 , @xmath585 . choose sections @xmath586 of @xmath587 over @xmath584 such that @xmath588 meets @xmath589 transversally in @xmath584 for @xmath585 . now , @xmath590 . let @xmath591 be the corresponding embedding . restricting @xmath586 to @xmath592 we get three normal vector fields @xmath593 along @xmath594 and @xmath595 . let @xmath596 be the universal cover . then @xmath597 is an immersion . let @xmath598 . choose morse functions @xmath599 such that @xmath600 . let @xmath276 be small . then @xmath601 , given by @xmath602 is a generic immersion with @xmath603 . hence , @xmath604 . 99 v. i. arnold , _ plane curves , their invariants , perestroikas , and classifications _ , in singularities and bifurcations , edited by arnold , adv . sov . math . * 21 * ( 1994 ) 39 - 91 . * 2 * , ( 1963 ) 181 - 195 . j. eccles , _ characteristic numbers of immersions and self - intersection manifolds _ , bolayi soc . * 4 * , topology with applications , ( szekszrd ) ( 1993 ) 197 - 216 . t. ekholm , _ differential 3-knots in 5-space with and without self intersections _ , in immersions and their self intersections ( phd - thesis , uppsala ) ( 1998 ) . t. ekholm , _ immersions in the metastable range and spin structures on surfaces _ , math . scand . * 83 * ( 1998 ) 5 - 41 . t. ekholm , _ vassiliev invariants and regular homotopy of generic immersions @xmath605 , @xmath606 . _ , j. knot theor . ramif . * 7 * ( 1998 ) 1041 - 1064 . j. f. hughes and p. m. melvin , _ the smale invariant of a knot _ , comment . math . * 60 * ( 1985 ) 615 - 627 . j. w. milnor and j. d. stasheff , _ characteristic classes _ , ann . of math . stud . * 76 * , princeton university press ( 1974 ) . m. a. kervaire , _ sur le fibr normal une sphre immerge dans une espace euclidien _ , 33 * , ( 1959 ) 121 - 131 . s. smale , _ classification of immersions of spheres in euclidean space _ , ann . of math . * 69 * ( 1959 ) , 327 - 344 . v. a. vassiliev _ cohomology of knot spaces _ in theory of singularities and its applications , edited by arnold , adv . sov . math . * 1 * ( 1990 ) 9 - 22 .

this paper presents an analysis of a keck hires spectrum of the qso first j104459.6 + 365605 , covering the rest wavelength range from 2260 to 2900 . the line of sight towards the qso contains two clusters of outflowing clouds that give rise to broad blue shifted absorption lines . the outflow velocities of the clouds range from 200 to 1200 and from 3400 to 5200 , respectively . the width of the individual absorption lines ranges from 50 to more than 1000 . the most prominent absorption lines are those of , , and , and is also present . the low ionization absorption lines occur at the same velocities as the most saturated lines , showing that the , and line forming regions must be closely associated . many absorption lines from excited states of are present , allowing a determination of the population of several low lying energy levels . the populations of the excited levels are found to be considerably smaller than expected for lte , and imply an electron density in the line forming regions of . modelling the ionization state of the absorbing gas with this value of the electron density as a constraint , we find that the distance between the and line forming region and the continuum source is parsec . from the correspondence in velocity between the , and lines we infer that the lines must be formed at the same distance . the absorption fulfills the criteria for broad absorption lines defined by weymann et al . ( 1991 ) . therefore the distance we find between the line forming region and the continuum source is surprising , since bals are generally thought to be formed in outflows at a much smaller distance from the nucleus .

astro-ph0010230

many qso rest - frame uv spectra exhibit blue shifted absorption lines that are indicative of an outflow . the properties of these absorption lines vary over a wide range , from very broad ( @xmath3 10,000 ) absorption lines , usually observed in high ionization species like and to fairly narrow ( @xmath4 30 ) lines from the same high ionization species , or from low - ionization species like and . the absorption lines are often divided in two classes . the first class contains the broad absorption lines ( bals)(weymann , turnshek & christiansen 1985 , turnshek 1988 ) which are broad absorption troughs , and can span a large range of outflow velocities up to several times @xmath5 . the second class refers to associated absorption lines ( aals ) ( weymann et al . 1979 , foltz et al . 1986 ) , which have individual absorption components less than a few hundred wide , and span a range of outflow velocities up to several thousand . it is not obvious whether this division reflects a real bimodal distribution of absorption line properties caused by two different underlying formation mechanisms . there are many examples of qsos that exhibit both types of lines simultaneously , or have properties that are a mixture between the two classes , eg . high velocity narrow lines or low velocity broad lines . it is generally thought that bals are formed relatively close to the galactic nucleus ( @xmath6 parsec ) , although the direct observational evidence for this is not very strong . the minimum distance separating the region in which broad absorption lines arise ( the balr ) from the source of the continuum radiation is set by the observation that the broad emission lines are often partly or completely absorbed by the outflow ( turnshek 1988 ) . this places the balr outside the broad emission line region ( belr ) , the size of which is now relatively well established with reverberation mapping techniques at @xmath7 pc for an average luminosity qso . the maximum distance of the balr from the central source is more difficult to constrain . adopting typical , if uncertain , physical parameters for the bal clouds it can easily be shown that unrealistically large mass and energy loss rates are implied if the lines are formed more than a few tens of parsecs from the central engine . further support for the conjecture that the balr is not much larger than the belr comes from consideration of the acceleration mechanisms , which generally require that the flow starts at a distance comparable to the size of the belr in order to reach the high speeds observed ( e.g. arav & li 1994 ) . the physical conditions inside the balr are difficult to constrain because of the extremely wide and deep absorption lines in classical balqsos . the width complicates the analysis because the optical depth at a certain wavelength is the convolution of the column density as a function of velocity and the multiplet structure of the line , so that the column density as a function of velocity can only be derived by a deconvolution . the large depth makes the derived column densities very sensitive to partial covering of the continuum light source . aals appear to be more diverse , and could arise from gas located in the agn environment , the host galaxy , or intervening gas with a smaller cosmological redshift than the qso . some recent studies of aals are hamann , netzer & shields ( 2000 ) , petitjean & srianand ( 1999 ) , srianand & petitjean ( 2000 ) , barlow & sargent ( 1997 ) and ganguly et al . the analysis of aals does not suffer as much from the problems of blending and saturation encounterd with bals , and can give quite detailed results on the physical conditions and geometry of the absorbing gas , as demonstrated in the references above . from these studies it has become clear that the great majority of aals show the effects of partial covering in their observed multiplet line ratios , ( e.g. hamann et al . 1997 , arav et al . 1999 ) which argues against a cosmological origin , and is most easily explained if the lines are formed in the agn environment . however , to our knowledge no firm limits on the distance of the absorbers to the active nucleus have been derived so far . here we investigate the z=0.700 qso first j104459.6 + 365605 ( hereafter fbqs1044 + 3656 ) . this object was discovered in the first bright quasar survey ( white et al 2000 ) which consists of 636 radio - selected quasars distributed over @xmath8 square degrees . the absorption lines in the spectrum of this qso have several unique features , allowing us to constrain the density and ionization parameter in the outflow . the spectrum contains absorption lines of , and . some of the lines arise from low - lying excited states . most of the lines are not very wide nor heavily saturated , behaving roughly as expected from their relative oscillator strength , making quantitive estimates of the column density possible . the derived relative populations of several low - lying energy levels of allow us to put significant constraints on the physical conditions in the absorbing clouds . the characteristic velocity and width of the absorption lines in fbqs1044 + 3656 are intermediate between bals and aals . the absorption is very broad , and just fulfills the strict criterion applied to the line by weymann et al . 1991 to decide whether a qso is a balqso . the lower ionization lines of and are narrower , and more characteristic of aals . however , we will argue that the excellent correspondence in velocity between the strongest components and the and lines leave little doubt that they arise from the same outflowing cloud complex . our analysis is similar to the study of balqso q0059 - 2735 by wampler , chugai & petitjean 1995 ( hereafter referred to as wcp ) . q0059 - 2735 is a classical balqso with very deep and wide bal troughs in both high and low ionization species . its spectrum also contains several narrow , low ionization absorption line systems , such as , , and , at velocities of a few hundred . another recent study with some similarities to ours considers narrow absorption line systems in the balqso apm 08279 + 5255 ( srianand & petitjean 2000 ) , which show absorption from excited states of and . on december 26 , 1998 we used the high resolution echelle spectrometer ( hires , vogt et al . 1994 ) on the keck i 10-m telescope to obtain @xmath9 second exposures of fbqs1044 + 3656 covering 3825 - 6280 using a 11 wide slit . the orders overlap up to 5128 , beyond which small gaps occur between orders . the slit was rotated to the parallactic angle to minimize losses due to differential atmospheric refraction . the observing conditions were excellent with sub - arcsecond seeing and near - photometric skies . the spectra were extracted using routines tailored for hires ( barlow 2000 ) . the resolution of the final spectrum is 4 pixels pixels fwhm , or 8.8 in velocity space . the redshift of fbqs1044 + 3656 ( z=0.700 ) was determined by fitting a gaussian to the broad emission line . because the blue wing of the bel is strongly affected by absorption , the redshift of the qso rest frame is uncertain by several hundred . since all outflow velocities quoted in this paper are calculated from the difference in redshift between the absorbing gas and the qso rest frame , they are uncertain by the same amount . in this section we will analyze the absorption lines in the spectrum of fbqs1044 + 3656 , deriving the column densities of the observed species . in the next section , these column densities will be used to derive the physical condition in the outflow . an overview of the hires spectrum is shown in figure [ totspec ] . a low resolution spectrum of fbqs1044 + 3656 covering a larger wavelength range has been published in white et al . 2000 . the line at @xmath10 2852.96 is a singlet transition from the ground level . the line profile as a function of velocity is shown in figure [ mgplot ] , together with the profile of the absorption . the velocity scale of the absorption is based on the rest wavelength of the red component of the doublet . we divide the absorption features into two clusters , the first ranging from @xmath11 to @xmath12 ( cluster 1 ) , the second from @xmath13 to @xmath14 ( cluster 2 ) . some of the components in the low - velocity cluster are not connected by any absorption , so this definition of these two clusters is purely ad hoc at this point . it will turn out however that our division appears to have a physical basis since the components of cluster 2 exhibit significant absorption from excited levels of , whereas the detection of absorption from excited levels from the components of cluster 1 is only marginal . this indicates that the physical conditions differ between the two clusters . in each cluster we have identified the two strongest components so that we can refer to these later when identifying the corresponding absorption lines of other species . based on the assumed redshift of 0.700 for the qso , these components correspond to an outflow speed of -205 , -1227 , -3557 and -3860 km s@xmath15 for components a , b , c and d respectively . a lower limit on the column density of can be obtained in the optically thin approximation , valid if the line is fully resolved and the absorbing gas is covering the entire continuum source : @xmath16 the column density derived in this way is @xmath17 for cluster 1 and @xmath18 for cluster 2 . from the analysis of the lines below , we will find that the absorbing clouds do not cover the entire continuum source . it is not clear if the same applies to the line , since the continuum at the position of the lines contains a large fraction of broad emission which comes from a much more extended region than the continuum from the accretion disk . thus it is possible that the outflow covers the accretion disk source completely but the emission line regions only partially , which would lead to partial covering being observed for absorption lines in the part of the spectrum where broad line emission contributes significantly to the background source . even if the same partial covering would apply to the line , however , this would increase the column density only by a factor @xmath19 since the lines are significantly shallower than the saturated lines . larger deviations could occur if the covering factor for the producing clouds is very different from that of the producing clouds . this possibility can not be excluded , but the fact that the line profile can be used as template for both strong and weak lines ( as shown below ) argues against it . the presence of in fbqs1044 + 3656 stands in important contrast with the absence of absorption in q0059 - 2735 , which was used by wcp to obtain an upper limit on the distance of the clouds to the continuum source . from the comparison of and absorption as a function of velocity in figure [ mgplot ] , it can be seen that the absorption extends to much higher velocities than . there is a strong component centered around @xmath20 that does not appear in , and further contiguous absorption is visible up to @xmath21 . at even higher velocities , some very shallow absorption systems at -6500 , -7310 and -7410 are probably due to since both components of the doublet are visible . the lines are heavily saturated allowing only a lower limit on the column density . over the velocity range covered by cluster 1 and 2 , we estimate lower limits on the column density of @xmath22 and @xmath23 respectively , but the real columns are likely to be significantly higher . the differences between the and profiles do not imply that the ionization state has to change with velocity : the column density of associated with the high velocity components would be too small to be detected if the / ratio is the same as in the strongly saturated components . this interpretation is supported by the high - velocity absorption not being readily detectable in the weakest absorption lines , but being just visible in the strongest lines . most of the physical constraints on the absorbing gas can be extracted from the numerous absorption lines of in the spectrum of fbqs1044 + 3656 . because of the blending of the lines , most of our analysis must use a template for the column density as a function of velocity , which we derive from the unblended singlet transition of , an approach similar to that of arav et al . the template method is based on the assumption that the column density of in each energy level @xmath24 between velocity @xmath25 and @xmath26 is proportional to the column density of in the same velocity interval : @xmath27 the column density of is derived from the absorption line profile using @xmath28 where @xmath29 is the remaining flux in the normalized line profile , and @xmath30 the oscillator strength of the resonance line . by using equation [ mgprof ] , we implicitly assume that partial covering effects are not important for ( see above ) , and that the absorption lines are fully resolved . any quantitative analysis must make the second assumption , although it should be kept in mind that the thermal line width of fe and mg at the temperatures of a few thousand degrees that are expected for the line forming regions are only of order 1 - 2 , i.e. well below the resolution of our observations . when the constants of proportionality @xmath31 between the column densities of and @xmath32 are specified , we can calculate the optical depth in every line as a function of velocity , and construct a model absorption line spectrum that can be compared to the observed spectrum of fbqs1044 + 3656 . from a first inspection of the spectrum it is clear that the components of cluster 2 show significant absorption in excited lines , whereas the components of cluster 1 show up only very weakly in these lines . therefore , the two clusters appear to arise from physically distinct cloud complexes . to keep the number of free parameters in our fit manageable , we will assume that the level populations do not vary between different components within each cluster , but allow for different level populations between the two clusters . the lines considered in our study are listed in table 1 . we included all lines from the lowest energy ( @xmath33 ) term , and the strongest lines from the next two terms ( @xmath34 and @xmath35 ) with lowest levels at approximately 0.2 and 1 ev , respectively . in total , we consider 43 lines arising from 10 energy levels . the fit was obtained by first choosing values for the @xmath31 s for each cluster and then comparing the predicted and observed absorption line spectra . the values of the @xmath31 s were successively improved by adjusting them to reproduce the observed flux at the positions in the spectrum where the specific @xmath31 is best constrained . to be properly constraining , we required that the parts of the spectrum used to adjust the fit contain a cluster component that is not blended with another line , and is not very deep so that partial covering effects play as small a role as possible . although many lines are blended , all levels give rise to some absorption components that occur in a part of the spectrum where blending is not a problem . note that this method of fitting a model to the observed spectrum does not try to fit individual _ lines _ , but rather _ level populations _ and thus attempts to fit all lines from a given lower level with one parameter , the column density of in the state corresponding to this lower level . because isolated , medium depth lines arising from the level are the most accurate indicators of this column density , these are the main drivers for the fit . the model resulting from the fit obtained in this way is compared with the observed spectrum in figure [ fe2plot ] . the thin line represents the reconstructed spectrum based on the model parameters in table 2 . we express the model parameters in terms of relative level populations and a total column density . @xmath36 is the level population per unit statistical weight of level @xmath24 relative to that of the ground level , @xmath37 @xmath36 is equal to 1 if the levels are in lte with @xmath38 . the total column density is @xmath39 for cluster 1 and @xmath40 for cluster 2 . the upper limits and error estimates were obtained by visually comparing model spectra with the observations and estimating how much deviation from either 0 or the measured value could be tolerated . systematic effects such as errors in oscillator strength or continuum level are thus not included . the heavily blended regions of the spectrum are well reproduced using the @xmath31 s obtained from the isolated lines , proving the consistency of the result . the figure shows that for most of the absorption components the fit is satisfactory , considering that the problem is heavily overconstrained and that the number of lines that have to be reproduced is several times larger than the number of energy levels included in the fit . the most significant differences between the observed spectrum and the model that can not be removed by changing the model parameters occur in the strongest absorption lines from the ground level . these strong lines show absorption at high velocity , extending beyond the highest velocity seen in cluster 2 of . this extra absorption is also seen as the very deep component at @xmath20 in . the other major deviation between model and observation is that the strongest lines are much deeper in the model . it is impossible to obtain a good fit to the five lines originating from the ground state simultaneously . the most likely explanation for this discrepancy is partial covering of the continuum source , which fills in the bottom of the strongest lines . note that the underprediction of flux in the deepest lines is the result of our choice to optimize the fit for medium depth lines ( which are the weakest lines when considering lines from the ground level ) . there are several other sources of uncertainty that may also contribute to discrepancies between the model and the observed spectrum . one is that there are still some uncertainties in the oscillator strengths of the uv absorption lines studied here . table 1 shows a comparison of these oscillator strengths from different literature sources . in this work we have used the oscillator strengths of fuhr , martin & wiese 1988 ( nist database ) , which are generally in reasonable agreement with recent experimental values of bergeson et al . 1996 . the only data source that has significant outliers are the theoretical values from the iron project ( nahar 1995 ) , with some discrepancies as large as a factor of 5 . for one of the important transitions for this study ( @xmath10 2374.46 , a line from the ground level ) the nahar value is almost 2 times larger than the others . using observations of absorption lines from the interstellar medium cardelli & savage 1995 derived an oscillator strength of 0.0326(14 ) for this transition , in good agreement with the laboratory experiments of bergeson et al . thus the theoretical value of 0.0527 derived by nahar 1995 is unlikely to be correct . for this transition only , we have not used the oscillator strength of fuhr , martin & wiese 1988 but instead a compromise between the two recent experimental values of 0.032 . more recent theoretical calculations ( raassen & uylings 1998 ) appear to reproduce the experimental values quite well , but are not as extensive as the iron project results . another source of uncertainty is the exact level of the continuum . the continuum contains several undulations , that are likely to be due to the presence of broad emission , as can be seen from figure [ totspec ] , or even more clearly from the low resolution spectrum of fbqs1044 + 3656 published in white et al . 2000 which covers a larger wavelength range . the relative contribution of the bels to the continuum can be estimated from the broad trough in the spectrum around 2650 which is actually a gap between emission line complexes ( see e.g. , verner et al . this trough is also present in the lbqs composite spectrum ( francis et al . 1991 ) , but is especially deep in fbqs1044 + 3656 , indicating that at least 30% of the apparent continuum is actually broad emission . the strongest emission lines are expected at the same wavelengths where the absorption lines occur ( verner et al . 1999 ) , and since the main blended absorption complexes themselves are also quite broad , it might well be possible that there is emission at the position of the absorption complexes that has not been accounted for in our continuum normalization . the spectrum of j121442 + 280329 published in white et al . 2000 is a clear example of a qso spectrum in which a complex interplay between absorption and emission is shaping the spectrum , and modelling of the absorption would be impossible without considering the emission lines . in the case of j121442 + 280329 the problem is more obvious because the emission lines are narrower . note that the lines from the ground level are distributed over a wavelength range several 100 wide , i.e. they are separated by much more than the width of a single broad emission line . this means each absorption line will be affected by a different set of emission lines , unlike the case of most doublet lines where the separation is much smaller than the emission line width , and the presence of broad emission is unlikely to have a strong differential effect on the continuum level between the positions of the two doublet components . the five strong lines arising from the ground level can not be fitted simultaneously using the preceding template approach when it is assumed that the continuum source is completely covered by the outflow . partial covering has been shown to be important in several other qsos with absorption lines ( e.g. , hamann , barlow , & junkkarinen 1997 , barlow & sargent 1997 , arav et al . 1999b ) , usually by considering the relative depth of the two lines of one of the major doublets , such as and . when relatively narrow absorption components are present so that the two lines of the doublet are visible ( which implies that this type of analysis can generally only be applied to aals and not bals ) , it is possible to derive a model for the outflow in which at each velocity a fraction @xmath41 of the continuum source is covered with a homogeneous layer of absorbing gas with optical depth @xmath42 in line 1 of the doublet . this is done by solving the set of coupled equations @xmath43 for every velocity @xmath25 . here @xmath44 is the residual intensity in the normalized spectrum , the subscripts 1 and 2 refer to the two lines of the doublet , and @xmath45 is the known ratio of the optical depths in two lines that originate from the same lower level . the same analysis can be applied to every pair chosen from the 5 lines arising from the ground level , and if the partial covering model used is correct , then each pair should yield the same covering factor as a function of velocity , providing a strong confirmation of the simple partial covering model . several complications crop up when analysing the lines in this way , the most serious being line blending . if one of the two lines used to solve for optical depth and covering factor is contaminated by another absorption line , the solution will be erroneous and often no solution is possible at all . a second problem is that the oscillator strengths of the lines are not accurately known , in contrast to the case of the doublet lines where the ratio is exactly 2 . when the optical depth is very low , small continuum errors also cause problems . in spite of these difficulties , we can still derive some insights from such a covering factor analysis . we first consider the @xmath41 that can be derived from the two lines at 2600 and 2586 ; these suffer the least from blending with other lines and therefore give the cleanest solutions . in figure [ covfac ] , we show what fraction of the continuum source must be uncovered to make the profiles of these two lines consistent , both for cluster 1 and cluster 2 . at some velocities , contamination by other lines or noise do not allow a solution for @xmath41 , and we have indicated these points with asterisks above the spectrum . to reduce the effect of noise , the spectrum was smoothed with a 5-point moving average before solving equation [ coveq ] . the shape of the strongest line is almost entirely determined by the covering factor , as is expected on the basis of our previous result that the template fitting procedure predicts a much smaller flux at the bottom of the strongest lines than is actually observed . this applies both in cluster 1 and cluster 2 . thus , the residual flux at the bottom of the strong lines is either due to a small part the continuum source not being completely covered by the outflow , or due to scattering of continuum light around the absorbers . in the case of fbqs1044 + 3656 we consider the first possibility more likely , because the strong variations of covering factor as a function of velocity are not expected if the remaining flux is due to scattered light . the covering factor solutions based on other line pairs suffer more from all of the problems mentioned above . nevertheless , it is very important to check whether the solutions obtained from different pairs yield more or less consistent covering factors . cluster 1 is too much affected by blending to make a useful comparison . in figure [ covcomp ] , we compare the solutions for the covering factor in cluster 2 for the 5 most suitable line pairs . in view of all the uncertainties , the solutions are consistent . there is a systematic difference between the pairs including the line 2344.2 and the other pairs , in the sense that they yield a smaller covering factor outside the deepest components in the profile . such behavior could be caused by an error in the oscillator strength of the line 2344.2 , an error in the continuum level , or by a wavelength dependent continuum source size . since the line is close to the blue cut - off of the spectropgraph , the s / n ratio at the position of the 2344.2 line is considerably worse than at the position of the other lines , which may also contribute to the discrepancy . since our continuum consists of a combination of a smooth underlying continuum generated in a very small region and broad undulating emission lines from a much larger region , we consider a wavelength dependent continuum source geometry to be the most likely explanation . in principle , the covering factor solution also yields the optical depth as a function of velocity , and hence the column density . the velocity integrated column densities derived in this way for each cluster are typically 2 - 3 times higher than those derived from the template fitting . since the column densities derived from template fitting mainly depend on the weakest non - saturated lines , this difference can not be attributed to the effects of partial covering . instead , the cause of the discrepancy is that the integrals of optical depth over velocity are dominated by large peaks that occur when the residual fluxes in the two lines used to derive the covering factor solution become almost equal , which can occur because of small continuum errors when the lines are weak , or because of another absorption line falling into the profile of the weaker line . this forces a solution with very high optical depth at this velocity . since this is not physical , we consider the column densities derived from the template fitting more reliable . we now verify that the results for the level populations obtained by the covering factor approach and the template fitting procedure are consistent . the foregoing analysis shows that the covering factor completely determines the profiles of the two strongest lines , so the covering factor as a function of velocity can be estimated from the remaining flux in these lines as , @xmath46 \label{covmean}.\ ] ] the thick line in figure [ fe2plot ] again shows the model fit based on the parameters in table 2 , this time including the effect of the covering factor defined by equation [ covmean ] . all the lines are now well - reproduced by the model . note that if multiple lines from gas at different velocities ( and hence with different covering factors ) contribute to the absorption at a given wavelength , the total remaining flux depends on whether the gas at different velocities covers the same region of the background source or not . for the model presented here , we have assumed that there is no correlation between the positions of the covered or uncovered regions for gas with different velocities . numerically , this means that each line contributing to the absorption at a given wavelength reduces the flux by a factor @xmath47 equal to @xmath48 the only other feature that we can identify with certainty is absorption from cluster 2 . all three lines from the ground state ( 2576.877 , 2594.499 and 2606.462 ) are detected . the depth of the features is well fit if we use the absorption template , and assume that the column density of is @xmath49 times the column density of ( @xmath50 ) , in good agreement with standard abundance ratios . in figure [ fe2plot ] the positions of the absorption lines are indicated . absorption lines of are absent from the spectrum , providing an important constraint on the ionization parameter of the absorbing gas . the strongest line from the ground level ( 2484.0209 rest wavelength ) is not detected ( see figure [ fe2plot ] ) . we have determined an upper limit on the column density by trying to fit model spectra using the absorption template , and estimating by eye how much could be tolerated without the line being clearly detected . because of the relatively smooth continuum in the relevant part of the spectrum , we are able to constrain the column density of in the ground state to @xmath51 . it would be useful to have some information on the abundance ( wcp ) , but our wavelength coverage does not include any lines . our most important result is the finding that absorption from excited levels of is detected , and that the level populations are smaller than expected for lte . this shows that the electron density in the outflowing clouds in fbqs1044 is lower than that in the lower velocity clouds in spectrum of q0059 - 2735 studied by wcp , which were found to be in lte . an accurate electron density estimate based on the observed level populations is difficult to obtain , since only a small fraction of the radiative transition probabilities for the strongly forbidden transitions within and between the @xmath33 , @xmath34 and @xmath35 terms are known ( nussbaumer & storey 1980 , quinet , le dourneuf & zeippen 1996 ) . a very rough estimate of the electron density can be obtained by considering only the relative populations of the ground level and the lowest excited level ( @xmath52 ) . this level can only decay radiatively to the ground level and not make other radiative decays with unknown transition probabilities . the electron density can be estimated assuming that there is equilibrium between collisional excitation into this level from the ground level and radiative de - excitation to the ground level . for this transition the radiative decay probability @xmath53 is 0.00213 ( quinet , le dourneuf & zeippen 1996 ) , and collision strength @xmath54 is 5.5 ( zhang & pradhan 1995 ) . taking into account the formal errors on the level populations in table 2 , and constraining the temperature of the absorbing gas to lie in the range between 300 and 10.000 k , we obtain @xmath55 for cluster 2 and @xmath56 for cluster 1 . using the observed level populations and the collision strengths from zhang & pradhan 1995 , it is easily verified that for @xmath57 and a temperature of 5000 k the total collisional excitation rate out of the lowest excited level is 0.24 times the radiative decay rate to the ground level , so that its neglect is justified for this rough estimate . we can also derive an estimate of the electron density by using the most recent attempt at modelling the level populations of as a function of density using a full multilevel calculation by verner et al . 1999 . in these models the unknown transition probabilities have been set to such a small value that they are effectively neglected . based on these results we can make an estimate of the electron density by comparing the departure coefficient of the ground level ( the only level among the ones relevant here for which results are presented in verner et al . 1999 ) with the predicted value as a function of electron density . we define an `` observed '' departure coefficient for the ground level as @xmath58 where the sum is over the levels @xmath59 given in table [ levpop ] , and the temperature was taken to be 5000 k for consistency with the verner et al . model . for this temperature , the contribution of the higher energy levels to the partition function will be very small because of the boltzmann factor , and their neglect is well justified . the observed departure coefficients are @xmath60 for cluster 2 and @xmath61 for cluster 1 . using figure 3 in verner 1999 we can convert these limits to @xmath62 for cluster 2 and @xmath63 for cluster 1 . these values are in reasonable agreement with the results obtained above by treating the ground and lowest excited level of as a two - level atom . we have used the photoionization code cloudy ( ferland 1996 ) to model the ionization state of the outflow to derive additional constraints on the absorbing gas . our observational data yield the following constraints on the outflow in fbqs1044 + 3656 : a column density @xmath64 , a lower limit on the / ratio of @xmath65 , an column density @xmath66 , an upper limit on the column density of @xmath67 and an estimate of the electron density of @xmath68 . it turns out that the lower limit on the / ratio is not very useful , so we will concentrate on the , , and column densities . we will first consider simple models of the absorbing clouds , investigating their structure as a function of ionization parameter , density and shape of the ionizing spectrum . for the moment we ignore the possible effects of dust , which can introduce significant complications . the clouds are modelled as thin constant density slabs . in order to satisfy the observational constraints , the slabs are truncated when a value of the column density of @xmath69 is reached . furthermore , to satisfy the electron density constraint we have adjusted the density of the slab in each model so that the weighted mean electron density is @xmath70 . in this way the ionization parameter , the abundances and the shape of the ionizing spectrum remain as parameters that can be adjusted to satisfy the constraints on the and column densities . the parameters and properties of the ionization models presented here are given in table [ ionmods ] . figure [ uplot ] illustrates the structure of the slab as a function of ionization parameter for gas with solar composition irradiated by a standard matthews & ferland ( 1987 ) agn spectrum . when modelling the very low ionization species considered here with the cloudy code , one encounters the difficulty that , and can occur in gas with a temperature below the limit at which cloudy results are reliable . we therefore had to stop the calculation of the cloud structure at the depth where the temperature falls below 1000 k , if this occurs before the required column density is reached . at temperatures below 1000 k is still the main form of fe . this means that when the calculation is stopped because of the temperature constraint ( which occurs for low ionization parameters ) , the column densities found are only lower limits . the run of temperature and electron density through the slab is similar to that found in the detailed bel models of e.g. wills et al . ( 1985 ) and collin - souffrin et al . ( 1986 ) , with an almost fully ionized strmgren layer at the illuminated side of the side , followed by a partially ionized zone in which most of the column density of resides . to understand the behaviour of the column density as a function of the model parameters it is important to realize that the ionization equilibrium of mg behaves slightly counter - intuitively , in the sense that the fraction of neutral mg drops sharply _ behind _ the hydrogen ionization front . the reason for this behaviour is that the photoionization rate of is hardly affected by the front because the ionization potential of is considerably lower than that of hydrogen , whereas the recombination rate is much reduced because of the sharp drop in electron density . therefore , the column density of the slab resides almost entirely in the ionized strmgren layer . the three models with @xmath71 all have column densities less than @xmath72 ( see table [ ionmods ] ) , between 3 and 10 times less than observed in fbqs1044 + 3656 . the three models with @xmath73 equal to 4 , 5 and 6 run into the 1000 k temperature limit before a high enough column density is reached , so that the column densities of and are lower limits only . the model with @xmath74 is excluded , since it already has an column density higher than our upper limit when the column density is only @xmath75 . thus of the models presented in figure [ uplot ] , only those with @xmath76 and @xmath77 may explain the observed column densities in fbqs1044 + 3656 . although the column densities may be consistent with such a very low ionization parameter model , they are not likely to provide a good explanation for the observed lines in detail , since most of the column density would have to reside in the region with a temperature below 1000 k , which is difficult to reconcile with the observed population of the excited state at 1873 cm@xmath15 because the boltzmann factor for this level should be @xmath79 . thus we conclude that simple solar composition ionization models are not able to reproduce the observed column densities . model 2 ( @xmath80 ) only fails to reach the observed column density by a factor of 3 . such a relatively small discrepancy may be fixed by changing the abundances from the solar values . the simplest solution is to increase the mg abundance by a factor of 3 , which brings the column density very close to the observed value . increasing the fe abundance reduces the thickness of the slab for a given column density , but because most of the resides in the strmgren layer , this hardly affects the column density . thus with an overall increase in metallicity to 3 times solar ( model 7 ) the observed column densities can be reproduced . although the column densities can be reproduced by this high metallicity model , there are two problems that remain to be resolved . the first is that these models are rather contrived in the sense that extreme fine tuning of the cloud geometry is required to obtain the correct column density . most resides in a thin layer at the back side of the cloud , with a thickness much smaller than the thickness of the strmgren layer , a problem also pointed out by wcp ( see below ) . since the thickness of the slab is set by the column density , this problem can be remedied by reducing the fe abundance by a very significant factor . this leads to slabs with a much thicker partially ionized zone . the larger extent of the partially ionized zone does not contribute appreciably to the column density however , so that an enhanced mg abundance is still required . the second problem is that the and reside in different zones of the slab . in fbqs1044 + 3656 , the gas along the line of sight is broken up into many clouds at different velocities , and we would not expect to see such a good correspondence between the absorption components of and if these lines were not formed in the same gas . in figure [ spplot ] we investigate the sensitivity of these results to our assumptions about the shape of the ionizing spectrum . the agn spectrum used above has a very strong euv bump . some more recent observation ( e.g. laor et al . 1997 ) suggest that in at least some active galaxies the euv flux is considerably smaller than predicted by the matthews & ferland spectrum . to investigate this effect , we have repeated some models with a simple power law spectrum with slope -1.4 between 10 microns and 50 kev , which effectively removes the euv bump . figure [ spplot ] compares the models with @xmath73 equal to -2 , -4 and -6 for the two choices of ionizing spectrum . although the temperature and electron density in the partially ionized zone are higher in the case of the power law spectrum , the column densities are very similar , so that our conclusions from figure [ uplot ] are not very dependent on the exact shape of the ionizing spectrum . from the ionization models above we have found that the low ionization species we have been studying are mainly found in gas with a temperature of a few thousand kelvin . this means it is very well possible that some form of dust will be present in the outflowing clouds . as we will demonstrate , this can change the ionization models significantly . the presence of dust will have three main effects : 1 . fe is likely to be strongly depleted . this means that the hydrogen column density required to obtain the observed column density is much larger . it is possible that fe and mg are depleted by _ different _ depletion factors . thus the column density ratios between fe and mg ions will be changed . 3 . the dust absorbs ionizing radiation , which changes the ionization equilibrium in the slab . the following models are based on the orion dust model in cloudy , which is in turn based on the work on abundances and dust properties in the orion nebula h ii regions ( ferland 1996 , baldwin et al . 1991 , osterbrock et al . the properties of this dust model that are important for this work are the depletions of fe and mg ( @xmath4 0.1 for both ) and the dust opacity , which reflects the absence of small grains . the physical conditions in the clouds we consider are more similar to those in h ii regions than those in the local ism , for which a different depletion pattern and grain size distribution has been derived(e.g cowie & songaila 1986 ) . the results of the ionization models including dust are illustrated in figure [ udust ] . the most obvious effect of the presence of dust is the depletion of fe , which leads to thicker slab models with a much thicker partially ionized zone , as discussed above . however , the most important physical difference between these models and the models without dust is that the fraction no longer drops as steeply behind the ionization front , so that the largest contribution to the column density now comes from the partially ionized zone rather than the strmgren layer . the reason for this behaviour is that the ionizing flux ( including trapped ly @xmath81 ) is absorbed by dust , so that the reduction in recombination rate due to the drop in electron density is partially compensated . thus in the models with dust the and column density reside in the same region of the slab , whereas they are separated in the models without dust . since the partially ionized zone now contributes to the column , the observed column can be explained with a much smaller mg abundance than in the models without dust . the largest column density for the given column density is produced for @xmath82 ( model 13 ) . even for this model it is still necessary to assume that mg is three times less depleted than in the standard orion dust model ( i.e. by a factor of 0.33 instead of 0.1 ) to reproduce the observed column . for lower ionization parameters the models run into the 1000 k temperature limit again , which means that most would be too cold to explain the excitation , as discussed above . higher ionization parameters can also reproduce the column density if we allow mg to be even less depleted . however , the models with larger ionization parameters are effectively excluded by another constraint , which is that the spectrum of fbqs1044 + 3656 does not appear to suffer from very strong extinction . the extinction at 2500 predicted by the models with dust is given in table [ ionmods ] . a comparison of the spectrum of fbqs1044 + 3656 with the mean qso template of weymann et al . 1991 shows that the extinction at 2500 is not likely to be larger than 1.0 magnitude , unless the spectrum is intrinsically very steeply rising towards shorter wavelengths . this excludes the models with higher ionization parameters than @xmath82 . all models with @xmath83 have column densities that lie well below our upper limit . we conclude that models with dust can explain the observed properties of the outflow reasonably well , and do not suffer from the problems with an improbable cloud geometry and separation of the and producing zones encountered in the high metallicity models without dust . for the cosmological parameters @xmath84 , @xmath85 and @xmath86 , the absolute b magnitude of fbqs1044 + 3656 is @xmath87 . assuming further that the shape of the continuum is close to the typical agn spectrum described by matthews & ferland ( 1987 ) , we estimate the luminosity of fbqs1044 + 3656 to be close to the canonical qso value of @xmath88 . in table [ ionmods ] , the distance between the source of ionizing radiation and the absorbing cloud corresponding to each model is tabulated assuming this luminosity . both the high metallicity models without dust and the dusty models require that the outflow is situated at a distance of several hundred parsecs from the source of ionizing radiation , i.e typical of the qso narrow line region . this distance is about two orders of magnitude larger than the distance derived by wcp for the absorbing clouds in the balqso q0059 - 2735 . this large size of the outflow raises several questions . the velocity of the outflow is considerably larger than that of material in the narrow line region , and is more typical of the bel clouds that are situated at only 0.1 parsec from the nucleus . the combination of a large distance and the very small cloud size implied by partial covering re - emphasizes the problems associated with the stability of small , high speed clouds with a very low filling factor , which have been discussed extensively in the context of the formation of broad absorption lines ( weymann , turnshek & christiansen 1985 , begelman , de kool & sikora 1991 , weymann 1997 ) . although the small implied sizes of the clouds are hard to understand theoretically , we do point out that the small transverse cloud size implied by partial covering is consistent with the physical thickness of @xmath89 cm of the slabs in our ionization models as derived from the hydrogen densities and column densities . one of the main objections to large distances for bal - like outflows comes from constraints on the mass and energy involved using the typical parameters of high - ionization bals . the low ionization absorption lines studied in this paper , however , have rather low velocity and equivalent width compared to typical high - ionization bals . we can estimate the mass loss rate and kinetic energy of the outflow for the specific case of fbqs1044 + 3656 to see if they are unreasonably large . the mass involved in the low ionization outflow can be estimated from @xmath90 where @xmath91 is the hydrogen column density normalized to @xmath92@xmath93 , @xmath94 the size of the outflow in units of @xmath92 cm and @xmath95 is the global covering factor of the outflow . since absorption is not very common in qsos , @xmath95 could be as small as @xmath49 for absorption - producing outflows . alternatively the outflows may be short lived phenomena , in which case the constraints on the mass loss rate are also less severe . taking the parameters of our preferred models 7 or 15 and estimating the flow time to be @xmath96 , the implied mass outflow rate is @xmath97 yr@xmath15 and the kinetic power of the outflow is @xmath98 erg s@xmath15 . these are values that do not argue strongly against the large size of the outflow . as mentioned in the introduction , the characteristics of the absorption lines in fbqs1044 + 3656 are intermediate between those of bals and aals . weymann et al . ( 1991 ) classified a qso as a balqso if the line profile showed a continuous absorption of at least 10% in depth spanning more than 2000 in velocity , discounting absorption closer than 3000 bluewards of the emission peak . the absorption in fbqs1044 + 3656 meets these criteria , even when a correction is made for the fact that the doublet separation in is larger than in . the total width of the absorption is about the same as in q0059 - 2735 which exhibits extremely strong bals in the higher ionization lines , and it is an established property of low ionization balqsos that the absorption troughs of higher ionization species like or are generally much wider than the absorption ( weymann et al . thus , considering the properties of the absorption alone fbqs1044 + 3656 should be classified as a balqso . in the standard picture of balqsos , the absorption lines are formed in an outflow with a typical size of a few parsec . the characteristics of the very low ionization lines of and are typical of aals ( weymann et al . 1979 , foltz et al . 1986 ) , for which no firm distance scale has been established so that the distance of a few hundred parsec found in this paper is perhaps not a surprise . the remarkable fact lies in the excellent correspondence in velocity between the strongest features and the and absorption lines , which establishes without a doubt that the very low ionization lines and the lines are formed in the same outflow . the association of the and absorption with the bal is much clearer in fbqs1044 + 3656 than in q0059 - 2735 , which only shows a number of relatively narrow components with outflow speeds of a few hundred kilometer per second , and a very smooth bal profile . the combined properties of the , and lines seem to indicate that fbqs1044 + 3656 contains an outflow with bal - like characteristics situated at a distance of a few hundred parsec from the nucleus , a scale very much larger than would have been attributed to the outflow on the basis of the lines only . at this time , fbqs1044 + 3656 is a unique object and we do not know how general these results are . in view of the importance of establishing the relation between the large scale outflow found here and the classical high ionization bal phenomenon , the essential next step in the study of fbqs1044 + 3656 is to obtain a uv spectrum of the source to compare the absorption profiles of the low and high ionization species . part of this work was performed under the auspices of the u.s . department of energy by university of california lawrence livermore national laboratory under contract no . w-7405-eng-48 . arav , n. , & li , z. y. 1994 , apj , 427 , 700 arav , n. , korista , k.t . , de kool , m. , junkkarinen , v.t . & begelman , m.c . 1999a , , 516 , 27 arav , n. , becker , r.h . , laurent - muehleisen , s.a . , gregg , m.d . , white , r.l . & de kool , m. 1999b , , 524,566 . baldwin , j. , ferland , g.j . , martin , p.g . , corbin , m. , cota , s. , peterson , b.m . , & slettebak , a. , 1991 , apj 374 , 580 barlow , t.a . 2000 , aj , in preparation barlow , t.a . & sargent , w.l.w . 1997 , aj , 113 , 136 begelman , m.c . , de kool , m. & sikora , m. 1991 , , 382,416 bergeson et al . 1996 , apj , 464 , 1044 cardelli , j.a . & savage , b.d . 1995 , , 452 , 275 collin - souffrin , s. , joly , m. , pequignot , d. & dumont , s. 1986 , a & a , 166 , 27 ferland , g.j . , 1996 , hazy , a brief introduction to cloudy , university of kentucky department of physics and astronomy internal report . foltz , c.b . 1986 , , 307,504 francis , p.j . , hewett , p.c . , foltz , c.b . , chafee , f.h . , weymann , r.j . & morris , s.l . 1991 , , 373 , 465 fuhr , r.j . , martin , g.a . & wiese , w.l . 1988 , j. phys . chem . ref . data , 17 , suppl . 4 , 108 ganguly , r. , eracleous , m. , charlton , j.c . & churchill , c.w . 1999 , , 117 , 2594 giridhar , s. & ferro , a.f . 1995 , rev . , 31 , 23 hamann , f. , korista , t. k. , & morris , s. l. 1993 , 415 , 541 hamann , f. , barlow , t.a . , beaver , e.a . , burbidge , e.m . , cohen , r.d . , junkkarinen , v.t . & lyons . , r. 1995 , , 443 , 606 hamann , f. , barlow , t.a . , & junkkarinen , v.t . 1997 , , 478 , 87 hamann , f.w . , netzer , h. & shields , j.c . 2000 , , 536 , 101 laor , a. , fiore , f. , elvis , m . , wilkes , b.j . & mcdowell , j.c . 1997 , , 477 , 93 mathews , w.g . & ferland , g. 1987 , , 323 , 456 murray , n. , chiang , j. , grossman , s.a . , & voit , g.m . 1995 , , 451 , 498 nahar , s.n . 1995 , a & a 293 , 967 osterbrock , d. e. , tran , h.d . , & veilleux , s. , 1992 , apj 389 , 305 petitjean , p. & srianand , r. 1999 , a & a , 345 , 73 raassen , a.j.j . & uylings , p.h.m . 1998 , a & a , 340 , 300 srianand , r . & petitjean , p. 2000 , a & a , 357 , 414 turnshek , d. a. 1988 , in space telescope sci . 2 , qso absorption lines : probing the universe , ed . s. c. blades , d. a. turnshek , & c. a. norman ( cambridge : cambridge univ . press ) , 17 verner , e.m . , verner , d.a . , korista , k.t . , ferguson , j.w . , hamann , f. , & ferland , g.j . 1999 , , 120 , 101 vogt et al . 1994 voit , g.m . , weymann , r.j . & korista , k.t . 1993 , , 413 , 95 wampler , e.j . , chugai , n.n . & petitjean , p. 1995 , , 443 , 586 ( wcp ) weymann , r.j . , williams , r.e . , peterson , b.m . & turnshek , d.a . 1979 , , 234 , 33 weymann , r. j. , turnshek , d. a. , & christiansen , w. a. 1985 , in astrophysics of active galaxies and quasi - stellar objects , ed . j. miller ( oxford : oxford univ . press ) 333 ( wtc ) weymann , r. j. , morris , s. l. , foltz , c.b . & hewett , p.c . 1991 , , 373 , 23 weymann r.j . 1997 , in asp conf . 128 , masss ejection from agn , ed . n. arav , i. shlosman , & r.j . weymann ( san francisco : asp),3 white , r.l . 2000 , apjs , 126 , 133 wills , b.j . , netzer , h , & wills , d. 1985 , , 288 , 94 zhang , h.l . & pradhan , a.k . 1995 , a & a , 293 , 953 h. nussbaumer & p. j. storey , 1980 , a & a , 89 , 308 quinet p. , le dourneuf m. & zeippen c.j . , 1996 , a & a suppl . 120 , 361 rrrrrrrrr @xmath33 & 2632.10 & 667.683 & 6 & 0.084 & 0.125 & 0.076 & 0.085 & 0.084 + & 2631.83 & 862.613 & 4 & 0.120 & 0.175 & 0.112 & 0.131 & 0.125 + & 2629.07 & 977.053 & 2 & 0.180 & 0.227 & 0.173 & 0.173 & 0.173 + & 2626.45 & 384.790 & 8 & 0.043 & 0.068 & 0.043 & 0.044 & 0.043 + & 2622.45 & 977.053 & 2 & 0.050 & 0.065 & 0.056 & 0.056 & 0.055 + & 2621.19 & 862.613 & 4 & 0.004 & 0.003 & 0.004 & 0.004 & 0.004 + & 2618.39 & 667.683 & 6 & 0.045 & 0.050 & 0.045 & 0.050 & 0.048 + & 2614.60 & 862.613 & 4 & 0.100 & 0.114 & 0.104 & 0.109 & 0.112 + & 2612.65 & 384.790 & 8 & 0.110 & 0.131 & 0.106 & 0.125 & 0.125 + & 2607.86 & 667.683 & 6 & 0.110 & 0.118 & 0.110 & 0.118 & 0.124 + & 2600.17 & 0.000 & 10 & 0.230 & 0.240 & 0.224 & 0.240 & 0.245 + & 2599.14 & 384.790 & 8 & 0.099 & 0.095 & 0.099 & 0.109 & 0.109 + & 2586.65 & 0.000 & 10 & 0.068 & 0.055 & 0.065 & 0.069 & 0.071 + & 2414.04 & 977.053 & 2 & 0.190 & 0.175 & 0.186 & 0.177 & 0.186 + & 2411.80 & 977.053 & 2 & 0.210 & 0.220 & 0.208 & 0.208 & 0.223 + & 2411.25 & 862.613 & 4 & 0.190 & 0.203 & 0.194 & 0.208 & 0.223 + & 2407.39 & 862.613 & 4 & 0.140 & 0.161 & 0.141 & 0.147 & 0.161 + & 2405.61 & 667.683 & 6 & 0.200 & 0.243 & 0.196 & 0.235 & 0.252 + & 2405.16 & 862.613 & 4 & 0.031 & 0.031 & 0.031 & 0.026 & 0.028 + & 2399.97 & 667.683 & 6 & 0.120 & 0.131 & 0.118 & 0.118 & 0.124 + & 2396.35 & 384.790 & 8 & 0.270 & 0.288 & 0.261 & 0.286 & 0.307 + & 2396.14 & 667.683 & 6 & 0.019 & 0.023 & 0.019 & 0.015 & 0.016 + & 2389.35 & 384.790 & 8 & 0.089 & 0.097 & 0.088 & 0.083 & 0.088 + & 2383.78 & 384.790 & 8 & 0.012 & 0.012 & - & - & - + & 2382.76 & 0.000 & 10 & 0.364 & 0.343 & 0.417 & 0.316 & 0.347 + & 2381.48 & 667.683 & 6 & 0.036 & 0.010 & 0.036 & 0.034 & 0.036 + & 2374.46 & 0.000 & 10 & 0.028 & 0.053 & 0.039 & 0.032 & 0.033 + & 2367.59 & 0.000 & 10 & 0.004 & 0.004 & - & - & - + & 2365.55 & 384.790 & 8 & 0.051 & 0.055 & 0.051 & 0.050 & 0.053 + & 2359.82 & 862.613 & 4 & 0.038 & 0.038 & - & - & - + & 2349.02 & 667.683 & 6 & 0.100 & 0.077 & 0.103 & - & 0.090 + & 2345.00 & 977.053 & 2 & 0.130 & 0.126 & 0.138 & 0.158 & 0.169 + & 2344.21 & 0.000 & 10 & 0.113 & 0.126 & 0.110 & 0.115 & 0.126 + & 2338.72 & 862.613 & 4 & 0.087 & 0.087 & 0.087 & 0.093 & 0.100 + & 2333.51 & 384.790 & 8 & 0.091 & 0.081 & 0.097 & - & 0.077 + & 2328.11 & 667.683 & 6 & 0.032 & 0.040 & 0.031 & 0.036 & 0.037 + @xmath34 & 2360.72 & 1873 & 10 & 0.020 & 0.040 & 0.028 & 0.028 & 0.023 + & 2348.83 & 1873 & 10 & 0.034 & 0.086 & 0.037 & 0.043 & 0.037 + & 2332.02 & 1873 & 10 & 0.019 & 0.004 & 0.021 & 0.021 & 0.018 + & 2369.32 & 2838 & 6 & 0.033 & 0.064 & 0.033 & 0.034 & 0.030 + & 2361.02 & 2430 & 8 & 0.037 & 0.073 & 0.037 & - & 0.031 + @xmath35 & 2564.24 & 8392 & 6 & 0.085 & 0.098 & 0.084 & - & 0.094 + & 2563.30 & 7955 & 8 & 0.110 & 0.143 & 0.111 & - & 0.134 + cccc @xmath33 & 0 . & 1.00 & 1.00 + & 384 . & 0.04 ( 0.02 ) & 0.12 ( 0.03 ) + & 667 . & 0.04 ( 0.02 ) & 0.10 ( 0.03 ) + & 862 . & 0.04 ( 0.03 ) & 0.10 ( 0.03 ) + & 977 . & @xmath990.1 & 0.10 ( 0.03 ) + + @xmath34 & 1873 . & @xmath990.3 & 0.50 ( 0.1 ) + & 2430 . & @xmath990.6 & @xmath990.2 + & 2838 . & @xmath990.6 & @xmath990.2 + + @xmath35 & 7955 . & @xmath990.1 & @xmath990.03 + & 8392 . & @xmath990.2 & @xmath990.1 + ccccccccccc 90 1 & -1.0 & 3.90 & agn & solar & no & 22.00 & 12.45 & 10.44 & - & 0.24 + 2 & -2.0 & 4.00 & agn & solar & no & 21.01 & 12.89 & 11.23 & - & 0.67 + 3 & -3.0 & 4.10 & agn & solar & no & 20.25 & 12.75 & 11.66 & - & 1.9 + 4 & -4.0 & 4.20 & agn & solar & no & ? & @xmath312.47 & @xmath312.09 & - & 5.3 + 5 & -5.0 & 4.50 & agn & solar & no & ? & @xmath312.82 & @xmath312.56 & - & 19 + 6 & -6.0 & 4.90 & agn & solar & no & ? & @xmath312.82 & @xmath313.20 & - & 94 + 7 & -2.0 & 4.00 & agn & 3 @xmath100 solar & no & 20.97 & 13.35 & 11.48 & - & 0.67 + 8 & -2.0 & 4.00 & power law & solar & no & 21.03 & 12.74 & 11.21 & - & 0.53 + 9 & -4.0 & 4.20 & power law & solar & no & 20.00 & 12.34 & 12.00 & - & 4.2 + 10 & -6.0 & 4.90 & power law & solar & no & ? & @xmath312.79 & @xmath313.06 & - & 74 + 11 & -1.0 & 4.60 & agn & orion & yes & 21.70 & 12.61 & 11.88 & 4.2 & 0.10 + 12 & -2.0 & 5.00 & agn & orion & yes & 21.26 & 12.49 & 11.67 & 1.5 & 0.21 + 13 & -3.0 & 5.00 & agn & orion & yes & 21.06 & 12.86 & 12.30 & 0.96 & 0.66 + 14 & -4.0 & 5.00 & agn & orion & yes & ? & @xmath312.65 & @xmath312.42 & ? & 2.1 + 15 & -3.0 & 5.00 & agn & orion mg @xmath100 3 & yes & 21.06 & 13.34 & 12.30 & 0.96 & 0.66 +

a relativistic helicity has been formulated in the four - dimensional minkowski space - time . whereas the relativistic distortion of space - time violates the conservation of the conventional helicity , the newly defined relativistic helicity conserves in a barotropic fluid or plasma , dictating a fundamental topological constraint . the relation between the helicity and the vortex - line topology has been delineated by analyzing the linking number of vortex filaments which are singular differential forms representing the pure states of banach algebra . while the dimension of space - time is four , vortex filaments link , because vorticities are primarily 2-forms and the corresponding 2-chains link in four dimension ; the relativistic helicity measures the linking number of vortex filaments that are proper - time cross - sections of the vorticity 2-chains . a thermodynamic force yields an additional term in the vorticity , by which the vortex filaments on a reference - time plane are no longer pure states . however , the vortex filaments on a proper - time plane remain to be pure states , if the thermodynamic force is exact ( barotropic ) , thus , the linking number of vortex filaments conserves .

1308.2455

the helicity of a vector field is a topological index measuring the _ link _ , _ twist _ and _ writhe _ of the field lines . here we say ` index ' because it is invariant under the action of diffeomorphism groups generated by some ideal dynamics . conventionally , the helicity is defined for three - dimensional vectors ; that the dimension of space is three is , in fact , essential to define a helicity ( because , as to be formulated in a general setting , the helicity is the integral of a 3-form ) . in this work , however , we study four - dimensional vectors in the minkowski space - time . in the relativistic dynamics , the conventional helicity is no longer an invariant . one might connect the non - conservation of the helicity with the topological fact that loops ( 1-cycles ) do not link in four - dimensional space@xcite . however , this is not true ; we will show that the field lines obeying an ideal ( barotropic ) equation of motion is still subject to the topological constraint ; two field lines do link in the spatial subspaces ( temporal cross - sections of the space - time ) , and the linking number conserves . we will formulate a lorentz - covariant helicity , show its invariance , and delineate its relation to the linking number . on the other hand , we will find the reason why the conventional helicity fails to describe the link in the relativistic space - time . we start with a short review of the conventional helicity . let @xmath0 be a three - dimensional vector field defined on a domain @xmath1 . we call @xmath2 the _ vorticity _ of @xmath0 . if @xmath3 has a boundary @xmath4 , we assume @xmath5 where @xmath6 is the unit normal vector onto @xmath4 . we define ( assuming integrability ) @xmath7 and call it the _ helicity _ of @xmath8 ( or , sometimes , of @xmath0 ) on @xmath3 ; the integrand is called the _ helicity density_. the pioneering use of the helicity in the classical field theory was made by woltjer@xcite in order to characterize twisted magnetic field lines ; for the magnetic field @xmath8 and its vector potential @xmath0 , obeying the ideal magnetohydrodynamics ( mhd ) equation ( with the scalar potential @xmath9 ) @xmath10 as well as the boundary condition ( [ bc ] ) , the corresponding _ magnetic helicity _ is a constant of motion . for a fluid , we define the _ fluid helicity _ by putting @xmath11 ( fluid velocity ) and @xmath12 ( vorticity ) . in an ideal fluid , @xmath13 obeys the evolution equation @xmath14 where @xmath15 is the total enthalpy ( we may write @xmath16 with the static enthalpy @xmath17 ; in a homentropic flow , @xmath18 with the density @xmath19 and the pressure @xmath20 ) . if the boundary condition ( [ bc ] ) holds , the fluid helicity is conserved . the helicity is a topological index characterizing the twists of the bundle of field lines ; since it is an integral over a volume , some different geometrical characteristics of field lines are summed up in @xmath21 ; see@xcite . to delineate its topological meaning in the simplest form , let us consider a pair of _ vortex filaments _ ( to be identified as _ pure - state _ differential forms ; see sec.[subsec : pure - state ] ) . suppose that @xmath22 and @xmath23 are a pair of disjoint loops ( closed curves bounding disks ) of class @xmath24 in @xmath25 . the unit tangent vectors on these curves are denoted by @xmath26 and @xmath27 , respectively . the directions of @xmath26 and @xmath27 are arbitrarily chosen , and they are attributed to the loops @xmath22 and @xmath23 as their _ orientations_. we regard both @xmath26 and @xmath27 as @xmath28-measures on the loops @xmath22 and @xmath23 , and consider a pair @xmath29 . for an arbitrary smooth loop @xmath30 that singly links with either @xmath22 or @xmath23 , we find , using stokes s formula ( in the generalized sense for the singular @xmath8 ; see remark[remark : generalize_stokes_formula ] ) and the definition @xmath31 , @xmath32 where @xmath33 is the unit tangential vector on @xmath30 , @xmath34 is the length element on @xmath30 , @xmath35 is a surface bounded by @xmath30 , @xmath6 is the unit normal vector on @xmath35 , and @xmath36 is the surface element on @xmath35 . if @xmath30 and @xmath37 do not link , @xmath38 . we may consider a more complex link of @xmath30 with @xmath37 ; by a homotopical deformation of the path @xmath30 of the integral ( [ stokes ] ) , we find that @xmath39 evaluates the _ linking number _ of the pair of loops @xmath30 and @xmath37 ( the sign of link is determined by the orientations of the loops ) . by the definition , we may write @xmath40 . using ( [ stokes ] ) , we obtain @xmath41 the linking number in ( [ stokes2 ] ) can be evaluated by the _ biot - savart integral_. the relation @xmath42 is inverted as @xmath43 inserting @xmath44 into ( [ biot - savart ] ) , and using the resultant @xmath0 in ( [ helicity_conventional ] ) , we obtain @xmath45 the right - hand - side integral ( @xmath46 ) is called the _ gauss linking number_. the relation ( [ stokes2 ] ) derived by stokes formula gives the proof that @xmath47 evaluates an integer number counting the link of two loops @xmath22 and @xmath23 ( see @xcite for generalization to a higher dimension ; see for the gauss integral for three - component links ) . we will build a topological theory of relativistic field lines around two new constructions ; the first is an appropriate _ relativistic helicity _ in the four - dimensional minkowski space - time , and the second is the notion of _ pure - state vorticities _ by which the helicity reads as the linking number of vortex filaments . in sec.[sec : preliminaries ] , we will start by reviewing the basic equations that describe the relativistic ideal ( barotropic ) dynamics of a plasma ( charged fluid ) . the helicity will be defined for the vorticity of the canonical momentum that combines the mechanical momentum and the electromagnetic field . in sec.[sec : r - helicity ] , we will formulate the relativistic helicity , and show its conservation . section[sec : topology ] is devoted for the delineation of the topological implication of the helicity conservation . to this end , we will consider the link of vortex filaments which are formally the aforementioned @xmath28-measures on co - moving loops . we will justify them as the pure - states of banach algebra , and show that they are the generalized ( weak ) solutions of the equation of motion . for a pair of pure - state vortex filaments , the relativistic helicity evaluates their linking number in the spatial subspace , and its conservation parallels the relativistically corrected kelvin s circulation law . [ remark : generalize_stokes_formula ] to derive ( [ stokes2 ] ) , we evaluated the integral @xmath48 ( so called _ circulation _ ; see lemma[lemma : circulation_law ] ) along the loop @xmath37 on which the @xmath28-measure vorticity @xmath49 is supported , and picked up the contribution from the other @xmath28-measure vorticity on the loop @xmath50 ( @xmath51 ) . however , the integrand @xmath0 is not a continuous function , because it is generated by the @xmath28-measure @xmath8 . here are two mathematical issues pertinent to the use of stokes formula : let us decompose @xmath52 with @xmath53 ( @xmath54 ) . 1 . on the loop @xmath22 , @xmath55 is a smooth function , thus @xmath56 can be evaluated in the classical sense . to relate this integral with the `` source '' @xmath27 of @xmath55 , however , we invoke stokes s formula ( [ stokes ] ) in the generalized sense . to justify ( [ stokes ] ) for a @xmath28-measure vorticity @xmath57 , we first approximate @xmath27 by a smooth vector field @xmath58 that gives the same surface integral @xmath59 independent of @xmath60 , and path the limit of @xmath61 ( @xmath62 ) to define @xmath63 2 . to obtain ( [ stokes2 ] ) , we estimated @xmath64 ( @xmath54 ) , which means that @xmath65 . in the neighborhood of each loop @xmath37 , however , @xmath66 is not continuous . to justify these integrals , we consider a homotopy sequence of loops @xmath67 , and define @xmath68 we denote the minkowski space - time by @xmath69 . on a reference frame , we may decompose @xmath70 with @xmath71 ( time ) and @xmath72 ( space ) . following the standard notation , we write @xmath73 where @xmath74 is the speed of light . by a metric tensor @xmath75 we can write @xmath76 . the space - time gradients are denoted by @xmath77 the relativistic 4-velocity ( normalized by @xmath74 ) is defined by the proper - time derivative : @xmath78 where @xmath79 and @xmath80 . obviously , @xmath81 . the 4-momentum of a particle is @xmath82 , where @xmath83 is the rest mass of the particle . for a fluid , the _ effective rest mass _ is given by @xmath84 with a proper ( static ) molar enthalpy @xmath17 . we may write , on a rest frame , @xmath85 where @xmath19 is the number density , @xmath20 is the pressure , and @xmath15 is the internal energy that includes the rest mass energy @xmath86 as well as the thermal energy@xcite . the fluid 4-momentum is @xmath87 obviously , @xmath88 [ remark : nr ] in the non - relativistic ( nr ) limit ( @xmath89 ) , the 4-velocity ( [ relativistic_4-velocity ] ) coincides with the reference - frame velocity : @xmath90 the nr particle 4-momentum is @xmath91 ( in the 0-component , we have subtracted the rest - mass energy @xmath86 from the energy : @xmath92 ) . the nr fluid momentum 4-vector is @xmath93 where @xmath94 we need to assume @xmath95 to approximate @xmath96 . the energy - momentum tensor of a fluid is @xmath97 the quasi - static ( entropy - conserving ) equation of motion is derived by @xmath98 , which reads @xmath99 where @xmath100 is the _ matter field tensor_. by the thermodynamic first and second laws , we may write , for an isentropic flow , @xmath101 with the entropy @xmath35 . hence , the equation of motion ( [ eqm - tensor ] ) is rewritten as@xcite @xmath102 contracting both sides of ( [ eqm - tensor ] ) with @xmath103 , we obtain @xmath104 , implying the entropy conservation ( @xmath105 on a reference frame ) . in the present work , _ barotropic fluids _ will be at the center of discussions ; when @xmath35 is a function of @xmath106 , we may write @xmath107 , and then , ( [ eqm - tensor ] ) reads @xmath108 evidently , @xmath109 . notice that the heat term @xmath110 is reduced into an exact differential @xmath111 . this is the root cause of various topological constraints in a barotropic flow . for example , kelvin s circulation theorem ( see lemma[lemma : circulation_law ] ) is a consequence of vanishing heat cycle @xmath112 ( see @xcite for the relation between the circulation law and heat cycles , as well as its relativistic generalization ) . however , the exactness of @xmath111 is now in the relativistic space - time ; its representation on a reference frame may be non - exact . this difference causes an interesting `` relativistic effect '' which is , indeed , the target of the present exploration . henceforth , we will assume a barotropic relation , and discuss the equation of motion in the form of ( [ eqm - tensor - barotropic ] ) ; one may generalize the following formulations to a _ fluid by replacing @xmath111 by @xmath110 . [ remark : nr - barotropic ] it is remarkable that , in the nr limit , the barotropic heat term @xmath113 can be absorbed by the energy term ( 0-component ) of the fluid 4-momentum : modifying ( [ fluid-4-momentum - nr ] ) as @xmath114 and redefining the field tensor @xmath115 by the modified 4-momentum , the nr equation of motion ( [ eqm - tensor - barotropic ] ) can be reduced to @xmath116 hence , a nr barotropic fluid equation is equivalent to a homentropic one ( @xmath117 ) . however , such reduction is not possible in the relativistic regime . for a charged fluid ( plasma ) , we have to dress the momentum 4-vector with the electromagnetic ( em ) potential @xmath118 , and replace the 4-momentum by the _ canonical 4-momentum _ @xmath119 where @xmath120 is the charge . the matter - em field tensor ( 2-form ) is@xcite @xmath121 by which the equation of motion ( [ eqm - tensor - barotropic ] ) modifies to include the lorentz force as @xmath122 the relation ( [ enthalpy ] ) of a neutral - fluid is generalized as @xmath123 henceforth , we will consider the em - dressed ( or canonical ) momentum @xmath124 and its derivatives ( field tensor , helicity , etc . ) . synonymously , we may call it a matter - dressed em potential , i.e. , the em potential @xmath125 ( multiplied by the charge @xmath120 ) coupled with the matter motion @xmath126 , which will be denote by @xmath127 . the latter em - based notation will be useful when we compare newly defined helicities with the familiar magnetic helicity ( see sec.[subsec : tensor ] ) . the conventional mhd equation ( [ ideal_mhd ] ) is the zero mass and heat ( @xmath128 and @xmath129 ) limit of the spatial component of ( [ eqm - tensor - plasma ] ) . for the convenience of the forthcoming calculations , we write down the components of the field tensors and the equations of motion on a reference frame . here we invoke an analogy of the familiar faraday tensor to denote the elements of the field tensors . we write the spacial components in bold - face symbols . denoting the canonical momentum ( or the dressed em potential ) by @xmath130 we define @xmath131 then , the field tensors are written as @xmath132 we may identify @xmath133 as the minkowski - hodge dual of @xmath134 ; see remark[remark : mh - dual ] . we have an obvious relation @xmath135 the equation of motion ( [ eqm - tensor - barotropic ] ) reads @xmath136 in the following analysis , the differential - geometric formulations of the field tensors and the equation of motion will play an essential role . the 4-velocity @xmath137 is regarded as a vector field @xmath138 ( @xmath139 is the minkwski space - time , and @xmath140 is the tangent bundle on @xmath139 ) . we denote the corresponding covector ( 1-form ) by @xmath141 ( the cotangent bundle on @xmath139 ) . the 4-momentum is a 1-form : @xmath142 , and its exterior derivative is the field tensor : @xmath143 the components of @xmath144 ( the minkowski - hodge - dual of @xmath145 ) will be denoted by @xmath146 , i.e. @xmath147 ; see ( [ * m - components_down ] ) . using the definition ( see remark[remark : mh - dual ] ) @xmath148 we may write @xmath149 denoting by @xmath150 the interior product with @xmath151 , we write @xmath152 . the equation of motion ( [ eqm - tensor - plasma ] ) is now written as @xmath153 operating the lie derivative @xmath154 on the momentum 1-form @xmath155 , we obtain @xmath156 where we have used the relation ( [ enthalpy2 ] ) . hence ( [ eqm - dif - form ] ) reads @xmath157 the right - hand side is the exact differential of the _ free energy_. in the following discussion , we will use the form ( [ eqm - lie - form ] ) of the equation of motion . the 2-form @xmath158 may be called the `` four - dimensional vorticity '' of the canonical momentum . the relativistic vorticity equation is derived from ( [ eqm - dif - form ] ) : @xmath159 [ remark : mh - dual ] combining the hodge - duality ( between @xmath20-forms and @xmath160-forms ) and the minkowski - duality ( between @xmath161 and @xmath162 ) , the `` star operator '' maps the basis as @xmath163 for example , if @xmath145 is a 2-form based on @xmath164 , the minkowski - hodge dual @xmath144 is a 2-form based on @xmath165 ; see ( [ * m - components_down ] ) , ( [ 2-form basis identity ] ) and ( [ * m ] ) . the ideal fluid / plasma motion is represented by a diffeomorphism @xmath166 that is generated by a vector @xmath151 on the minkowski space - time @xmath167 , i.e. , @xmath168 for every initial point @xmath169 , @xmath170 and @xmath171 . we assume that @xmath172 is a one - parameter group of @xmath173-class diffeomorphisms that are continuously differentiable with respect to @xmath174 . by @xmath166 , we can define various geometrical objects ( chains ) `` co - moving '' with the fluid . for the latter convenience , we introduce two different temporal cross - sections of the space - time @xmath139 ( which we call planes , although they have three dimensions ) . the `` @xmath175-plane '' is , for a fixed parameter @xmath176 , @xmath177 in the relativistic theory , the proper time @xmath174-plane plays a more essential role , which is a three - dimensional hyper - surface in @xmath139 defined by @xmath178 evey object that is initially contained in @xmath179 stays in @xmath180 such an object is said _ co - moving_. we note that a co - moving object does not stay on a @xmath175-plane @xmath181 ; the synchroneity ( simultaneity ) is broken when @xmath182 is inhomogeneous . a co - moving object @xmath183 is _ temporally - thin _ , i.e. , @xmath184 for every @xmath185 , or @xmath186 has zero measure with respect to @xmath187 . this is because the diffeomorphism @xmath166 is dynamical in the sense that @xmath188 , and thus , @xmath166 ( @xmath189 ) does not have any fixed point in @xmath139 . before starting analyses , we make a remark on the framework of our discussions . the vector @xmath151 , representing the velocity of fluid elements , must be consistent with the evolution of other physical quantities @xmath155 , @xmath17 , and @xmath190 . we do not know , however , a theorem guaranteeing the global existence of solutions to the nonlinear system of equations ; it is not possible to construct self - consistent fields @xmath151 , @xmath155 , @xmath145 , etc . by solving the evolution equations . yet , we may discuss _ a priori _ properties of the solutions , i.e. , the mathematical relations that must be satisfied by every solution ( of certain class ) whenever it exists . the aim of this study is to derive a topological conservation law pertinent to the diffeomorphism group @xmath172 . the helicity will guide our exploration . conventionally , a helicity is defined by a three - dimensional integral @xmath191 with three - dimensional vectors @xmath0 and @xmath42 defined on space @xmath192 ; see ( [ helicity_conventional ] ) . here , we consider the helicity of `` matter - dressed em field '' ( or , synonymously , `` em - dressed fluid momentum '' ) by inserting @xmath193 ( @xmath194 ) into @xmath0 , and @xmath195 into @xmath8 ; see sec.[subsec : tensor ] . by straightforward extension of the well - known magnetic - helicity conservation in an ideal fluid obeying ( [ ideal_mhd ] ) ( or the fluid - helicity conservation by ( [ ideal_fluid ] ) ) , we find that @xmath196 is conserved in a non - relativistic barotropic charged fluid ( plasma ) . in relativistic dynamics , however , @xmath21 is no longer a constant ; the relativistic effect yields a `` relativistic baroclinic effect '' by space - time distortion@xcite . the aim of the latter discussion is to formulate a more appropriate helicity that conserves in relativistic motion , and use it to elucidate a fundamental topological property of relativistic dynamics . our construction starts by relating the helicity to a 3-form such as @xmath197 explicitly , we may write @xmath198 where we used the identity @xmath199 . inserting ( [ p - components ] ) and ( [ * m - components_down ] ) , we may write @xmath200 with @xmath201 hence , we may write the conventional helicity as @xmath202 as the domain of integration ( the @xmath175-plane @xmath192 ) is not lorentz covariant , @xmath21 depends on the choice of a reference frame , therefore , we call @xmath21 a _ semi - relativistic helicity_. a fully relativistic helicity must be defined by both covariant integrand and a domain of integration this will be done in the next subsection . here we examine @xmath203 explicitly . using ( [ div mstar ] ) and ( [ eqm - vector-1 ] ) , we obtain @xmath204 we thus find @xmath205 obviously @xmath21 is invariant in the non - relativistic limit ( @xmath206 ) ; we now have a unified non - relativistic helicity conservation law that combines the magnetic - helicity conservation and fluid - helicity conservation introduced in sec.[sec : introduction ] . interestingly , in a homentropic fluid ( @xmath207 ) , @xmath208 , thus , @xmath21 is constant regardless of the relativistic effect . otherwise , the semi - relativistic helicity @xmath21 is not a constant of motion . we define a _ relativistic helicity _ @xmath209 in the minkowski space - time by the integral of the 3-form @xmath210 over a co - moving three - dimensional volume @xmath211 ( @xmath212 ; see sec.[subsec : diffeo ] ) : @xmath213 notice that the integrand 3-form @xmath214 includes space - time coupled terms ( @xmath215 ) , and the synchroneity ( simultaneity ) may be broken on @xmath186 . [ theorem : helicity_conservation ] suppose that @xmath216 is a solution of ( [ eqm - lie - form ] ) such that @xmath217 ( a bounded set ) . the helicity @xmath218 evaluated on a co - moving domain @xmath219 is a constant of motion ( i.e. @xmath220 ) . by the definition of @xmath186 , we obtain @xmath221 operating by @xmath222 on both sides of the equation of motion ( [ eqm - lie - form ] ) , which reads @xmath223 we find @xmath224 . using this , we obtain the vorticity equation @xmath225 the integrand of ( [ 4dhelicity_conservation-1 ] ) reads @xmath226 . \label{4dhelicity_conservation_lie - derivative}\end{aligned}\ ] ] we thus have @xmath227 = c^{-1}\int_{\partial v(s ) } ( h+q\varrho -\theta ) \rmd \mathcal{p}. \label{4dhelicity_conservation}\ ] ] if @xmath228 on the boundary @xmath229 , @xmath230 . now we study the relation between @xmath231 and @xmath186 . by the assumption , @xmath232 . the following lemma[lemma : support ] , then , shows @xmath233 we thus find that @xmath231 is contained in a column that has compact @xmath174-cross - sections . since @xmath186 is temporally thin ( see sec.[subsec : diffeo ] ) , @xmath229 is purely spacial , i.e. for every @xmath234 , @xmath235 ( @xmath236 ) . on each @xmath174-plane @xmath180 , @xmath237 on @xmath229 . < 1.5em - 1.5em plus0em minus0.5em height0.75em width0.5em depth0.25em to complete the proof of theorem[theorem : helicity_conservation ] , we have yet to prove lemma[lemma : support ] . we prepare the following lemma : [ lemma : circulation_law ] let @xmath238 be an arbitrary co - moving loop ( 1-chain bounding a disk ) of class @xmath24 . in a tubular neighborhood of @xmath239 , we assume that a 1-form @xmath155 is continuously differentiable and satisfies the equation of motion ( [ eqm - lie - form ] ) . then , the _ circulation _ @xmath240 is conserved , i.e. @xmath241 . in sec.[sec : topology ] , we will use this lemma for singular solutions of ( [ eqm - lie - form ] ) , so we assume that @xmath155 is a classical solution of ( [ eqm - lie - form ] ) only in a tubular neighborhood the loop @xmath239 where we need to evaluate the circulation . for the existence of a tubular neighborhood of a @xmath24-submanifold , see hirsch@xcite . the derivation is straightforward : @xmath242 < 1.5em - 1.5em plus0em minus0.5em height0.75em width0.5em depth0.25em it is often convenient to rewrite the circulation ( [ r - circulation ] ) in terms of the _ vorticity _ @xmath243 and a co - moving surface @xmath244 such that @xmath245 ; by stokes formula , @xmath246 [ lemma : support ] suppose that @xmath247 is a solution of the equation of motion ( [ eqm - lie - form ] ) . let @xmath158 , and @xmath248 . then , @xmath249 by lemma[lemma : circulation_law ] and ( [ r - circulation-2 ] ) , any co - moving surface @xmath244 that does not intersect with @xmath250 at @xmath251 yields zero circulation for every @xmath252 . on the other hand , if @xmath253 and @xmath250 intersect to yield a finite circulation , it conserves through the motion of @xmath244 . therefore , @xmath254 only on @xmath244 that starts from @xmath253 containing a point on @xmath250 , implying ( [ support_of_m-2 ] ) . < 1.5em - 1.5em plus0em minus0.5em height0.75em width0.5em depth0.25em [ remark : kelvin ] lemma[lemma : circulation_law ] is the special - relativistic correction of kelvin s circulation theorem ( see @xcite for a general - relativistic treatment ) . the point is that the circulation must be evaluated on a co - moving loop @xmath255 , on which the synchroneity ( simultaneity ) with respect to a reference - frame time @xmath175 is broken ; if we evaluate the circulation on a synchronic loop @xmath256 , the corresponding _ semi - relativistic _ circulation is not invariant@xcite . in lemma[lemma : support ] , we used the circulation law to show that the vorticity co - moves with the fluid . the same argument applies to show that every vortex line is frozen in the fluid element . the motion of a vortex line ( or a magnetic field line in an mhd system ) can be described explicitly by the _ connection equation _ that governs the separation vector connecting infinitesimally close fluid elements@xcite . pegoraro@xcite formulated the lorentz - covariant connection equation for a relativistic mhd system . interestingly , the _ magnetic flux _ on a synchronic surface @xmath257 is invariant in the mhd system , because the right - hand side of ( [ e : div k ] ) vanishes in the mhd model ( see also sec.[sec : discussion ] ) . in parallel to this fact , the connection equation of mhd may be written in a seemingly non - covariant form . pegoraro s formulation considers a surface orbit in space - time to overcome the lack of simultaneity on the co - moving magnetic field line . in the next section , we will consider a `` singular '' vortex surface in space - time , which , however , is different from the virtual surface of the separation vectors ; it is physically the orbit in space - time of a vortex filament carrying a unit vorticity , and is mathematically the pure state of banach algebra . link is definable ( as a homotopy invariant ) for a pair of geometric objects ( chains ) having codimension less than or equal to one@xcite ; for example , two loops ( 1-chains bounding disks ) may link in three - dimensional space , while they do not in four - dimensional space . therefore , it seems that the helicity ceases to be related to linking numbers in the four - dimensional relativistic space - time . however , the relativistic helicity conservation , derived in sec.[subsec : r - helicity ] , does impose a topological constraint . the aim of this section is to elucidate the topological meaning of the relativistic helicity conservation by generalizing the conventional relation between the helicity and the link of vortex filaments ( see sec.[sec : introduction ] ) in the three - dimensional space to the minkowski space - time . since @xmath258 is a 2-form , we may consider a _ ( a differential form with hyper - function coefficients ) supported on a two - dimensional surface . a pair of two - dimensional surfaces can link in the four - dimensional space - time . the reason why the link of surfaces yields a topological constraint on loops ( vortex - filaments ) is because such surfaces can be chosen as _ orbits _ of loops . the conventional vortex - filaments are , then , the temporal cross - sections of the surfaces . when we consider a dynamical process in space - time , represented by a diffeomorphism @xmath166 , geometrical manipulations caused by @xmath166 is constrained by the causality ( on any reference frame , @xmath175 can not go to negative ) . for instance , consider a one - dimensional space @xmath259 . two points on the two - dimensional space - time ( @xmath260-@xmath175 plane ) do not link , and the spacial projections of them , @xmath260 and @xmath261 , may exchange their positions on the space - axis without intersecting their orbits in the space - time . however , such exchange is only possible if one particle goes to the direction of negative @xmath175 . otherwise , @xmath260 and @xmath261 can not change their order without causing an intersection of their orbits in the @xmath260-@xmath175 plane . similarly , a topological constraint on the link of surfaces in the four - dimensional space - time causes a constraint on the link of their temporal cross - sections , loops , if the surfaces are the orbits of the loops . to explore the basic topological constraint due to the relativistic helicity conservation , we consider the link of vortex filaments or surfaces that are formally the @xmath28-measures , supported on loops or surfaces , with a _ unit magnitude_. to formulate them on a rigorous mathematical footing , we invoke the the notion of _ pure states_. let us start with a well - known example of commutative @xmath262 ring . a `` point '' on a compact space @xmath192 is equivalent to a pure state of the banach algebra @xmath263 , which is a linear form @xmath264 such that @xmath265 and @xmath266 ( it is `` pure '' if @xmath264 is not a mixture @xmath267 of two distinct states ) . by gelfand s theorem , every point @xmath268 can be represented by a pure state @xmath269 such that @xmath270 for every @xmath271 . algebraically , @xmath269 is the quotient of @xmath263 by the maximum ideal @xmath272 generated by @xmath273 . we may represent a pure state by a @xmath28-function , i.e. , @xmath274 . here , we generalize `` points '' to @xmath20-chains . [ definition : pure_state ] let @xmath139 be a smooth manifold of dimension @xmath19 ( in the present application , @xmath139 is the four - dimensional minkowski space - time ) , and @xmath275 be a @xmath20-dimensional connected null - boundary submanifold of class @xmath24 . each @xmath3 can be regarded as an equivalent of a _ pure - sate functional _ @xmath276 on the space @xmath277 of continuous @xmath20-forms : @xmath278 which can be represented as @xmath279 with an @xmath160-dimensional @xmath28-measure @xmath280 , where @xmath281 are local coordinates , and @xmath282 we call @xmath283 a _ pure state _ @xmath160-form , which is a member of the hodge - dual space of @xmath277 . on compact manifolds , the pure state of a submanifold @xmath3 is obtained as a limit of the thom form of @xmath3 , and the cohomology class of a pure state of @xmath3 corresponds to the poincar dual of @xmath3@xcite . we also note that the duality between submanifolds and pure state parallels the duality between infinite chains in homologies with closed support ( or borel - moore homology ) and cochains in cohomologies@xcite . [ remark : pure - state ] the notion of pure state is commonly used in quantum theory , where the observables are self - adjoint operators , and their duals are wave functions , members of a hilbert space @xmath284 . the pure states , which are the `` points '' on the unit sphere of @xmath284 , constitute the spectral resolution of the _ identity _ on @xmath284 , the aforementioned @xmath262 ring of scalar functions ( classical mechanical observables ) is _ quantized _ by replacing the ring by a commutative ring of unitary operators ; then , the classical state , a point on coordinates , is replaced by a wave function , a point in the function space @xmath284 . in definition[definition : pure_state ] , observables are differential forms , and their duals are the hodge - dual @xmath160-forms . the notion of a point is , then , generalized to a @xmath20-chain . pure - state on 2-forms in four - dimensional space - time , which is represented by a 2-chain @xmath285 . @xmath286 is the tangential surface element on @xmath285 ( i.e. @xmath287 and @xmath288 are the local coordinates on @xmath285 ) , and @xmath289 is the normal 2-form ( i.e. @xmath290 ) . in these local coordinates , @xmath291 . the initial - time cross - section of @xmath285 is denoted by @xmath292 . an @xmath174-cross - sections of @xmath285 is denoted by @xmath293 . , scaledwidth=50.0% ] let @xmath285 be a 2-dimensional connected null - boundary submanifold of class @xmath24 embedded in the four - dimensional space - time @xmath139 ( see fig.[fig : pure_state ] ) . a pure - state functional @xmath294 is represented by a pure - state 2-form @xmath295 such that @xmath296 where @xmath297 is the two - dimensional @xmath28-function supported on @xmath285 , @xmath298 is a local two - dimensional surface element on @xmath285 ( @xmath299 and @xmath300 are tangent to @xmath285 ) , and @xmath301 is decomposed in terms of @xmath164 ; @xmath302 is a certain antisymmetric tensor characterizing the surface @xmath285 , which we denote @xmath303 in sec.[subsec : orbit_of_loop ] , we will identify @xmath285 as the orbit of vortex filaments , and then , @xmath302 of ( [ m - components_down ] ) will be compared with the vorticity 2-form @xmath134 of ( [ m - components_down ] ) . the @xmath175-cross - section of @xmath285 is @xmath304 which is assumed to be a loop ( 1-cycle bounding a disk ) in the space @xmath181 . denoting by @xmath305 the @xmath28-function supported on @xmath256 ( i.e. @xmath306 ) , we define a `` @xmath307-filament '' on a @xmath175-plane by @xmath308 where @xmath309 with @xmath310 , i.e. @xmath311 is the minkowski - dual of the magnetic - field - like 3-vector @xmath312 ; notice the flip of the sign by the representation the 1-form @xmath311 on the basis @xmath313 . the projector @xmath314 is written as @xmath315 notice that @xmath295 has other three components including @xmath316 on a @xmath175-plane , which we call an `` @xmath317-filament '' : @xmath318 where @xmath319 with @xmath320 . the projector @xmath321 is @xmath322 in the relativistic formulation , the proper - time @xmath174-cross - section of @xmath285 is more important : @xmath323 generalizing ( [ t - projector_to_b ] ) , the `` relativistic @xmath307-filament '' is a singular 3-form such that @xmath324 where @xmath325 . similarly , the relativistic @xmath317-filament is defined as ( denoting @xmath326 ) @xmath327 or , we may write @xmath328 explicitly , we may write @xmath329 with a 3-form @xmath330 and a 1-form @xmath331 given by ( denoting @xmath332 and @xmath333 ) @xmath334 we find that the 3-vector parts of @xmath330 and @xmath331 are , respectively , the lorentz transformations of the em - like fields @xmath335 and @xmath336 ; compare ( [ m - components_down ] ) and ( [ m - components_down ] ) . in general , neither @xmath337 nor @xmath338 is a pure - state 3-form ( we use the lower - case @xmath339 to denote a non - pure - state singular filament ) . iff @xmath340 , then @xmath337 is a pure state , and it is the case when @xmath295 belongs to the orthogonal complement of @xmath341 . we will denote by @xmath342 the pure - state @xmath307-filament . here we prove the following lemma : [ lemma : pure - state-1 ] let @xmath343 be a 2-chain of class @xmath24 , whose proper - time cross - section @xmath344 @xmath345 is a single loop ( 1-cycle bounding a disk ) . suppose that @xmath346 is a pure - state 2-form . iff @xmath347 , then @xmath348 is a pure - state 3-from . if @xmath295 is a pure - state 2-form , we may write , for every continuous 2-form @xmath349 , @xmath350 let us put @xmath351 , where @xmath352 is a scalar and @xmath353 is a 1-form . consider a sequence @xmath354 ( @xmath355 ) . if @xmath356 parallels @xmath285 , i.e. @xmath357 ( implying that @xmath347 ) , the right - hand side of ( [ pure - state - filament - proof-1 ] ) yields @xmath358 on the other hand , the left - hand side of ( [ pure - state - filament - proof-1 ] ) reads @xmath359 \wedge a.\ ] ] denoting @xmath360 , we may write , for every continuous 1-form @xmath353 , @xmath361 proving that @xmath362 is a pure - state 3-form . next , we examine the normalization of the pure - states . let us invoke the representation @xmath363 in terms of the tensor ( [ m - components_down ] ) . then , we find @xmath364 which must be unity for @xmath295 to be a pure - state . by ( [ lorentz - transformation - b ] ) and ( [ lorentz - transformation - e ] ) , we obtain @xmath365 \nonumber \\ & = & |\bm{\mathfrak{e}}|^2 - |\bm{\mathfrak{b}}|^2 + |\tilde{\mathfrak{e}}|^2 , \label{normalization-2}\end{aligned}\ ] ] where @xmath366 ( by writing ( [ normalization-2 ] ) as @xmath367 , which implies the lorentz invariance of the lagrangian , we notice that the sign changes because of the minkowski metric ) . by ( [ normalization-1 ] ) , we have shown @xmath368 . therefore , only if @xmath369 ( i.e. @xmath347 ) , @xmath370 so that @xmath371 is a pure state . < 1.5em - 1.5em plus0em minus0.5em height0.75em width0.5em depth0.25em in lemma[lemma : pure - state-1 ] , we found that the projection of a pure - state 2-form @xmath295 onto a proper - time plane @xmath180 yields a pure - state 3-form @xmath362 ( @xmath307-filament ) iff @xmath372 , implying that the surface @xmath285 must be the _ orbit _ of the loop @xmath293 . here we construct @xmath295 from @xmath362 along the orbit . the components ( @xmath302 ) of the pure - state 2-form @xmath295 ( which is the singular counterpart of the vorticity 2-form @xmath158 ) must be consistent with the dynamics equation . as remarked in sec.[subsec : diffeo ] , however , we may not solve the full set of equations to determine @xmath295 . what we are going to construct are the elements of @xmath295 that are consistent with given @xmath151 and @xmath190 ( whereas @xmath151 is related to @xmath155 , and is an unknown variable in the fluid / plasma equations ; @xmath190 must be consistent with @xmath151 through a relation @xmath373 ) . we also note that the equation of motion must be generalized when we consider pure - state vorticities that are not regular functions obeying differential equations in the classical sense ; accordingly , the helicity conservation law must be reformulated to be amenable to the singular vorticities this will be the task of sec.[subsec : toplogy * ] . before the construction of @xmath295 , we generalize the equation of motion ( [ eqm - dif - form ] ) in order to incorporate singular vorticities of pure states into the solutions . let us first rewrite it as @xmath374 = 0 , \label{eqm - dif - form-2}\ ] ] where @xmath375 with @xmath376 . by @xmath377 and @xmath373 , it is evident that ( [ eqm - dif - form ] ) and ( [ eqm - dif - form-2 ] ) are equivalent . the _ adjoint form _ of ( [ eqm - dif - form-2 ] ) is @xmath378 = 0 . \label{eqm - adjoint}\ ] ] in the present context , it is natural to consider the left - hand side of ( [ eqm - adjoint ] ) to be a linear form on continuous 1-forms . if @xmath145 satisfies ( [ eqm - adjoint ] ) for every continuous 1-form @xmath353 , we say that @xmath145 is a _ generalized solution _ of the equation of motion ( [ eqm - dif - form ] ) . as we have shown in lemma[lemma : support ] , the support of the vorticity @xmath158 of a regular solution is co - moving with the four - dimensional flow @xmath166 ( i.e. frozen into the fluid ) . therefore , a _ generalized solution _ of the equation of motion ( [ eqm - lie - form ] ) , which may be regarded as some limit of regular solutions , must be co - moving . let @xmath379 be an initial ( @xmath251 ) single vortex filament which is a 1-cycle of class @xmath24 bounding a disk in @xmath380 . orbit _ of the single vortex filament is , denoting @xmath381 , @xmath382 or , the family @xmath383 is the motion picture of the loop along @xmath151 , which is a representation of a surface link@xcite . we construct ( using given @xmath151 and @xmath384 ) a pure - state vortex @xmath295 from an initial pure - state @xmath307-filament @xmath385 . in sec.[subsec : pure - state ] , we derived @xmath362 from @xmath295 by operating the projector @xmath386 . here , we construct @xmath295 from @xmath387 by an inverse map of @xmath386 , which , however , is not injective ; we must choose an appropriate inverse that is consistent with the ( adjoint ) equation of motion . [ lemma : orbit ] let @xmath285 be an orbit of a single loop @xmath388 . on @xmath293 @xmath345 , a pure - state @xmath307-filament @xmath389 and an @xmath317-filament @xmath390 are given . we define a pair of 2-forms , in the vicinity of @xmath293 , by @xmath391 where @xmath356 is the 1-form such that @xmath392 . 1 . the @xmath174-plane projections of the singular 2-forms @xmath393 and @xmath394 yield @xmath395 2 . let us put @xmath396 if @xmath397 , then @xmath145 together with @xmath398 satisfy the adjoint equation of motion ( [ eqm - adjoint ] ) . 3 . @xmath399 is a pure - state 2-form . the relations ( [ gamma - orbit - b - projection ] ) and ( [ gamma - orbit - e - projection ] ) follows directly from the definitions ( [ gamma - orbit - b ] ) and ( [ gamma - orbit - e ] ) . evidently , @xmath397 matches the definition ( [ eqm - dif - form-2 ] ) of @xmath398 . we observe @xmath400 hence , @xmath401 , satisfying ( [ eqm - adjoint ] ) for every continuous 1-form @xmath353 . by lemma[lemma : pure - state-1 ] , @xmath295 is a pure state , iff the @xmath317-filament @xmath402 . by ( [ gamma - orbit - b - projection ] ) , @xmath403 , thus , the third statement of this lemma is proven . < 1.5em - 1.5em plus0em minus0.5em height0.75em width0.5em depth0.25em in the definition ( [ gamma - orbit - m ] ) , the total vorticity @xmath145 consists of two parts , @xmath404 and @xmath405 . as we have shown , the first part ( denoted by @xmath295 ) is a pure state ; hence , the total @xmath145 is not a pure state , if the second part ( denoted by @xmath406 ) is non - zero , i.e. @xmath407 . this is the reason why the conventional helicity conservation is broken in a relativistic fluid ( see sec.[subsec : semi - relativistic ] ) . however , the relativistic helicity does conserve ; we have yet to prove this fact for singular vorticities . by lemma [ lemma : orbit ] , we have constructed the vorticity @xmath408 satisfying the adjoint equation of motion ( [ eqm - adjoint ] ) . the support @xmath285 of the vorticity is the orbit of the vortex filament @xmath293 that is transported by the diffeomorphism @xmath166 . hence , different vortex filaments do not intersect in space - time ; given disjoint 1-cycles @xmath409 and @xmath410 , the corresponding surfaces @xmath411 and @xmath412 do not intersect , thus their temporal ( @xmath174 or even @xmath175 ) cross - sections conserve the linking number . to put this fact in the perspective of the helicity conservation law , we need to re - formulate the helicity for the singular vorticity ( see remark[remark : generalize_stokes_formula ] ) . here we consider a pair of disjoint loops ( 1-cycles bounding disks ) @xmath409 and @xmath410 , and their orbits @xmath413 and @xmath414 . the total vorticity is the combination of twin vorticities : @xmath415 where @xmath416 and @xmath417 , respectively , stem from pure - state @xmath307-filaments @xmath418 and @xmath317-filaments @xmath419 ( @xmath420 ) , as constructed in lemma[lemma : orbit ] . let @xmath186 be a co - moving temporally - thin volume , i.e. @xmath421 ( see sec.[subsec : r - helicity ] ) . for an arbitrary continuous 1-form @xmath353 , we obtain , using ( [ gamma - orbit - b - projection ] ) , @xmath422 the final expression is nothing but the _ circulation _ of the 1-form @xmath353 along the cycles @xmath409 and @xmath410 . to use the formula ( [ circulation-0 ] ) in order to evaluate the helicity @xmath423 , we have to insert @xmath155 into @xmath353 , and then , we have to relate @xmath155 with @xmath145 by inverting the defining relation @xmath424 ; let us formally write @xmath425 the operator @xmath426 will be explicitly defined in lemma[lemma : lw - operator ] . here , we remark that the vorticity @xmath145 of ( [ twin - vorticity ] ) consists of @xmath28-measures , thus @xmath155 is not continuous at @xmath427 . this difficulty can be removed by decomposing @xmath155 as @xmath428 and putting the `` self - field helicity '' zero ( see remark[remark : generalize_stokes_formula]-2 ) , i.e. @xmath429 then , the relativistic helicity of the twin vorticity ( [ twin - vorticity ] ) evaluates as @xmath430 let us make the operator @xmath426 explicit . [ lemma : lw - operator ] we denote @xmath431 , and @xmath432 ( dalembertian ) . we invert @xmath433 by the linard - wiechert integral operator , which we denote by @xmath434 . in ( [ lw - formal ] ) , we can define @xmath435 first we transform @xmath436 with a scalar function @xmath437 such that @xmath438 . operating @xmath28 on the both sides of @xmath439 , we obtain @xmath440 . the left - hand side reads @xmath441 , since @xmath442 . we obtain @xmath443 transforming back to @xmath155 , we obtain @xmath444 , thus we may write @xmath445 since @xmath446 for every @xmath158 such that @xmath447 at @xmath229 , we may replace @xmath426 by @xmath448 in ( [ lw - formal ] ) . < 1.5em - 1.5em plus0em minus0.5em height0.75em width0.5em depth0.25em now the following conclusions are readily deducible : [ theorem : singular_vortex ] let @xmath449 be a twin vortex generated by a pair @xmath450 of pure - state @xmath307-filaments @xmath451 , and @xmath317-filaments @xmath452 supported on co - moving loops @xmath453 . 1 . the relativistic helicity @xmath454 is a constant of motion . the constant @xmath455 is the linking number @xmath456 , which may be represented as ( generalizing the gauss integral ) @xmath457 from the foregoing derivation , it is clear that ( [ circulation-2 ] ) is an appropriate expression of the relativistic helicity generalizing ( [ 4dhelicity ] ) . since , @xmath458 and @xmath459 are smooth ( holomorphic ) functions in the vicinities of @xmath460 and @xmath461 , respectively , and satisfies the equation of motion ( [ eqm - lie - form ] ) ( lemma[lemma : orbit ] ) , we can apply lemma[lemma : circulation_law ] ( circulation law ) to prove the constancy of @xmath218 . or , we can calculate directly as @xmath462 where @xmath463 or @xmath464 . for every loop @xmath465 bounding a disk @xmath466 , we observe @xmath467 where @xmath468 is the orbit of @xmath469 . since each @xmath451 is a pure state , the right - hand side yields @xmath470 , iff the loops @xmath453 and @xmath471 link in @xmath180 ( the sign depends the orientations of the loops @xmath472 and @xmath465 ) , i.e. @xmath473 . hence , we conclude that @xmath455 is the linking number of loops contained in an @xmath174-plane @xmath180 . < 1.5em - 1.5em plus0em minus0.5em height0.75em width0.5em depth0.25em [ remark : subtraction_of_baroclinic_term ] from ( [ circulation-4 ] ) , it is evident that a non - exact @xmath317-filament in a baroclinic fluid ( i.e. @xmath474 with a temperature @xmath106 and an entropy @xmath35 ; see sec.[subsec : field - tensor ] ) brings about a change in the helicity ( or the circulation ) . it is also clear that the helicity ( or the circulation ) evaluated only by the pure - state part @xmath475 of the vorticity @xmath476 is conserved even in a baroclinic fluid . conservation of such a reduced helicity ( or a reduced circulation ) has been noticed in nr formulations of fluid mechanics ; see @xcite this work was given its motivation by the finding of non - conservation of helicity ( or circulation ) in a relativistic fluid@xcite . the relativistic distortion of space - time , measured by @xmath477 , yields relativistic baroclinic effect on a thermodynamically barotropic fluid , and violates the conservation of helicity . since the vorticity is dressed by a magnetic field in a high - energy charged fluid , the breaking of the helicity constraint gives rise to a seed ( cosmological ) magnetic field . the aim of this work was set to unearth an alternative , generalized conservation law that dictates a deeper topological constraint beneath the superficial ( or reference - frame dependent ) non - conservation . we have introduced the relativistic ( lorentz invariant ) helicity ( [ 4dhelicity ] ) , which is conserved ( with respect to the proper time ) in a thermodynamically barotropic fluid ( theorem[theorem : helicity_conservation ] ) . considering a pair of pure - state vorticity filaments ( relativistic @xmath307-filaments ) , we have shown that the helicity - conservation law means the constancy of the linking number in the proper - time cross - section of space - time ( theorem[theorem : singular_vortex ] ) . as shown in sec[sec : r - helicity ] , the semi - relativistic helicity @xmath478 ceases to be a constant of motion in a relativistic fluid with @xmath479 . the reason why it can change is not because vortex loops change their link ( the linking number of the @xmath175-cross - sections of @xmath480 and @xmath481 does not change ) , but is because the circulation changes on the loops . in another word , a @xmath307-filament on a @xmath175-plane ( @xmath482 ) is not a pure state ; instead , the pure state is the relativistic @xmath307-filament on an @xmath174-plane ( lemma[lemma : orbit ] ) . interestingly , however , @xmath478 does conserve if @xmath207 ( i.e. in a homentropic fluid ) ; @xmath478 must be , then , a linking number on a @xmath175-plane . to see how @xmath478 conserves , we may replace the co - moving volume @xmath186 in the definition of @xmath218 by @xmath483 , where @xmath484 is the diffeomorphism generated by the reference - frame 4-vector @xmath485 ; see ( [ nr_4-velocity ] ) . then , @xmath218 converts to the semi - relativistic @xmath478 , and theorem[theorem : helicity_conservation ] modifies to conclude @xmath486 ; to prove this , we just replace @xmath174 by @xmath175 and @xmath151 by @xmath487 in the proof of theorem[theorem : helicity_conservation ] as well as in lemma[lemma : circulation_law ] and lemma[lemma : support ] . it is evident that these replacements applies as far as @xmath207 . since the modified lemma[lemma : circulation_law ] shows the conservation of the circulation on every loop @xmath256 on the @xmath175-plane , the constant @xmath478 can be made to measure the linking number of twin vortex filaments . however , we have to apply a different normalization of the @xmath307-filaments on the @xmath175-plane ; we set @xmath488 , instead of @xmath489 , to let @xmath490 be a pure state . by ( [ normalization-2 ] ) , these two different normalizations conflict with each other , because @xmath491 whenever @xmath492 ; hence two constants @xmath478 and @xmath218 have different values . it is needless to say that @xmath490 can remain as a pure state only if @xmath207 . we end this paper with a short summary of helicities in different systems and their comparisons . for the em potential @xmath493 , the helicity density is the 3-form @xmath494 ( @xmath495 the dual of the faraday tensor ) , which may be viewed as a 4-current in the minkowski space - time . the 0-component @xmath496 is the familiar magnetic helicity density @xmath497 . the divergence of the current @xmath498 reads @xmath499 with the standard em fields @xmath317 and @xmath307 . the total `` charge '' @xmath500 , the magnetic helicity , is invariant if @xmath501 ( remember the discussion in sec.[subsec : semi - relativistic ] ) . for example , a _ _ null em field__@xcite , such that @xmath502 , propagates in the vacuum with conserving the helicity ( see @xcite for the examples of knotted em fields in light ) . also in an ideal mhd system , @xmath503 , thus the conventional magnetic helicity conserves ( hence , the link of magnetic field lines is invariant ; cf.@xcite ) . when dressed by the fluid - mechanical momentum , however , the helicity density @xmath504 is no longer divergence - free ( excepting the simplest homentropic fluid ) , and thus , the total charge is not conserved in the relativistic regime ( sec.[subsec : semi - relativistic ] ) . in the non - relativistic limit ( @xmath206 ) , however , the heat term @xmath505 of a barotropic fluid may be absorbed by the enthalpy term to modify the helicity density to be divergence - free ( remark[remark : nr - barotropic ] ) ; by subtracting @xmath506 from the spacial part @xmath507 , we define a modified helicity density @xmath508 because the right - hand side of ( [ e : div k ] ) may be written as @xmath509 , we obtain @xmath510 . the closed 3-form @xmath511 is a _ noether current _ pertinent to the _ relabeling symmetry _ of the action @xcite . the 0-component @xmath512 is the noether charge , and its spatial ( @xmath175-plane ) integral is the conventional ( non - relativistic ) helicity @xmath21 . the relativistic effect ( @xmath513 ) , however , makes the right - hand side of ( [ e : div k ] ) non - exact , thus we can not introduce such a modified divergence - free helicity density . yet , the relativistic helicity is made invariant by integrating @xmath210 on a co - moving domain , because @xmath514 is an exact 3-form . whereas @xmath210 is not divergence - free ( because of the relativistic length contraction ) , we find that the lagrangian representation of the helicity density is divergence - free . the lagrangian coordinates labeling the position of each fluid element have the redundancy ( symmetry ) of relabeling , and the consequent noether current turns out to be the present relativistic helicity density after transforming into the eulerian coordinates ; detailed analysis of the symmetry of the relativistic lagrangian will be published elsewhere . we acknowledge a debt to the isaac newton institute for mathematical sciences , university of cambridge ; this work was given a chance to start during the workshop _ topological fluid dynamics_. we are grateful to professor k. moffatt and professor y. mitsumatus for their suggestions and discussions . the work of zy was partially supported by grant - in - aid for scientific research from the japanese ministry of education , science and culture no . the work of yk was supported by grant - in - aid for jsps fellows 241010 . the work of ty was partially supported by the jst crest program at department of mathematics , hokkaido university . loops do not link in four - dimensional space . in general , we have the following theorem which is readily deducible by the thom transversality : let @xmath515 and @xmath516 be , respectively , @xmath517 and @xmath518-dimensional submanifolds embedded in @xmath519 . if @xmath520 , then @xmath516 is contractible in @xmath521 and @xmath515 is contractible in @xmath522 . see m. w. hirsch , _ differential topology _ , graduate texts in mathematics no . 33 ( springer - verlag , new york - heidelberg , 1976 ) . b. iversen , _ cohomology of sheaves _ , universitext ( springer - verlag , berlin , 1986 ) . s. kamada , _ braid and knot theory in dimension four _ , mathematical surveys and monographs 95 ( american mathematical society , providence , 2002 ) . c. eckart , _ hydrodynamics of oceans and atmospheres _ , ( pergamon press , london , 1960 ) .

a new method of polarimetric calibration is presented in which the instrumental response is derived from regular observations of psrj0437 based on the assumption that the mean polarized emission from this millisecond pulsar remains constant over time . the technique is applicable to any experiment in which high - fidelity polarimetry is required over long time scales ; it is demonstrated by calibrating 7.2 years of high - precision timing observations of psrj1022 made at the parkes observatory . application of the new technique followed by arrival time estimation using matrix template matching yields post - fit residuals with an uncertainty - weighted standard deviation of 880 ns , two times smaller than that of arrival time residuals obtained via conventional methods of calibration and arrival time estimation . the precision achieved by this experiment yields the first significant measurements of the secular variation of the projected semi - major axis , the precession of periastron , and the shapiro delay ; it also places psrj1022 among the ten best pulsars regularly observed as part of the parkes pulsar timing array ( ppta ) project . it is shown that the timing accuracy of a large fraction of the pulsars in the ppta is currently limited by the systematic timing error due to instrumental polarization artifacts . more importantly , long - term variations of systematic error are correlated between different pulsars , which adversely affects the primary objectives of any pulsar timing array experiment . these limitations may be overcome by adopting the techniques presented in this work , which relax the demand for instrumental polarization purity and thereby have the potential to reduce the development cost of next - generation telescopes such as the square kilometre array .

1212.3446

high - precision pulsar timing exploits the exceptional rotational stability of millisecond pulsars to measure orbital dynamics in binary systems and perform unique tests of general relativity in the strong field regime ( e.g. * ? ? ? the basic measurement in timing analyses is the pulse time - of - arrival ( toa ) , the epoch at which a fiducial phase of the pulsar s periodic signal is received at the observatory . the difference between the measured toa and the arrival time predicted by a best - fit model is known as the timing residual . timing residuals with a standard deviation of the order of 100ns have been achieved for a growing number of millisecond pulsars ( e.g. * ? ? ? * ; * ? ? ? * ) , which has renewed both theoretical and experimental interest in the detection of gravitational radiation using a pulsar timing array ( pta ; e.g. * ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the sensitivity of a pta and the confidence limits that may be placed on any gravitational wave detection directly depend upon the precision and accuracy with which pulse arrival times can be estimated . therefore , an important part of the ambitious pta effort is the quantification and reduction of systematic error through the development of improved methods of arrival time estimation ( e.g. , * ? ? ? * ; * ? ? ? * ) , instrumental calibration @xcite , radio - frequency interference ( rfi ) mitigation @xcite , compensation for the effects of propagation through the interstellar medium ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , statistical analysis of the stochastic nature of the pulsar signal @xcite , and the characterization of timing noise @xcite . at the parkes observatory , the majority of pulsar timing data are obtained from observations of psrj0437@xmath04715 , the closest and brightest millisecond pulsar known . discovered in the parkes 70-cm survey @xcite , psrj0437@xmath04715 has a spin period of @xmath2 ms ; at 20 cm , it has a pulse width at half maximum of about 130 @xmath3s @xcite and an average flux of 140 mjy @xcite . owing to its sharp peak and large flux density , it is an excellent target for high - precision pulsar timing studies . it is also fortuitously located away from the galactic plane ( @xmath4 , @xmath5 ) , which permits long observing sessions during periods of local sidereal time when competition for the telescope is relatively low . early timing of this pulsar highlighted the need for more accurate polarimetric calibration @xcite . the systematic timing error due to instrumental polarization artifacts is readily observed in the dramatic variation of arrival time residuals as a function of parallactic angle ( see figure [ fig : uncalibrated ] ) . this issue motivated the discovery and development of the polarimetric invariant profile @xcite . arrival times derived from the invariant profile proved to be significantly more accurate than those of the total intensity profile , a breakthrough that led to the first detection of annual - orbital parallax and a new test of general relativity @xcite . despite its demonstrated advantages over the total intensity profile , there remain a number of drawbacks to the use of the invariant profile . first , the signal - to - noise ratio @xmath6 of the invariant profile is lower than that of the total intensity , reaching zero for completely polarized sources . second , the noise in the invariant profile is neither normally distributed nor homoscedastic with respect to pulse longitude ( see appendix [ app : invariant ] ) ; these two properties are assumed , either explicitly or implicitly , by most template matching algorithms commonly used for pulsar timing . finally , the computation of the invariant profile requires estimation and correction of the bias due to noise , and the relative precision of the bias estimate is inversely proportional to the square root of the number of discrete samples within off - pulse regions of pulse longitude . for a typical observation of psrj0437@xmath04715 , the relative error in the bias correction is of the order of 8% , an unacceptably large uncertainty when aiming for arrival time uncertainty that is four orders of magnitude smaller than the pulse width . these concerns motivated the development of matrix template matching ( mtm ; * ? ? ? * ) , a technique that exploits the additional timing information in the polarized component of the pulsar signal while simultaneously eliminating residual polarimetric calibration errors . for some pulsars , the mean polarization profile exhibits sharper features than observed in the total intensity ; these features generate more power at higher harmonics of the pulsar spin frequency , yielding greater arrival time precision than might be expected from the polarized flux alone ( e.g. see figure 1 of * ? ? ? * ) . furthermore , mtm improves arrival time accuracy by mitigating systematic timing errors due to instrumental polarization artifacts . although the development of mtm was primarily driven by the high - precision timing of psrj0437@xmath04715 , @xcite predicted that psrj1022@xmath11001 would benefit most from this technique , including a @xmath7 increase in timing precision and a significant decrease in systematic timing error . psrj1022@xmath11001 was contemporaneously discovered in both the princeton - arecibo declination - strip survey @xcite and the green bank northern sky survey @xcite . @xcite observed variations in the shape of its mean pulse profile on a timescale of the order of minutes and noted that the accuracy of arrival time estimates was limited by this effect . mean profile variations over such timescales would be distinct from pulse - to - pulse shape variations such as giant pulses , microstructure , and drifting sub - pulses ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , which when averaged over many pulses can be described as phase jitter @xcite or stochastic wide - band impulse - modulated self - noise ( swims ; * ? ? ? as it is typically assumed that the mean pulse profile does not change with time , detection of systematic profile variations in a millisecond pulsar would have a serious impact on the field of high - precision timing . accordingly , the stability of the psrj1022@xmath11001 pulse profile has been the subject of a number of detailed studies @xcite . these are reviewed in [ sec : systematic ] , following a mathematical treatment of the systematic timing error due to instrumental distortion of the total intensity profile . [ sec : observation ] describes the observing system and the novel procedures used to calibrate the psrj1022@xmath11001 data for this study , including a new scattered power correction algorithm that is described in more detail in appendix [ app : spc ] . as part of this analysis , a 7.2-year history of the polarimetric response at parkes is presented and its impact on systematic timing error is discussed . in [ sec : discussion ] , the results of this study are compared and contrasted with previous work , the relevance of the calibration method to other pulsars is discussed , and potential future directions are considered . in the following discussion , the polarization of electromagnetic radiation is described by the second - order statistics of the transverse electric field vector @xmath8 , as represented by the four stokes parameters @xmath9 the angular brackets denote an ensemble average ; @xmath10 is the conjugate transpose of @xmath8 ; @xmath11 is the @xmath12 identity matrix ; @xmath13 , @xmath14 , and @xmath15 are the pauli spin matrices ; @xmath16 is the total intensity , and @xmath17 is the polarization vector ( e.g. * ? ? ? the reception and propagation of polarized radiation is described by linear transformations of @xmath8 , as represented using complex @xmath12 jones matrices . any non - singular matrix can be decomposed into the product of a unitary matrix and a self - adjoint matrix . unitary transformations result in a three - dimensional euclidean rotation of the stokes polarization vector , leaving the total intensity unchanged and preserving the degree of polarization . self - adjoint jones matrices effect a lorentz boost of the stokes 4-vector , mixing total intensity and polarized flux @xcite . the boost component , also known as the _ poldistortion _ @xcite , of the instrumental response distorts the shape of the total intensity pulse profile and introduces systematic arrival time errors . physically , instrumental boost transformations are caused by differential amplification of the signals from the two orthogonally - polarized receptors and by cross - coupling between these receptors . in @xcite , the systematic timing error induced by a given level of polarization distortion is predicted for a number of millisecond pulsars using a numerical simulation in which the template profile for each pulsar is copied and subjected to a boost transformation . the orientation of the boost axis in poincar space is varied , and the minimum and maximum induced phase shifts between the distorted total intensity profile and the template profile are recorded . it is also possible to predict the systematic timing error due to polarization calibration errors by computing the rate at which the best estimate of the phase shift varies with respect to the instrumental boost transformation parameters . beginning with equation ( 12 ) of @xcite , the best estimate of the phase shift @xmath18 derived using only the total intensity is given by @xmath19 here , @xmath20 is the cross - spectral power of the observed total intensity @xmath21 and the template total intensity @xmath16 in the fourier domain , @xmath22 is the argument of @xmath23 , and the summations in the numerator and denominator are performed over @xmath24 to @xmath25 , where @xmath26 is the maximum harmonic at which the fluctuation power spectrum exhibits significant power . now consider an observed total intensity profile that is a copy of the template subjected to an instrumental boost transformation , as parameterized in section 4.1 of @xcite such that @xmath27 . referring to equation ( 10 ) of @xcite , the transformed total intensity is given by @xmath28 where @xmath29 . the partial derivatives of @xmath30 with respect to the boost parameters @xmath31 are used to compute the partial derivatives of the phase shift with respect to these parameters when the boost parameter @xmath32 is zero , @xmath33 \nu_m } \over { \pi \sum |s_0(\nu_m)|^2 \nu_m^2 } } . \label{eqn : susceptibility}\ ] ] the above expression defines the components of a three - dimensional gradient @xmath34 . the magnitude of the gradient @xmath35 provides a measure of the susceptibility of arrival time estimates from a given pulsar to instrumental distortion . to aid in the physical interpretation of the susceptibility @xmath36 , the first order approximations to equations ( 25 ) and ( 32 ) of @xcite are used to define the systematic timing error induced by either a 1% differential gain error or receptor non - orthogonality of 0.01 radians ( @xmath37 0.6 ) , @xmath38 where @xmath39 is the pulsar spin period . values of @xmath40 for each of the 20 millisecond pulsars that are regularly observed as part of the parkes pulsar timing array ( ppta ; * ? ? ? * ) project are reported in table [ tab : theory ] . llll pulsar & @xmath41 & @xmath42 & @xmath40 ( ns ) + j0437@xmath04715 & 0.85 & 1.43 & 207 + j0613@xmath00200 & 0.92 & 1.46 & 59 + j0711@xmath06830 & 0.88 & 1.54 & 81 + j1022 + 1001 & 0.68 & 1.65 & 282 + j1024@xmath00719 & 0.74 & 2.11 & 34 + j1045@xmath04509 & 0.88 & 1.48 & 338 + j1600@xmath03053 & 0.90 & 1.39 & 115 + j1603@xmath07202 & 0.85 & 1.55 & 142 + j1643@xmath01224 & 0.91 & 1.40 & 266 + j1713 + 0747 & 0.85 & 1.58 & 6 + j1730@xmath02304 & 0.71 & 1.70 & 198 + j1732@xmath05049 & 0.96 & 1.38 & 185 + j1744@xmath01134 & 1.56 & 6.43 & 105 + j1824@xmath02452a & 0.88 & 2.56 & 18 + j1857 + 0943 & 0.89 & 1.43 & 124 + j1909@xmath03744 & 1.02 & 1.51 & 22 + j1939 + 2134 & 0.95 & 1.49 & 44 + j2124@xmath03358 & 0.85 & 1.45 & 127 + j2129@xmath05721 & 1.15 & 1.61 & 211 + j2145@xmath00750 & 0.95 & 1.44 & 147 + [ tab : theory ] the quantities in this table are derived from the high @xmath6 polarization profiles presented by @xcite . for the majority of pulsars in the ppta , a modest degree of instrumental distortion induces systematic timing error of the order of 100 ns , which is sufficient to seriously impede progress toward the detection of the stochastic gravitational wave background @xcite . to first order , the systematic timing error due to instrumental distortion is given by the inner product , @xmath43 . in the polar coordinate system best suited to describing linearly polarized receptors , @xmath44 , where @xmath45 parameterizes the differential receptor gain , and @xmath46 and @xmath47 describe the non - orthogonality of the receptors ( see section 3.3 of * ? ? ? * ) . these properties of the instrument may vary as a function of time for a variety of reasons . diurnal variations are introduced by the parallactic rotation of the receiver with respect to the sky , as shown in figure [ fig : uncalibrated ] . furthermore , step changes in differential gain are introduced when the levels of attenuation applied to each of the two orthogonal polarizations are reset at the start of each observation and/or periodically updated during the observation . in contrast , the cross - coupling of the receptors is typically assumed to remain relatively stable over timescales of the order of months . of particular interest to a long - term timing experiment such as a pulsar timing array are any step changes in @xmath48 due to modifications of the instrumentation and/or slow variations in @xmath48 due to gradual degradation of one or more system components . the temporal variations in distortion - induced error due to the long - term evolution of a given instrument will be correlated between the different pulsars observed using that system . in an array of @xmath26 pulsars , the spectral structure of the correlated systematic error due to instrumental distortion is described by @xmath49 where @xmath50 is the @xmath51 matrix of the auto- and cross - power spectra of the timing residuals ( e.g. * ? ? ? * ) , @xmath52 is the @xmath53 jacobian matrix , @xmath54 ( @xmath55 is the spin period of the @xmath56th pulsar ) and @xmath57 is the @xmath58 matrix of the auto- and cross - power spectra of the instrumental distortion parameters ( the components of @xmath48 ) . in practice , an optimally - weighted sum of the cross - power spectral harmonics is used to define a detection statistic with maximum sensitivity to a stochastic background of low - frequency gravitational waves ( e.g. * ? ? ? * ; * ? ? ? given only the mean polarization profile of each pulsar in the timing array , it is possible to predict the impact of correlated systematic timing error due to instrumental distortion on such a detection statistic . first , only the dominant harmonic of the cross - power spectrum of two pulsars with distortion gradients @xmath59 and @xmath60 is considered . next , it is assumed that the instrumental distortion parameters are uncorrelated and homoscedastic , such that @xmath61 , where * i * is the @xmath58 identity matrix . following these simplifications , a first - order approximation of the coefficient of correlated systematic timing error is given by @xmath62 this correlation coefficient depends only on the shape of the mean polarization profile and has no modal structure on the celestial sphere . that is , the correlated timing error due to instrumental distortion will corrupt all of the moments in a multipole expansion of timing delay , which will impact on all of the primary goals of any pulsar timing array experiment @xcite , including the long - term measurement of time ( monopole moment ; * ? ? ? * ) , corrections to the solar system ephemeris ( dipole moment ; * ? ? ? * ) , and detection of the gravitational wave background ( quadrupole moment ; * ? ? ? * ) . to illustrate the potential importance of correlated systematic timing error , the high @xmath6 polarization profiles presented by @xcite are used to compute the cross - correlation coefficients for each of the 15 pulsar pairs presented in figure 4 of @xcite . the cross - spectral power measurements of @xcite are compared with the predicted values of @xmath63 in figure [ fig : ych+11 ] ; the coefficient of correlation between predicted and measured values is @xmath64 . the agreement between predicted and experimental measures of correlation indicates that the systematic error due to instrumental distortion is a plausible contributor to the apparent anti - correlation between the results of @xcite and the expected quadrupolar signature predicted by @xcite . also shown in table [ tab : theory ] are the relative arrival time uncertainties @xmath41 and @xmath42 defined in @xcite . the former is the ratio between the theoretical error in arrival time estimates derived using matrix template matching ( mtm ) and the error in estimates derived using only the total intensity ; the latter is the ratio between the uncertainties in arrival times derived from the invariant interval and from the total intensity . the statistical uncertainty in the invariant profile arrival times is underestimated because the effects of heteroscedasticity and bias correction error are not considered . regardless , for every pulsar in the ppta , precision is lost through use of the invariant interval . the increase in uncertainty is most dramatic for psr j1744@xmath01134 , which emits almost 100% linearly polarized radiation . for this pulsar , mtm is also predicted to yield arrival times with greater error than those derived from the total intensity owing to the large coefficient of multiple correlation ( @xmath65 ) between the phase shift used to compute the toa and the model parameters that describe the polarization transformation . table [ tab : theory ] also predicts that matrix template matching will yield the greatest relative improvement in arrival time precision for psrj1022@xmath11001 . this pulsar also has the second - highest value for predicted systematic timing error owing to the susceptibility of the total intensity profile to polarization distortion . this may explain the profile shape variations and excess arrival time error first noted by @xcite . in a detailed study of the spectrum and polarization of these variations using the effelsberg 100-m radio telescope , @xcite assert that the observed fluctuations in pulse shape can not be explained by instrumental effects . however , the data used in this analysis were calibrated using the `` ideal feed '' assumption that there is no cross - coupling between receptors and that the reference source used for gain calibration produces an equal and in - phase response in each receptor . given that receptor cross - coupling of the order of 10% to 20% is commonly observed when more accurate calibration is performed ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , an instrumental origin of the variability deserves consideration . for instance , the temporal variation of the total intensity profile observed by @xcite is reproducible using a simple model in which the polarized signal is rotated by the parallactic angle of the receiver and then subjected to an instrumental boost that mixes @xmath3710% of the linearly - polarized flux with the total intensity . the similarity between the simulated observations plotted in figure [ fig : kxc+99 ] and the total intensity profiles presented in figure 2 of @xcite supports the argument that the variations are the result of uncalibrated instrumental distortion . similarly , the small frequency - scale variations observed by @xcite may be readily explained by spectral variation of the instrumental response ( e.g. figure 3 of * ? ? ? the observations used to demonstrate the narrow - band variations and presented in figure 8 of @xcite were recorded using the effelsberg - berkeley - pulsar - processor with a bandwidth of 56 mhz . in this mode , only the fluxes in each of the circularly polarized signals ( i.e. two out of four stokes parameters ) are retained ; consequently , only the mixing between circularly - polarized flux and total intensity can be calibrated in these data . therefore , frequency - dependent mixing between linearly - polarized flux and total intensity is a plausible explanation for the observed profile shape variations . @xcite analyzed the variability of psrj1022@xmath11001 using multi - frequency observations made at the westerbork synthesis radio telescope ( wsrt ) . they argue that , because the wsrt is composed of equatorially - mounted antennas , it is possible to rule out pulse shape variations due to instrumental distortion combined with parallactic rotation of the feed . however , the polarimetric response of a phased array can vary in both time and frequency for a wide variety of reasons , including imperfect phase calibration and ( as for any single - dish antenna ) instrumental gain variations . indeed , the automatic gain controllers used to prevent instrumental saturation @xcite also make it impossible to accurately calibrate the polarimetric response of the wsrt phased array @xcite . the high degree of susceptibility of the psrj1022@xmath11001 pulse profile to polarization distortion and the inadequate instrumental calibration employed in these previous works cast doubt on the conclusion that the profile variations are intrinsic to the pulsar . furthermore , @xcite find no evidence of significant profile shape variations in psrj1022@xmath11001 observations made at the parkes observatory . this study achieved arrival time residuals with a standard deviation of the order of 2.3 @xmath3s and @xmath66 per degree of freedom @xmath67 , whereas @xcite obtained a reduced @xmath66 close to unity only after systematic errors of the order of 10 @xmath3s were added in quadrature to the formal uncertainties . the analysis by @xcite is based on observations integrated over 5 min and 64 mhz and spanning 1.3 years . increasing the integration length to 60 min and the time span to 2.3 years yielded arrival time residuals with a standard deviation of the order of 1.5 @xmath3s and reduced @xmath68 @xcite . with the same instrumentation and similar integration lengths , @xcite achieved 1.6 @xmath3s residuals in data spanning 5.1 years . in the absence of systematic error and other sources of noise ( e.g. dispersion variations ) , the one - hour integrations used in these two studies should have yielded timing residuals of the order of 660 ns . this is of the same order as the systematic timing error introduced by a differential gain error of @xmath69 or by receptor non - orthogonality of @xmath70 ( cf . table [ tab : theory ] ) . however , the arrival times used in these studies were derived from either uncalibrated @xcite or inaccurately calibrated @xcite total intensity pulse profiles , which are highly susceptible to instrumental distortion . therefore , it is reasonable to anticipate that high - fidelity polarimetric calibration will improve the accuracy of psrj1022@xmath11001 arrival time estimates . this conjecture is verified in the following section , which describes a new method of deriving the instrumental response by using psrj0437@xmath04715 as a polarized reference source . the method is used to calibrate high - precision timing observations of psrj1022@xmath11001 and demonstrated to reduce the systematic error due to instrumental polarization distortion . dual - polarization observations of psrj0437@xmath04715 and psrj1022@xmath11001 were made at the parkes observatory using the h - oh receiver and the center element of the parkes 21-cm multibeam receiver @xcite . two 64-mhz bands , centered at 1341 and 1405mhz , were digitized using two bits per sample and processed using the second generation of the caltech - parkes - swinburne recorder ( cpsr2 ; * ? ? ? * ; * ? ? ? * ) . to maintain optimal linear response during digitization , the detected power was monitored and the sampling thresholds were updated approximately every 30 seconds . the baseband data were reduced using dspsr @xcite , which corrects quantization distortions to the voltage waveform using the dynamic level setting technique ( * ? ? ? * hereafter ja98 ) , then performs phase - coherent dispersion removal @xcite while synthesizing a 128-channel filterbank @xcite . the stokes parameters are then formed and integrated as a function of topocentric pulse phase , producing uncalibrated polarization profiles with 1024 discrete pulse longitudes . during analog - to - digital conversion , the radio signal is subjected to non - linear distortions that significantly alter the observed pulse profile ( ja98 ; * ? ? ? ja98 introduce a dynamic output level setting technique that is employed by dspsr to correct the underestimation of undigitized power . to implement this technique , the digitized data are divided into consecutive segments of @xmath71 samples and , for each segment , the number of low - voltage states @xmath72 is counted . a histogram of occurrences of @xmath72 is archived with the pulsar data and used offline to evaluate the quality of the digital recording . when the voltage input to the digitizer is normally distributed , the ratio @xmath73 has a binomial distribution as in equation ( a6 ) of ja98 . the difference between this theoretical expectation and the recorded histogram of @xmath72 provides a measure of the degree to which either the input signal is not well described by a normal distribution or the sampling thresholds diverge from optimality ( e.g. at the start of an observation ) . this difference , called the _ two - bit distortion _ , is given by @xmath74 ^ 2 \label{eqn : d2bit}\ ] ] where @xmath75 is the expected binomial distribution and @xmath76 is the recorded distribution of @xmath72 . separate histograms of @xmath72 are maintained for each polarisation , and the reported distortion is simply the sum of the distortion in each polarisation . for each 16.8-second integration output by cpsr2 , the two - bit distortion @xmath77 was computed and all data with @xmath78 were discarded . this threshold was chosen using the histograms plotted in figure [ fig : d2bit ] ; beyond the range of plotted values , a sparse tail of outliers extends to a maximum value of @xmath79 . approximately 3.3% of the observations were rejected ; the remaining data were corrected for scattered power using the method described in appendix [ app : spc ] and implemented in the psrchive software package for pulsar data archival and analysis @xcite . the corrected data were then summed until the integration length was of the order of five minutes . to use a pulsar as a polarized reference source requires a high @xmath6 , well - calibrated , template polarization profile . from over 700 hours of psrj0437@xmath04715 observations were selected only those sessions with low levels of two - bit distortion and of sufficient duration to achieve a precise estimate of the instrumental response . about 50 sessions , each @xmath37 8 hours in duration and with corresponding pulsed noise diode observations , were calibrated using measurement equation modeling ( mem ; * ? ? ? * ) . the observed stokes parameters were normalized to account for pulsar flux variations as described in appendix [ app : invariant ] . as in @xcite , the two degenerate model parameters ( a boost along the stokes @xmath80 axis and a rotation about this axis ) were constrained by assuming that observations of 3c 218 ( hydra a ) have negligible circular polarization and that the orientation of the receiver is known . although these assumptions are not necessarily valid , absolute certainty in the stokes parameters is not required for the purposes of a high - precision timing experiment . it is sufficient that the observed stokes parameters can be mapped to the template profile by a single jones transformation . this prerequisite may be evaluated using the objective merit function of the mtm least - squares fit ; the reduced @xmath66 statistic will be greater than unity whenever the transformation between the template and the observed pulse profile is not well - described by a single jones matrix . various phenomena may contribute to a poor model fit . for example , if the mean polarized emission from the pulsar evolves over time , then the difference between the template and the observed profile will vary as a function of pulse longitude in a manner that can not be adequately modeled by a single jones transformation . furthermore , as jones matrices describe only linear transformations of the electric field , matrix template matching will fail to model any non - linear component of the instrument . analog - to - digital conversion using only two bits per sample is an intrinsically non - linear process ; therefore , variations in the response of the digitizer adversely affect the mtm goodness - of - fit , as shown in figure [ fig : chisq_d2bit ] . here , 5 of the @xmath81 mem - calibrated psrj0437@xmath04715 observations were integrated to produce a template profile and the mean polarization profiles from each of the original @xmath81 observations were fit to this template using mtm . the reduced @xmath66 of each mtm fit ( averaged over all frequency channels ) is highly correlated with the median value of the two - bit distortion @xmath77 in each session ; i.e. greater distortion reduces the merit of the mtm fits . for the 5 observations used to create the template profile , @xmath82 because covariance between the observation and the template is not included in the definition of the mtm objective merit function . figure [ fig : chisq_d2bit ] justifies the experimental design decision to regularly update the two - bit sampling thresholds during each observation . independent sampling thresholds were applied to each of the two orthogonal polarizations ; consequently , the astronomical signal was subjected to an unknown differential gain that varies with time . such differential gain variations can be modeled using mtm , whereas the non - linear response of the digitizer can not . therefore , when recording the baseband signal with a two - bit digitizer , it is more important to maintain optimal linear response than it is to maintain a constant flux scale . out of the @xmath83 sessions that were calibrated using mem , 16 were selected with low median two - bit distortion and mtm reduced @xmath66 within 4% of unity . the mtm - corrected mean profiles from these 16 sessions were integrated to form a polarimetric template profile with a total integration length of @xmath37100hours , shown in figure [ fig:0437 ] . the over - polarization seen from pulse phase 0.88 to 0.94 in this figure is an instrumental artifact that is currently not understood . it may be that the unpolarized quantization noise is overestimated by the two - bit scattered power correction algorithm described in appendix [ app : spc ] . however , a very similar artifact is observed in polarization data obtained with the second - generation of the parkes digital filter bank ( pdfb2 ; * ? ? ? * ) , which employs an 8-bit digitizer . these facts might point to the existence of a non - linear component in the signal path shared by cpsr2 and pdfb2 . however , the third and fourth generations of the pdfb also regularly exhibit @xmath84% over - polarization when it is not observed using the latest generation of baseband recording instrumentation at parkes . therefore , it may be only coincidence that a similar artifact is observed in both pdfb2 and cpsr2 data . matrix template matching has been previously utilized to calibrate and derive arrival time estimates from observations of psrj0437@xmath04715 ( e.g. * ? ? ? however , this method can not be applied directly to psrj1022@xmath11001 because the mean flux of this pulsar at 20 cm ( @xmath376 mjy ; * ? ? ? * ) is much lower and the precision of calibrator solutions derived from similar integration lengths is inadequate . it is not possible to integrate over longer intervals because the instrumental response varies with time , primarily due to the parallactic rotation of the receiver . therefore , to calibrate the psrj1022@xmath11001 data , a new technique was developed that combines mtm with mem . the new method , called measured equation template matching ( metm ) , matches multiple pulsar observations to the template profile , thereby increasing the precision of the solution . as with mem , variations of the instrumental response ( e.g. the parallactic rotation of the receiver ) are included in the model . metm also incorporates measurements of an artificial reference source ( e.g. a noise diode coupled to the receptors ) , enabling the backend component of the calibrator solution to be later updated as in @xcite so that the solution may be applied to other astronomical sources . in contrast with mem , the metm method does not require any additional observations of a source of known circular polarization to constrain the boost along the stokes @xmath80 axis because the template profile provides this information . using the new metm technique , a detailed history of the instrumental response at parkes was derived from the psrj0437@xmath04715 data and used to calibrate the psrj1022@xmath11001 data . first , the high @xmath6 template profile ( created as described in the previous section ) and the new metm method were used to derive a calibrator solution from every observation of psrj0437@xmath04715 with more than 3.5 hours of integration . a total of @xmath85 solutions were produced , spanning 2003 april to 2010 june . the instrumental response was modeled using equation ( 19 ) of @xcite , which includes three independent lorentz boost transformations described by the differential gain and the non - orthogonality parameters , @xmath46 and @xmath47 . the derived non - orthogonality parameter estimates are plotted in figure [ fig : coupling ] . in addition to smooth variations in @xmath47 and @xmath46 , there are a number of distinct steps that correspond to physical changes in the 21-cm multibeam receiver and downconversion system at parkes . the gaps in the data correspond to maintenance periods during which the multibeam receiver was unavailable ( 2003 november 9 to 2004 september 7 , and 2006 december 16 to 2007 may 17 ) . immediately following the first maintenance period , the receptor cross - coupling remains stable for a period of approximately one year . the distinct changes in state that occur around mjd 53660 ( 2005 october 17 ) and mjd 53800 ( 2006 march 6 ) are due to changes in the downconversion system ( the signal paths for the two cpsr2 bands were swapped ) . following the second maintenance period , the cross - coupling parameters are notably less stable in both time and frequency . figure [ fig : boost_close ] provides a closer look at the variations of these parameters over a period of one week and a bandwidth of 20 mhz . in addition to significant quasi - periodic variations in these parameters as a function of radio frequency , the spectral structure appears to drift as a function of time . for example , from 11 to 12 february 2010 , the main features in the spectrum have shifted toward lower frequencies by @xmath86 mhz . after mjd @xmath37 54300 , it is invalid to assume that the receptor cross - coupling remains stable for any period of time longer than a day . figure [ fig : coupling ] plots the non - orthogonality parameters in units of centiradians ; these values can be multiplied by the estimates of @xmath40 listed in table [ tab : theory ] to approximate the variation in systematic arrival time error introduced by receptor cross - coupling . for example , for psrj1022@xmath11001 , the peak - to - peak variations of @xmath87 radians in both @xmath47 and @xmath46 are predicted to induce systematic timing errors of the order of 1.7@xmath3s . however , after integrating over all frequency channels , the net effect of the cross - coupling variations will be diminished . furthermore , receptor cross - coupling describes only two of the three forms of instrumental distortion . in addition to the presumption of zero receptor cross - coupling , calibration techniques based on the ideal feed assumption typically also incorporate the false premise that the artificial reference source illuminates both receptors identically . under this assumption , any intrinsic imbalance in the induced amplitude of the reference source signal in each receptor will be misinterpreted as differential gain . in contrast , the metm technique includes direct estimation of the stokes parameters of the reference source ( as in fig . 2 of * ? ? ? * ) . using these estimates , the systematic error in differential gain @xmath88 due to unbalanced reference source amplitudes may be derived . given estimates of @xmath88 , @xmath47 and @xmath46 as a function of frequency , it is possible to directly compute the systematic arrival time error at each epoch by applying the instrumental distortion transformation to a copy of the template profile and estimating the induced phase shift between the template and its distorted copy . application of the metm - derived instrumental distortion parameters to the psrj1022@xmath11001 template profile yields the systematic timing error shown in figure [ fig : systematic_error ] . the instrumental distortion transformation is applied independently in each frequency channel before integrating over frequency . the induced phase shift is computed using only the total intensity profile and the fourier - domain template matching algorithm described by @xcite . during the periods of maintenance on the parkes 21-cm multibeam receiver , the solutions derived from observations made with the receiver have been used . uncalibrated instrumental distortion introduces peak - to - peak variations in systematic timing error of the order of 600ns ; the largest jumps in timing error occur when switching between receivers . these systematic errors are present in arrival times derived from either uncalibrated or inaccurately - calibrated psrj1022@xmath11001 observations . as shown in the following section , these errors are significantly reduced after calibration with the metm - derived solutions . the solutions produced using the new metm method were applied to calibrate the 5-minute psrj1022@xmath11001 integrations as in @xcite . the calibrated data were then integrated in frequency and time to form approximately 64-minute integrations . the observations with the greatest @xmath6 were added to form a full - polarization template profile with an integration length of @xmath89 hours , shown in figure [ fig:1022 ] . to enable a quantitative comparison between the analysis described in this paper and previous work , the psrj1022@xmath11001 observations were also calibrated using the ideal feed assumption employed by @xcite . arrival times for the two datasets were then derived using only the total intensity profile @xcite and matrix template matching @xcite . these variations produced four sets of arrival time estimates , calibrated using either full measurement equation template matching ( metm ) or the ideal feed assumption ( ifa ) and timed using either matrix template matching ( mtm ) or standard total intensity ( sti ) methods . the post - fit arrival time residuals for these four sets are compared in table [ tab : residual ] . ccc method & @xmath90 ( @xmath3s ) & @xmath91 + ifa sti & 1.9 & 2.2 + metm sti & 1.4 & 2.1 + ifa mtm & 0.92 & 1.1 + metm mtm & 0.88 & 1.0 + [ tab : residual ] after calibration based on the ideal feed assumption ( ifa ) and arrival time estimation with the standard method of template matching using only the total intensity profile ( sti ) , the residuals are of similar quality to those presented by @xcite . that is , compared to uncalibrated data , the ifa - calibrated data are equally adversely affected by distortions to the total intensity profile . this is not surprising , given that the ifa does not apply to the 21-cm multibeam receiver at parkes @xcite . calibration via measurement equation template matching ( metm ) reduces both the standard deviation of the arrival time residuals and the reduced @xmath66 of the model fit . assuming that the systematic error removed by metm is uncorrelated with the remaining noise in the metm sti residuals , instrumental distortion of the ifa - calibrated total intensity profile gives rise to timing errors with a standard deviation of approximately 1.3 @xmath3s . for both metm- and ifa - calibrated data sets , matrix template matching ( mtm ) yields arrival times of similar quality . this indicates that mtm is able to correct the remaining instrumental calibration errors in the ifa data and produce arrival time estimates with reduced systematic error . it is not possible to test mtm on uncalibrated data because the variation of instrumental response with frequency leads to bandwidth depolarization , which can not be modeled using a single jones matrix . that is , although inaccurate , calibration based on the ifa may be sufficient to avoid depolarization . focusing on only the metm - calibrated data , comparison of arrival times generated using sti and mtm demonstrates that mtm reduces the standard deviation of arrival time residuals by approximately 35% , which is consistent with the relative uncertainty predicted in table [ tab : theory ] and by @xcite . when compared to the results of the typical data analysis employed in most high - precision timing experiments to date ( ifa sti ) , matrix template matching reduces arrival time residuals by over 50% . the corresponding improvement in the experimental sensitivity of this dataset is equivalent to that of quadrupling the integration length of each observation . the results presented in this section are based on the data calibrated using metm and arrival time estimates derived using mtm . table [ tab : parameters ] lists the best - fit timing model parameters obtained using the tempo2 analysis software @xcite . lr parameter name & value + reference clock & tt(tai ) + planetary ephemeris & de414 + @xmath39 epoch ( mjd ) & 53869.00016629324189 + pulse period , @xmath39 ( ms ) & 16.4529299518323(2 ) + period derivative , @xmath92 ( 10@xmath93 ) & 4.33399(6 ) + ecliptic longitude , @xmath94 ( ) & 153.865881(3 ) + ecliptic latitude , @xmath32 ( ) & @xmath00.06(4 ) + proper motion in @xmath94 ( mas yr@xmath95 ) & @xmath015.97(2 ) + proper motion in @xmath32 ( mas yr@xmath95 ) & 38(19 ) + annual parallax , @xmath96 ( mas ) & 1.4(2 ) + binary model & ddh + orbital period , @xmath97 ( days ) & 7.805135(1 ) + projected semi - major axis , @xmath98 ( lt - s ) & 16.76541(4 ) + orbital eccentricity , @xmath99 ( @xmath100 ) & 9.702(7 ) + epoch of periastron , @xmath101 ( mjd ) & 53876.1038(2 ) + longitude of periastron , @xmath102 ( ) & 97.757(9 ) + orthometric amplitude , @xmath103 ( @xmath3s ) & 0.52(7 ) + orthometric ratio , @xmath104 & 0.5(1 ) + periastron advance , @xmath105 ( yr@xmath95 ) & 0.009(3 ) + secular variation of @xmath98 , @xmath106 ( @xmath107 ) & 14(1 ) + [ tab : parameters ] these results include a significant detection of the shapiro delay ; after subtracting two degrees of freedom from the least - squares fit with the addition of the shape @xmath108 and range @xmath109 parameters , @xmath66 is reduced by @xmath110% ( from @xmath111 to @xmath112 ) . although the detection is significant , the elongated contours of constant @xmath113 shown in figure [ fig : m2sini ] demonstrate that @xmath114 and @xmath115 are highly covariant ; the coefficient of correlation between these two model parameters is @xmath00.99 . furthermore , the constraint on the companion mass is not very informative ; the 95% confidence interval given by the projection of the @xmath113 = 4 contour onto the @xmath116 axis ranges from 0.25 to @xmath117 solar masses . the shape and range parameters are highly covariant because , in nearly circular orbits that are only moderately inclined with respect to the line of sight , the shapiro delay is readily absorbed in the roemer delay by modification of the keplerian orbital parameters . the unabsorbed remnant of the shapiro delay is best described as a weighted sum of harmonics of the orbital frequency using the orthometric parameterization introduced by @xcite . accordingly , table [ tab : parameters ] lists the best - fit values of the ratio of the amplitudes of successive harmonics @xmath118 and the amplitude of the third harmonic @xmath119 the values of @xmath114 and @xmath115 derived from @xmath104 and @xmath103 are nearly identical to the values yielded by directly modeling the shape and range of the shapiro delay . however , the coefficient of correlation between @xmath103 and @xmath104 is only @xmath00.38 . there is close agreement between the orbital inclination angle , @xmath120 , constrained by the shapiro delay and previous estimates of the inclination of the pulsar spin axis , @xmath121 @xcite and @xmath122 @xcite , constrained using the rotating vector model @xcite . this agreement is compatible with the expectation that the pulsar spin and orbital angular momentum vectors are aligned during the period of mass accretion that led to the formation of the millisecond pulsar ( e.g. * ? ? ? the best - fit model parameters also include the first significant detections of the precession of periastron @xmath123 and the secular variation of the projected semi - major axis of the orbit , @xmath106 . the estimate of @xmath124yr@xmath95 is consistent with the value predicted by general relativity , @xmath125yr@xmath95 , based on the masses of the pulsar and its companion yielded by the shapiro delay detection . the measured value of @xmath106 is seven orders of magnitude larger than the general relativistic prediction , @xmath126 , because it is dominated by the secular variation due to proper motion @xcite . this geometric effect can be used to place an upper limit on the orbital inclination angle ; i.e. @xmath127 , where @xmath128 mas yr@xmath95 is the total proper motion . the resulting constraint , @xmath129 , rules out only 12% of possible orientations and is consistent with the stronger constraint yielded by measurement of the shapiro delay . the proper motion of the system also induces a quadratic doppler shift @xcite that contributes @xmath130 to observed period derivatives , where @xmath131 is the period ( e.g. spin or orbital ) , @xmath132 is the distance to the pulsar , and @xmath133 is the speed of light . the magnitude of the shklovskii effect is estimated using the distance @xmath134 parsec derived from the annual trigonometric parallax after correction for lutz - kelker bias , @xmath135 mas @xcite . the shklovskii contribution to the spin period derivative ( @xmath136 ) is slightly larger than its measured value , which would imply that the intrinsic @xmath137 of the pulsar is negligible . however , the predicted contribution to the orbital period derivative ( @xmath138 ) is around 20 times larger than the upper limit derived by including @xmath139 in the timing model . the relative uncertainties in the above estimates of the shklovskii effect are almost 100% , primarily due to the large uncertainty in the estimate of the proper motion in ecliptic latitude . using only the proper motion in ecliptic longitude , the predicted kinematic contribution to @xmath139 is only three times larger than the timing - derived upper limit . to explain the remaining discrepancy , three other contributions to the observed @xmath139 are considered ( as in * ? ? ? first , the loss of energy due to gravitational radiation as predicted by general relativity contributes around @xmath140 , approximately 4 orders of magnitude smaller than the shklovskii effect . second , the contribution due to differential rotation about the galactic centre ( approximately @xmath141 ) can also be neglected . finally , the acceleration of the psrj1022@xmath11001 system in the galactic gravitational potential is considered . using the parallax distance @xmath142 and the galactic latitude @xmath143 , the component of gravitational acceleration perpendicular to the galactic disk , @xmath144 m s@xmath145 @xcite contributes approximately @xmath146 to the observed @xmath147 . at large heights above the galactic disk , the relative uncertainty in @xmath148 is more than a factor of two ; therefore , this contribution is large enough to negate the shklovskii effect , resulting in an observed @xmath139 that is consistent with zero . instrumental distortion of the total intensity profile introduces significant systematic timing errors that are correlated between pulsars observed with the same instrument . therefore , high - fidelity polarimetry is inextricably linked with the long - term objectives of high - precision timing and the primary goals of pulsar timing array experiments . this paper presents a novel method of calibrating the instrumental response by exploiting the long - term stability of the polarized emission from a millisecond pulsar . full - polarization timing observations of psrj0437@xmath04715 are used to model variations in the 21-cm multibeam and h - oh receivers and down - conversion system at parkes . over a period of @xmath149 years , temporal and spectral variations of the receptor cross - coupling introduces systematic error of the order of 1@xmath3s in arrival time estimates derived from observations of psrj1022@xmath11001 . high - fidelity calibration of instrumental polarization and arrival time estimation using matrix template matching is demonstrated to correct systematic error and double the sensitivity of the experiment , yielding significant detections of the shapiro delay , the precession of periastron , and the secular variation of the projected semi - major axis of the orbit . the improvements in both timing accuracy and precision are consistent with the hypothesis that the average polarized emission from these millisecond pulsars is stable over the timescales of relevance to this experiment . with a modest instrumental bandwidth of 64 mhz and median integration length of 64 minutes , psrj1022@xmath11001 yields arrival time residuals with an uncertainty - weighted standard deviation of only @xmath90 = 880ns , roughly five orders of magnitude smaller than the pulsar spin period @xmath150 s. the remarkable timing precision of these observations supports the conclusions of @xcite , who argue that the psrj1022@xmath11001 profile variations observed by @xcite and @xcite are not intrinsic to the pulsar and are most likely due to instrumental calibration errors . the precision of the arrival time data presented in this paper places psrj1022@xmath11001 among the top ten sources regularly observed as part of the parkes pulsar timing array ( ppta ) project . other ppta sources may also benefit from the application of metm ; e.g. six of the sources listed in table [ tab : theory ] have instrumental distortion susceptibility factors ranging from 200 ns to 340 ns . referring to the measured values of the non - orthogonality parameters plotted in figure [ fig : coupling ] , these factors roughly correspond to systematic timing error variations of the order of 1 @xmath3s . therefore , the timing precision of these pulsars would likely be improved by application of the methods developed for this study . for psr j1744@xmath01134 and psr j2129@xmath05721 , matrix template matching is predicted to yield arrival times with greater uncertainty than those derived from the total intensity profile . in these two cases , the best results would be obtained by calibrating using metm and deriving arrival times using sti . in a wider analysis of all of the pulsars in the ppta , @xcite calibrated approximately 5.9 years of psrj1022@xmath11001 timing observations using mem - derived solutions . using typical integration lengths of 64 min and an instrumental bandwidth of 256 mhz , this study yielded arrival time residuals with an uncertainty - weighted standard deviation of 1.7 @xmath3s and reduced @xmath151 . the residual noise is approximately 2.4 times greater than expected based on simple extrapolation of the metm sti results presented in table [ tab : residual ] . this discrepancy may be explained by the fact that the pdfb instruments have a non - linear response that introduces over - polarization , which can not be calibrated using mem . the accuracy of mem - based calibration is also limited by the fact that there is no unique solution to the measurement equation when the only constraining transformation is the geometric rotation of the receptors with respect to the sky @xcite . to constrain the otherwise degenerate boost along the stokes @xmath80 axis , it is necessary to include observations of a source that is assumed to have negligible circular polarization ( e.g. , for the mem fits performed in [ sec : template ] , observations of hydra a were used to constrain @xmath46 ) . however , the use of an unpulsed source of radiation as a constraint is intrinsically susceptible to variability in other contributions to the system temperature , including receiver noise and ground spillover . that is , it is safer to assume that the polarized emission from psrj0437@xmath04715is constant than it is to rely on a source of unpulsed radiation during calibration . the results presented by @xcite may also be limited by the instability of the receptor cross - coupling , as shown in figure [ fig : coupling ] . the temporal variations of the non - orthogonality parameters are sufficiently resolved only via application of the metm method presented in this paper , which yields a seven - fold increase in the number of available calibrator solutions . as noted in previous studies @xcite , other phenomena may also currently limit the timing precision of psrj1022@xmath11001 . for example , owing to its low ecliptic latitude , the radio signal from this pulsar is subject to significant dispersive delays in the solar wind @xcite . no corrections for dispersion variations were applied to the data presented in this work ; however , observations made when the line of sight to psrj1022@xmath11001 passes near the sun ( around late august of each year ) were excluded from the data set . in addition to fluctuations in dispersion , the observed flux of the pulsar varies as a function of time and radio frequency due to scintillation in the interstellar medium . the average profile of psrj1022@xmath11001 varies significantly as a function of radio frequency ( e.g. see figure 1 of * ? ? ? * ) and , when modulated by interstellar scintillation , the frequency - integrated mean profile may fluctuate with time . the potentially significant arrival time estimation errors induced by this effect could be mitigated through the use of a frequency - dependent template profile . it is reasonable to expect that psrj1022@xmath11001 timing will be improved by the current generation of instrumentation at the parkes observatory . the data presented in this paper were observed using a system with low dynamic range that performed analog - to - digital conversion of the radio signal with only two bits per sample . two - bit quantization is an intrinsically non - linear process and the techniques applied to restore linearity ( dynamic output level setting ) and mitigate quantization noise ( scattered power correction ) are based on a linear approximation to the response of the digitizer @xcite . as discussed in appendix [ app : spc ] , the scattered power correction algorithm applied in this analysis is accurate only to first order and it is possible that overestimation of the unpolarized scattered power contributes to the over - polarization noted in figure [ fig:0437 ] . to overcome the limitations of two - bit sampling , a new baseband recording and processing system with greater dynamic range was commissioned at the parkes observatory in 2010 april . designed in collaboration with the center for astronomy signal processing and electronics research ( casper ) at berkeley , the casper - parkes - swinburne recorder ( caspsr ) digitizes a dual - polarization 400 mhz band with eight bits per sample and performs real - time radio frequency interference excision based on the spectral kurtosis estimator @xcite . this system currently operates in parallel with the third and fourth generations of the parkes digital filter bank ( pdfb ) instruments as part of the ppta program . the calibration techniques applied in this paper could potentially be improved by reducing the number of degrees of freedom in the estimates of the receptor cross - coupling parameters . for example , this could be achieved by fitting an analytic model to the temporal and spectral variations of the calibrator solutions or by employing a lossy compression algorithm that can be applied to irregularly sampled data . the smoothing effected by such a transformation would reduce the instantaneous noise in the applied calibrator solutions and might also provide a robust means of interpolating between solutions . however , further refinements of the calibration technique may yield only marginal improvements to arrival time accuracy and precision . table [ tab : residual ] demonstrates that , even when the data are calibrated using the inaccurate ideal - feed assumption , matrix template matching ( mtm ) yields arrival time estimates that are nearly as good as those derived from data calibrated using the new metm technique described in this paper . this seems to suggest that , as long as the signal is not depolarized by integrating over time and/or frequency , the fidelity of the calibration technique has negligible impact on arrival times derived using mtm . this conclusion may have a significant impact on the design of the next generation of instrumentation for high - precision timing . for example , it has been asserted that in order to achieve timing precision of the order of 100 ns , the square kilometre array ( ska ) must achieve net polarization purity corresponding to @xmath152 @xcite . however , even when non - orthogonality as large as @xmath153 is ignored , as is the case when the 21-cm multibeam receiver at parkes is assumed to be ideal , mtm yields accurate arrival time estimates . that is , by relaxing the requirement for polarization purity , the application of mtm has the potential to reduce the cost of ska development for high - precision timing . all of the software required to perform measurement equation modeling , polarimetric calibration , and matrix template matching is freely available as part of psrchive @xcite ; the use of this software is demonstrated by @xcite and more fully documented online . i am grateful to matthew bailes , paul demorest , mike keith , michael kramer , stefan osowski , and john reynolds for helpful discussions during this research project and for insightful comments that greatly improved this report . joris verbiest implemented and assisted with tempo2 support of ecliptic coordinates and the orthometric parameterization of the shapiro delay . the parkes observatory is part of the australia telescope which is funded by the commonwealth of australia for operation as a national facility managed by csiro . the invariant profile is formed by computing the lorentz interval @xmath154 as a function of pulse longitude , where @xmath155 . assuming that the noise in each of the stokes parameters is independent and normally distributed with standard deviation @xmath104 , define the normalized stokes parameters @xmath156 . the noise in @xmath157 has a noncentral @xmath66 distribution with one degree of freedom and noncentrality parameter @xmath158 . likewise , the noise in @xmath159 has a noncentral @xmath66 distribution with three degrees of freedom and @xmath160 . the distribution of the noise in @xmath161 is given by the cross - correlation of the distributions of @xmath162 and @xmath159 ; examples are shown in figure [ fig : invint ] . the variance of @xmath161 is equal to @xmath163 , where @xmath164 is the euclidean norm of the stokes four - vector , defined by @xmath165 . as all four stokes parameters vary as a function of pulse longitude , the noise in @xmath166 is heteroscedastic , which violates one of the basic premises of most template - matching algorithms . both the distribution of the invariant interval and its tendency toward zero for highly polarized radiation adversely affect its usefulness in template matching ( as a replacement for the total intensity ) and as a normalization factor during measurement equation modeling ( mem ; * ? ? ? gain normalization is necessitated by interstellar scintillation , which causes the received flux of the pulsar to vary as a function of time and radio frequency . to address the problems with the invariant interval during the application of mem in [ sec : template ] , the stokes parameters were normalized by the sum of the invariant interval over all on - pulse longitudes . for the typical observation of psrj0437@xmath04715 , this summation reduces the relative error of the normalization factor by @xmath167 , thereby yielding more normally distributed normalized stokes parameters . the data presented in this paper were processed using a new scattered power correction algorithm that can be used to correct digitization distortions to folded pulsar profiles . in the case of two - bit sampled voltages , the scattered power ( or quantization noise ) is @xmath168% of the total power . the derivation of the method starts with the mean digitized power @xmath169 given by equation ( a5 ) of ( * ? ? ? * hereafter ja98 ) , @xmath170 where @xmath171 is the fraction of samples that fall between the chosen thresholds , @xmath172 is the digitized power as a function of @xmath171 , given by equation ( a4 ) of ja98 , and @xmath75 is the discrete probability distribution for @xmath171 , given by equation ( a6 ) of ja98 . the parameter @xmath171 is eliminated by the summation in the above equation ; however , both @xmath172 and @xmath75 are also parameterized by the expectation value @xmath173 . therefore , equation ( [ eqn : a5 ] ) represents the relationship between the mean digitized power and the mean value of @xmath171 . that is , given the mean digitized power @xmath169 , equation ( [ eqn : a5 ] ) can be inverted to compute @xmath173 , which can in turn be used to estimate the mean undigitized power @xmath174 and the mean scattered power @xmath175 via equations ( 45 ) and ( 43 ) of ja98 , respectively . equation ( a5 ) of ja98 can be inverted using the newton - raphson method and the partial derivatives of equations ( a4 ) through ( a6 ) of ja98 with respect to @xmath173 . these are simplified in the case of two - bit sampling by noting that equation ( a4 ) of ja98 reduces to @xmath176 where @xmath177 and @xmath178 are given by equations ( 41 ) and ( 40 ) of ja98 . the folded profile scattered power correction algorithm is based on the following assumptions and approximations . first , at the time of recording the astronomical signal , it is necessary that the baseband voltages are sampled using the optimum two - bit input thresholds defined in table 1 of ja98 . to first order , this condition is satisfied by excluding data with excessive two - bit distortion as defined by equation ( [ eqn : d2bit ] ) of this paper . second , equations ( 45 ) and ( 43 ) of ja98 only approximately relate the expectation values of @xmath171 , @xmath174 and @xmath179 . for example , the relationship between @xmath179 and @xmath174 defined by equation ( 43 ) of ja98 is concave down ( see figure 3 of ja98 ) ; therefore , jensen s inequality dictates that the value of @xmath179 estimated from the mean undigitized power will be greater than the expectation value of @xmath179 . consequently , the mean fractional scattered power may be systematically underestimated . this limitation can not be overcome without prior knowledge of the distribution of total intensity fluctuations intrinsic to the pulsar signal . finally , it is assumed that the signal in each frequency channel has not been significantly altered , such that there is a well - defined relationship between the mean undigitized power and the mean digitized power . for example , after phase - coherent dedispersion removal , the flux density over a given time interval no longer represents the voltage fluctuations in the digitizer over that interval ; the previously smeared pulsar signal will be recovered and the scattered power will be smeared . to estimate the scattered power using coherently - dedispersed digitized power , the dispersion smearing in each frequency channel must be less than or of the order of the time resolution of the folded profile . this condition is satisfied in the 20-cm timing observations of psrj0437@xmath04715and psrj1022@xmath11001 presented in this paper . , d. j. , hobbs , g. b. , manchester , r. n. , edwards , r. t. , backer , d. c. , bailes , m. , bhat , n. d. r. , burke - spolaor , s. , coles , w. , demorest , p. b. , ferdman , r. d. , folkner , w. m. , hotan , a. w. , kramer , m. , lommen , a. n. , nice , d. j. , purver , m. b. , sarkissian , j. m. , stairs , i. h. , van straten , w. , verbiest , j. p. w. , & yardley , d. r. b. 2010 , apj , 720 , l201 manchester , r. n. , hobbs , g. , bailes , m. , coles , w. a. , van straten , w. , keith , m. j. , shannon , r. m. , bhat , n. d. r. , brown , a. , burke - spolaor , s. g. , champion , d. j. , chaudhary , a. , edwards , r. t. , hampson , g. , hotan , a. w. , jameson , a. , jenet , f. a. , kesteven , m. j. , khoo , j. , kocz , j. , maciesiak , k. , oslowski , s. , ravi , v. , reynolds , j. r. , sarkissian , j. m. , verbiest , j. p. w. , wen , z. l. , wilson , w. e. , yardley , d. , yan , w. m. , & you , x. p. 2012 , pasa , accepted , r. , levin , y. , janssen , g. h. , lazaridis , k. , kramer , m. , stappers , b. w. , desvignes , g. , purver , m. b. , lyne , a. g. , ferdman , r. d. , jessner , a. , cognard , i. , theureau , g. , damico , n. , possenti , a. , burgay , m. , corongiu , a. , hessels , j. w. t. , smits , r. , & verbiest , j. p. w. 2011 , mnras , 414 , 3117 , j. p. w. , bailes , m. , coles , w. a. , hobbs , g. b. , van straten , w. , champion , d. j. , jenet , f. a. , manchester , r. n. , bhat , n. d. r. , sarkissian , j. m. , yardley , d. , burke - spolaor , s. , hotan , a. w. , & you , x. p. 2009 , mnras , 400 , 951 , w. m. , manchester , r. n. , van straten , w. , reynolds , j. e. , hobbs , g. , wang , n. , bailes , m. , bhat , n. d. r. , burke - spolaor , s. , champion , d. j. , coles , w. a. , hotan , a. w. , khoo , j. , oslowski , s. , sarkissian , j. m. , verbiest , j. p. w. , & yardley , d. r. b. 2011 , mnras , 414 , 2087 , d. r. b. , coles , w. a. , hobbs , g. b. , verbiest , j. p. w. , manchester , r. n. , van straten , w. , jenet , f. a. , bailes , m. , bhat , n. d. r. , burke - spolaor , s. , champion , d. j. , hotan , a. w. , oslowski , s. , reynolds , j. e. , & sarkissian , j. m. 2011 , mnras , 414 , 1777 , x. p. , hobbs , g. , coles , w. a. , manchester , r. n. , edwards , r. , bailes , m. , sarkissian , j. , verbiest , j. p. w. , van straten , w. , hotan , a. , ord , s. , jenet , f. , bhat , n. d. r. , & teoh , a. 2007 , mnras , 378 , 493